Volume 2 Number 2 Jul - Dec 2001press.utp.edu.my/wp-content/uploads/2018/12/Platform-v2n...NOTES FOR...

Relay Feedback Auto-tuning Controller For Waste Water Treatment

V R Radhakrishnan

Integrating Analysis And Design Improvement

In A Reverse Engineering Framework

Ahmad Majdi Abdul Rani

Sustaining Students’ Interest – Sharing Of Experience

Azizan Zainal Abidin

A Multimedia Approach To Facilitate The Studying

Of A Physics Concept: Motion In 2-D

Balbir Singh Mahinder Singh and Hasnah Mohd Zaid

Fatigue Behaviour Of Fibre Reinforced Bituminous Mixtures

From The Indirect Tensile Test

Ir Dr Ibrahim Kamaruddin

Contamination Of Phosphate Glasses Upon Melting

Jariah Mohamad Juoi

New Radiation Grafted And Sulfonated Membranes

For PEM Fuel Cell

Mohamed Mahmoud Nasef, Hamdani Saidi and Hussin Mohd Nor

Neural Fuzzy Based 3d Anti-sway Modelling And

Control Design For Overhead Cranes

M Mahfouf

Scanning Electron Microscopy of Anisotropic Etching in Fabrication

of VMOS (Vertical Metal-Oxide Semiconductor) Transistor

Norani M Mohamed, Kamarulazizi Ibrahim & Leong Yew Wei

Jet Impingement Cooling Of Microelectronic Systems

Mohd Shiraz Aris, I Rushyendran, G A Quadir & K N Seetharamu

P L A T F O R M

Volume 2 Number 2 Jul - Dec 2001

2

6

14

20

24

29

33

38

47

53

VO

LU

ME T

WO

N

UM

BER

TW

O J

ULY

- DEC

EM

BER

200

1

VOLUME TWO NUMBER TWO JULY - DECEMBER 2001

NOTES FOR CONTRIBUTORS

Instructions to Authors

Authors of articles that fit the aims,scopes and policies of this journal areinvited to submit soft and hard copiesto the editor. Paper should be writtenin English. Authors are encouragedto obtain assistance in the writing andediting of their papers prior tosubmission. For papers presented orpublished elsewhere, also include thedetails of the conference or seminar.

Manuscript should be prepared inaccordance with the following:1. The text should be preceded by

a short abstract of 50-100 wordsand four or so keywords.

2. The manuscript must be typedon one side of the paper, double-spaced throughout with widemargins not exceeding 3,500words although exceptions willbe made.

3. Figures and tables have to belabelled and should be includedin the text. Authors are advisedto refer to recent issues of thejournals to obtain the format forreferences.

4. Footnotes should be kept to aminimum and be as brief aspossible; they must benumbered consecutively.

5. Special care should be given tothe preparation of the drawingsfor the figures and diagrams.Except for a reduction in size,they will appear in the finalprinting in exactly the sameform as submitted by the author.

6. Reference should be indicatedby the authors’ last names andyear of publications.

Publisher

Universiti Teknologi PETRONAS

Bandar Seri Iskandar

31750 Tronoh

Perak Darul Ridzuan

MALAYSIA

Univers i t i Teknologi Petronas • http://www.utp.edu.my

1PLATFORM • Volume 2 Number 2 • July – December 2001

PlatformPlatformContentsAdvisors

Dr Rosti Saruwono

Ir Dr Ahmad Fadzil Mohamad Hani

Editor-in-Chief

Dr Puteri Sri Melor Megat Yusoff

Co-Editors

Suziah Sulaiman

Zulqarnain Abu Bakar

Yap Vooi Voon

Editorial Board

Dr Mohammed Halib

Dr Abas Md Said

Dr Mohamed Ibrahim Abdul Mutalib

Dr Abd Rashid Abd Aziz

Dr Mohd Noh Karsiti


Dr Nasiman Sapari

Dr Azmi Mohd Shariff

Azrai Abdullah

Address

Editor-in-Chief

PLATFORM



31750 Tronoh

Perak Darul Ridzuan

Malaysia

ht tp: / /www.utp .edu.my

kamarular i f f in_amminudin

@petronas.com.my

Telephone +(60)5 367 8018

+(60)5 367 8019

+(60)5 367 8055

Facsimile +(60)5 367 8252

PLATFORM is a biannual, peer-reviewed journal of Universiti Teknologi PETRONAS. It servesas a medium for faculty members, students and industry professionals to share their

knowledge, views, experiences and discoveries in their areas of interest and expertise.

It comprises as collection of, but not limited to, papers presented by the academic staffof the university at various local and international conferences, conventions and seminars.

The entries range from opinions and views on engineering, technology and social issuesto deliberations on the progress and outcomes of academic research.

Opinions expressed in this journal need not necessarily reflect the official views of theuniversity.

All materials is copyright of Universiti Teknologi PETRONAS. Reproduction in whole or in

part is not permitted without the written permission of the publisher.

I S S N 1 5 1 1 - 6 7 9 4

Relay Feedback Auto-tuning Controller For Waste Water

Treatment by V R Radhakrishnan



by Ahmad Majdi Abdul Rani


by Azizan Zainal Abidin

A Multimedia Approach To Facilitate The Studying Of A

Physics Concept: Motion In 2-D

by Balbir Singh Mahinder Singh & Hasnah Mohd Zaid



by Ir Dr Ibrahim Kamaruddin


by Jariah Mohamad Juoi

New Radiation Grafted And Sulfonated Membranes For PEM

Fuel Cell by Mohamed Mahmoud Nasef, Hamdani Saidi

& Hussin Mohd Nor

Neural Fuzzy Based 3d Anti-sway Modelling And Control

Design For Overhead Cranes by M Mahfouf

Scanning Electron Microscopy of Anisotropic Etching in

Fabrication of VMOS (Vertical Metal-Oxide Semiconductor)

Transistor by Norani M Mohamed


by Mohd Shiraz Aris, I Rushyendran, G A Quadir & K N Seetharamu

2

6

14

20

24

29

33

38

47

53

PLATFORM • Volume 2 Number 2 • July – December 2001

2 Univers i t i Teknologi Petronas • http://www.utp.edu.my

INTRODUCTION

In order to make the waste treatmentefficient, practical and cost effective,the operating parameters in eachsectional unit of the treatment systemmust be brought under appropriatecontrol within the set- point.Unbounded operations, such asparameters running away from the set-point, even for a short period interval,can be detrimental to the plantoperation. For example, pH in theaerobic treatment should becontrolled within 6 - 8. A pH driftingfrom the optimum value will alter thenature of bacterial activities, deters the

desirable enzymatic behaviour. Theoverall results could be the stoppageof the aerobic oxidation process thatconverts hazardous food into treatedbio-mass. This in turn affects theoperation of downstream units orresults in unauthorized discharge ofindustrial waste.

The control of waste water operatingparameters, such as pH, had long beenregarded as a difficult task (Shinskey,1973; Buchholt and Kummel, 1979;Jacobs et al., 1980; Gustafsson andWalker, 1983; Piovosso and Williams,1985; Williams et al., 1990; Wrightand Kravaris, 1991; Kulkarni et al.,

1991). The difficulties arise from thesevere process non- linearity andfrequent load changes, such as changesin the influent composition and flowrate which give rise to its time varyingcharacteristics. The non-linearity canbe seen from the S-shaped static pHresponses with the addition of titrant.Moreover, a chemical plant, in mostoccasions, consists of a differentnumber of processing units thatdischarge different types of waste andamounts of acid or base. These leavethe plant as a hardly definable wastemixture. Besides the non-lineardynamic behaviour, there are oftenseveral other substantial factors such

Relay Feedback Auto-tuning Controller

For Waste Water Treatment

V R Radhakrishnan

School of Chemical Engineering

Universiti Sains Malaysia

31750 Tronoh, Perak

ABSTRACT

The control of waste water operating parameters, especially pH, had long been regarded as a difficult task. This difficultyarises from its severe process non-linearity, tremendous dead time and frequent load changes, such as changes in theinfluent composition and flow rate. The non-linearity can be seen from the S-shaped static pH responses with theaddition of titrant where the process gain is low in the buffered zone and extremely high around the neutrality point. Inorder to make the pH control effective, for example, we need a controller which is rapid in the buffered zone with largecontroller gain and sensitive around neutrality (pH 6-8 where that is a sudden change from acid to base condition). Thecontroller gain in this region must be small enough in order to maintain a constant gain in the feedback loop to avoidsustained oscillation around the set-point. This difficulty have led to the various alternatives being introduced to solvethe problems, such as predictive combined with traditional control, such as modified PI controller, adaptive control,predictive-adaptive control, fuzzy logic and gain scheduling. In this study, a relay feedback auto-tuning controller is usedto regulate the pH. The auto-tuner will excite the process output into limit cycles from where the ultimate gain andperiod can be determined. The PID values can be calculated based on these parameters and imparted to the controller.

Keywords:

relay feedback, auto-tuning, pH control, PC auto-tuner

This paper was presented at the 14th Symposium of Malaysian Chemical Engineers, SOMCHE 2000, Putrajaya, 30-31 October, 2000.



as type and strength of the acid andbase (different pK values), exactvolumetric flow of influent(particularly the upstream processplants are a mixture of continuous,semi-continuous, batch process),buffer capacity of the entire system,effectiveness of the stirring system,exact pumping rate of neutralizingsolution delivered by the meteringpumps and etc. Those practicalproblems make the effective controlof waste water treatment rather hectic,if not ineffective, by conventionalmeans.

In order to make the pH controleffective, we need a controller whichis rapid in the buffered zone (wherepH varies slightly with titrant, in theregion of pH 4 - 5, 9 - 10) andsensitive around neutrality (pH 5 - 9where that is a sudden change fromacid to basic, with great pH sensitivityto changes in the base concentration)(Moore, 1978).

These difficulties have led to thevarious alternatives being introducedto solve the problems, such aspredictive combined with traditionalcontrol algorithms (Riggs, 1990;Gaulian, 1990), modified PIcontroller (Costello, 1994), adaptivecontrol (Kurtz, 1985), predictive-adaptive control (Aragon, 1993;Palancar, 1993; Proll, 1994), fuzzylogic.

The subject of this research is acontinuation of these previousresearch works, that is to apply relayfeedback auto tuning controller toregulate the intensities of pH.

RELAY FEEDBACK AUTO-

TUNING

The Ziegler-Nichols ultimate cyclingor frequency response method is a wellknown technique for tuning a PID

controller manually. It is designed togive quarter amplitude damping forthe load disturbance response. Thismethod, however, is not very practicalas it is difficult to operate the processon the brink of instability. In orderto use the Ziegler-Nichols ultimatecycling tuning formula to tune a PIDcontroller manually, the first step isto switch off the I and D part to tunethe P controller. The proportionalgain is increased slowly until sustainedoscillation is obtained. Anyway, ifsufficient care is not taken, theamplitude of the oscillation may growunbounded, creating upset to plantoperation. Also, it may take a longertime to obtain the sustained oscillationor even if it has been obtained, it isdifficult to control its amplitude.

Refer to Figure 1, during auto-tuning,the relay feedback is activated and thePID controller is disconnected. Whena stable limit cycle is established, thecorresponding parameters such asultimate gain and period are foundand the PID parameters arecomputed. The PID controller is thenconnected back to the process and isused to control the process using thenewly auto-tuned PID values. Besideseliminating the time consumingmanual procedure, this methodachieves controlled oscillation ratherrapidly, thereby allowing auto-tuningto be performed in a very short time.

Moreover, the amplitude of thesustained oscillation is well controlledwithin permissible limit.

THE METHOD OF HARMONIC

BALANCE (DESCRIBING

FUNCTION)

The excitation of the process outputinto sinusoidal limit cycle by thefeedback relay auto-tuner can bedescribed by the method of harmonicbalance or describing function. In thismethod, it is assumed that the limitcycle has a period Tu and frequency

wu = 2πTu

. The relay output is a

periodic symmetrical square wave. Ifthe relay amplitude is d, a simpleFourier series expansion of the relayoutput shows that the first harmonic

component has the amplitude 4dπ .

Assume further that the processdynamics are of low-pass characterand the contribution from the firstharmonic dominates the output. Theerror signal then has the amplitude

a = 4dπ

G( jwu) (1)

The condition for oscillation is thusthat

arg G(jwu) = –π and

Figure 1: Relay Feedback Auto- Tuning Controller

ε Process

– 1

Relay

PID



Ku = 4dπa

= 1

G( jwu)(2)

where Ku can be regarded as theequivalent gain of the relay fortransmission of sinusoidal signals withamplitude a.

DETERMINATION OF PID FROM

RELAY FEEDBACK AUTO-

TUNING

A very simple rule for choosing theparameters of PID controllers (KC,

Ti1 , TD) is to determine them from

Ku and Tu. These correlations aregiven by the Ziegler-Nichols closed-loop method.

KC = K u

1.7 (3)

Ti = Pu

2 (4)

TD = Pu

8 (5)

Alternatively, the PID values can becalculated from the Standard Astrom-Hagglund (1984) formula (Tan et al.,1996):

KC = K u

Amcosφm (6)

Td =tanφm + 4

α+tan2 φm

2ωu

(7)

Ti = αTd (8)where φm, Am, α are the desired phasemargin, gain margin and arbitariconstant. For most applications inprocess control, the specification ofAm = 2 and φm = 0.7854 isrecommended (Hagglund andAstrom, 1991).

HYSTERESIS

In practice, hysteresis is installed in therelay to prevent relay switchover due

to presence of noise. The hysteresiswidth can be selected base on the noiselevel, for example, 2 times larger thanthe noise amplitude. The ultimategain in this case is then:

Ku' = 4d

π a2 −ζ 2 (9)

where ζ is the hysteresis width.

EXPERIMENTAL TECHNIQUES

The PC, data acquisition system andCSTR are assembled to establishclosed loop operation. The auto-tuneris implemented on the PC with AD/DA card (Opto 22). The relayfeedback auto-tuner is programmedusing Visual Basic. This program isconnected by means of ObjectLinking and Embedding (OLE) withthe Paragon. The auto-tuner isprogrammed in such a way that it willauto-tune the closed loop withrectangular wave input, having aamplitude d being 180° out of phasewith the output signal. The auto-tuning time, square input amplitudeand period, upper limit and lowerlimit, hysteresis width can be changedto any values. The amplitude a andperiod, Tu of the output oscillationsare then computed.

The calculation of PID values can becoded onto the PC itself, so that thePID controller parameters, KC,Ti, TDare at once found. The correlationsbetween KC,Ti, TD with Ku and Tu aregiven by Ziegler-Nichols (1942) andStandard Astrom-Hagglund (1984)method.

The relay is then disconnected, andthe PID controller with the just foundparameters is given charge of thesystem regulation. This is done forboth load and set-point changes(regulatory and servo problem), ateach reactor operating point.

The developments of the software

program to incorporate the relaysystem into the pH process controlcan be best described by the followingsoftware flow diagram:

pHProcess

Hardware Optomus(ADC, DAC)

Paragon TNT ControlSupervisor System

OLE

Visual Basic Programfor auto-tuning

Figure 2: Software Flow Diagram

The performance of the PIDcontroller tuned using the auto-tunerprogrammed with PC is comparedwith the commercial auto-tuningcontroller.

CONCLUSIONS

The PID controller tuned with the PCprogrammed relay feedback auto-tuner provides better performancewith those tuned with conventionalZiegler-Nichols or commercial auto-tuners (Yokogawa Ut350). Theadvantages lie in the fact that the auto-tuner allows the PID tuning in acontrolled manner and the small limitcycles generated is what we need toprevent unbounded operating cyclesin real tuning environment.Conventional open-loop tests are bothtime consuming and less practical.The Ziegler-Nichols closed loopmethod, on the other hand, could giveraise to unbounded sustainedoscillations during tuning.



A commercial software is written toimplement the control strategies. Theauto-tuner parameters, such asamplitude, hysteresis width andperiod can be changed at will. pHcontrol is important for the successfuloperation of many chemical,biochemical and waste watertreatment systems. This in turncontributes to the development inMalaysia, which is experiencing rapidgrowth in the process industries. Thisproject will reinforce the study ofadvanced control strategies in USMand Malaysia. The relay feedbackauto-tuning under study gives an easyalternative as it is an on-line tuningmethod plus the fact that it generatessmall sustained oscillations duringtuning make sure the process isoperating within permissible controllimits.

NOTATION

a Amplitude of relay oscillation, %Am Amplitude margin, dimensionlessd Amplitude of process output, %KC Proportional gainKu Ultimate gain, dimensionlessTu Ultimate period, secondsTi Integral time, secondsTD Derivative time, secondsφm Phase margins

ACKNOWLEDGEMENT

The authors gratefully acknowledge UniversitiSains Malaysia which has supported thisresearch through a short-term grant.

REFERENCES

[1] J M Aragon, M C Palancar and J AMiguens. Adaptive Control System forThe Regulation of The pH of WasteWater. European Meeting On ChemicalIndustry And Environment, Palahi:Girona, Spain, 2: 69-77. 1993.

[2] F Buchholt and M Kummel. Self-tuningControl of A pH Neutralization Process.Automatica, 15: 665. 1979.

[3] D J Costello. Evaluation of Model- BasedControl Techniques for Buffered Acid-Base Reaction System. Trans. Inst. Chem.Engng, 72, 47- 54. 1994.

[5] R Garrido, M Adroer and M Poch.Wastewater Neutralization Control basedIn Fuzzy Logic: Simulation Results. Ind.Eng. Chem. Res., 36: 1665-1674. 1997.

[6] T K Gustafsson and K V Waller.Dynamic Modelling And ReactionInvariant Control of pH. Chem. Engr.Sci., 38: 389. 1983.

[7] C C Hang, K J Astrom and W K Ho.Relay Auto-timing In The Presence ofStatic Load Disturbance. Oxford:Pergamon Press Ltd, pp. 563 - 564. 1993.

[8] O L R Jacobs, P F Hewkin and C While.Online Computer Control of pH in AnIndustrial Process. IEE Pt D, 127 (4):161. 1980.

[9] B D Kulkarni, S S Tambe, N V Shukla,and P B Deskpande. Non-linear pHControl. Chem. Engng Sci., 46: 995.1991

[10] H Kurtz. Adaptive Control of AWastewater Neutralization Process. IFACProc. Serv., 3257-3261. 1985.

[11] R L Moore. Neutralization of WasteWater by pH Control. PittsburghInstrument Society of America, pp. 10-17, 37-38. 1978.

[12] Z J Palmor and M Blau. An Auto-TunerFor Smith Dead Time Compensator. IntJ Control, 60 (1): 117-135. 1994.

[13] M J Piovoso and J M Williams. Self-tuning pH Control: A Difficult Problem,An Effective Solution. In Tech., 32 (5):45. 1985.

[14] F G Shinskey. pH and pION Controlin Process and Waste Stream. New York:John Wiley & Sons. 1973.

[15] G Stephanopoulos. Chemical ProcessControl: An Introduction to Theory andPractice. New Jersey: Prentice-Hall, Inc.1984.

[16] W S Su and I B Lee. pH Control UsingAn Identification Reactor. Ind Eng.Chem. Res., 34: 2418-2426. 1995.

[17] K K Tan, T H Lee and Q G Wang.Enhanced Automatic Tuning Procedurefor Process Control of PI/PlDControllers. AIChE Journal, 42 (9):2555-2562. 1996.

[18] G L Williams, R R Rhinehart and J BRiggs. In-line Process-Model-BasedControl of Wastewater pH Using DualBase Injection. Ind Eng. Chem. Res., 29:1254. 1990.

[19] R A Wright and C Kravaris. NonlinearControl of pH Processes Using TheStrong Acid Equivalent. Ind Eng. Chem.Res., 30: 1561. 1991.



ABSTRACT

The purpose of this research is firstly to develop an improved method for implementing reverse engineering by integratingengineering analysis within the reverse engineering framework. Secondly, the literature publication of this research willexpose mechanical engineering students taking Mechanical Design Technology course to the larger perspective of reverseengineering.

Instead of creating just the CAD model, a FEA model complete with nodes and meshes, loading and boundary conditionsis generated allowing for critical part performance information to be predicted. Costly and time consuming destructiveor non-destructive testing on the part is eliminated.

The results of this study will make the task of reverse engineering an existing part and producing an improved part greatlysimplified and structured. A significant benefit that is provided by the new RES is the total cost savings due to moreefficient manufacturing and prototype testing of the clone part, thus improving productivity.

Index Terms

CAD/CAM: computer-aided design/computer-aided manufacturing, CMM: coordinate measuring machine, RES: reverseengineering system, NRES: new reverse engineering system.

INTRODUCTION

Reverse engineering is the process bywhich an existing part or a physicalmodel is recreated or cloned. Thereverse engineering system starts witheither a contact or non-contact dataacquisition technique, followed byCAD system for model regeneration.The model is subsequently integratedinto a CAD/CAM system for tool-path generation and completed withautomated manufacturing system.Unavailable or missing geometricdata for existing parts can be acquiredby contact or non-contact data

acquisition techniques. Using thisacquired geometric data, the CADmodel of the part can be regenerated.Based on the CAD model, the partcan be reproduced using a CAD/CAM system and a numericalcontrol machine tool.Mechanical Design Technology is acourse offered for fourth yearmechanical engineering students inUniversiti Teknologi Petronas,Malaysia. The course cover topicsfrom design to manufacture andreverse engineering being one topic.Reverse engineering’s applicationsare expanding rapidly from

manufacturing to medicine. RES isalso an effective tool forimplementing concepts such asconcurrent or simultaneousengineering since it helps shortendesign-to-manufacture lead time. Inrelation to this research, reverseengineering can be further defined asthe process of building a CAD modelfrom an existing part or prototype,allowing for engineering analysis anddesign improvement, before othermanufacturing processes such ascomputer-aided manufacture andmachining to obtain a cloned ormodified part.

INTEGRATING ANALYSIS AND DESIGN IMPROVEMENT IN A

REVERSE ENGINEERING FRAMEWORK

Ahmad Majdi bin Abdul Rani


31750 Bandar Seri Iskandar, Tronoh, Perak, Malaysia.

[email protected]

This paper was presented at the International Conference on Engineering Education, Oslo, 6-10 August, 2001.



PROBLEM STATEMENTS

Two notable shortcoming in relationswith reverse engineering currently:(1) the existing RES fails to providecritical performance information forthe part being remanufactured, and(2) available literatures andpublications on reverse engineeringare not only very limited forengineering students but thoseavailable mainly focus on dataacquisition rather than the overallperspective.

Current available reverse engineeringsystems lack the ability to analyze apart before its remanufactured. Itcould be a waste of effort to digitize abroken or worn part and laterremanufacture the exact duplicate.For often times, breaks and excessivewear in parts are the results of poordesign. Instead, it is feasible toanalyze and redesign the part model,if necessary, eliminating inferiorcharacteristics before remanufacture.This is extremely effective foridentifying areas of concern such assafety and reliability before significantamounts of effort and money arespent on remanufacturing parts.Costly and time consumingdestructive or non-destructive testingon the part are also eliminated. Sucha situation would increaseproductivity and efficiency.

RESEARCH SIGNIFICANCE

Integrating engineering analysis in thenew reverse engineering system,NRES, provides data on theperformance of the part beforeremanufacture, thus minimizing thepossibility of remanufacturing partswith defects or deficiencies.Additionally, this effort will provideengineering students with a literaturethat encompass the larger perspectiveof reverse engineering. Students

taking similar courses to MechanicalDesign Technology which coversreverse engineering will be exposedfrom data acquisition, CAD modelgeneration, engineering analysis, tool-path generation and manufacture.

Rules and procedures are formulatedfor conducting engineering designchanges for design improvement.The task of reverse engineering anexisting part and reproducing animproved part will be now be greatlysimplified and structured.

BRIEF PROCEDURE

The front-end of the reverseengineering process involves thereacquisition of CAD data from a partor prototype. Based on geometricdata obtained by the contact methoddigitizing system from theCoordinate Measuring Machine, awire-frame model is generated. Thewire-frame model is divided into amesh and node network. Load andboundary conditions are assigned tothe model, transforming it into afinite element model. Engineeringanalysis is then conducted, and theperformance results are analyzed. Ifnecessary, modification for designimprovement can be executedutilizing the procedure and rulesdeveloped in this research.

APPLICATION AREA

A large amount of research has beendone in the area of reverseengineering. Although theapproaches differ in many respects,none have integrated engineeringanalysis and procedures for designimprovement. Here, some of the pastand present research dealing withRES application areas, the frameworkof existing RES, and data acquisitiontechniques are reviewed.

Huang and Tai [2000] presents amethod for pre-processing data pointsfor curve fitting in their RES. Datapoints of an existing part is measuredby a CMM and processed beforefitting into a B-spline form. Thismethod is implemented for a numberof practical application.

Carbone et al. [2001] proposed a RESfor complex, free form surfaces, basedon the integration of themeasurement information from a 3Dvision sensor and a CMM. CADmodel of complex geometry can bereconstruct with high accuracy andminimum human intervention.

Quality Machine of Looveland,Colorado developed andimplemented a RES to remanufacturea thermoplastic elastomer handle.The handle attaches to a lever thatcontrols a valve on a hydraulicmachine. In their RES, QualityMachine utilized a non-contact dataacquisition technique, a Digibot laserscanner, to digitize the handle. Basedon the digitized geometric data, a 3-D CAD model of the handle iscreated. A CAD/CAM system,SurfCAM, in the RES is used togenerate the NC codes required tomachine the electrode to produce themold. Before cutting expensivematerial, the tool paths were testedusing NC proofing polyurethanefoam. Their RES lacks criticalperformance information for themold which could have beengenerated through engineeringanalysis by utilizing the samegeometric data.

Most other research in RES focusesmainly on data acquisition whilemaking a passing statement on theCAD/CAM component of theirRES; none took the extra step offurther utilizing the CAD modelgenerated from the digitizing system



to perform engineering analysis.Thus, this research attempts tocontribute in the reverse engineeringframework by developing a procedurefor integrating engineering analysissuch that the NRES can provideadditional information by predictingthe performance of the part to bereverse engineered. Additionally, amethodology is developed to assist inconducting design changes forproduct performance and reliabilityimprovement by providing rules andprocedures for engineering designchanges.

RESEARCH PROCEDURE

The framework developed forgenerating performance informationthrough engineering analysis from thenew reverse engineering system(NRES) is described and the structureis shown in Figure 1.

The geometric data of an existing partis acquired by a contact dataacquisition method using a CMM.A wire-frame model of the digitizedpart is constructed based on thegeometric data acquired. The criticalperformance information of the partto be remanufactured can begenerated by performing engineeringanalysis. Instead of producing aprototype and testing it with either adestructive or non-destructivemethod, in this research, a finiteelement method is utilized as a tool.A procedure is developed totransform the wire-frame model intoa format suitable for conducting finiteelement analysis. The part model isrepresented as a finite element modelby breaking-up the wire-frame modelinto a finite number of smallerelements. Each element is in theshape of a hexahedral brick with eightnodes. Actual loading and boundaryconditions are assigned to the model.Critical performance information

regarding the linear stress anddeflection data is collected. This datais compared to the known yield stressvalue of the intended material of thepart. The yield stress value acts as thethreshold for the linear stress valueobtained from conducting theperformance analysis. Any valueexceeding the yield stress is anindication that the part has yieldedunder that particular loadingcondition. Considering safetyfactors, any stress value less than theyield stress but exceeding the safetymargin is deemed as unsuitable.

Engineering changes are required toovercome deficiencies by makingdesign modifications or selecting asuperior material. A set of rules isdeveloped in this research providingguidelines for making designmodifications in the NRES. Oncedesign modification is completed,engineering analysis is againconducted to analyze theperformance of the modified design.The iteration between designmodification and engineering analysisis executed until a suitable design isaccomplished.

DATA ACQUISITION OF THE

SAMPLE PART

The isometric view of the sample partis shown in Figure 2. For this partthe line/line intersection alignment isused since the part is of a prismaticshape with planar surfaces.

The digital measurement capabilitiesof a CMM allows the extraction oracquisition of engineering design datasuch as Cartesian coordinates,surfaces, and orthographic drawingsfrom the existing part. The geometricdata acquisition for the test part isimplemented using a Brown &Sharpe MicroVal PFx coordinatemeasuring machine. MicromeasureIV software is used for writing the

Improved CAD Model

Part DigitizationUsing CMM

2DModeling

3DModeling

Engineering Analysis& Design Modification

Figure 1Structure for integrating engineering analysis and design modification.

Figure 2

Isometric View of Sample Part.



data acquisition program. Thecomplete digitizing program writtenfor the geometric data acquisitiongenerated for the test part is writtenfor each elevation.

The mappings of all the views are usedto generate the CAD model of thesample part, using the modelingmodule in the Algor system.

PROCEDURE FOR

INTEGRATING ENGINEERING

ANALYSIS AND DESIGN

IMPROVEMENT

The most crucial step in integratingengineering analysis, in reverseengineering, is transforming the wire-frame model into a proper finiteelement model. In this research, aprocedure is developed fortransforming the CAD model into aFEA model. The aim is to developthe most suitable mesh and nodalpattern that provides enough elementsto obtain accurate results withoutwasting data interpretation andprocessing time.

With a 3D model, the choice of finiteelement is limited to volume elementssuch as tetrahedron, pentahedron,

hexahedron, and solid or brickelements (Refer Figure 4). Three-dimensional analysis is preferred overtwo-dimensional analysis to analyzethe result through-out the entiremodel rather than just a cross-sectionof the part. The brick element isselected over the other element types.

It was chosen based on its suitabilityand ease in dividing the prismatic partinto finite elements. Tetrahedral,pentahedral, or hexahedral are usuallynecessary to accommodateirregularities in geometry.

Although finite element meshesshould be uniform throughout themodel, mesh refinement is performedat regions of rapid change in geometry[Champion; 1992]. Mesh refinementis required to obtain more accurateresults especially of stresses rather thandeflections. The reason being thatstresses are calculated using derivativesof the displacements. Stressescalculated at adjacent elements maydiffer substantially if the finiteelement mesh is not adequatelyrefined. A factor to be consideredwhen conducting mesh refinement isthe element aspect ratio. This isdefined as the ratio between theelement’s longest and shortestdimension. Regions with smallvariation of stresses could have a 40to 1 aspect ratio and still yield goodresults. As a general rule, an aspectratio of about or under 10 and 3 fordeflection and stress analysis,respectively, should be followed[Spyrakos, 1994].

In the part model being analyzed, thestress output at critical locations iswhere there is an abrupt change in thegeometry. Corner nodes are difficultlocations to compute stresses.Unfortunately, in many cases peakstress occurs at corners, and theirmagnitudes may govern the design.

Engineering design changes need tobe executed to overcomeconcentrated stresses that may resultin local plastic deformation. On theother hand, even with ductilematerials, areas of stress concentrationare possible sites for fatigue if thedesign part is cyclically loaded. Filletradius may be introduced to reducethe stress concentration well belowthe material yield point withconsideration to the safety factor.

In making design changes byintroducing a fillet (Refer Figure 5and 6), it is necessary to start withsharp fillet, gradually increasing tofull or even blunt fillet if necessaryTetrahedron Pentahedron

Hexahedron Solid or Brick

Figure 4Wire-frame Model of the DigitizedSample Part.

D

r

h

h

d

Sharpness of fillet = h/rDepth of fillet = h/d

Figure 5Sharpness and Depth of Fillet.

Sharpness of fillet, h/r > 1.0

Full fillet, h/r = 1.0

Blunt fillet, h/r < 1.0

Figure 6Various Sharpness of Fillet



[Avallone et al., 1986]. These stepsare necessary in order to achievesignificant reduction in stressconcentration. In application areassensitive even to minimal deflectionof the part, buttresses can beintroduced to provide support againstdeflection.

ENGINEERING ANALYSIS

PROCEDURE

The idea of implementing theprocedure to integrate engineeringanalysis in the new reverse engineeringsystem, NRES, is demonstrated usingthe sample part shown earlier in Figure2. The terms sample part and test partare used interchangeably in thisresearch referring that part.

The aim of this procedure is todevelop the most suitable mesh andnodal pattern that provides enoughelements to obtain accurate resultswithout wasting data interpretationand processing time. Considering thatthe sample part is prismatic in shapewith planar surfaces, brick elementsare selected, uniformly dividing thewire-frame model into a finite elementmodel as shown in Figure 7. Thenodes are all defined at load andsupport points.

Observing that the model as shownin Figure 7 consist of corners withrapid change in geometry between

two coplanar surfaces, meshrefinement needs to be conducted(Refer Figure 8). To ensure that theanalysis yield good results, elementaspect ratio is kept under 3 for stressanalysis (Spyrakos, 1994).

LOADING AND BOUNDARY

CONDITIONS

The assumed actual uniform staticload on the test part is 1 000 lb. Theuniform static load is acting on theleft vertical surface of the step withthe whole base of the sample part fullyfixed. An example source of theuniform static load can be from amating part acting on the surface. Abase frame that is required to supporta dead weight which is essentiallyconstant overtime, and the base framedoes not move can be a good example(Refer Figure 9).

The vector of the loading conditionsand specified boundary conditions areshown in Figure 10. The direction ofuniform static load is represented bythe horizontal arrows and the fixedbase boundary conditions arerepresented by the triangles at the baseof the part as shown in Figure 10.

MATERIAL PROPERTIES

In this research, Aluminum Alloy(Aluminum Association Number2024 - O) is selected as the material

for the sample part. This aluminumalloy is easily machined andconsidering the cost and strength ofaluminum alloys, they are among themost versatile materials from thestandpoint of fabrication. The tableof materials properties shown in TableI provides the required data forconducting engineering analysis.

Table I: Material Properties of

Aluminum Alloy [ASM Metals

Reference Book].

Material Aluminum

Aluminum Association Number 2024

Modulus of elasticity E, Mpsi 10.3

Modulus of rigidity G, Mpsi 3.85

Poisson’s ratio v 0.33

Density δ , lb/in3 0.10

Yield strength Sy , kpsi 11.0

Figure 7Side Elevation of Model with

Uniform Brick Elements.

Figure 8

Side Elevation of Model after MeshRefinement.

Figure 9Load Source and ApplicationExample.

Figure 10Loading and Boundary Conditions

from Side Elevation.



TESTING AND RESULTS

Once the finite element model iscompleted with loading and boundaryconditions, a test run is executed toanalyze the stress concentrations anddeflection on the sample part. Theactual engineering analysis isconducted on the Algor finite elementsystem. The output of the linear stressanalysis and deflection is reported ina dithered plot of the stress anddeflection of the sample part. Adithered plot is a graphical display ofthe stress variation in the form ofcolored areas. Dithered plots providea vivid and impressive presentation ofstresses and deflection as shown inFigures 11 and 12. The maximumstress and deflection under 1 000 lb.is 7 216.38 psi and 3.5e–4 in.respectively.

SAFETY FACTOR AND FAILURE

CRITERIA

A safety factor is a unitless ratio thatis necessary to calculate and estimatethe likelihood of failure. Inengineering design the ratio is usuallybetween yield strength/stress output,both having the same units of poundper square inch (psi). Otheracceptable ratios include critical load/applied load and load-to-fail part/expected service load. In this research,the safety factor selected is betweenyield strength/stress output since they

are a function of the applied loads andthe part’s geometry.

Several theories exist in explainingfailure, but the most accurateapproach that agrees closely withexperimental data is the von Misestheory (Norton, 1996). This theoryis the best choice for predicting failurein the case of static loading of ductilematerial in which the tensile andcompressive strength are equal. Thevon Mises theory was adopted in thisresearch. The direct comparison ofthe von Mises stress with yield stressallows identification of areas that haveyielded.

Considering the safety factor of N =3, the calculation so that the stressstate will be safely inside the failure-stress value:

N =σ y

σ ' (1)

σ ' = 110003

= 3 666.67 psi (Threshold stress value)

Critical performance informationregarding the linear stress anddeflection data is collected. This datais compared to the known calculatedthreshold stress value of 3 666.67 psi.Any value exceeding the yield stress isan indication that the part has yieldedunder that particular loadingcondition.

DATA ANALYSIS AND DESIGN

MODIFICATION

Under the assumed load of 1 000pounds, peak stress is located at thecorners between the coplanar surfacesas shown in Figure 11. Althoughconsiderable small deflection occurs,3.5e–4 inch, the stress output is foundto be 7216.38 psi. This value is wellbeyond the specified threshold valueof 3666.67 psi. Design modificationis conducted by generating a sharpfillet with radius Ri between coplanarsurfaces. The highest sharpness offillet ( Hi / Ri ) value of 6.0 is selected.

Sharpness of Fillet:6.0 ≥ [Hi / Ri] > 1.0Hi / Ri = 6.0Ri = 0.5 / 6.0 = 0.0833 inch

A fillet with a 0.0833 inch radius isintroduced to the test part, stress anddeflection are again analyzed. Underthe same loading and boundaryconditions the test data collectedshows a maximum stress output of4 937.88 psi and maximum deflectionof 2.5e–4 in. The mating part forcingon the test part is assumed to have theequivalent fillet radius.

The element shape selected for mostparts of the model is a solid or brickelement, while the region with filletare assigned with hexahedronelements. As previously, the elementaspect ratio is again kept under 3 forstress analysis.

Figure 13a and 13b show the ditheredplot of the stress on the test part undera 1 000 pound load.

The analysis result shows that themaximum stress output of 4 937.88psi exceeded the safety threshold valueof 3 666.67 psi calculated for load of1 000 lb. Further iterations of designmodification is conducted by reducing

Figure 11Side View of Stress Concentration

at 1J000 lb.

Figure 12

Side View of Deflection at 1J000 lb.



the sharpness of fillet until the stressoutput is equal or lower than thethreshold stress value as shown inTable II.

Figure 14a and 14b show the ditheredplot of the stress on the test part undera 1 000 pound load.

Under the load of 1 000 pounds, thetest results identified that themaximum stress output on the testpart with a sharp fillet of 0.2 inchs is3 170.40 psi and a minimal deflectionof 0.00018 inchs. Comparing thestress value with the specifiedthreshold value of 3 666.67 psi, it iswithin the safety range.

Thus, it can be concluded that it ishighly unlikely for the improveddesign test part to fail by yielding orbreaking under the assumedconditions. Even though a lowerstress output can be obtained byfurther modification with a full fillet,this design is unsuitable given the loadposition, also the design with a sharpfillet had achieved the target.

The improved design of the test part,shown in Figure 15, can then betransferred to a CAD/CAM systemusing either data exchange format(DXF) or initial graphic interchangespecification (IGES) format fortoolpath generation. The NC codesgenerated from the CAD/CAMsystem can be downloaded onto anumerical control machine for theproduction of the improved part.

SUMMARY OF RESEARCH

WORK

The objectives of this research wereto integrate analysis and designimprovement in a reverse engineeringframework by further utilizing thegeometric data acquired through thecontact digitizing method of a CMM.

Figure 13a3D Dithered Plot of Stress with6.0 Sharpness of Fillet.

Figure 13bSide View Dithered Plot of Stress.

Iterations Fillet Fillet Max. Max. Threshold

Sharpness Radius Deflct’n Stress Value,

(in) (in) (psi) psi

1 - - 3.5e4 7216.3 > 3666.67

2 6.0 0.0833 2.5e-4 4937.8 > 3666.67

3 5.0 0.1000 2.4e-4 4599.3 > 3666.67

4 4.0 0.1250 2.3e-4 4132.3 > 3666.67

5 3.0 0.1670 2.0e-4 3736.9 > 3666.67

6 2.5 0.2000 1.8e-4 3170.4 < 3666.67

Table IIDeflection and Stress Output for Various Design Iterations.

Figure 14a3D Dithered Plot of Stress with2.5 Sharp Fillet.

Figure 14bSide View of Stress Plot with 2.5Sharpness of Fillet.

0.500

0.491

1.997

1.499

0.486

1.998

All filletradii 0.2

Figure 15Partially Dimensioned Drawing

of Modified Part.

* All dimensions same as in

Figure 4.1b except fillet.(All dimensions in inch.)



Critical performance information ofa part to be reverse engineered wassought without conducting costlydestructive or non-destructive testingof the part. Instead, the criticalperformance information wasacquired by conducting engineeringanalysis on the generated CAD model.

To accomplish these objectives thewire-frame model generated in themodeling module based on thedigitized data of the test part istransformed into a finite elementmodel. The finite elementrepresentation of the test part is inputto a finite element analysis system,which, based on the provided loadingand boundary conditions, generatesthe stress and deflection data on thetest part. The generated test data onstress is compared to the calculatedthreshold stress value, whichincorporates a safety factor. Designimprovement on the test part isexecuted and tested until a suitableand improved design is accomplished.

CONCLUSIONS

By providing the design engineer withthe NRES, the task of reverseengineering an existing part andproducing an improved and superiorpart is greatly simplified andstructured. It is very easy for theengineer or designer to analyze andpredict the performance of the partinstead of merely reproducing a cloneof the existing part. Changes in thecomponent or part design can beconducted with ease and engineeringanalysis can predict how the changeswould affect factors such as stress anddeflection. This eliminates the needfor conducting costly and timeconsuming destructive or non-

destructive testing on the part. Insteadof reverse engineering a part thatmight have failed due to inferiordesign, the design engineer has thecapability of producing a superiordesigned part.

Past research and development in thearea of reverse engineering hascontributed tremendously to taskssuch as contact data acquisition, non-contact data acquisition, andintegrating computer-aided designand computer-aided manufacturing.This research has integrated analysisand design improvement in the reverseengineering framework to assist in thetask of generating performanceinformation and design improvementprocedures for a prismatic part. Thisarea of reverse engineering will helpreduce manufacturing lead time andpromote consistency in integration ofdesign and engineering analysis in thereverse engineering framework.

The publication of this research workwill provide engineering students thelarger perspective of reverseengineering. It encompass from initialdata acquisition, CAD modelgeneration and finite element analysisbefore the part is re-manufactured.

The prototype new reverseengineering system developed in thisresearch has the following limitations:• the test part is limited to a

prismatic part with planarsurfaces.

• the workpiece material is limitedto ductile materials with no crackformation.

• only static loading conditions areconsidered.

• only room-ambient environmentare considered.

REFERENCES

[1] Eugene A Avallone and TheodoreBaumeister III. Mark’s StandardHandbook for Mechanical Engineers (9thed). New York; McGraw Hill, 1986.

[2] C Bradley, V Chan. “A ComplementaryApproach to Reverse Engineering.”Journal of Manufacturing Science andEngineering 123, 1 (Feb. 2001): 74-82.

[3] V Carbone, M Carocci, E Savio, GSansoni and L De Chiffre.“Combination of a Vision System and aCoordinate Measuring Machine for theReverse Engineering of FreeformSurfaces.” The International Journal ofAdvanced Manufacturing Technology 17,4 (Jan.’ 2001): 263-271.

[4] Edward R Champion Jr. Finite ElementAnalysis In Manufacturing Engineering: APC Based Approach. New York; McGrawHill, 1992.

[5] Fred Hansen, Elias Pavlakos and EricHoffman. “PARES: A Prototyping andReverse Engineering System forMechanical Parts-On-Demand on theNational Network.” Journal ofManufacturing Systems 12, 4 (1993): 269-81.

[6] Ming-Chih Huang and Ching-Chih Tai.“The Pre-processing of Data Points forCurve Fitting in Reverse Engineering.”The International Journal of AdvancedManufacturing Technology 16, 9 (July2000): 635-42.

[7] Robert L Norton. Machine Design: AnIntegrated Approach. New Jersey; PrenticeHall, 1996.

[8] Constantine C Spyrakos. Finite ElementModeling in Engineering Practice.Pittsburgh; Algor Publishing, 1994.



INTRODUCTION

A topic of interest in the teachingprofession is related to how to gainand attain the students’ interestthroughout the lecture. Sustainingstudents’ interest for a whole periodof lecture is by no means an easy task.To capture and hold students’ interestrequire creativity, skill and experience.“Teachers must face the problem of howto maintain curiosity and interest as thechief motivation forces behind thelearning. Sustained interest leads thestudent to set himself realistic standardsof achievement.” [1].

Keller described six strategies forgaining and maintaining attention.[2]i. Concreteness, i.e using concrete

examples in tying up the topic tograb the learner’s attention.

ii. Incongruity and conflict, i.e.posing facts contrary to learner’sexperiences to stimulate thelearner’s interest.

iii. Humour, i.e. using a fair amountof jokes maybe as an introductionto a topic or while deliveringlectures to break monotony.

iv. Variability , ie. combiningmethods of presentation andlearner’s activities to stimulate andsustain student’s interest.

v. Participation, ie. involvinglearners in hands-on learning.

vi. Inquiry, i.e. asking questions forthe students to solve or involvingstudents brainstorming to comeup with solutions to the lesson.

Studies have shown that thetraditional format of lecturing posesdifficulties to students where attentionspan is concerned. Adult learners canconcentrate well for no more than 15to 20 minutes at the beginning of alecture [3]. At this period of time thestudents are still fresh. JoanMiddendorf and Alan Kalish, in theirarticle, ‘The “Change-up” in lectures’

Sustaining Students’ Interest –

Sharing Of Experience




[email protected]

ABSTRACT

The problem faced by most instructors is in sustaining the interest of students during lectures. The objective of this paperis to share teaching and learning experiences from the mistakes made which contributed to the problem of sustainingstudents attention throughout the lecture.

At both the college and university levels, the evaluation of lecturers by their students has provided a good feedback toimprove teaching methods. The paper focuses on the writer’s personal experience of sustaining students’ interest not onlyat the college level where the class sizes do not exceed fifty but also at the university level where the number reaches closeto two hundred.

The paper describes the steps taken by the writer in trying to improve her teaching and results from the evaluation by thestudents have shown positive improvement although there is still room for improvement yet in the same problem area.

Keywords:

sustaining interest, evaluation, experience, college , university.

This paper was presented at the international Conference on Challenges & Prospects in Teacher Education, Shah Alam, 16-17 July, 2001



suggest that lectures should beinterrupted with periodic activities.[4] A study by A.H. Johnstone and F.Percival in 1976 observed a patternthat shows the lapse of attentionoccurring at intervals; first lapsewithin the first three to five minutesafter the students settled down, onestudy noted that the next lapse ofattention usually takes place ten toeighteen minutes later, and as thelecture continued, the attention spanreduced to as low as three or fourminutes towards the end of thelecture.[5]

BACKGROUND

“Effective learning in the classroomdepends on the teacher’s ability … tomaintain the interest that broughtstudents to the course in the first place.”(Ericksen, 1978, p.3) Whatever levelof motivation your students bring to theclassroom will be transformed, for betteror worse, by what happens in thatclassroom. [5].

The ability of an instructor insustaining students interest is of coursea great concern to Institutions ofHigher learning and universities tohelp students learn. One methodemployed by two institutions ofhigher learning A, a private college inSelangor and B, a private universityin Perak to measure this criteria is viaan evaluation process. The evaluationis an assessment of lecturers by theirown students. Students responded toa questionnaire employed at the endof each semester. The studentsinvolved in this evaluation processwere mainly those in the foundationlevel and the courses taught by thewriter were more or less the same typeat both A and B. Each evaluationexercise was done before the finals andin the absence of the instructorconcerned. Some importantcomponents to be assessed by the

students were; quality of presentation,knowledge on subject matter,interaction with students, ability tosustain students’ interest, pace ofinstruction and value of activities andexercises. An important componentof the feedback that we will look intois the ability in sustaining students’interest. This particular componentis a common feature for both A andB. This paper focuses at the rating ofthis component assessed by herstudents. The writer had spent sevenyears of teaching Mathematics at A,is currently a lecturer at B. She aimedat achieving at least a 60% ratingfavouring above average or satisfactoryfor her ability in this aspect. Theobjective of this paper is to share herexperience in sustaining her students’interest at both the learning centres.This paper further describes the stepstaken by her in an effort to improveher ability in capturing and holdingher students’ interest throughout astandard lecture. Although there isstill room for improvement, it ishoped that the experience shared willbe beneficial to teachers andinstructors in their teachingprofession.

At A, the number of students per classwas small as small as 11 and thelargest, a group of 47 students.Lectures were done in classrooms andmost of the time the overheadprojector and the whiteboard were themain teaching aids then that wereutilised by the writer. Each standardlecture at the college took 40 minutesfollowed by a 10 minutes coffee-breakand proceeded with the lectures foranother 40 minutes.

At B, lectures were carried out in thelecture theatres as the numbersreached close to 200. The durationfor each lecture or tutorial is 50minutes. Due to the large class size,the students were divided into smaller

groups of about 30 for tutorialsessions. Teaching load was threelecture hours per week and fivetutorial hours fortnightly. Teachingaids for lectures used by the writerwere the microphone, overheadprojector and whiteboard. For thepurpose of tutorials, only thewhiteboard was fully used. The writerhad tried to adapt a somewhat similartechnique to sustaining studentsinterest at the university as what shehad used whilst at the college.

The following describes some of herapproaches in an attempt to overcomesome problems encountered ingaining students interest. These stepshad already been implemented by herwhile teaching at A and B.

1. Randomly, starting off a

lecture with a short quiz

The writer observed particularly thatthe 8.00 o’clock and the 2.00 o’clocklectures required a technique to keepstudents awake. Starting offimmediately with a lecture may wellsend them off to slumber. One wayof hastening students to shorten theirsettling down time and give theirimmediate attention is by giving thema short pop quiz. The 10-minute quizrequired only brief answers to ease herburden of grading. The marksrewarded contributed 10% in thecontinuous assessment. Throughoutthe 15-week semester sheadministered 10 such quizzes.

2. Describing the

objective(s) or the

topic(s) for the day

At the end of one Algebra andTrigonometry lecture at A, a studentcame forward to ask the writer whythe particular topic was done and thathe could not appreciate the use of it.The mistake, the writer realised, was



starting off the lecture by merelymentioning,

“OK, the new topic we will be dealingwith today is Linear Programming.”and then went on straight to the topicof discussion. The topic has not beenproperly introduced. A moreinformative introduction could havesparked some curiosity. “One cause ofteaching disappointment can be that thestudent did not recognise the reasons whythe material being presented was of valueto him/her: the value of the material asperceived by the student is a prime factorin the student’s motivation to learn.”[5].“In ‘selling’ lectures, the first threeminutes are most critical – they open orclose the prospect’s mind. If you don’testablish the prospect’s needs for yourproduct in the first three minutes, allthe rest of the time is wasted.”[5].During the following semester thewriter started off the topic LinearProgramming by asking a question,“Did you know that in the businessworld, there is a mathematicalprocedure in deciding how many ofeach model of a product amanufacturer should produce in orderto maximise his profit?” From herobservation, it was an eye-opener!

3. Providing ready-made

notes

Having to cope with writing downnotes was bad enough, let alone thetask of trying to listen and understandat the same time. Many students cameforward to suggest that they preferredhandouts to copying notes from theboard during lectures. Havingprovided those materials well before alecture, the writer personally foundthat the students had time to readbefore coming to class and gave a moreundivided attention as far as trying tolisten to what she was saying wasconcerned. Their responses andparticipation proved better. While

providing ready-made notes helped ingaining students’ attention, it shouldalso be cautioned that this practicecould initiate a habit of absenteeism.So to discourage such circumstances,some of the solutions to examples wereleft out as empty spaces for in-class-activity purposes. At both A and Bwhere class attendance was monitoredduring every lecture, the writer foundno changes.

4. Controlling voice and

speech

When the writer first started lecturesat A, the students complained that shespoke too fast for them to catch up.In trying to deliver as much aspossible, the lectures delivered werewith minimum pause and there wasno variety in the intonation of speech.It must have been the monotone thatmade many of her students felt bored.In learning to stress on only the mainideas rather than everything she knew,the writer managed to slow down soshe could pronounce each wordclearer and paused more often. Clarityand loudness of voice, the writerfound, is a very important asset inholding students’ attention. Thewriter observed that the studentsbecame more inquisitive, most likelybecause there was time to ponder overwhat had been said. “Use your voiceand gestures to reinforce the content ofthe lecture.” [6]

5. Observing the students

It is vital to have a good view of thestudents’ facial expressions and bodylanguage to detect any signs ofboredom, confusion or evensleepiness. At A, when the studentsshowed these signs the writer wouldgive them a few minutes for stretchingout or even allow them to go to thewashroom to freshen up, sincewashrooms were located on every

floor. For a less convenient situation,such as in lecture halls at B the writerwould purposely switch the topic ofdiscussion entirely and talked aboutsome local or world news, asked theirviews regarding some issues, or givethem a riddle. At times she merelypractised a moment of silence and justobserved the students, particularlythose who had actually dozed off. Therest will start turning around andperhaps silently nudged a snoozingneighbour. From the students’ facialexpressions, she could also detectblank looks and to avoid anyembarrassment, she often asked thisquestion, “Anyone not happy with theexplanation? Would anybody like torequest for a replay? Do I see any showof hands ?”

6. Preparing and

delivering lectures

At A notes flashed on the overheadwere hand-written. The writings werebig and multicoloured. At B, thewriter started writing notes in PowerPoint and added in some cartoons orpictures to create more interest duringthe lectures. The contents per slidewould be minimum. Detailedexplanation would be discussed andwritten on the whiteboard. Whereformulas were concerned, she createda story in making it easy for herstudents to memorise a certainformula. For example while teachingTrigonometry, students needed torecall the formulas for Sin (A±B) orCosine (A±B). she would start off bypersonalising Mr Sine and Mr Cosine.Mr Sine has the character of anobedient and sociable person whilstMr Cosine a total opposite;disobedient and will only clique withhis own type. Hence the formulasSin (A + B) = SinA CosB + Cos A SinBSin (A – B) = SinA CosB – Cos A SinBCos( A + B) = CosA CosB – SinA SinBCos(A – B ) = CosA CosB + SinA SinB



could be recalled easily.Demonstrations were inevitable incapturing students interest. When shetaught the students the topic ProjectileMotion in Calculus, she used todemonstrate the idea of zero velocityachieved at the highest point bythrowing up a duster in the air acouple of times. When studentsfound difficulties in understandingthe definition of the Ellipse in theAlgebra and Trigonometry class, shebrought along some thread andrequested two students to help herhold the two ends acting as the foci,while she held the thread taut with amarker pen to sketch the locusformed.

7. Punctuating lectures

At A, the first ten minutes would betaken up by either a short pop quiz ora recap. The writer would ask studentsverbally to tell the class the gist of whathappened in the previous lesson. Inthe next 10 minutes she would covera small section of a topic, with a coupleof examples, all worked out. The thirdproblem would be for the students totry out, where she would allocate fiveminutes at the most, after which shewould request for a volunteer to workthe problem out on the board. Theworst part was when there were novolunteers, so she tried a way whichworked wonders! Even the most timidstudents were willing to contributetheir ideas! She mentioned to thestudents that she allocated marks forclass participation. Although it maynot be much, but it will still contributeto their continuous assessment. Toprove her point, she brought along aname list for each lecture to jot downmarks for participation. Anyquestions and answers will be dealtwith by the volunteer. This wouldnormally consume at the most tenminutes. Another short lectureproceeded for the next 10-15 minutes

before she released the students fortheir coffee break. As soon as coffee-break was over, the writer usuallycontinued lecturing for another 15-20 minutes and then requested thestudents to break up into small groupsof four for group discussion andbrainstorming session. She wouldguide them by writing hints on theboard. The class may be a little noisywhen they started their discussion, butthat was when they were learning fromtheir peers. There will be some groupswith questions which will requiresome help, so she will be movingaround the classroom to check ontheir progress. Solutions will then bedisplayed on the whiteboards byrepresentatives from each respectivegroups. Any questions or explanationswere dealt with by the presentersthemselves. The writer was then onlya facilitator.

The writer used the same techniqueof giving pop quizzes or recap followedby brief lectures at B except thatinterruptions for hands-on werelimited to solving the problems onindividual basis. So in order to adaptto the situation where movement isrestricted, she developed the cultureof working out the more difficultproblems , step by step, together withthe students. She would ‘rely’ on thestudents suggestions for each move inthe working by writing them on theboard. Literally she would be writingaccording to what the students said.‘Working together’, they would arriveat the final solution and often, thestudents would give themselves around of applause, gleefully. Thesestudents will not gather marks for suchmassive participation, but they had anequal chance of getting marks for theircontribution during tutorial sessions.

Methodology

Data was collected through 12

semesters, from August 1995 untilAugust 1999 at A. There were threesemesters per year. The data collectedwhist at A is based on the summarisedresult of the particular component‘ability to sustain students’ interest’.The questionnaires employed for thisevaluation purpose was the same oneissued to students throughout thisperiod of time. Students respondedto the item by choosing one of thefollowing:

A = ExcellentB = GoodC = AverageD = Poor

The total number of studentsregistered for the courses taught by thewriter was 1498 and 1099 wereinvolved in the survey. These studentswere either in their first, second, thirdand sixth semester.

Data obtained from the evaluationexercise at B was from semesterJanuary 2000, June 2000 and January2001. Students’ ratings were based onthe a scale of 1 - 7 as shown:

Poor Satisfactory Excellent

1 2 3 4 5 6 7

For the purpose of comparing theevaluation during the three semestersat B, the writer has categorised thepoints as follows:

A Excellent 7B Good 5, 6C Satisfactory 3, 4D Poor 1, 2

The total population registered for thecourses taught was 652 and 536responded to the questionnaire. Theserespondents were first semesterstudents. The same questionnaire wasused each semester.



RESULTS

Student evaluation at A

The graphs in Figure 1 reveal thestudents’ opinion of the writer’s abilityin sustaining their interest from theAugust 1995 through August 1999 atA.

Student Evaluation at B

Figure 2 shows how the students at Bhad evaluated the writer’s ability insustaining their interest, for Jan 2000,July 2000 and Jan 2001 semesters.

ANALYSIS

We may regard the students at A andB to be a constant component in theevaluation process, since they weremainly those in the foundation level.Courses taught at both the institutionswere basically Algebra and Calculus.The performance of the writer is thevariable factor. The writer aimed toachieve an improvement in theevaluation pertaining to her ability insustaining students’ interest. She hadhoped to maintain a 20% ‘excellent’rating and a 40% ‘good’ rating so thatshe could obtain a 60% rating of atleast a ‘good’ rating. Implementationof the steps in an attempt to improveher performance was sometimetowards the end of the year 1997.Although a 20% “excellent’ ratingoccurred prior to 1997, it wasprominently inconsistent. It isapparent that after implementing thesteps, the charts showed that sinceApril 1998, the writer managed toachieve her target. It is also noted thatthe percentage of ‘poor’ ratingdecreased considerably and achieved0% by August 1999. The similarscenario was maintained at B, exceptperhaps the ‘excellent’ rating wasslightly lower than 20% in the firstsemester, probably due to adjustmentsin overcoming a different teachingenvironment. It was however

FIGURE 1:DetailedVersion of theEvaluation OfThe Author’sAbility InSustainingStudents’Interest at A.

FIGURE 2:Students’Evaluation OfAuthor’s Abilityin Sustaining

StudentsInterest at B

A B C D

50

40

30

20

10

0

%

Aug 95

Dec 95

Apr 96

A B C D

50

40

30

20

10

0

%

Aug 96

Apr 97

Aug 97

A B C D

60

50

40

30

20

10

0

%

Dec 97

Apr 98

Aug 98

A B C D

60

50

40

30

20

10

0

%

Aug 98

Dec 98

Apr 99

Aug 99

A B C D

80

60

40

20

0

%

Jan 00

Jun 00

Jan 01



followed by an improvement for thesubsequent two semesters. Theincrease in the average score over 15semesters is significant. Thecorrelation coefficient between thesemester and the score is 0.7478,which is significant at 95% level.

The writer hopes in the near future,to achieve a 30% ‘excellent’ rating anda 50% ‘good’ rating. To achieve that,the writer intends to vary her teachingstyle by perhaps adapting the LCDprojector to deliver her lectures.

CONCLUSION

Apparently, the evaluation processprovided a useful tool in helping thewriter realise the students’ expectationand quality level of her teaching,particularly her ability in sustainingher students’ interest. She took timeto implement the changes in her styleof teaching and admitted thatadapting punctuated lectures and

getting the students involved activelywere more enjoyable for her and theresults reflected that her students werealso in favour of such techniques. Thegraphs indicated positive results afterthe writer implemented a moreorganised and a student-friendly wayof teaching. However there is stillroom for improvement. Nevertheless,it is with the hope that teachers andlecturers will find this experiencebeneficial.

REFERENCES

[1] Pedagogy: The teaching–learningsituation. 1999. [Online]. Available:http://britannica.com [Jan 10, 2001].

[2] D Wesselhoff. Keller’s ARC Model-Attention. 1998. [Online]. Available:http://coe.sdsu.edu/eet/Art ic les/attention/start.htm

(Dec 6, 2000)

[3] B G Davis. (1999).Tools for teaching.Motivating Students. [Online].Available: http://advising.berkeley.edu/sled/bgd/motivate.html (Feb 19, 2001)

[4] J Middendorf and A Kalish. Tools forteaching. The “Change-up” in lectures.1997. [Online]. Available: http://w w w. i n d i a n a . e d u / ~ t e a c h i n g /changeups.html (April 18, 2001)

[5] W Cordes. TSFC Newsletter. Sell YourLecture. 1995. [Online]. Available:http://www.uark.edu/misc/tfscinfo/news995.html. (Jan 10, 2001)

[6] Learning and Teaching. LectureNewsletter. Teaching By Lecture. (Feb1993) [Online] Available: http://w w w. p s u . e d u / c e l t / n e w s l e t t e r /ID_Feb93.html (Jan 10, 2001)

BIODATA:

Azizan Bt Zainal Abidin, a graduate fromUniversity of Salford and Ohio University iscurrently a lecturer in Mathematics atUniversiti Teknologi Petronas, Tronoh, Perak.She has been teaching for 19 years at variousinstitutions for higher learning , includingUPM (Serdang), UPM (Semenggok), UiTM(Kuching), UiTM (Shah Alam), OhioUniversity Academic Advancement Centre(Athens), Stamford College (Jalan Barat) andA College (Subang Jaya).



INTRODUCTION

The area related to the study ofmotion is referred to as mechanicswhereby the divisions of mechanics arekinematics and dynamics. The carefuldefinitions of these divisions bringabout clear indications of what areexpected from the students. Thestudents are taught to memorise andpractise as many problems as possibleand the trait observed in the students

who has been trained in this way isthat they try to match the problemsgiven to them with the problems thatthey have solved previously. Thecontinuation of this trend at thehigher level is critical because thestudents struggle to grasp the conceptseven in the kinematics area, where allthat is required from them is theunderstanding and application ofequations of motion. One area offailure that has been identified is the

application of these equations to themotions of objects in 2-dimensionalcases. A paradigm shift in the methodof learning is inevitable and the use ofmodern techniques was sought.

A careful observation of tests andexams indicated that the commonmistakes made by students showedtheir weaknesses in grasping theconcepts related to two-dimensionalmotion completely. Many do not see






e-mail: [email protected], [email protected]

ABSTRACT

The study of motion has always played an important and early role in the education of science generally and Physicsspecifically. In fact, the first area to be thoroughly investigated is the study of motion. Since the area related to the studyof motion is rather wide, categorization is always taken to be the best way out to systematically sort the concepts relatedto motion. Categories formed under the main header called mechanics are kinematics and dynamics. The carefuldefinitions of these divisions bring about clear indications of what are expected from the students. The conventional wayof teaching this rather important and significant area is by expecting the students to memorise and practise as manyproblems as possible. Normally the trait observed in the students who has been trained in this way is that they try tomatch the problems given to them with the problems that they have solved previously. The method is seen to be realisticat the introductory level, but the continuation of this trend at the higher level is critical. The students seem to fail evenin the kinematics area, where all that is required from them is the understanding and application of the equations ofmotion. One clear indication of failure in this area is related to the application of these equations to the motions ofobjects in two-dimensional cases. Since students have been taught only one-dimensional cases at the introductory leveland a paradigm shift in the method of learning is inevitable, the use of modern techniques has been looked into. Thesearch for suitable simulation software that can cater to the needs of our students in Malaysia was proven to be futile asthe bundled education packages were too general. The next best thing to do was to create our own simulation programme,by using a suitable programming language. MATLAB was chosen as the programming language and the final productwas a simulation package for the facilitation of the learning of motion in 2-dimensions. The package consists of aninteractive software, which incorporated multimedia instructional designs.

Keywords:

motion in two-dimensions, interactive software and on-line learning.

This paper was presented at the Seminar & Workshop On Multimedia Approach In Learning Physics & Engineering, Selangor, 12-14 June, 2001.



why the x- and y- components ofvelocities and displacement areindependent of each other. They can’tseem to relate the fact that while thevelocity in the y-direction changes, thevelocity in the x-direction staysconstant. The purpose of this projectis to design and produce a multimediainteractive learning module that willserve to help the students tounderstand the concept of motion intwo-dimension.

DEVELOPMENT OF MODULE

The module incorporates all the nineinstructional events based on theoriesdeveloped by Gagne and his colleagues(e.g. Gagne and Briggs 1979). Price(1991) states ‘In order to ensure thateach learning process happens, theCAI author should include a sequenceof nine instructional events that“teach” for each objective’ (ref. [4]).The nine instructional events aregaining attention, informing learnerof lesson objective, stimulating recallof prior learning, presenting stimuliwith distinctive features, guidinglearning, eliciting performance,providing informative feedback,assessing performance, and enhancingretention and learning transfer (ref.[5]).

BACKGROUND THEORY

For motion in one dimension, withv

0 and v denoting initial and final

velocities respectively and s, a and trepresenting displacement,acceleration and time elapsed:

v = v0 + at (1)

v 2 = v02 + 2as (2)

s = v0t + 12

at2(3)

A projectile’s motion with initialvelocity v at an angle θ with thehorizontal can be taken as good

example of motion in two dimensions.The simple equations of motion asshown above will be used to derive thefollowing equations that gives valueto certain parameters related to thiskind of motion. In order to apply thenecessary conditions and assumptions,a visual diagram as shown in Figure 1is necessary which shows the trajectorypaths of projectiles at different anglesto the horizontal.

0 2 4 6 8 10 12 14 16 18 20

10

8

6

4

2

y

x

Range

Ve

rtic

al D

ista

nc

e (

me

ters

)

75°

60

45

30°

15°

Figure 1The trajectory paths of projectilesfired at different angles withrespect to the horizontal axis.

The components of the initial velocityare given as equation (4) and (5).

v0 y = v0 sinθ (4)

v0 x = v0 cosθ (5)

The above resolution requires a soundknowledge in vector algebra. Aftermanaging that, the next step will bein selecting the right equations tofurther our investigations into findingout the maximum height reached andthe range of projectile. The ultimatewill be in proving that the path takenis parabolic. Figure 3 shows that themaximum height, h, occurs when vy

is equals to zero. It is observed thatthe object in the projectile changedits direction from going upwards tocoming downwards. This clearlyindicates that it reached a pointwhereby the vertical velocity was zero.But the horizontal velocity remainedconstant, or else you will haveobserved a point where the motion

would have been frozen for a veryshort period of time.

vy2 = v0y

2 − 2gh (6)

h =v0 y

2

2g= v0

2 sin2 θ2g

(7)

The range, X, is the horizontaldistance reached by the projectile fromthe original point of projection and isgiven by the following equation.

X = v02 sin 2θ

g (8)

The trajectory equation is used torelate the vertical and horizontaldisplacements, which normally showsthat the relationship between thesetwo parameters indicates that the pathtaken is parabolic. The equationshown below is an equation of aparabola.

Y = X tanθ −kX 2 (9)

k =g

2v02 cos2 θ (10)

RESULTS

The equations derived earlier weretested by using the Matlab software.Figure 2 shows that the maximumheight increases as the angle ofprojection is increased. At 90 degreesto the horizontal, the path reduces toa straight line, thus enablingverification by using equations ofmotion in one-dimension.

Figure 3 is a good visualizationdiagram, which demonstrates themotion in two-dimension by givingthe real coordinates of the entire path,based on the equations derived earlier.And this validates the model to beused as the results shown prove the



values are aligned to the theoreticalcalculations. The path taken is indeeda parabolic path, and equations (9)and (10) successfully describes themotion of an object in two-dimensions.

The authors went on to use theMATLAB software to design amultimedia software module toincorporate all the equations and inthe process create a user-friendlyenvironment for providing an extraalternative for the understanding ofthe concepts in this area. The tediouspart of programming is hidden, thusallowing for a comfortable interactionbetween the user and the computer.

Based on the flowchart shown inFigure 4, the software was designedand the main menu of the softwaredeveloped is shown in Figure 5. Theinstructional objectives of the modulewill include the ability to solve for theunknown quantities in motion in twodimension cases. The module has fourmain sections accessible through themain page with a menu consisting ofAnimations, Simulations, Workedexamples, and Practice Exercise. Thesimulation part contains thebackground theory, derivation ofequations and learning objectives.These pages will accommodate thefirst three instructional events stated.Clarification of the theory will beprovided using diagrams andequations while real life cases areintroduced to motivate and gainstudents’ attention.

The animations section as shown inthe special window in Figure 5, showsanimated diagrams tracing the path ofobjects in two-dimensional motions.It will show cases with various valuesof projection angles and initial andfinal values of vertical position, y. Themain objectives of this section are to

25

20

15

10

5

0M

axi

mu

m H

eig

ht,

m

Angle, θ in degrees

0 10 20 30 40 50 60 70 80 90

Figure 2At an initial velocity of 20 m/s, the graph above shows that as the angle ofprojection increases, the maximum height increases.

6

5

4

3

2

1

0

Ve

rtic

al D

isp

lac

em

en

t, m

Horizontal Displacement, m

0 5 10 15 20 25 30 4035

Figure 3Graph shows that the path taken by the projectile with an initial velocity of20 m/s, fired at an angle of 30° is a parabolic path.

guide learning, enhance retention andsustaining students’ interest. Theproceeding section includes workedexamples whereby calculations ofunknown quantities in trajectoryproblems are shown. Examples alsoinclude diagrams showing themotions to clarify them.

The final section consists of PracticeExercises to evaluate students’understanding and attainment of theconcepts. Multiple-choice questionsare provided for easy assessment.Immediate feedback will be providedinforming the students whether theanswers they have entered are corrector otherwise.



CONCLUSION

This software will be placed in thePhysics laboratory, where the studentswill be allowed to access this softwareby using the networked computersavailable in the lab. This will onlyhappen after the lecture sessions onmotion in two-dimensions is alreadyconducted and would be an additionalbenefit to students who requires extra

User suppliesinformation on v0

and angle θ

Equation tocalculatethe range

Simulation

Output shownas a multimedia

presentation

Another set of v0and angle θ

Stop

Equation to relatehorizontal & vertical

displacements

Equation tocalculate

maximum height

yes

no

Figure 4The Flowchart used in developing the Simulation module ofthe multimedia software

Figure 5The Main Menu of the Multimedia instructionsSoftware

Figure 6The Practice Exercises menu of the Multimediainstructions software, where the user just click onthe multiple choice buttons to submit thesuggested answer.

assistance in this area. The softwarewill allow the students to learn at theirown pace, promote understanding bydisplaying simulations of motion withvarious values of initial velocities andangle of projections and workedexamples. It also serves to stimulatetheir interest and motivation whileenhancing retention by animateddiagrams.

REFERENCES

[1] Tom Boyle. Design for MultimediaLearning. Prentice Hall 1997.

[2] James E Shuman. Multimedia in action.Intergrated Media Group. 1998.

[3] Walter Dick and Lou Carey. TheSystematic Design of Instruction.

[4] Michael D Williams. IntergratingTechnology into Teaching and Learning,Concept and Aplication.

[5] R Gagne, L Briggs & W Wage. Principlesof Instructional Design (4th Ed), FortWorth, TX. HBJ College Publisher 1992.

[6] Douglas C Giancoli. Physics ForScientists & Engineers. 3rd edition 2000.Prentice Hall International Editions.



1.0 INTRODUCTION

Bituminous pavements are subjectedto a variety of loading conditionswhich result in the development ofinternal tensile stresses. Bituminousmixes have low tensile strength andthis has been recognised as acontributor to their performanceproblems. One source of failure whichis likely to be induced in bituminousmixtures as a result of this is fatigueor cracking.

Investigations into the reinforcementof bituminous mixtures to improve itstensile properties have been attemptedin the past with different materials andwith varying degree of successes.Cotton fibres have been used but theseare degradable and was not suitableas a long term reinforcement. Metalwires have also been proposed bur theyare susceptible to rusting with thepenetration of water. Asbestos wasonce used until it was determined asa health hazard. Other effortsundertaken include the use of fabricsas a stress-absorbing interlayer(Woodside and Rogan, 1994) and theincorporation of homogeneouslydispersed short synthetic fibres(Freeman et. al, 1989) within thebituminous mix.

This paper describes and presents theresults of a laboratory investigation toassess the influence of polymer fibreson the fatigue behaviour ofbituminous mixtures using theIndirect Tensile Test.

Fatigue Behaviour of Fibre Reinforced Bituminous Mixtures

from the Indirect Tensile Test




2.0 MATERIALS USED IN

THE INVESTIGATION

2.0.1 Mineral Aggregates,

Filler and Bitumen

Limestone aggregates and OrdinaryPortland cement (OPC) filler and abinder of normal pen. Grade 50 wereused. Some relevant properties ofthese material are shown in Table 1.

2.0.2 Synthetic Fibres

Two types of synthetic fibres namelypolypropylene and polyester were usedin this study. The fibres were used asa partial replacement of the filler; onan equal volume basis; at twoconcentrations of 0.5% and 1% byweight of the mix. The chopped fibreswere the by-products of the textileindustry and thus their potential usewas desirable on environmentalgrounds.

Table 2:Characteristics of Fibres Used

Poly- Poly-

propylene ester

(PP) (POL)

–––––––––––––––––––––––––––––––––––––––Specific Gravity 0.91 1.41

–––––––––––––––––––––––––––––––––––––––Denier 6 3

–––––––––––––––––––––––––––––––––––––––Length (mm) 6 6

–––––––––––––––––––––––––––––––––––––––Average 22* 17*

Diameter (µm)

–––––––––––––––––––––––––––––––––––––––Degradation 160 250Temp. (°C)

–––––––––––––––––––––––––––––––––––––––* Values obtained from 20 readings

using a light microscope at 400Jxmagnification.

Some characteristics of the fibres usedare shown in Table 2. As thepolypropylene fibres had a lowdegradation temperature of around

Material Percentage Relative Absorption BS Specification

by Weight Density (%)

––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––CoarseAggregate 35 2.75 0.47 BS 594: Part 1:1992

––––––––––––––––––––––––––––––––––––––––––––––––––––––– Table 3, type FSand 55 2.65 1.37 wearing coarse

––––––––––––––––––––––––––––––––––––––––––––––––––––––– designation 30/14Filler (Ordinary 10 3.15Portlandcement)

––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

Penetration Softening Penetration

(0.1mm) Point (°C) Index (PI)

Bitumen 52 48.5 -0.37

Table 1: Properties of Mineral Aggregates, Filler and Bitumen Used in Study



160°C, it was decided that the mixingtemperatures when preparing theHot-Rolled Asphalt (HRA) will notexceed 140°C and compaction will bedone at 130°C, in order to maintainthermal stability of the mixes.

3.0 Fatigue Relationship

Before the fatigue performance of abituminous material can be assessed,the failure of the specimen tested mustbe defined in a consistent manner.Defining the failure criterion in theconstant stress mode is relatively easyas the specimens undergo a relativelyshort crack propagation period.Hence, the failure point is taken aswhen the specimen has completelyfailed. However in the constant strainmode of loading, the failure point isnot very well defined due to the largeamount of crack propagation includedin the test. An arbitrary point offailure must thus be assumed. This isnormally defined as the point whenthe specimen has reached a reductionin its initial stiffness of 50% or inpractical terms is the point when thestress applied has been halved toachieve a constant strain.

Having defined the point of failure forthe test, the interpretation of theresults can then be considered. It hasbeen found that a linear relationshipexists between the log of stress σ orstrain ε and the log number of loadrepetitions Nf to failure. The failurecriteria can therefore be expressedeither as:

Log.stress against log.load applications

Log.strain against log.load applications

This gives an equation in the generalform:Log (σ or ε) = a + b log Nf

For the strain-controlled tests, theresults are normally presented in thelatter manner. This generates an

equation of the form:

N f = A1ε

b

Equation (1)

For the stress controlled tests, theresults are normally presented usingthe log. stress against log. loadapplications. This generates anequation of the form:

N f = A1σ

d

Equation (2)

where:Nf – number of load

applications to failureε1 – tensile stress or strain

repeatedly appliedload

A, b, C, d – material coefficients

Pell (1963) demonstrated that thetensile strain is an importantparameter for fatigue cracking andhence the results from the controlledstress mode of loading are generallypresented in a form similar to that ofequation (1) but with differentmaterial coefficients as shown below:

N f = K 11ε1

K 2

Equation (3)

where:Nf – number of load

applications to failure ata particular level ofinitial strain

ε1 – initial tensile strainK1, K2 – material coefficients

4.0 Indirect Tensile

Fatigue Test (ITFT)

This test is an indirect tensile testwhere the specimen tested is subjectedto repeated loading. In this study, thetest was conducted in the UMATTA(Universal Material Testing Apparatusor in short MATTA) with theappropriate force, rise time and pulserepetition period being selected. Foreach loading pulse, the accumulated

displacement are continuously beingcalculated and displayed.

Read and Brown (1994) is of theopinion that the Indirect TensileFatigue test is able to characterise thefatigue life of a bituminous mixtureby testing a small number ofspecimens (less than 10) at hightemperatures (in excess of 25°C) andat high stress levels (greater than 450kPa). This means that the fatiguetesting time needed to produce anadequate fatigue relationship for abituminous material is significantlyshorter than the other traditionallaboratory fatigue testing methods.

A schematic diagram of the IndirectTensile Fatigue test is given in Figure1. The test involves a repeated line ofloading (generally controlled stress)applied along the vertical diameter ofa cylindrical specimen. This verticalload produces both a verticalcompressive stress and a horizontaltensile stress on the diameters of thetested specimen. The magnitude ofthe stresses changes along the diameterof the specimen with the maximumoccurring in the centre of thespecimen. The maximum strain at thecentre of the specimen can then becalculated. The assumptions used inthis test are that the:

• Specimen is subjected to planestress conditions

• Material is homogeneous andbehaves in an isotropic and linearelastic manner

• Poisson’s ratio (υ) for the material

is known• Force (P) is applied as a line

loading

With these assumptions, the stressconditions in the specimen are thengiven by:

σ x max= 2P

πdt Equation (4)



and

σ ymax= – 6P

πdt Equation (5)

where:σ xmax – maximum horizontal

tensile stress at the centreof the specimen

σ ymax – maximum verticalcompressive stress at thecentre of the specimen

d – diameter of speciment – thickness of specimen

By simple linear elastic stress analysis(Hooke’s law)

ε x max=

σ x max

Sm

=υσ y max

Sm

Equation (6)where:

ε x max – maximum initial

horizontal tensile strain atcentre of specimen

υ – Poisson’s ratioSm – stiffness modulus of

specimen

Substituting equation (4) and (5) into(6) gives:

ε x max= 2P

πdtSm

+ υ6PπdtSm

Equation (7)

Substituting equation (4) into (7)gives:

ε x max=

σ x max

Sm

1 + 3υ( )Equation (8)

Equation (8) was used for thecalculation of the maximum tensile

strain ( ε x max ) at the centre of the

specimen. It is evident that this isdependent on the stiffness modulusof the material. This parameter wasobtained from the Indirect TensileStiffness Modulus test at the samestress level and test temperature as theITFT.

The specimen geometry selected forall the ITFT was 100 mm in diameterby 40 mm thick. The specimens werefirst conditioned at the testtemperature of 20°C ± 0.5°C beforetesting commenced. A cyclic loadpulse was applied to the specimenwith the time to peak of the load pulsebeing 120 ms. A loading rate of 40 ±1 pulse/minute at the test stress levelwas employed and the permanentvertical deformation was measured bylinear variable differential transducers(LVDT).

In addition to the measurement ofpermanent vertical deformation, theLVDT also measured the transientvertical deflection. The failure pointof the test is defined as 9 mm ofpermanent vertical deformation, thusensuring that all the materials testedhave reached the point of crackinitiation (N1) under the controlledstress conditions. The point of crackinitiation (N1) is defined to be themaximum point on a plot of thenumber of cycles (N) divided by thetransient vertical deflection (∆p)against the number of cycles (N) asshown in Figure 2. The figure relatethe strain-number of cycles to failurethat was typical of the output obtainedfrom the MATTA during the test.The figure shows an initial period ofcomparatively large displacementfollowed by a part that represents aconstant rate of strain amplitude.Finally the curve started to concaveindicating the point of failure.Nikolaides (1997) defined the pointof failure in the ITFT as the pointwhere the straight line obtainedbetween the number of loading cyclesto transient deformation started toconcave. In this test however, thepoint of failure was taken as the pointon the curve that indicated that thetest specimen had ruptured.

4.0.1 Target Test Stress Level

BS DD ABF/95 proposed that thetarget stress level for the testing of thefirst specimen in the Indirect TensileTest be 600 kPa. Unless this cannotbe reliably obtained, a stress level of500 kPa was proposed. Table 1presents the proposed test stress levelfor the subsequent test specimens. Inthis test the target test stress levels wereselected so as to give as wide a rangeof lives as possible. The minimumspread of lives must be one order ofmagnitude so that the maximum valueof N600 or N500 is at least ten timesgreater than the minimum value.

5.0 DISCUSSION OF RESULTS

The fatigue characteristics of thecontrol and the fibre incorporatedmixes were tested in the IndirectTensile Fatigue test at their optimumbitumen content. The number ofspecimens tested for each mix type inorder to determine its fatiguecharacteristics ranges from 12-15specimens. BS DD ABF/95 requirethat a linear regression analysis usingthe Least Squares method beemployed to the paired data oflog10( ε x max

) and log10(Nf) and thedetermination of its correlationcoefficient (R2). It recommends thatif the value of R2 is below 0.90,additional specimens should be testedand incorporated into the results ofthe data set until on repeating thelinear regression analysis, the value ofR2 is greater than 0.90.

Fatigue tests on each mix type wasconducted at the optimum bitumencontent. The fatigue characteristicsrelating the strains (microstrains) withthe number of cycles to failure is givenin Figure 2. Table 2 presents asummary of the fatigue characteristicsof the mixes that were tested at the



optimum bitumen content. For thecontrol mix, the life at 100microstrains was 172,232 loadapplications and the strain at 106cycles was 52 microstrains. This isconsiderably poorer than the fibrereinforced mixes where the life at 100microstrains varies from 700,026 loadapplications (0.5PP) to 2,589,372load applications (1POL) and thestrains at 106 cycles varies from 93microstrains (0.5PP) to 110microstrains (1POL). The resultsfrom the ITFT also show that the 1%fibre mixes have a better fatiguebehaviour than those of the 0.5% fibremixes. The higher bitumen contentof the 1% fibre mixes could possiblyhave been responsible for the superiorfatigue characteristics of these mixes.

The polyester (POL) mixes alsoexhibited superior fatigue propertiesas compared to that of thepolypropylene (PP) mixes. This maypossibly be due to the higher bitumencontent in the polyester mixes ascompared to the polypropylene mixes.In addition, the higher viscosity of thepolyester incorporated binder resultedin harder bitumen which wasresponsible in a higher stiffness for thepolyester fibre mix. This increase instiffness may have resulted in greaterfatigue life of the polyester mixes ascompared to those of thepolypropylene mixes. The fatigueequation together with the coefficientof regression obtained from theIndirect Tensile Test for the mixes attheir optimum bitumen content andthe corresponding material constantsfor the mixes are shown in Table 3.

6.0 COMPARISON OF THE

INDIRECT TENSILE FATIGUE

TEST (ITFT) AND THE BEAM

FLEXURAL FATIGUE TEST

Results from the ITFT was comparedwith the conventional beam flexural

fatigue test in determining the fatiguebehaviour of the mixes studied toallow a comparison be made withrespect to the fatigue behaviour of themixes using different modes of testing.These are presented in graphical formin Figures

The ITFT data set appear to have thesame slope as that obtained from thebeam flexure fatigue test for all themixes tested. Data sets for the controlmix and 0.5% fibre concentrationmixes appear to also fall on the samefatigue line. The results show goodagreement between the two sets of testmethods with an overall correlationcoefficient of 0.974 for the control,0.947 for the 0.5% POL mix and0.976 for the 0.5PP mix. However,the data sets from the beam fatiguetest in the 1% fibre mixes exhibitedhigher fatigue behaviour than thatobtained from the ITFT and arelatively lower correlation coefficientof 0.901 for the 1PP mix and 0.821for the 1POL mix. Comparison ofthese results seems to indicate theversatility of the ITFT that was carriedout in the MATTA. The test isrelatively quick with excellent controlof temperature during the test. As thetest correlates well with the moreconventional beam fatigue test whichwould normally require longer timeto run the test, it therefore meant thatthe fatigue behaviour of bituminousmaterials can be rapidly determinedby the ITFT.

7.0 CONCLUSION

Based on a fatigue test conducted, thefollowing conclusions can be drawn:

1. It appears that the incorporationof synthetic fibres in thebituminous mixtures have thepotential of improving the fatigueperformance of the mix. Fatiguetesting confirmed the high strain

capacity of the fibre-modifiedmixes owing to their higherbitumen content and the thickerbitumen film coating theaggregates.

2. The air void content of the fibrespecimens were however greaterthan that of the control. This issignificant in that the fatigueperformance would usually sufferwhen the void content isincreased.

3. The test results indicate that thefibre mixtures provide about thesame fatigue performance as thecontrol at low strain levels, but athigh strain levels, the fibre mixesprovided superior fatigueperformance indicative that thefibre mixtures appear to be mostbeneficial at high strain levels.

4. Comparisons of the fatiguerelationship obtained from theIndirect Tensile Fatigue test andthat of the Beam Flexural testrevealed that the fatigue lines forboth the tests do results in a singleline defining the fatiguebehaviour of the mix tested.

REFERENCES

[1] N A Bjorklund. “PermanentDeformation and Resistance to Fatigueof Resurfaced Pavements – A LaboratoryInvestigation Performed on Beams TakenAcross the Wheel Path Resurfaced in theLaboratory”, Proceedings of theAssociation of Asphalt PavingTechnologists, Vol. 54, 1985, pp. 551-586.

[2] British Standard Institution, Draft forStandard Development forDetermination of the Indirect TensileStiffness Modulus of BituminousMixtures DD213, 1993.

[3] H W Bushing. and J D Antrim. “FiberReinforcement of Bituminous Mixtures”,Proceedings of the Association of AsphaltPaving Technologists, Vol. 37, 1968, pp.629-659.



[4] J W Button and G H Hunter. “SyntheticFibres in Asphalt Paving Mixtures”,Report No. FHWA/TX-85/73, TexasTransportation Institute, November1984.

[5] R D Freeman, J L Burati, S NAmirkhanian and W C Bridges, W.C,“Polyester Fibres in Asphalt PavingMixtures”, Proceedings of the Associationof Asphalt Paving Technologists, Vol. 58,1989, pp. 388-409.

[6] R T N Goddard, W D Power and M WApplegate “Fatigue Resistance of DenseBituminous Macadam, the Effect ofMixture Variables and Temperatures”,TRRL Report SR 410, Department ofEnvironment, UK, 1978.

[7] I Kamaruddin, “The Properties andPerformance of Polymer Fibre ReinforcedHot-Rolled Asphalt”, Unpublished PhDthesis, University of Leeds, 1998.

[8] J M Read and S F Brown, “FatigueCharacterisation of Bituminous MixesUsing a Simplified Test Method”,Proceedings of Symposium onPerformance and Durability ofBituminous Materials, University ofLeeds, March 1994, pp.158-172.

[9] A F Nikolaides, “Effect of BinderContent Variation on the Stiffness andFatigue of Asphaltic Concrete”,Proceedings of the Second EuropeanSymposium on Performance andDurability of Bituminous Materials,Leeds, April 1997, pp. 227-240.

[10] A R Woodside and C Rogan, “The Roleof Fabrics in Upgrading the Durabilityof Bituminous Treatments”, Proceedingsof Symposium on Performance andDurability of Bituminous Materials,University of Leeds, March 1994, pp.113-122.



INTRODUCTION

Glass-ceramics based on phosphateglasses have been of interest forbiomedical applications where highstrength, biocompatibility andmachineability are required. Theseapplications include dental implants,skin treatment, gum treatment, jaw-bone reconstruction, as well as

orthopaedic, facial, ear, nose, throatand spinal surgeries [1]. Differentcrystalline phases of calciumphosphate have been designed andutilised for various uses depending onthe properties to be maximised. Forinstance, the hydroxyapatite(Ca10(PO4)6(OH)2) phase has beencrystallised out from a parent calciumphosphate glass to produce a bioactive

implant material whilst the tricalciumphosphate (Ca3(PO4)2) phase iscrystallised out for use as a resorbableimplant material [2]. This flexibilityand control of the crystalline phase (ora combination of phases) can beachieved by proper selection of theCaO/P2O5 ratio in the starting glassmaterial. Consequently, glasses withdifferent CaO/P2O5 ratios were

Contamination Of

Phosphate Glasses Upon Melting

Jariah Mohamad Juoi



Radzali Othman

Universiti Sains Malaysia, 14300 Nibong Tebal, Penang, Malaysia

ABSTRACT

Calcium phosphate glasses with CaO/P2O5 molar ratios of 0.85, 0.95, 1.10 and 1.20 were melted at 1200oC in bothalumina and platinum crucibles. It was observed that the product of melting in all cases is a crystal clear and pore-freeglass. Analyses by x-ray diffraction (XRD) on all glass samples confirmed that all the glasses are amorphous. However, x-ray fluorescence (XRF) analyses revealed that 0.4 - 1.5 weight percent of alumina (Al2O3) was present in the set of glassesthat had been produced by melting in alumina crucibles, whilst no traces of Al2O3 was detected in those set of glassesmelted in platinum crucibles. This clearly indicates that molten calcium phosphate glasses has a corrosive effect onalumina crucibles which leads to contamination of the melt composition. Eventhough the contamination in the melt isrelatively small, it is significant enough in the subsequent heat-treatment process to produce calcium phosphate glass-ceramics by controlled crystallisation.

Kaca kalsium fosfat dengan empat nisbah mol CaO/P2O5 yang berbeza, iaitu 0.85, 0.95, 1.10 dan 1.20, telah dileburpada suhu 1200oC menggunakan mangkuk pijar alumina (Al2O3) dan platinum. Pada penglihatan mata kasar, hasildaripada kesemua proses peleburan adalah suatu kaca pejal yang jernih dan bebas daripada gelembung udara. Ujianpembelauan sinar-x (XRD) ke atas kedua-dua set kaca menghasilkan corak pembelauan yang mengesahkan kaca-kacatersebut adalah amorfus. Namun, ujian pendarfluor sinar-x (XRF) membuktikan kehadiran sebanyak 0.4 - 1.5 peratusberat Al2O3 dalam kaca yang dilebur menggunakan mangkuk pijar alumina. Sebaliknya, tiada sebarang Al2O3 dikesandalam set kaca yang dilebur menggunakan mangkuk pijar platinum. Ini jelas menunjukkan bahawa pencemaran telahberlaku bila peleburan dilakukan dalam mangkuk pijar alumina akibat daripada sifat mengkakis leburan kalsium fosfatke atas alumina. Kehadiran bahan cemar ini, walaupun dalam peratusan yang amat kecil, adalah signifikan dalam prosesolahan haba seterusnya untuk menghasilkan bahan seramik kaca kalsium fosfat menerusi proses penghabluran terkawal.

Keywords:

glass, glass-ceramics, calcium phosphate, alumina, platinum, contamination



melted in this work in order to studythe effect of this ratio on the crystallinephase(s) that can be produced bysubsequent heat-treatment of the glassproduced. The presence of otheroxides in small quantities (even lessthan 1% by weight) can affect thecrystallisation process upon heat-treatment by acting, for instance, asnucleating agents [3]. Hence, thenecessity to produce a good qualityglass (in terms of purity) cannot beoveremphasized. Contamination ofthe glass can be narrowed down to twopossible sources, viz. the starting rawmaterials and the reactions duringmelting. Contamination from rawmaterials can be eliminated orminimised by a judicious choice ofreagent grade chemicals. This leavesthe possibility of reactions duringmelting especially possible corrosivereactions between the melt and thecrucibles used. In this work themelting of calcium phosphate glass offour different CaO/P2O5 ratios wascarried out using both alumina andplatinum crucibles. An aluminacrucible is a common choice for glassmelting since it is normally inert andthe cost is lower [4], whilst platinumware is much more expensive. Theglass melts produced using both typesof crucibles were characterised usingx-ray diffraction (XRD) and x-rayfluorescence (XRF).

EXPERIMENTAL

Four batches of raw materials wereprepared using different amounts ofcalcium bis-dihydrogen phosphatemonohydrate (Ca(H2PO4)2.H2O)and phosphoric acid (H3PO4) toachieve a CaO/P2O5 molar ratio of0.85, 0.95, 1.10 and 1.20. Theprepared batch mixture was thenstirred for five minutes to ensurehomogeneity of mixing. It was thenmelted in a calcined alumina crucibleat 1200oC for a one-hour soakingduration. The procedure was repeated

for melting in a platinum (5% AuZGS Pt) crucible. When the glass wascompletely molten, it was cast onto astainless steel plate in air. After coolingto room temperature, the casts wereexamined with the naked eye. It wasthen powdered using an agate mortarfor XRD and XRF analyses. XRDanalysis was performed using a PhilipsXRD machine equipped with a PW1729 generator, PW 1820diffractometer and a PW 1710controller. The CuKα (1.54 _)radiation was used during the analysis.XRF analysis was performed on aRigaku RIX3000 equipment.

RESULTS AND DISCUSSION

The cast produced from each of themelt was found to be crystal-clear andpore-free. This indicates that a goodglass had been produced from thebatching, melting and castingprocedures adopted in this work.Generally, a glass system such ascalcium phosphate consists of twomain oxide components, viz. calciumoxide (CaO) and phosphoruspentoxide (P2O5). This binary systemcan easily form a glass due to the glass-forming ability of P2O5 [5]. The roleof P2O5 is similar to the role of silica(SiO2) in silicate glasses. Establishedx-ray diffraction patterns forcrystalline and amorphous solids, aswell as those for liquids andmonoatomic gases are shown in Figure1 [6].

The XRD results for each set of glassesproduced in alumina and platinumcrucibles are shown in Figure 2. Bycomparing the results in Figure 2 tothe theoretical predictions of Figure1, it is confirmed that the glassesproduced in this work are amorphous.This is characterised by the x-raydiffraction pattern that shows a broadmaximum in the range of 20o to 40o

value of 2θ. This pattern representsthe diffraction of a material with no

perfect periodic arrangement. Hence,it is confirmed that the crystal-clearcasts produced are good calciumphosphate glasses.

Results from XRF analysis are shownin Table 1. Apparently, there issurprising major difference betweenthe chemical composition of the castsproduced by melting in aluminacrucibles as compared to the castsproduced in platinum crucibles. Theglasses melted in alumina cruciblesindicate the presence of Al2O3 in therange of 0.4 - 1.5 weight percent,whilst there is no trace of Al2O3 inthe glasses melted in platinumcrucibles. Since the raw materialsbatch had been judiciously selected soas to be free from significantimpurities, Al2O3 is thus believed tobe the result of contamination duringthe melting process itself.

180900

a)

180900

b)

180900

c)

Figure 1:The XRD curve for a) crystallinesolid b) amorphous solid c)liquidand monoatomic gas.



Calcium phosphate glass with XRD patterns for glass melt XRD patterns for glass meltdifferent CaO/P2O5 molar ratio in Alumina crucible in Platinum crucible

0.85

0.95

1.10

1.20

Figure 2:

XRD patterns of calcium phosphate glass produced in alumina and platinum crucibles



The difference between the two setsof glasses is in the type of cruciblesbeing used. Thus, it can be reasonablyargued that the trace of Al2O3 mustbe the result of corrosive action of thephosphate glass melt on the internalsurface of the alumina crucible. Thisis in complete agreement withfindings from previous works on thecorrosive action of phosphoric acid onalumina [7]. The contamination isfound to be worst in the melt with aCaO/P2O5 molar ratio of 0.85, i.e.there is a 1.5 weight % of Al2O3 inthe glass. This can be attributed tothe higher amount of H3PO4 used inthe batch mixture which renders themelt to be more corrosive. Inaddition, observations made on theinternal surface of the aluminacrucibles corroborate the statementregarding the corrosive nature ofphosphoric acid on alumina. Thecrucible used for melting the glasswith a 0.85 CaO/P2O5 molar ratioexhibits the most corroded internalsurface compared to crucibles formelting glasses of other CaO/P2O5

ratios.

Calcium phosphate glasses producedby melting in a platinum crucible donot indicate any contamination. Thiscan be attributed to the inert natureof platinum in general. Perhaps muchmore significant is the fact that the

platinum crucible used in this work isof the 5% Au ZGS type. This type ofplatinum crucible has been sodesigned to increase the corrosionresistance especially at hightemperatures [8].

The presence of Al2O3 even in traceamounts, is significant since it canaffect the crystallisation process duringheat-treatment of phosphate glass toproduce phosphate glass-ceramic [9].This in turn will affect the bioactivityand mechanical properties of calciumphosphate glass-ceramic [10].

CONCLUSIONS

In conclusion, this work has revealedthat whilst an apparently goodphosphate glass can be produced bymelting in either alumina or platinumcrucibles, the subtleties of tracecontamination in the glass meltcomposition itself can be of utmostimportance when the glass is intendedfor heat-treatment to produce a glass-ceramic material. Therefore, the usageof platinum crucible is the mostsuitable in the work done in thisresearch. In addition, a judiciouschoice of analytical techniques needsto be made to ensure that the researchfindings can lead to a sound scientificinterpretation.

ACKNOWLEDGEMENTS

The authors would like to thank The NipponSheet Glass Foundation of Japan for thefinancial support (awarded to Professor RadzaliOthman) in carrying out this work.

REFERENCES

[1] D Day. Using glass in the body, TheAmer. Ceram. Soc. Bull., 74(7-12), 64-68.1995.

[2] L L Hench. Bioceramics – a clinicalsuccesss, The Amer. Ceram. Soc. Bull., 77,67-72. 1998.

[3] P W McMillan. Glass-ceramics,Academic Press, London, 1-36. 1964.

[4] P F James. Glasses and glass-ceramics,Chapman and Hall, 59-61. 1989.

[5] Z Strnad. Glass-ceramic materials,Elsevier, Amsterdam. 1986.

[6] U Hoppe. A structural model forphosphate glasses, J. Non-crystallineSolids, 195, 38-47. 1996.

[7] H P Robert. Chemical Engineers’Handbook, McGraw Hill Inc., Section 3.

[8] Johnson Matthey Catalogue on PlatinumWares. 1984.

[9] I M Reaney, P F James and W E Lee.Effect of nucleating agents on thecrystallisation of calcium phosphateglasses, J. Amer. Ceram. Soc., 79(7), 1934-1944. 1996.

[10] P F James. Glass ceramics: newcompositions and uses, J. of NonCrystalline Solids, 181, 1-15. 1995.

CaO/P2O5 molar ratio 0.85 0.95 1.10 1.20

Type of crucible Al2O3 Pt Al2O3 Pt Al2O3 Pt Al2O3 Pt

CaO weight % 25.0 22.0 31.0 23.0 34.0 27.0 36.0 30.0

P2O5 weight % 73.5 78.0 68.7 77.0 65.5 73.0 63.5 70.0

Al2O3 weight % 1.5 0.0 0.4 0.0 0.5 0.0 0.6 0.0

Table 1:Chemical composition of calcium phosphate glass with various CaO/P

2O

5 molar ratios produced in alumina and

platinum crucibles.



1. INTRODUCTION

Polymer electrolyte membrane (PEM)fuel cells are being considered as oneof the most promising power sourcesfor electric vehicle and on-siteapplications. This is due to their highpower density, efficiency and lowemissions (Strasser,1992; Shoesmith etal. 1994). The solid electrolyterepresented by the proton exchangemembrane is the vital component inthe PEM fuel cells where it conductsprotons from the anode to cathodeand prevents the bulk mixing of H2

and O2 (Scherer, 1990).

Nafion® membranes are the mostwidely proton exchange membranestested for PEM fuel cell applications[4]. However, these membranes areexcessively expensive (800 US$/m2)and as a result, their large-scale

application was limited [Hietala et al.,1998]. Hence, new proton exchangemembranes having a combination ofhigh conductivity, stability and lowcost are needed to enhance theemergence of commerciallycompetitive PEM fuel cells.

Several new experimental andcommercial membranes such aspolytrifluorostyrene-based BAMmembranes (Ballard AdvancedMaterials), PEEK-based Aventismembranes (Hoechst) and Dais 585membranes (DAIS, Co.) have beendeveloped recently using chemicalpolymerization techniques and testedin PEM fuel cell (Savadogo, 1998).Among the alternative membranes,radiation grafted and sulfonated onesare advantageous in terms of theability to control their compositionsand properties by variation of grafting

conditions (Chapiro, 1962). Thepossibility of using commercialradiation grafted sulfonic acidmembranes as polymer electrolytes inPEM fuel cells was explored byGuzman-Garcia et al. (1992). Thesemembranes have shown an initialstability up to 1000 h in PEM fuelcell (Wang and Capuano, 1998).Experimental radiation graftedsulfonic acid membranes based oncommercial poly(tetrafluoroethylene-co-hexafluropropylene) (FEP) filmswere developed by Scherer and co-workers and found to be stable for1400 h in PEM fuel cell attemperatures up to 80 oC (Rouily etal., 1993; Gupta et al. 1994, 1996;Büchi et al., 1995). Similar sulfonicacid membranes based on FEP,polytetrafluoroethylene (PTFE) andpolyvinylidene fluoride (PVDF),poly(ethylene-co-tetrafluoroethylene)

New Radiation Grafted and Sulfonated

Membranes for PEM Fuel Cell

Mohamed Mahmoud Nasef



Hamdani Saidi & Hussin Mohd Nor

Membrane Research Unit, Faculty of Chemical and Natural Resources Engineering,

Universiti Teknologi Malaysia, Jalan Semarak, 54100 Kuala Lumpur, Malaysia

E-mail: [email protected]

ABSTRACT

New PFA-g-polystyrene sulfonic acid membranes were prepared by radiation-induced graft copolymerization of styreneonto poly(tetrafluoroethylene-co-hexafluoropropylene), (PFA) film followed by sulfonation reaction. The properties ofthese membranes were found to be controlled by variation of the amount of polystyrene grafted therein when it issubsequently sulfonated i.e the degree of grafting. The membranes were found to have a good combination of wateruptake and ion exchange capacity. Moreover, their proton conductivities are in the same order of magnitude (10-2 Ω-1cm-1)of Nafion 117 membranes. The preliminary results of the performance of these membranes in PEM fuel cell were foundto be promising. However further testing has to be carried out to encourage the use of these membranes for the futuredevelopment of alternative low cost proton exchange membranes.

This paper was presented at the Regional Symposium on Chemical Engineering, Singapore, 11-13 December, 2000.



(ETFE) films were prepared in variousoccasions (Nasef et al. 2000 a,b,c,d;Holmberg et al., 1996; Brack et al.,1997)

Po l y ( t e t r a f l uo roe thy l ene - co -hexafluoropropylene) (PFA) is one ofthe commercial fluorinated polymers,which are well-known for their highthermal, mechanical and chemicalresistance. The use of PFA films as apolymer backbone for preparation ofnew sulfonic acid membrane byradiation induced graftcopolymerization has beeninvestigated (Nasef et al., 1999, 2000e). The obtained membranes werecharacterized by measuring theirphysico-chemical properties and theirmorphological as well as thermalproperties were also studied (Nasef etal. 2000 f, g).

In this paper some of the importantproperties of these membranes areevaluated and correlated with thedegree of grafting. The results of theshort-term performance of themembrane in PEM fuel cell are alsoinvestigated. Nafion 117 membraneis used a reference.

2. EXPERIMENTAL

2.1 Membrane Preparation

The membranes were prepared by atwo-step preparation procedure. Inthe first step, styrene (> 99 %, Fluka)was grafted onto PFA film (Porghof,USA) having a thickness of 120 µmusing simultaneous irradiationtechnique at room temperature. PFAfilm (5 cm x 5 cm) was washed withacetone, dried at 60oC in a vacuumoven, weighed and then immersed instyrene diluted with methylenechloride in a glass ampoule. Theampoule was flushed with nitrogen(99.9 % purity) for 10 minutes andthen sealed. The ampoule wasirradiated at a dose rate of 1.32 kGy/

h using g-rays from a Co-60 sourcelocated at Malaysian Institute forNuclear Technology (MINT).Kinetics of the grafting reactions andproperties of the grafted PFA filmswere reported elsewhere (Nasef et al.,1999). The degree of grafting wasgravimeterically determined using thefollowing equation:

Degree of grafting (%)

=Wg −W0

W0

×100

where, Wg and W0 are the weights ofthe grafted and the original PFA films,respectively.

In the second step, The grafted filmwas sulfonated with a mixturecomposed of chlorosulfonic acid(Fluka) and 1,1,2,2-tetrachloroethane(Fluka) (30:70 v/v) for 5 h at atemperature of 90 oC. The sulfonatedmembrane was hydrolyzed with 1 MKOH solution and regenerated with3.5 M HCl, then washed acid freeusing deionized water several times.More details on sulfonation reactioncan be found in Nasef et al. (2000 e).

2.2 Membrane Properties

2.2.1 Ion Exchange Capacity

The membrane samples in acid formwere equilibrated in 0.5M KClsolution for 15 h. The amount ofprotons (H+) released in the solutionwere titrated with standerdized 0.05MKOH solution by automatic titrator(Metrohom, Switzerland) until pH 7was reached. Form the volume ofKOH solution consumed in titrationand volume changes in the swollenmembranes, ion exchange capacity perunit volume (meq/cm3) wascalculated.

2.2.2 Swelling Behavior

Vacuum dried membrane samples

were boiled in deionized water (18MΩcm) until swelling equilibriumwas reached. The surfaces of themembrane samples were quicklywiped with an absorbent paper toremove the excess of water adheringto the surfaces and the samples werethen weighed. The membrane wateruptake was calculated using thefollowing equation:

Water uptake (vol%)

= Vw −Vd

Vd

×100

where, Vw and Vd are the weights ofwet and dried films, respectively.

2.2.3 Proton Conductivity

Proton conductivity of themembranes was measured at roomtemperature using AC impedancespectroscopy. Measurements werecarried out using a frequency responseanalyzer (Solartron, 1250) incombination with an electrochemicalinterface (EG&G Princeton AppliedResearch) at 0.01-100 kHz frequencyrange as described in Nasef et al(2000e).

The Nafion 117 (Du Pont, USA)membrane samples were initialized bysoaking in 20 % HCl at 80 oC for 4hours followed by washing severaltimes with excess of deionized waterbefore use.

2.2.4 Testing the Membrane

Performance

The membrane performance wastested in a single fuel cell test fixture(Globe Tech., Inc.) having ageometrical area of 5 cm2. The fuelcell was heated by two 240 W heatingcartridges inserted in the end of thecopper plates. A temperaturecontroller (Shinko, Japan) attached tothe thermocouple (type K) was usedto control the output of the cartridges.



The flow of gases was typically in arange of 5-10 cm3/min and wascontrolled by mass flow meters (ColePalmer). Gas-diffusion referenceelectrodes having high platinumloading of 1.6 g/ cm2 (E-Tech, Inc.)were used with the membranes toformulate the membrane/electrodeassembly (MEA). The dry membranesample in acid form was pressedbetween the two electrodes. TheMEA was then hosted in the fuel cellin which the pressure at each backsidewas maintained at 1 bar. The fuel cellwas operated using pure hydrogen andoxygen gases (MOX, Malaysia) atatmospheric gas pressure and at atemperature of 50 oC. Gases werehumidified at temperature of 15-20 oC higher than the cell temperature(70 oC for H2 and 65 oC for O2)before they were supplied to the cell.

3 RESULTS AND

DISCUSSION

3.1 Membrane Preparation

Grafting of styrene onto PFA films bysimultaneous irradiation techniqueresulted in grafted films having variousdegrees of grafting (6-49 %). Figure1 shows the relationship between thedegree of grafting and the monomerconcentration. The degree of graftingwas found to increase with the increase

in the monomer concentration withinthe range of the applied monomerconcentration (20-50 vol %). Thisbehavior was attributed to the increasein the styrene diffusion and itsconcentration in the grafting zone ofthe PFA film and consequently, thegrafting increased.

The grafted films were sulfonated ina subsequent step and the degree ofsulfonation in all of the resultingmembranes were found to be close to100 % [21]. Therefore, the propertiesof the membranes are only discussedin correlation with the degree ofgrafting. The membrane propertiessuch as the ion exchange capacity andthe water uptake was expressed interms of volume in order to have a faircomparison with Nafion 117membrane, which has a differentequivalent weight.

3.2 Ion Exchange Capacity

Figure 2 shows the relationshipbetween the ion exchange capacity(IEC) and the degree of grafting. Ascan be seen the IEC depends stronglyon the degree of grafting of themembranes. The IEC of themembranes per unit volume (meq/cm3) increases with the increase in thedegree of grafting. This can bereasonably attributed to the increase

in the number of the polystyrene sidechain grafts formed in the graftedfilms, which led to incorporation ofmore sulfonic acid groups.

3.3 Swelling Behavior

The relationship between the wateruptake and the degree of grafting ofthe membranes is shown in Figure 3.The water uptake in terms of volumeis found to increase linearly with theincrease in the degree of grafting. Thisis due to the increase in thehydrophilicity imparted to themembranes by the incorporation ofmore hydrophilic sulfonic acid groupswith the increase of the degree ofgrafting.

3.4 Proton Conductivity

Figure 4 shows the relationshipbetween the proton conductivity andthe degree of grafting of themembranes at room temperature.The proton conductivity increaseswith the increase in the degree ofgrafting. This can be understoodbased on the fact that ionicconductivity is a function of ionexchange capacity and water uptakeof the membranes which were foundto be strongly dependant upon thedegree of grafting. However, theincrease in proton conductivity is

Figure 1The relationship between thedegree of grafting and themonomer concentration forgrafting of sytrene onto PFA films.

Figure 2The relationship between the ionexchange capacity and the

degree of grafting of PFA-g-polystyrene sulfonic acidmembranes.

Figure 3The relationship between the wateruptake and the degree of grafting

of PFA-g-polystyrene sulfonic acid.

60

50

40

30

20

10

00 605040302010

De

gre

e o

f g

raft

ing

(%

)

Monomer concentration (vol %)

4.00

3.50

3.00

2.50

2.00

1.50

1.00

0.50

0.000 605040302010

Ion

exc

ha

ng

e c

ap

ac

ity (

me

q/c

m3)


80

70

60

50

40

30

20

10

00 605040302010

Wa

ter

up

take

(vo

l%)




found be drastic at the beginning andtends to level off at a degree of graftingof 12 %. Such behaviour can beexplained by taking the distributionof the sulfonated polystyrene graftsinto consideration. At low degrees ofgrafting, the polystyrene grafts aredistributed only near to the surface ofthe membranes while its middleremains ungrafted and subsequentlyexerts high local resistance to protontransport. As the degree of graftingincrease up to a value of 12% moregrafts tend to be formed near themiddle part of the membranes,resulting in a sharp decrease in thelocal resistance and the protonconductivity greatly increases. Furtherincrease in the degree of grafting doesnot bring considerable changes to theproton conductivity due to thepossible achievement of homogenousdistribution of the sulfonatedpolystyrene grafts in the membranes.

A comparison between some of theimportant properties of PFA-g-polystyrene sulfonic acid membranescompared to that of Nafion 117membranes is shown in Table 1. Thethree membranes show a goodcombination of high IEC and lowwater uptake compared to that ofNafion 117. Moreover, they haveacceptable thicknesses and theirproton conductivities are in the sameorder of magnitude (10-2 W-1 cm-1)of Nafion 117 membranes.

3.5 Short-term Membrane

Performance

Figure 5 shows the polarisation curvesof PEM fuel cell with PFA-g-polystyrene sulfonic acid membraneshaving degrees of grafting of 16 and26%, respectively. Nafion 117membrane is used as a reference withthe same electrodes at the sameoperating conditions. Theperformance of the cell with the twoPFA-g-polystyrene sulfonic acidmembranes was found to be lowerthan that of Nafion 117. The cellshowed open circuit voltages of 840and 850 mV with these membranescompared to 950 mV for Nafion 117membrane. Moreover, the currentdensity measured at cell voltage of0.500 mV is found to be 20 and 40%lower than that of Nafion 117 for 16and 26% grafted membranes,respectively as depicted from Table 2.This finding can be mainly attributedto the differences in the protonconductivity between Nafion 117 andthe membranes prepared in this study.It should be emphasized that theobjective of the present study was totest the membrane electrolyticcapability in the fuel cell rather tooptimize the cell performance. Inaddition, the performance of the PEMfuel cell with Nafion 117 membranesin this study was found to be lower

Figure 4The relationship between theproton conductivity and thedegree of grafting of PFA-g-polystyrene sulfonic acidmembranes at room temperature.

Figure 5Polarization characteristic curves ofPEM fuel cell with PFA-g-polystyrenesulfonic acid and Nafion 117membrances. Cell operatingconditions: Tcell = 50°C; pressure,atmospheric; Thydrogen = 80°C;Toxygen = 65°C.

Membrane type Degree of grafting Water uptake IEC Proton conductivity Thickness

(%) (vol %) (meq/cm3) (ΩΩΩΩΩ-1cm-1) x 10-2 (µµµµµm)

PFA-g-PSSA 16 27 2.4 3.3 160

PFA-g-PSSA 26 48 2.9 3.8 170

PFA-g-PSSA 38 60 3.2 4.3 180

Nafion 117 (Du Pont) _ 39 1. 9 5.6 180

PSSA = Polystyrene sulfonic acid

Table 1.Comparison between the important properties of PFA-g-polystyrene sulfonic acid membranes and that of Nafion117 membrane.

0.045

0.035

0.025

0.015

0.005

0 605040302010

Pro

ton

co

nd

uc

tivity (

Ω–1c

m–1)

Degree of grafting (wt %)

0.003

0.002

0.001

0

1000

900

800

700

600

500

400

300

200

100

0

0 14010080604020

Vo

lta

ge

(m

V)

Current density (mA/cm2)

120



than that reported in literature despitethe use of high platinum loadingelectrodes (Gupta et al., 1993; Büchiet al., 1995). Such difference issuggested to be due to the highinterfacial resistance i.e. ohmic lossresulted from the poor contact in themembrane/electrode assembly.

5 ACKNOWLEDGEMENT

The authors wish to thank the Ministry ofScience, Technology and the Environment forfunding this work. Dr Khairul Zaman MDahlan and Dr Kamaruddin Hashim fromMalaysian Institute for Nuclear TechnologyResearch (MINT) are also thanked for theirco-operation during irradiation of the samples.

6 REFERENCES

[1] F N Büchi, B Gupta, O Haas and G GScherer. Study of Radiation-GraftingFEP-g-Polystyrene Membrane asPolymer Electrolyte in Fuel Cells.Electrochemica Acta, 40 (3), 345-353.1995.

[2] H P Brack, F N Büchi, M Rota and G GScherer. Development of RadiationGrafted Membranes for Fuel CellApplication Based on Poly(ethylene-alt-tetrafluoroethylene). Polym. Sci., Eng.,77, 368-369. 1998.

[3] A Chapiro. Radiation Chemistry ofPolymeric Systems. pp 481. John Wileyand Sons, New York. 1962.

[4] B Gupta, F N Büchi and G G Scherer.Cation Exchange Membranes byPreirradiation Grafting of Styrene ontoFEP. I. Influence of SynthesisConditions. J. Polym. Sci., Part A:Polym. Chem., 32, 1931-1938. 1994.

[5] B Gupta, F N Büchi, G G Scherer and AChapiro. Materials Research Aspects ofOrganic Solid Polymer ProtonConductors. Solid State Ionics, 61, 213-218. 1993.

[6] B Gupta, F N Büchi, M Staub, D Grmanand G G Scherer. Cation ExchangeMembranes by Preirradiation Grafting ofStyrene onto FEP. II. Properties ofCopolymer Membranes. J. Polym. Sci.,Part: A: Polym. Chem., 34, 1873-1880.1996.

[7] A G Guzman-Garcia and P Pintauro, MW Verbrugge and E W Schneider.Analysis of Radiation-GraftedMembranes for Fuel Cell Electrolytes. J.Appl. Electrochem., 22, 204-214. 1992.

[8] S Hietala, S Holmberg, M Karjalainen, JNäsman, M Paronen, R Serimaa, FSundholm and S Vahvaselka. StructuralInvestigation of Radiation Grafted andSulfonated Poly (vinylidene fluoride)PVDF Membranes. J. Mater. Chem., 5,721-727. 1997.

[9] S Holmberg, T Lehtinen, J Näsman, DOstrovskii, M Paronen, R Serimaa, FSundholm, G Sundholm, L Torell andM Torkkeli. Structure and Properties ofSulfonated Poly [(Vinyl Fluoride)-g-Styrene] Porous Membranes. J. Mater.Chem., 6, 1309-1317. 1996.

[10] M M Nasef, H Saidi and H M Nor.(2000 a). Proton Exchange MembranesPrepared by Simultaneous RadiationGrafting of Styrene onto FEP Films. I.Effect of Grafting Conditions. J. Appl.Polym. Sci., 76 (2), 220-227.

[11] M M Nasef, H Saidi, H M Nor and M FOoi. (2000 b). Proton ExchangeMembranes Prepared by SimultaneousRadiation Grafting of Styrene onto FEPFilms. II. Properties of the SulfonatedMembranes. J. Appl. Polym. Sci., Dueto appear.

[12] M M Nasef, H Saidi, A M Dessouki andE M El-Nesr. (2000 c). Radiation-induced Grafting of Styrene ontoPoly(tetrafluoroethylene), PTFE Films.I. Effect of Grafting Conditions andProperties of the Grafted Films. Polym.Inter.., 49, 399-406.

[13] M M Nasef, H Saidi, H M Nor and M FOoi. (2000 d). Radiation-inducedGrafting of Styrene ontoPoly(tetrafluoroethylene), PTFE Films.II. Properties of Grafted and SulfonatedMembranes. Polym. Inter. In press.

[14] M M Nasef, H Saidi, H M Nor and M FOoi. (2000 e). Cation ExchangeMembranes by Radiation-induced GraftCopolymerization of Styrene onto PFACopolymer Films. II. Characterizationof the Sulfonated Graft CopolymerMembranes. J. Appl. Polym. Sci., 76 (1),1-11.

[15] M M Nasef, H Saidi and H M Nor.(2000 f ) Cation Exchange Membranesby Radiation-induced GraftCopolymerization of Styrene onto PFACopolymer Films. III. Thermal Stabilityof the Membranes. J. Appl. Polym. Sci.,77 (9), 1877-1885.

[16] M M Nasef, H Saidi and M A Yarmo.(2000 g). Cation Exchange Membranesby Radiation-induced GraftCopolymerization of Styrene onto PFACopolymer Films. IV. Morphologicalinvestigations. J. Appl. Polym. Sci. Dueto appear.

[17] M M Nasef, H Saidi, H M Nor, K MDahlan, K Hashim. Cation ExchangeMembranes by Radiation-induced GraftCopolymerization of Styrene onto PFACopolymer Films. I. Preparation andCharacterization of the GraftedCopolymer. J. Appl. Polym. Sci., 73 (11),2095-2102. 1999.

[18] O Savadogo. Emerging of Membranesfor Electrochemical Systems: (I) SolidPolymer Electrolyte Membranes for FuelCell Systems. J. New Mater. forElectrochem. System, 1, 47. 1998.

[19] G G Scherer. Polymer Membranes forFuel Cells. Ber. Bunsenges. Phys. Chem.,94, 1008-1014. 1990.

[20] J P Shoesmith, R D Collins, M J Oakleyand D K Stevenson. Status of SolidPolymer Fuel Cell System Development.J. Power Sources, 49, 129-142. 1994.

[21] K Strasser. Mobile Fuel CellDevelopment at Siemens. J. PowerSources, 37, 209-217. 1992.

[22] H Wang. and G A Capuano. Behaviorof Raipore Radiation-Grafted PolymerMembranes in H2/O2 Fuel Cells. J.Electrochem. Soc., 145, 780-788. 1998.

Table 2 .The current density of the cell obtained at 0.500 mV of PEM fuel cell withPFA-g-polystyrene sulfonic acid membranes. The open circuit voltage is alsoincluded.

Membrane sample Open circuit voltage Current density

(mV) (mA/cm2) at 500 mV

PFA-g-PSSA (16 %) 830 40

PFA-g-PSSA (26 %) 850 48

Nafion 117 950 60



1. Introduction

Material handling systems are widelyused in many places such as shipyardsand power stations where large andheavy payloads or cargoes are handled.Despite efforts to improve andautomate the process of loading andunloading, the mode of operationcontinues to be manually intensiveand time-consuming. Acceleratingand decelerating an overhead craneusually induces swinging of thesuspended payload. The principal

contributing factor in the inefficiencyof the operation happens at the endof each loading operation when theoperator attempts to pick up or placethe load. The operator, during themove itself or at the end, must be ableto dampen the sway induced while theload is transported. As a result, thetime spent waiting for the load to stopswinging and fine positioning it at theend of the loading operation willdefinitely have serious financialrepercussion.

As a response towards easing the taskof crane operators, researchers haveresearched and implemented varioustechniques over the past decades toachieve an automatic anti-sway cranecontrol system. Initially, most of thework focused on the use ofconventional PID controllers whichfailed due to the non-linear behaviourof the crane system [Nalley and Trabia,2000]. Others opted for model-basedcontrol, which used a fifth-degreedifferential equation to describe thesystem behaviour [Altrock, 1996].

Neural Fuzzy Based 3D Anti-Sway Modelling

and Control Design for Overhead Cranes

M Mahfouf

Department of Automatic Control and Systems Engineering,

University of Sheffield, Mappin Street, Sheffield S1 3JD, United Kingdom

I Ismail



G Zissis

Formerly with the Department of Automatic Control and Systems Engineering,

University of Sheffield, United Kingdom

ABSTRACT

A non-linear 3D-model for an overhead crane system which takes into account a combination of a trolley and a pendulumis derived. The overall mathematical model hence obtained is simulated using MATLAB-SIMULINK. Open-loopsimulations suggest the validity of this model by reflecting similar trends in industries, which are concerned with materialhandling equipment. A handcrafted fuzzy controller, which includes two rule-bases, one for position control, the otherone for sway-angle control, was first designed and successfully implemented using the above simulated model. The studywas later extended to include the design of an Adaptive Neuro-Fuzzy Inference System (ANFIS) approach which provedto be effective in controlling the load oscillations and accurate in positioning the load itself along two axes.

Keywords:

neural fuzzy, fuzzy, 3-D overhead crane, modelling, anti-sway

This paper was presented at the 2001 International Conference on Artificial Intelligence, Las Vegas, USA, 25-28 June, 2001.



This approach works well in theorybut failed in practice due to non-linearand uncertain factors such as windgust and variations in payload’sweights.

Further work was carried-out towardsproviding accurate and robust controlschemes for overhead cranes. Sakawaand Shindo (1982) employed optimalcontrol to minimise the swing of thesuspended payload and Auernig andTroger (1987) used time optimalcontrol with a hoisting effect.Limitations to the above designbecame apparent when consideringexternal disturbances such as wind,length variations of the wire rope andweight variations of the load.Moustafa et al (1996) proposed a non-linear model for load sway control ofoverhead cranes with load hoisting viastability analysis techniques.

Fuzzy logic was applied to theoverhead crane control by severalresearchers. Yasunobu and Hasegawa(1996) applied predictive fuzzy to ashipyard crane control problem.Kimiaghalam et al (1999) adopted afuzzy controller for pendulationsuppression of a shipboard crane.Both previous papers concentrated ona simple two-dimensional model, andonly motion in one plane wasconsidered. Nalley and Trabia (2000)used a three-dimensional model tocompare a PD controller with a fuzzycontroller. Their model is a linearisedone with the assumption of anoverhead crane that carriesconsiderably long and massivepayloads. Ho (1998) has also focusedon the control of three-dimensionaloverhead cranes, where the load swing,crane motion, and load hoisting areconsidered all together in themodelling and control. He used adecoupled control scheme to controlload swings.

The work presented in this paperrepresents an extension to the earlierwork carried by our group in which asimple fuzzy logic based controller wasdesigned for standard overhead cranes(Mahfouf et al, 2000). The systemwas based on a trolley and pendulumcombination with motions along oneaxis only.

This current paper is organised asfollows: in Section 2 the 3D-mathematical modelling of anoverhead crane is overviewedincluding the cases of a constant andvariable rope length. In Section 3 thehandcrafted Fuzzy Logic based controlarchitecture is reviewed and itsimplementation on the model isanalysed and discussed. In Section 4the method is extended to include anadaptive architecture using the well-known Adaptive Neuro-FuzzyInference System (ANFIS) (Jang andMizutani, 1997) to elicite the rules forseveral scenarios relating to thepositioning of the load along two axes.Finally, in Section 5 conclusionsrelating to the overall study will bedrawn.

2. The 3D-Modelling of an

Overhead Crane System

The general co-ordinate system of a3-Dimensional Overhead Crane andits loads is depicted in Figure 1.

XYZ is the fixed coordinate systemand XTYTZT is the trolley co-ordinatesystem that moves with the trolley.The origin of the trolley co-ordinatesystem is parallel to the counterpartof the fixed co-ordinate system. YT isdefined along the girder, which is notshown in the figure. The trolley moveson the girder in the YT (traverse)direction and the girder and YT axismove in the XT (travel) direction. θis the swing angle of the load in anarbitrary direction in space and hastwo components: θx and θy, where θxis the swing angle projected on theXTZT plane and θy is the swing anglemeasured from the XTZT plane.

The position of the load (xm, ym, zm)in the fixed co-ordinate system is givenby:xm = x + l sinθx cosθy (1)

ym = y + l sinθy (2)

zm = – l cosθx cosθy (3)

where l denotes the rope length.

θx θ

θy

l

Fh

m

Load (xm , ym , zm)

XT

Fy

Trolley

ZT

0

Z

X

Y

(x, y, 0)

FxYT

Figure 1

Coordinate System of a 3-D Overhead Crane



The purpose of this is to control themotion of both the crane and its load.Hence x, y, l, θx, and θy are defined asthe generalised coordinates to describethe trolley and load motions.

The equations of motion of the cranesystem are derived by means ofLagrange’s equation of motion(Meirovitch, 1970). In deriving thedynamic model, the load is consideredas a point mass. The stiffness of therope is neglected.

The kinetic energy K of the crane andits load and the potential energy P ofthe load are given as:

K =12

M x x 2 + M y y 2 + Ml l2( ) + m

2vm

2

(4)

P = mgl (1 – cosθxcosθy) (5)where Mx, My and Ml are the x(travelling), y (traversing) and l(hoisting) components of the cranemass and the equivalent masses of therotating parts such as motors and theirdrive trains, respectively. m, g and vmdenote the load mass, the gravitationalacceleration and the load speedrespectively. The load speed vm isobtained as:

vm = xm2 + ym

2 + zm2

∴ vm2 = x 2 + y 2

+ l 2 cos2 θ yθx2 + l 2θ y

2

+2(sinθx cosθy l

+l cosθ x cosθ yθx

−l sinθx sinθ yθy )x

+2(sinθ yl +l cosθ yθy ) y

(6)

The Lagrangian L and Rayleigh’sdissipation function F are defined asfollows:

L = 12

M x x 2 + M y y 2 + Ml l2( )

+ m2

vm2 + mgl cosθx cos θy −1( )

(7)

F = 12

Dx x 2 + Dy y2 + Dl l2( ) (8)

where Dx, Dy and Dl denote theviscous damping coefficientsassociated with the x, y and z motionsrespectively.

The equations of motion for X-direction, Y-direction, and Z-directionare obtained by using equations (7)and (8) into Lagrange’s equation,equation (9):

ddt

∂L∂q

− ∂L

∂q+ ∂F

∂q= f i

where

( q = x ,θx , y,θ y , l )

( q = x ,θx , y,θ y , l ) (9)

where fi = fx, fy and fl which are thedriving forces for the x, y and zmotions respectively. Thus, theequations of motion can be derivedas follows [2]:

(M x + m)˙x + ml cosθx cosθ y˙θx

−ml sinθ x sinθy˙θ y +m ˙l sinθx cosθ y

+Dx x + 2ml cosθx cosθyθ x

−2ml sinθx sinθyθ y

−ml sinθ x cosθ yθx2

−2ml cosθx sinθ yθx θy

−ml sinθ x cosθ yθy2

= f x

(10)

ml 2 cos2 θ y˙θx

+ml cosθx cos θy ˙x

+2mll cos2 θ yθx

−2ml 2 sinθ y cosθ yθx θ

+mgl sinθx cosθ y

= 0

(11)

(M y +m)˙y +ml cosθy˙θ y

+m ˙l sinθ y + Dy ˙y

+2ml cosθyθ y + mgl sinθx cosθy

= 0

(12)

ml 2˙θy −ml sinθx sinθy ˙x

+ml cosθy ˙y + 2mllθ y

+ml 2 cosθ x sinθy xθx2

+mgl cosθ x sinθ y

= 0

(13)

(M l +m)˙l + msinθx cos θy ˙x

+m sinθ y ˙y + Dl l −ml cos2 θ yθx2

−mlθy2 −mg cosθ x cosθy +mg

= f l

(14)

Constant rope length

This is the case where the length ofthe rope l for the hoisting motion isassumed to be constant. Based on aconstant rope length, this will lead to

the velocity ( l ) and the acceleration

( ˙l ) of the hoist being both zero.

Thus

l = constant ⇒ l = ˙l = 0

Therefore, expanding andmanipulating of equations (10) - (14)lead to the followings equations:



˙x =f x − Dxx +mg sinθx cosθy

M x

(15)

˙θ x =[− f x cosθ x + Dx cosθ x x

−mg sinθ x cosθ x cosθy

+2 M xl sinθ yθxθ y

− M x g sinθx ]/[M xl cosθ y ]

(16)

˙y =f y +mg sinθ y − Dy y

M y(17)

˙θ y =

[ M y sinθ x sinθ y

( f x − Dx x + mg sinθx cosθ y )

− M x cosθy ( f y − Dy y +mg sinθy )

− M x M y sinθy

(l cosθ yθ x2 )]/[M y M xl ]

(18)

Variable rope length

This is the case where the length ofthe rope l for the hoisting motion willbe varied. Expanding andmanipulating of equations (10) - (14)leads to the followings equations:

x =[ f x − Dxx + sinθx cos θy

(M l˙l + Dl l + mg + f l )]/[ M x ]

˙

(19)

θx =[cos x (− f x + Dxx − mg sin θx cos θx

(M l˙l + Dl l + mg + f l ))

+ M x (−2 cos θyl θx + 2l sinθy θx θ y

− g sin θx )]/[M xl cosθy ]

˙

(20)

˙y =[ f y − sinθy

( f l − M l˙l − Dl l − mg)

−Dy y]/[M y ] (21)

˙θy =

[ M y sin θx sinθ y

( f x − Dx x + sinθx cosθ y

(M l˙l + Dl l + mg + f l ))

− M x cosθ y ( f y − sin θy

( f l − M l˙l − Dl l − mg) − Dy y)

− M x M y (2θ yl +l sinθ y cos θy θx2

+ g cos θx sin θy )]/ M y M xl ]

(22)

˙l =

[ M x yM ( f l +ml cos2 θy θx

2 +mlθy2

+Dl l + mg cos θx cos θy −mg)

− M ymsin θx cos θy

( f x + sinθx cos θ y

( f l + Dl l +mg) − Dx x)

− M xm sinθy ( f y − Dy y − sinθy

( f l − Dl l −mg))]/

[ M x M y ( Ml +m)

+ Ml m(M y sin2 θx cos2 θ y

+ M x sin2 θy ](23)

The above models will form the basisof an experimental design workrelating to anti-sway control as thenext section will endeavour to show.

3. A SIMPLE FUZZY LOGIC

BASED ANTI-SWAY CONTROL

DESIGN

A fuzzy logic control (FLC) systemcan be configured as shown in theblock diagram of Figure 2. Basically,a fuzzy controller can be divided into4 stages: fuzzification, fuzzy inferencesystem, fuzzy rule base anddefuzzification.

3.1 FLC Linguistic Control

Strategy

The following points describe thelinguistic control strategy obtainedfrom the analysis of an expert humanoperator’s action in controlling theload sway and trolley position.• Pick up the container and start

moving the crane head with amedium power.

• Adjust and increase the motorforce while observing the loadsway. The load shall be justbehind the crane head.

• Reduce the motor force whenabout to reach target. The loadshall be slightly ahead of the cranehead.

• Increase or decrease the motorforce to adjust to final position.

Fuzzifier Inference Engine Defuzzifier

Knowledge Base

Overhead Crane

Process Output & State Crisp Control Signal

Position, Angle

(real-time)

Applied Force

(real-time)

Fuzzy Logic Controller

Figure 2Block Diagram of the Crane Fuzzy Controller



• When the load is at targetposition and the load sway is zerostop the motor.

3.2 FLC Design

Fuzzy model may basically fall intotwo categories. One is the linguisticmodel that are based on the collectionof if-then rules with vague predicatesand that use a fuzzy reasoning such asMamdani’s model. The form of thismodel for a two-input-single- outputsystem is [5]

if x is A and y is B, then z is C(24)

where A, B, and C are fuzzy sets ofthe universe of discourse X, Y, and Z,respectively.

On the other hand, Sugeno’s model ischaracterised with functional typeconclusion. It has the form for a two-input-single-output system as

if x is A and y is B, then z = f(x,y)(25)

where A, B, and C are fuzzy sets ofthe universe of discourse X, Y, and Z,respectively. x and y are values of theinput variables.

A first order Sugeno’s model has theform of

if x is A and y is B,then z = px +qy + r (26)

The simulation of FLC is carried outusing MATLAB-Simulink software.In the case of a constant rope length,four inputs are required for each X andY directions which are:1) Cross beam and crane head

displacement, x(t), y(t)2) Cross beam and crane head

velocity, x (t), y (t)3) Load sway angle, θx(t), θy(t)4) Load sway angle velocity,

θx (t), θy (t)

The outputs are:1) Applied force in the X-direction,

fx2) Applied force in the Y-direction,

fy

For the case of a variable rope length,apart from the inputs and outputsdefined in a constant rope length, thefollowing additional inputs andoutput are required:1) Rope length, l(t)

2) Rope length velocity, l (t)

3) Applied force in the Z-direction,Fh

For both cases of the constant ropelength and the variable rope length,each input and output signal ismapped into five membershipfunction (MFs) of a Gaussian-typelabelled as Negative Big (NB),Negative Small (NS), Zero (Z),Positve Big (PB) and Positive Small

(PS). The span of each MF isdetermined arbitrarily and they arefine tuned later to improve thecontroller performance.

The three input FLC is found to bethe best configuration based on thefastest time reaching target positionand the least sway angle experiencedby the load [Zissis]. Thus, in this theproject the three input FLC waschosen. The type of controller usedfor this system was of a Mamdani-typerather than of a Sugeno-type.

The input signals to the FLC for the

X-direction are xe, x and θx. This

controller is a combination of 25 rulesfor position control based upon theinput signals xe and x , and 5 rulesfor the anti-sway control based upon

the input signals xe and θx. The

membership functions of the inputsand their fuzzy rule bases are tuned togive the desired closed-loop response.The rule bases are shown in Tables 1and 2. Similar rule bases are adopted

for the Y-direction using ye, y and θy .

3.3 Controller Performance

The closed loop responses for the caseof a constant rope length and for thecase of a variable rope length areshown in Figure 3 and 4. The detailsof other parameters response areshown in Figure 9.

Displacement Error xe

NB NS Z PS PB

NB NB PS PB PB PB

Crane Head NS NB Z PS PS PB

Velocity Z NB NS Z PS PB

x PS NB NS NS Z PB

PB NB NB NB NS PB

Displacement

Error xe , Z

NB NB

Angular NS NS

Velocity Z Z

θx PS PS

PB PB

Table 1Fuzzy Control Rule-base for the Cross Beam Position Control

Table 2Fuzzy Control Rule-base for the LoadSway Angle Control



Generally the performance of thecontrollers are very satisfactorily as itmanaged to get to the target positionand control the sway in less than 0.5minutes. These enable the operatorto effectively control the load sway inless than 1 minute. The GaussianMFs generate a smoother controlsurface and facilitates theimplementation of FLC for the cranesystem. However, the performancesof the controller are improved byadjusting the gain inputs factor andchanging the rule bases. Here, theparameters are tuned by trial and errorwhich are very time consuming.Thus, a better system is required. Thiswill be discussed in Section 4.

4. ANFIS-BASED ANTI-

SWAY CONTROL DESIGN

4.1 The ANFIS Design and

Training

Neural networks (NN) and fuzzysystems are widely used for modellingnon-linear systems. Theapproximating capability of NN’s,such as Multilayer Perceptrons (MLP),Radial-Basis Function Networks

(RBFN), or dynamic recurrent NNhave been investigated by manyauthors [Barada and Singh, 1999].The neuro fuzzy (NF) tool refers tothe way of applying learningtechniques offered by NN’s forparameter identification of fuzzymodels. Using a given input/outputdata set, ANFIS constructs a fuzzyinference system (FIS) whosemembership function parameters areadjusted using either a back-propagation algorithm alone, or in acombination with a least-squares typeof method. This allows the system tolearn from the data they aremodelling. In the ANFIS Editor,fuzzy inference is generated using twopartition methods, which are gridpartitioning, and subtractiveclustering. After generating the fuzzyinference system, the generatedinformation describing the model’sstructure and parameters of both theinput and output variables were passedon to the ANFIS training phase. Thisinformation will later be fine-tunedby applying the hybrid learning or theback-propagation schemes. After thisstage, the MF’s will be adjusted tooptimise the controller actions.

4.2 Training Data &

Checking Data Generations

The various strategies used forgenerating the data relied on thefollowing points:

1) Single set point (SSP) data inputi.e., +10m, +5m, +2m, +1m, -1m,-2m, -5m, and -10m.

2) Multiple set points (MSP) datainput.

For Figure 8, the following set pointsis predetermined randomly:

X-motion:-10 → 2 → 4 → -4 → -9 → 1 → 9→ 7 → 6

Y-motion:5 → -3 → -5 → 0 → 4 → 1 → -4 →4 → 5

Initial trial runs showed that MSPgave a better data distribution patternfor training.

Figure 5 shows the simulation dataselected for training ANFIS. The setpoints for X and Y were varied

12

10

8

6

4

2

0

Dis

pla

ce

me

nt

(m)

0 5 10 15 20 25 30 35 40

Time (seconds)

XY

Controller response for X = 10 m and Y = 5 m Controller response for X = 10 m and Y = 5 m

12

10

8

6

4

2

0

Dis

pla

ce

me

nt

(m)

Time (seconds)

0 5 10 15 20 25 30 35 40

Figure 3Closed Loop Response for the Crane ControlSystem (constant rope length)

Figure 4Closed Loop Response for the Crane ControlSystem (variable rope length)



randomly. Note that at each set pointchange, the fixed fuzzy gain inputfactors were tuned to obtain the bestperformance.

4.3 The ANFIS-based

Control Performance

Case1:

FIS: grid partioning and training:hybrid

Simulation data for ANFIS training

10

5

0

–5

–10

Dis

pla

ce

me

nt

(m)

Time (seconds)

0 50 100 150 200 250 300 350 400 450 500

Figure 5ANFIS Training Data Generated using Multiple Setpoints (MSP)

10

5

0

–5

–10

Dis

pla

ce

me

nt

(m)

Time (seconds)

0 5 10 15 20 25 30 35 40 45 50

X

sp

sp = +10

sp = +7

sp = +3

sp = –2

sp = –5

sp = –10

200

100

0

–100

Fy

Thydot

10

5

0

–5

–10 –10

–50

5

10

Ye

Figure 7SurfaceView fro the Y-motion Controller using SubtractiveClustering

Figure 7Closed Loop Responses for the Y-directions usingANFIS with Subtractive Clustering

6

4

2

0

–2

–4

–6

Dis

pla

ce

me

nt

(m)

Time (seconds)

0 5 10 15 20 25 30 35 40 45 50

Y

sp

sp = +5

sp = +3

sp = +1

sp = –2

sp = –4

sp = –5

Figure 8Closed Loop Responses for the Y-directions usingANFIS with Subtractive Clustering

The FIS output surface for thecontroller generated using gridpartitioning for the X-target and theY-target are shown in Figure 10.

As the ANFIS controller for Y-motiondid not give a good performance ascompared to the fixed fuzzy system,the subtractive clustering method isused instead in the hope of improvingthe controller performance.

Case2:

FIS: subtractive clustering andtraining: hybrid

Figure 7 shows the new generatedoutput surface view.

The closed loop responses for thecontroller for the X and Y-motions areshown in Figure 8 where theperformances can be judged as good



2000

1000

0

–1000

Fo

rce

(N

)

0 5 10 15 20 25 30 35 40

Time (seconds)

40

0

–20

–40

0 5 10 15 20 25 30 35 40

10

0

–10

–20

0 5 10 15 20 25 30 35 40

1

0

–00 5 10 15 20 25 30 35 40

An

g.r

Ve

l. (d

eg

/se

c)

Sw

ay a

ng

. (d

eg

ree

)V

el.

(m/s

ec

)

Fx

Fy

X

Y

X

Y

X

Y

a) Cross Beam &Trolley Velocity

Figure 9Closed Loop Responses for a Constant Rope Length Crane Control System

b) Sway Angle

c) Angular Velocity

b) Applied Force

5

0

–5

Fy

Thxdot

20

0

–10

0

10

x 104

–20Xe

(a)

2000

1000

0

–1000

–2000

Fy

Thxdot

10

0

–50

5

–10Ye

(b)

Figure 10

Surfaces Views for the Controller: (a) for the X-motion (b) for the Y-motionby Grid Partitioning

since ANFIS allowed the generationof a set of fuzzy rules which wereeffective over a wide range of positionset-points, this would not have beenpossible to obtain with the simplefuzzy control scheme, above described,as we would have needed to retune thescaling factors and possibly the rulesat every set-point change.

5. CONCLUSIONS

A non-linear dynamic model of athree-dimensional overhead crane hasbeen derived based on two-degrees offreedom swing angle. Simulationresults using open loop impulse tests(not shown here) demonstrated thevalidity of the model.



A new fixed fuzzy logic controller wasdeveloped and tested on the non-linear model for a constant rope lengthand a variable rope length. Simulationresults showed accurate positioncontrol and effective anti-swaycontrol. However, the system iseffective for specific operationalparameters only.

Difficulties in tuning the fixed fuzzysystem led to the use of ANFIS. Thelearning capabilities of the systemallow the new fuzzy system to operateover wider operational requirements.Simulation results have shown thatANFIS can emulate the fixed fuzzysystem, but its effectiveness dependson the quality of the training data usedprior to the rule-base development.

6. REFERENCES

[1] C V Altrock. Practical Fuzzy LogicDesign, http://www.fuzzytech.com.1996.

[2] J W Auernig and H Troger. Time OptimalControl of Overhead Cranes with Hoistingof the Load, Automatica, Vol. 23, No. 4,p. 437. 1987.

[3] S Barada and H Singh. GeneratingOptimal Adaptive Fuzzy-Neural Models ofDynamical Systems with Applications toControl, IEEE Transactions On Systems,Man, and Cybernetics – Part C:Applications and Reviews, Vol. 28, No.3, p. 371-390. 1999.

[4] Ho-Hoon Lee. Modeling and Control ofa Three-Dimensional Overhead Crane,Journal of Dynamics System,Measurement, and Control, Vol. 120, p.471-476. 1998.

[5] J S R Jang, C T Sun and E Mizutani.Neuro-Fuzzy and Soft Computing: AComputational Approach to Learning andMachine Intelligence, Prentice Hall.1997.

[6] B Kimiaghalam, A Homaifar and MBikdash. Pendulation Suppression of aShipboard Crane using Fuzzy Controller,Proceedings of the American ControlConference, p. 586-590. 1999.

[7] M Mahfouf, C H Kee, M F Abbod andD A Linkens. Fuzzy logic based anti-swaydesign for overhead cranes, Journal ofNeural Computing and Applications – ASpecial issue on Fuzzy Systems Applications,9, 38-43. 2000.

[8] L Meirovitch. Methods of AnalyticalDynamics, Prentice Hall Inc. 1970.

[9] K A Moustafa, T G Abou-El-Yazid. LoadSway Control of Overhead Cranes withLoad Hoisting via Stability Analysis, JSMEInternational Journal: Series C, Vol. 39,No. 1, p. 34-40. 1996.

[10] M J Nalley and M B Trabia. Control ofOverhead Crane using a Fuzzy LogicController, Journal of Intelligent andFuzzy System, Vol. 8, p. 1-17. 2000.

[11] Y Sakawa and Y Shindo. Optimal Controlof Container Cranes, Automatica, Vol. 18,No. 3, p. 257. 1982.

[12] G Zissis. Anti-Sway Modelling andControl Design for Overhead Cranes, MScThesis, The University of Sheffield, U.K.1999.



INTRODUCTION

V-groove is one of the most importantcomponents in the structure ofVMOS and greatly influenced theperformance of the device. The keymethod in producing V-groove is touse the large orientation dependencein the etch rates. Some etchants etcha given plane of a semiconductormuch faster than other planes,resulting in an orientation-dependentetching. In diamond and zincblend

lattices, the (111)-plane is more closelypacked than the (100)-plane makingthe etch rate to be slower for the (111)-plane. A precise V-shaped grooves canbe created by orientation dependentetching of <100> oriented siliconthrough a patterned SiO2 mask. Theedges of the groove being (111) planesat an angle of 54.7° from the (100)surface as shown [2] in Figure 1(a).At the temperature of 70-80 °C, theetch rate is 0.6 µm/min for the (100)-plane, 0.1 µm/min for the (110)-

plane, and only 0.006 µm/min (60Å/min) for the (111)-plane [2].Therefore the ratio of the etch ratesfor the (100)-, (110)-, and (111)-planes is 100:16:1. If the window issufficiently large or if the etching timeis short, then a U-shaped groove willformed instead (see the right side ofFigure 1(a)). Etching done on the<110>-oriented silicon will producestraight-walled grooves with sides of(111)-planes, as shown in Figure 1(b).

Scanning Electron Microscopy of

Anisotropic Etching in Fabrication of VMOS

(Vertical Metal-Oxide-Semiconductor) Transistor

Norani M Mohamed



Kamarulazizi Ibrahim & Leong Yew Wei

Pusat Pengajian Sains Fizik, Universiti Sains Malaysia,

11800 Pulau Pinang, Malaysia.

ABSTRACT

MOSFET (metal-oxide-semiconductor field effect transistor) is one of the variations of the MOS transistor. It is themost important device for very-large-scale integrated (VLSI) circuits such as microprocessors and semiconductor memories.Since the first fabrication of MOSFET in 1960 [1], the minimum channel length has been shrinking continuously. Theforce for miniaturization is coming from the need for dense circuits for high-density memory and logic applications aswell as from the need for high frequency microwave devices. VMOS technology is one of the methods of producing shortchannel device. The study here focused on the etching mechanisms and the effectiveness of the etchant to form a perfectV-groove in the VMOS structure.

The work began by fabricating the VMOS transistor on <100>-oriented Si wafer by using a series of fabrication processessuch as oxidation, pre-deposition, drive-in deposition, photolitography, etching and metallization. The final step ofetching is to produce the V-groove by using a heated mixture of KOH and deonized water through SiO2 mask. Thefabricated VMOS structures were then cut cross-sectional and examined under SEM.

As evidence from the SEM image, only the right alignment of the mask to the substrate and the right size of the windowof the mask can produce perfect V-groove as required.

This paper was presented at the 10th Scientific Conference of Electron Microscopy Society Malaysia, Selangor, 8-10 November, 2001.



Figure 2 shows the cross section of aVMOS. The basic concept behind theMOS device is quite simple and canbe illustrated in this diagram. Thedevice consists of an active channelthrough which electrons flow from thesource to the drain. The source anddrain contacts are ohmic contacts.The width of the channel ismodulated by the potential applied tothe gate. The channel widthmodulation then results in themodulation of the current flowing inthe channel. An importantconsideration in this process is toisolate the gate from the channel. Ifthe gate is not well isolated from thechannel, it draws a lot of current,leading to a device that has a poorgain. Here, in the VMOS transistor,the gate is isolated from the channelby an oxide.

Fabrication of this device starts withan n-type silicon wafer. The wafer isthen oxidized, and p-diffusion isperformed through a window in theoxide into n-layer. This is followedby the n+-source diffusion through asmaller window in the oxide. In themiddle of the source region a V-grooveis etched down to the n-layerselectively using an anisotropicetchant. After groove etching thesurface is oxidized, and Al contacts areevaporated to form source, drain andgate.

EXPERIMENTAL DETAILS

Fabrication of VMOS involvedrepetitions of a series of processesincluding oxidation, photolitography,etching, doping, pre-deposition,drive-in deposition, ending withmetallization and etching of theunwanted aluminum. The substrateused was the wafer which was initiallycharacterised in order to determine thetype of wafer (n or p), thickness andresistivity. Five masks were preparedin order to create the windows for thesubsequent processes such as doping,etching of V-groove andmetallization. Prior to any fabricationprocess, the cut wafers were cleanedby RCA clean using solution 1 whichconsist of H2O + NH4OH + H2O2

(5:1:1), followed by solution 2: HF +H2O (1:50) and finally solution 3:H2O + HCl + H2O2 (6:1:1).

Oxidation was the first fabricationprocess carried out. In this process,passing dry oxygen through the waferin the thermal furnace, normally at1000 °C for 11⁄2 hours can produce alayer of oxide on top of the substrate.The purpose of photolitographyprocess is to transfer patterns on amask to a thin layer of radiation-sensitive material, resist, covering thesurface of the substrate. The exposuresystem used in this process is theultraviolet light. In the fabrication,wet chemical etching proceeds byoxidation, followed by dissolution ofthe oxide by a chemical reaction. Themain purpose of this process is tocreate windows in the oxide layer,which is not protected by the resist.The etchant used here was the mixtureof nitric acid (HNO3) andhydrofluoric acid (HF) in water. Toproduce V-groove, a different etchant

Wb

l

SiO2

Wa

(100)(111)54.7°

Si

SiO2

(111)

Si

(110)

SiO2

p

n+ - substrate < 100 >

Metal

n+

L

n+

SiO2

S G

D

Figure 1Orientation-dependent etching conducted through window patterns on(a) <100>-oriented silicon and (b) <110>-oriented silicon.

Figure 1 (a) Figure 1 (b)

Figure 2:Cross-section of aVMOS [3] obtainedthrough anisotropicetching.



Figure 3:Flow chart of the fabrication of VMOS

Secondaryflat

<100> n-type

wafer

180°

Primary flat

Substrate

Substrate

SiO2

Substrate

Resist

Substrate

Wafer characterisation:to determine type of wafer,thickness and resistivity.

RCA clean:Soln 1: H2O + NH4OH + H2O2 (5:1:1)Soln 2: HF + H2O (1:50)Soln 3: H2O + HCl + H2O2 (6:1:1)

1st Oxidation:to grow a thin layer of SiO2.Oxidation furnace, dry oxygen,1000 °C, 11⁄2 hours

1st Photolitography:to create a window for p-doping– sputtering of resist – 100 rpm for 5 secs– expose to UV for 55 secs– developing with the developer– drying in the furnace for 10-15 mins at 80 °C

1st etching:to etch the oxide notprotected by the resist.Etchant: HF + HNO3 + H2O

continued



Substrate

Boron

Substrate

Drive-in diffusion

Substrate

Phosphorus

Drive-in

Figure 3 (continuation):Flow chart of the fabrication of VMOS

Doping of boron;to create p region.– sputtering 100 rpm for 5 secs– pass H2 in the furnace for 10-15 mins at 80-90°C

continued

Pre-deposition; to diffuse dopants into the substrate– N2 gass at 1000 °C for 10 mins. in the furnace.Drive-in diffusion: to distribute dopants throughoutthe substrate – in pure o2 gas at 1000 °C fo one hr.

2nd Oxidation:same as the 1st – 1 hr.

2nd Photolitography:to create window for n doping– same as the 1st.

2nd Etching: same as the 1st.

Doping of phosphorus:same as the 1st doping.

Pre-deposition followed bydrive-in diffusion.

3rd Oxidation:same as the 2nd.



Figure 3 (continuation):Flow chart of the fabrication of VMOS

3rd Photolitography:to create window for the formationof V-groove

3rd Etching: same as the 1st

Anisotropic etching

4th Oxidation:to grow oxide gate– same as the 2nd

4th Photolitography:to create windows formetallization of source and drain.

4th Etching: same as the 1st

Metallization: to form an aluminum

5th Photolitography:to identify the regions for metal contact.Aluminum etching: Al etchant + deionised water(1:1:5), 40-60 °CCurve tracer test: to ensure the device is functioning.



was used consisting of a heated (70-80 ° C) mixture of KOH in water andpropanol.

Doping process involved introducingboron in order to create p region andphosphorus to create n region. Thisprocess is followed by heat treatment,normally known as pre-deposition, inN2 gas for 10 minutes at 1000 °C. Inthis way, the dopants can diffuse intothe substrate. Further distribution ofthe dopants can be achieved by thedrive-in diffusion process whereby inthis process, the substrate will beexposed to O2 gas at 1000 °C for onehour. The flow chart depicted inFigure 3 describes the processesinvolved in fabricating VMOS device.

RESULTS

SEM image in Figure 4(a) describes ashallow V-groove as a result of using asmall window in the mask and itsmisalignment with <111> orientation.With the right size of the window andthe mask aligned to <111>orientation, perfect V-groove can beformed as illustrated in Figure 4(b).With a large window, etching timeneed to be closely monitored in orderto avoid the penetration of the groovethrough the wafer (thickness of ~ 0.3mm).

(a) (b)

Figure 4:Cross-section images of (a) a shallow V groove and (b) 3 isolated Vgrooves.

CONCLUSION

The formation of V-groove dependson several factors: orientation of thewafer (substrate) used, size of thewindow in the mask for theanisotropic etching and the etchingtime. It was found that perfect V-groove can be achieved by using<100>oriented silicon wafer as thesubstrate with the mask aligned to theorientation of (111) plane of thesubstrate. Further requirementneeded would be the right size of themask with the etching done insufficient time.

ACKNOWLEDGEMENT

The authors would like to express theirgratitude to Mr Ragunathan, EM Unit atPostgraduate Centre, University Malaya for theelectron micrographs.

REFERENCES

[1] D Kahng and M M Atalla, IRE SolidState Device Res.Conf., Pittsburg, Pa.,1960, D.Kang, IEEE Trans ElectronDevices, ED-23,655, 1976.

[2] K E Bean, Anisotropic Etching in Silicon,IEEE Trans. Electron Devices, ED-25,(1978), p1185.

[3] H Martinot & P Rossel, Handbook OnSemiconductors, North-HollandPublishing Co., (1981), p910.



ABSTRACT

Electronic systems have now grown smaller resulting in very high heat generation compared to previous systems. Jetimpingement cooling has been identified to be useful in the electronics cooling due to its high heat removal capabilities.Jet impingement cooling in electronic packages is carried out numerically using a commercial finite volume code,FLUENTTM. The local heat transfer coefficients on a heat source due to a normally impinging, axisymmetric, confinedand submerged liquid jet are investigated. Numerical predictions are made for nozzle diameter (d) of 3.18 mm at severalnozzle to target plate spacing (H/d) ranging from 1 to 8. The turbulent jet Reynolds numbers considered are 8500,10000 and 13000 with a perflourinated dielecric fluid Flourinert-77 (FC77) as the working fluid. The flow field andheat transfer are solved using the standard high Reynolds number k-e turbulence model. A more detailed grid refinementcompared to previous investigations is utilized. The present predictions with the standard high Reynolds number k-eturbulence model with modified grid refinement are able to produce results with maximum errors of 4.6% and 9.9% forstagnation and averaged heat transfer coefficients respectively. In earlier prediction, the deviations from the experimentalresults are observed to be a maximum of 60.4% for the stagnation heat transfer coefficient and a maximum of 56.6% forthe averaged heat transfer coefficient. Numerical predictions are also carried out for those cases of H/d and Re for whichneither experimental data nor numerical predicted data are available in the literature. From the predicted results, correlationsare developed to determine the stagnation heat transfer coefficient, the average heat transfer.

Keywords

jet impingement cooling, electronic packaging, single circular nozzle, axisymmetric and stagnation heat transfer coefficient.

1 INTRODUCTION

Recent developments with increase inthe computer processing power, thetelecommunication industry and thegrowing need for better performanceand mobility; has resulted in increaseof heat generation per unit area.Traditional cooling methods areunable to handle such high

performance electronic packages. Asa result, new cooling techniques usingmicro heat exchangers, micro heatpipes, refrigeration cooling, liquidimmersion cooling and jetimpingement cooling are employed.Because of this and the simplicity ofthe equipment required, jetimpingement is utilized in electronicscooling.

Jet Impingement Cooling

Of Microelectronic Systems

Mohd Shiraz Aris

Department of Mechanical Engineering, Universiti Teknologi PETRONAS


Email: [email protected]

I Rushyendran, G A Quadir & K N Seetharamu

School of Mechanical Engineering, University Science of Malaysia,

Perak Branch Campus, 31750 Tronoh, Perak Darul Ridzuan.

email: [email protected]

Morris et al.1 predicted the local heattransfer distribution on a square heatsource for a normally impinging,axisymmetric, confined andsubmerged liquid (FC-77) jet using ahybrid wall treatment with fouralternative turbulent Prandtl numberfunctions. FLUENTTM was used tosolve the thermal and flow fields usinga standard high Reynolds number

This paper was presented at the ECCOMAS Computational Fluid Dynamics Conference 2001, Wales, 4-7 September, 2001.



k-ε turbulence model. Resultsobtained from FLUENTTM werefound to under predict the heattransfer coefficients. After postprocessing, all the data obtained usingthe Prandtl functions were reportedto provided better prediction. Behniaet al.2 numerically studied the heattransfer in an axisymmetric turbulentjet impinging on a flat plate using thenormal-velocity relaxation (V2F)turbulence model. Comparing withthe k-ε model, they concluded that theV2F predictions were in excellentagreement with experiments whereas,the k-ε model greatly over predictedthe heat transfer rates. They reportedthat computations using RNG k-εmodel produced results similar to thestandard k-ε model. Behnia et al.3

used an elliptic relaxation turbulencemodel to simulate the flow and heattransfer in circular confined andunconfined jet impingementconfigurations. The predicted resultshowed that confinement decreasedthe average heat transfer rates but,local stagnation heat transfercoefficient was not effected. Tanakaet al.4 conducted experiments andnumerically simulated theimpingement air-cooling withrectangular nozzles for Large ScaleIntegration (LSI) package with largeplate fins. They used a steady state 3dimensional incompressible flow

analysis based on two-equationturbulence model. The finite elementmethod with the penalty functionformulation was used. The calculatedvelocity vectors compared well withthe flow visualisation and thepredicted temperature distributionshowed good agreement with themeasured experimental temperatures.

Jet impingement heat transfer hasbeen widely studied by researchers formany years5. However, studiesregarding the use of jet impingementcooling in electronic packages startedonly in early 1990s. Generally, theexperimental data in the literature varygreatly. However, there is someconsistency in the work by somenotable researchers5.

It is obvious that jet impingement flowis very complex. Very thin boundarylayer develops at the stagnation regiondue to the impinging jet and the fluidflow direction changes from axial toradial. As the fluid flows away fromthe jet axis, the boundary layerbecomes thicker along the radialdirection. In many analyses, the gridsused are structured with gridsgradually becoming finer near thewalls. For the above mentioned flowa doubt arises, whether the grids usedare adequate and the closure model isable to predict the flow or not. This

indicates that a more detailed gridrefinement is required keeping in viewthe accuracy of the closure modelused. Therefore, a detailed gridrefinement is carried out in this workusing k-ε closure model to predict theheat transfer coefficients for differentimpingement cooling conditionsemployed.

Numerical studies carried out in theliterature suggest that k-ε model is notable to predict the flow field and theheat transfer that takes place in jetimpingement problems1-3. Nodetailed study on the effect ofmodifying the grid, at the stagnationand outflow region along theimpingement surface (which is theregion of interest in jet impingementcooling) has been conducted. Onlygeneral grid dependence study bydoubling the grid in both the radialand axial direction of the entire fluiddomain was carried out earlier5.

In this work, numerical predictions injet impingement flow field and heattransfer are carried out using CFDmethod. FLUENTTM Ver. 4.3 is used.A compatible pre-processor, PREBFCis used for geometry modelling andgrid generation. 2D Axisymmetric,confined and submerged models weremodelled. The geometries consideredas shown in Figure 1.1, are nozzle

Figure 1.1:Important Geometries Considered and Fluid Flow in Jet Impingement

Nozzle to

target plate

spacing, H

Nozzle

diameter, d

Orifice

Fluid Flow

Plenum

Nozzle

length, l

Target Plate



Table 2.1:Common and Varied Dimensions for the Cases Considered

a) Common b) Varied

Dimension Nozzle to target plate

Description (mm) Case Height, H (mm)

Plenum Chamber Height 12.72 H/d = 1 3.18

Plenum Chamber Radius 47.70 H/d = 2 6.36

Nozzle Radius, r 1.59 H/d = 3 9.54

Nozzle Length, l 6.36 H/d = 4 12.72

Outflow Region Radius 57.24 H/d = 8 25.44

diameter, d of 3.18 mm and nozzle-to-target plate spacing, H/d of 1, 2,3, 4 and 8 at Reynolds numbers of8500, 10000 and 13000.

Confinement was taken into accountsince in practical applications jets areusually confined and because it largelyaffects the turbulence level whichinfluences the heat transfer rate at theimpingement surface. Flourinert-77(FC-77), a perflourinert dielectricfluid commonly used liquid inelectronics cooling was considered.These geometries were selectedbecause of available reliable data inthese ranges for validation purposes.New data was generated for somecombinations of these parameters,which is not available in literature.

The standard high Reynolds numbersk-ε turbulence model was used tosolve the governing equations. Asystematic grid control to obtain goodapproximations to experimental datais explored.

2 ANALYSIS

2.1 The Governing

Equations

The basic equations describing theflow of fluid are conservation of mass,conservation of momentum andconservation of energy. The detailsof the equations along with the k-εturbulence model are available inFluent6.

2.2 Modeling Jet

Impingement Cooling

The jet impingement flow inelectronic cooling is mainly turbulentin nature. The jet produced from thenozzle impinges on the heat source,which represents the microelectronicsystem that requires cooling. Thefluid is decelerated at the stagnationregion and then accelerated again asthe fluid flows radially outward. Theheat transfer that takes place from theheat source on the impingement plate,is predicted in this simulation work.

A preprocessor called preBFC is usedto generate the model and creating thegrid. Then, the model is importedinto FLUENTTM where the boundaryconditions are applied, the solutioncriteria are selected and the model issolved.

Since the axisymmetric model hassymmetry at the centerline, a 2Dmodel of half of the cross section andspecifying the model to beaxisymmetric is sufficient to representthe jet impingement problemconsidered in this work. This sort ofmodeling capability is available in thesoftwares used here (only half of thecross section in Figure 1.1 is modeled).Tables 2.1 shows the common andvaried dimensions in the casesconsidered. Figure 1.2 shows themodel and grid generated for theH/d = 4 case.

Figure 2.2:Axisymmetric Model with Grid Generated



The number of grids in the plenumchamber and nozzle are same for allthe cases. Only the number of gridsat the outflow region is changed. Thegrid size used in each case is presentedin Table 2.2.

Table 2.2:Grid Size Used in Each Case

Case Grid Size

H/d = 1 113 x 220

H/d = 2 135 x 220

H/d = 3 158 x 220

H/d = 4 180 x 220

H/d = 8 225 x 220

2.3 Solution Procedure

In all the analysis carried out in thiswork, the solutions are obtained in 2parts. First, the flow in jetimpingement is obtained. Then heattransfer inputs are incorporated intothe flow field solution and simulationis continued to obtain the final heattransfer results.

2.3.1 Model, Boundary

Conditions and Physical

Constants Used

The standard high Reynolds numberk-ε turbulence model together withthe non-equilibrium wall function isused. In the non-equilibrium wallfunction local equilibrium assumption(production = dissipation), which isadopted in the standard (default wallfunction in FLUENTTM) wallfunction, is relaxed. The non-equilibrium wall function, partlyaccounts for the non-equilibriumeffects neglected in the standard wallfunction. Thus, the non-equilibriumwall function as recommended in theFluent6 for use in complex flowsinvolving separation, reattachment

and impingement is employed in allthe predictions done in this work.

All physical constant values forFlourinert-77 (FC-77) taken at 293K as listed in Table 2.3. These valuesare supplied in two stages; first flowfield followed by heat transfer.

The assumptions and boundaryconditions used are as detailed below,

a) Flow Field Solution

Inlet:For the cases considered, value of theimposed pressure is as suggested inliterature7. They obtained thepressure values by adjusting theimposed pressure values until thevolumetric flow rate is within 1%from the required value. They alsosuggested that pressure boundariesgive a better representation comparedto inlet and outlet velocities. Table2.4 shows the inlet pressure boundaryconditions imposed with theassociated errors for all the casesconsidered in this work. Turbulenceintensity at the inlet is given a valueof 0.5% as suggested by Morris andGarimella7. The characteristic lengthis chosen to be one jet diameter.

Walls:A no slip boundary condition isimposed along all solid surfaces.

Outlet:Due to the presence of toroidalrecirculation zone, the normally usedoutlet boundary does not providesufficient representation. Thus, astatic pressure boundary condition (0Pa) is applied (inlet boundary set asZone 2) as this allows flow to enter orexit the fluid domain. In the paperby Morris and Garimella7 they quotedthat Polat et al. (1989) recommendeda pressure boundary at the outlet forconfined jet impingement.

b) Heat Transfer Solution

In the heat transfer solution; the wallcells at the target plate are changed toZone 2 to represent the heat source.The wall cells changed are from thecell adjacent to the symmetry cell tillthe cell at a radial distance of 10 mm1.The remaining cells on the target plateare changed to Zone 3.

Inlet:The pressure boundaries applied inthe first part is maintained.Temperature is applied to be 293 K.

Walls:The no-slip boundary conditionapplied in the first part is maintainedfor all the wall zones. Temperature isapplied to be 293 K for Wall: Zone1. Wall: Zone 2; the heat source isapplied a heat flux of 250000 W/m2

and Wall: Zone 3 is applied with a

Table 2.3:Physical Constant Values of FC-77 at 293 K

Physical Constant Value Used

Operating Pressure 0 Pa

Density 1.789 x 103 kg/m3

Dynamic Viscosity 1.539 x 10-3 kg/ms

Thermal Conductivity* 6.340 x 10-2 W/mK

Specific Heat* 1.045 x 103 kg/m3

*Value keyed in before continuing with the heat transfer solution



heat flux of 0 W/m2 to representinsulated condition (no heatgenerated).

Outlet:0 Pa, boundary condition ismaintained and temperature is appliedto be 293 K.

2.3.2 Numerical Solution

Procedure

In both of the simulation parts, thesolving methods are as suggested byMorris et al.1 and Morris andGarimella7. However, some of therecommended methods are changedwhen they are not giving reasonablesolutions.

For the flow field simulation, thesolution is considered to be convergedwhen the sum of the normalisedresidual is 1 x 10-3. In the heat transfer

simulation, the convergence criteriaare the sum of the normalised residualis 5 x 10-4 and the enthalpy residual is5 x 10-5. The under relaxation factorsused are changed during simulationin order to facilitate convergence.

3 RESULTS AND

DISCUSSION

The jet impingement heat transferproblem, is numerically computedwith the commercial finite-volumecode FLUENTTM using the timeaveraged Navier Stokes and energyequations with the standard highReynolds number k-ε turbulencemodel. The k-ε model is adoptedsince it has been in use for a long timewith its merits and demerits wellunderstood and determined. It ismore robust than newer models andbetter suited for solving applicationoriented problems6.

3.1 Grid Refinement

The first task carried out is todetermine a proper grid for getting thenumerical heat transfer results. TheH/d = 4 case at Re = 10000 isconsidered to study the effect of gridon the numerical prediction. A gridsize of 160 x 160 is used to predictthe variation of heat transfercoefficient in the radial direction.

Then the grid size has been changedfrom 160 x 160 to 160 x 200, but theprediction did not improve. Grid isthen increased at the outflow region(200 x 220), resulting in betterpredictions near the outlet zone.Another trial by increasing the grid atthe nozzle as well as the outflow regionwas carried out. This grid size of 200x 260 did not improve the predictionscompared to the predictions obtainedwith the grid increased at the outflow

Table 2.3.1Inlet Pressure Boundary Imposed and the Associated Volumetric Flow Balance

–––––––––––– Volumetric Flow Rate ––––––––––––

Case Re No. Pressure, ∆∆∆∆∆P (Pa) Inlet Outlet Error (%)

H/d = 1 8500 *7116 2.487E-06 2.455E-06 1.3

10000 9850 2.962E-06 2.935E-06 0.9

13000 16646 3.935E-06 3.875E-06 1.5

H/d = 2 8500 7116 2.445E-06 2.424E-06 0.9

10000 9850 2.941E-06 2.888E-06 1.8

13000 16646 3.866E-06 3.837E-06 0.8

H/d = 3 8500 *7116 2.425E-06 2.417E-06 0.3

10000 9850 2.898E-06 2.853E-06 1.6

13000 *16646 3.854E-06 3.826E-06 0.7

H/d = 4 8500 *7116 2.424E-06 2.419E-06 0.2

10000 *9850 2.942E-06 2.919E-06 0.8

13000 *16646 3.819E-06 3.820E-06 0.0

H/d = 8 8500 7116 2.424E-06 2.423E-06 0.0

10000 9850 2.888E-06 2.889E-06 0.0

13000 16646 3.837E-06 3.825E-06 0.3

*Value from Morris and Garimella7



region alone. From the predictionsobserved, it is obvious that the grid atthe outflow region plays an importantrole in the prediction of the heattransfer coefficients.

Realising this, different trials werecarried out by concentrating at theoutflow region. Further gridrefinement is done by performing 2D-Interpolations at selected radialdistances from the jet axis and refiningthe grids using weighting factors.After several trials, (only some of themare shown) as shown in Figure 3.1 agrid size of 180 x 220 with suitablerefinement of grids at the outflowregion is selected which follows thefluid flow from the stagnation regionup to the outlet. It can be observedfrom Figure 3.2 that with this grid thevariation of heat transfer coefficientin the radial direction follows closelywith the experimental values.

3.2 Validation

The validation is carried out in termsof the predicted stagnation andaveraged heat transfer coefficients forthe H/d = 4 at Re = 10000 case. Thevariation of the heat transfercoefficients in the radial direction onthe target plate is also compared withearlier published numericalpredictions and experimental results.

Figure 3.3 shows the variation of heattransfer coefficients along the radialdirection for the geometry statedabove using various turbulent Prandtlnumber functions (TPF) as explainedby Morris et al.1 with the presentpredicted results. It can be seen thatthe predictions are not good,including the FLUENTTM predictiongiven by Morris et al.1 using simplek-ε model. For the geometricalconfiguration discussed above, theexperimental heat transfer coefficientat the stagnation region is 7803W/m2K and the predicted value from

the present grid refinement methodusing k-ε is 7807 W/m2K. Thepredicted value by Morris et al.1 usingthe standard high Reynolds numberk-ε model is 3540 W/m2K giving anerror of 54.6% with respect to theexperiment.

Morris et al.1 assumed that the gridthey selected is the proper one whereasthe turbulence model selected is notsuitable. Hence they tried with several

different TPF to improve theirpredictions. In this regard they triedout various post processingapproaches using different TPF givenby Wassel-Catton, Gibson-Launder(G-L TPF), Malhotra-Kang and Kaysand their results are shown in Table3.1. It can be seen that the Gibson-Launder TPF gave an error of 12%;Malhotra-Kang TPF, an error of 1.1%;Wassel-Catton, an error of 12% andKays, an error of 6.6%. Thus, the

5000

6000

7000

8000

9000

10000

11000

12000

0.00 0.50 1.00 1.50 2.00

r/d

hW

/(m

2k)

Experiment (Morris et al., 1996)

Grid 180x220 : Refinement Trial No. 5






d = 3.18 mmRe = 10000

H/d = 4l/d = 2

Figure 3.1:Variation of Heat Transfer Coefficients with Different Grid Refinement Trials

5000

5500

6000

6500

7000

7500

8000

8500

9000

0.00 0.50 1.00 1.50 2.00

r/d

hW

/(m

2k)

Experiment (Morris et al., 1996)

FLUENT (Present) : Final Grid

d = 3.18 mmRe = 10000

H/d = 4l/d = 2

Figure 3.2:Comparison of Experimental and FLUENTTM Results with Appropriate GridRefinement



predictions that they obtained didimprove. However, in the presentcircumstance, it can be observed thatthe simple standard high Reynoldsnumber k-ε model with non-equilibrium wall function gives betterpredictions, which are in closeagreement with the experimentalvalues.

The mesh gradation has beenmodified with a factor, which is

proportional to Re . Using the

specifically modified grids for both theRe = 8500 and Re = 13000 cases (theoverall grid size is the same but muchfiner or coarser depending on thecase), the predictions as presented inFigure 3.4 are obtained. It can beobserved that the predictions arereasonably good, both in terms of thestagnation heat transfer coefficientand the variation of the heat transfercoefficients along the radial line.

Morris et al. (1996) has also carriedout predictions for the H/d = 4 at Re= 8500 and Re = 13000 using thesimple k-ε model and found that thepredictions obtained were giving poorcomparison with experimental values.From Table 3.1, it is observed that theerrors for these cases using the k-εmodel were found to be 56.6% and46.9% respectively.

Observing this, Morris et al.1 thenproceeded to use the different TPF inthe post processing as discussed earlier(for the H/d = 4 at Re = 10000 case).The results that they obtained usingthis TPF are also shown in Table 3.1.It may be noticed that for H/d = 4 forRe = 8500, 10000 and 13000, thepredictions obtained by these TPFserrors vary from 2.9% to 16.3%.However, from the presentpredictions, error varies from 0.1% to4.6% with the simple standard highReynolds number k-ε model. Thisclearly shows that a proper grid sizeand pattern has to be selected for goodpredictions. Figure 3.5 shows thecomparison of experimental and heattransfer coefficients distributionpredicted by G-L TPF with presentpredictions for H/d = 4 at Re = 8500,10000 and 13000.

Table 3.2, shows the comparison ofpredicted and experimental averagedheat transfer coefficients for the sameearlier considered cases. Themaximum deviation in the case of G-L TPF is 20.5% whereas for Malhotra-Kang it is 15.6%. Wassel-Cattonshows a maximum deviation of 27.1%whereas Kays TPF shows a maximumdeviation of 23.8%. The results byMorris et al.1 with FLUENTTM usingk-e model gives a maximum error of60.4% and a minimum of 45% forall the cases considered.

However, the results obtained fromthe present predictions are superior to

Figure 3.3:Comparison of FLUENTTM (Present) Result, with Earlier Published Results andthe Experimental Results

r/d

hW

/(m

2k)

2000

3000

4000

5000

6000

7000

8000

9000

10000

11000

12000

0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00

Experiment (Morris et al., 1996)Malhotra & Kang (Morris et al., 1996)Kays (Morris et al., 1996)

Gibson & Launder (Morris et al., 1996)Wassel & Catton (Morris et al., 1996)FLUENT (Morris et al., 1996)FLUENT (Present)

d = 3.18 mmRe = 10000

H/d = 4l/d = 2

Figure 3.4:Comparison of Experimental and Predicted Heat Transfer CoefficientsDistribution Using the Graded Grid for H/d = 4 at Re = 8500, 10000 and 13000

4000

5000

6000

7000

8000

9000

10000

11000

0.00 0.50 1.00 1.50 2.00

Experiment : Re = 8500 (Garimella & Rice, 1995)

Experiment : Re = 10000 (Morris et al., 1996)


FLUENT (Present) : Re = 8500



d = 3.18 mm

H/d = 4l/d = 2

r/d

hW

/(m

2k)



all the earlier predictions using themodified turbulent Prandtl numberfunctions mentioned earlier. Theerror varies from 1.9% to 9.9%; inmajority of the cases it is less than2.7%. Thus the simple k-ε modelwith proper grid refinement and withthe non-equilibrium wall functiongives reasonably good predictions forthe averaged heat transfer coefficientsas well. This completes thequantitative validation exercise.

3.3 New Results Generated

for Cases Not Investigated

Earlier

Based on earlier simulations, thefollowing cases are carried out to

Table 3.1:Comparison of Experimental and Predicted Stagnation Heat Transfer Coefficients (W/m2k)

Gibson-Launder * Malhotra-Kang * Wassel-Catton * Kays * FLUENT TM * FLUENT TM (Present)

H/d Re Experiment * Prediction % error Prediction % error Prediction % error Prediction % error Prediction % error Prediction % error

1 8500 7078 7096 -0.3 8012 -13.2 7136 -0.8 7561 -6.8 3959 44.1 6795 4.0

8500 7142 6104 14.5 6880 3.7 4733 33.7 6500 9.0 3250 54.5 7350 -2.9

3 13000 8250 9533 -15.6 10754 -30.4 9559 -15.9 10155 -23.1 4866 41.0 8611 -4.4

8500 7001 5870 16.2 6607 5.6 5883 16.0 6246 10.8 3036 56.6 7326 -4.6

10000 7803 6846 12.3 7717 1.1 6863 12.0 7291 6.6 3540 54.6 7807 -0.1

4 13000 8544 8793 -2.9 9936 -16.3 8812 -3.1 9375 -9.7 4537 46.9 8663 -1.4

*Morris et al. (1996)

Gibson-Launder * Malhotra-Kang * Wassel-Catton * Kays * FLUENT TM * FLUENT TM (Present)

H/d Re Experiment * Prediction % error Prediction % error Prediction % error Prediction % error Prediction % error Prediction % error

1 8500 6607 5475 17.1 5577 15.6 5060 23.4 5278 20.1 2751 58.4 5950 9.9

8500 5972 4749 20.5 4800 19.6 4355 27.1 4550 23.8 2365 60.4 5463 8.5

3 13000 6528 7170 -9.8 7285 -11.6 6693 -2.5 6898 -5.7 3590 45.0 6609 -1.2

8500 5304 4467 15.8 4548 14.3 4120 22.3 4309 18.8 2208 58.4 5446 -2.7

10000 5793 5187 10.5 5307 8.4 4813 16.9 5023 13.3 2560 55.8 5904 -1.9

4 13000 6342 6649 -4.8 6816 -7.5 6214 2.0 6442 -1.6 3258 48.6 6649 -4.8

*Morris et al. (1996)

Table 3.2:Comparison of Experimental and Predicted Averaged Heat Transfer Coefficients (W/m2k)

Figure 3.5:Comparison of Experimental and Heat Transfer Coefficients DistributionPredicted by G-L TPF and Present Predictions for H/d = 4 at Re = 8500, 10000and 13000

r/d

hW

/(m

2k)

* Morris et al.

4000

5000

6000

7000

8000

9000

10000

11000

0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00

d = 3.18 mmH/d = 4

l/d = 2

Experiment* G-L* FLUENT(Present)

Re = 8500

Re = 10000

Re = 13000



generate new results where neitherexperimental data nor numericallypredicted results are available in theliterature;

a) H/d = 2 atRe = 8500 and Re = 10000

b) H/d = 3 at Re = 10000c) H/d = 8 at Re = 10000

These new cases are predicted usingthe same grid refinement method

employed with standard highReynolds number k-ε turbulencemodel and non-equilibrium wallfunction as discussed earlier. Theresults obtained in this section arepresented together with the results ofcases wherever appropriate forcomparison purposes.

Predictions were made first for the caseof H/d = 2 at Re = 8500 and 10000.In this case the grid refinements are

made taking into account theexistence of the secondary maximaobserved in experimental results in theliterature. Refinements are made sothat the primary minima and thesecondary maxima moves closer to thejet axis as the Reynolds number isdecreased; similar to the experimentalobservations. The new predictedresults obtained for these cases arepresented in Figure 3.6. From thefigure it can be observed that thestagnation heat transfer coefficientvariations along the radial line for H/d = 2 at both Re = 8500 and Re =10000 follow the trends as observedfor other cases for which experimentalresults are available.

The present methodology is used tomake predictions for the other twocases mentioned. Figures 3.7 and 3.8show the predicted local heat transfercoefficient variation for H/d = 3 andH/d = 8 at Re = 10000 respectively.In both the figures the predictionsobtained earlier for other casestogether with the correspondingexperimental results are also presented.It is observed that the present methodis able to predict the heat transfercoefficient variations for the H/d = 3at Re = 10000 as well as for the H/d =8 at Re = 10000, which falls inbetween the heat transfer coefficientvariations along the radial distancefrom the jet axis observed for the Re= 8500 and Re = 13000 cases. Thepredicted trends are also similar tothose obtained previously for othercases. However, for the H/d = 3 caseat r/d about 1.50, the local heattransfer predictions obtained are seento coincide with the experimentalvalues of the Re = 8500 case. Thus,with the consistency of the predictionsachieved in this analysis, it is believedthat the heat transfer coefficientvariation prediction for these cases isreasonably good.

r/d

hW

/(m

2k)

5000

5500

6000

6500

7000

7500

8000

8500

9000

9500

10000

0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25





d = 3.18 mmH/d = 2l/d = 2

Figure 3.6:New Result of Heat Transfer Coefficients Distribution with Present GridGrading Methodology for H/d = 2 at Re = 8500 and Re = 10000

r/d

hW

/(m

2k)

4000

4500

5000

5500

6000

6500

7000

7500

8000

8500

9000

9500

10000

0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00

Experiment : Re = 8500 (Morris et al., 1996)Experiment : Re = 13000 (Morris et al., 1996)FLUENT (Present) : Re = 8500FLUENT (Present) : Re = 10000FLUENT (Present) : Re = 13000

d = 3.18 mmH/d = 3l/d = 2

Figure 3.7:Comparison of Experimental and Heat Transfer Coefficients Distribution withPresent Grid Grading Methodology for H/d = 3 at Re = 10000



3.4 Correlations

The respective correlations obtainedfor the stagnation Nusselt number andthe area averaged Nusselt number arepresented below,

Nu0 = 0.117 Re0.418 Pr0.4 Hd

0.0481d

4.203

(3.1)

Nuave = 0.082Re0.477 Pr0.4 Hd

−0.0741d

3.747

(3.2)

The maximum percentage deviationof the correlated values from thepredicted results is found to be 1.7%and 2.2% for the stagnation Nusseltnumber and area averaged Nusseltnumber respectively.

4 CONCLUSIONS

The local heat transfer coefficientdistribution on a heated target platedue to a normally impinging,axisymmetric, confined and

submerged liquid jet is determinednumerically for a nozzle diameter of3.18 mm with turbulent jet Reynoldsnumbers ranging from 8500 to 13000for several nozzle to target platespacing from H/d = 1 to H/d = 8.Standard high Reynolds number k-emodel is used to obtain thepredictions. The predicted results forthe above cases are compared with theexperimental results as well as withnumerical results given by otherresearchers. Some new data aregenerated for those cases for whichneither experimental data nornumerical predictions are available inthe literature.

From the study the followingconclusions are made;i. Standard high Reynolds numbers

k-ε model provides gooddistribution of heat transfercoefficient as long as a well gradedgrid discretisation is used.

ii. The stagnation heat transfercoefficients with a maximumdeviation less than 5% and areaaveraged heat transfer coefficientwith a maximum deviation lowerthan 10% is predicted.

Figure 3.8:Comparison of Experimental and Heat Transfer Coefficients Distribution withPresent Grid Grading Methodology for H/d = 8 at Re = 10000

r/d

hW

/(m

2k)

3000

4000

5000

6000

7000

8000

9000

10000

11000

0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80 2.00

Experiment : Re = 8500 (Garimella & Rice, 1995)Experiment : Re = 13000 (Garimella & Rice, 1995)FLUENT (Present) : Re = 8500FLUENT (Present) : Re = 10000FLUENT (Present) : Re = 13000

d = 3.18 mmH/d = 8l/d = 2

iii. The heat transfer coefficientincreases as the Reynolds numberis increased.

iv. The nozzle to target plate spacing,H/d has a considerable effect onthe prediction of heat transfercoefficients.

The analysis carried out here showsthe importance of proper griddistribution in order to obtain goodnumerical prediction in jetimpingement cooling problems. Thenon-uniform structured grids utilizedin this work, were able to provide goodpredictions. A much better approachwould be to use unstructured gridwith similar grid grading.

From the analysis carried out, it isobserved that there is a need to go fora commercial code with adaptive andautomatic grid generation, which aidsin providing a proper grid for thesolution of the jet impingementproblem, once we start the solutionwith an initial coarse mesh.

REFERENCES

[1] G K Morris, S V Garimella and R SAmano, “Prediction of Jet ImpingementHeat Transfer Using a Hybrid WallTreatment with Different TurbulentPrandtl Number Functions”, J. HeatTransfer, ASME, 118, p. 562-569. 1996.

[2] M Behnia, S Parneix and P A Durbin,“Prediction of Heat Transfer in an JetImpinging on a Flat Plate”, Int. J. HeatMass Transfer, 41(12), p. 1845-1855.1998.

[3] M Behnia, S Parneix, Y Shabany and PA Durbin, “Numerical Study ofTurbulent Heat Transfer in Confined andUnconfined Impinging Jets”, Int. J. FluidFlow, 20, p. 1-9. 1999.

[4] T Tanaka, H Matsushima, A Uek and TAtarashi, “Numerical Simulation ofImpinging Air Cooling from LSIPackages with Large Plate Fins by thePenalty Finite Element Method”, J.Electronic Packaging, ASME, 119, p. 73-77. 1997.



[5] I Rushyendran. Jet Impingement Coolingin Electronic Packages UsingComputational Fluid Dynamics. Thesis(MSc). University of Science, Malaysia,Penang. 2001.

[6] Fluent, FLUENTTM User’s Guide, 1-4,Release 4.3, Fluent Inc., Lebanon, NewHampshire. 1995.

[7] G K Morris and S V Garimella, “Orificeand Impingement Flow Fields inConfined Jet Impingement”, J. ElectronicPackaging, ASME, 120(1), p. 68-72.1998.

[8] J A Fitzgerald and S V Garimella, “FlowField Effects on Heat Transfer inConfined Jet Impingement”, J. Heat MassTransfer, ASME, 119, p. 630-632. 1997.

[9] S V Garimella, and R A Rice, “Confinedand Submerged Liquid Jet ImpingementHeat Transfer”, J. Heat Transfer, ASME,117, p. 871-877. 1995.

[10] S V Garimella and B Nenaydykh,“Nozzle-Geometry Effect in Liquid JetImpingement Heat Transfer”, Int. J. HeatMass Transfer, ASME, 39(14), p. 2915-2923. 1996.

[11] J A Fitzgerald and S V Garimella,“Visualization of the Flow Field in aConfined and Submerged ImpingingJet”, Procs. 32nd National Heat TransferConf., Baltimore, Maryland, ASMEHTD-346(8), p. 93-99. 1997.

[12] J A Fitzgerald and S V Garimella, “AStudy of the Flow Field of a Confinedand Submerged Impinging Jet”, Int. J.Heat Mass Transfer, 41, p. 1025-1034.1998.

[13] G K Morris, S V Garimella and J AFitzgerald, “Flow Field Prediction inSubmerged and Confined JetImpingement Using the Reynolds StressModel”, J. Electronic Packaging, ASME,121, p. 255-262. 1999.

[14] L A Brignoni and S V Garimella, “Effectof Nozzle-Inlet Chamfering on PressureDrop and Heat Transfer in Confined AirJet Impingement”, Int. J. Heat MassTransfer, 43, p. 1133-1139. 2000.

[15] D W Colucci and R Viskanta, “Effect ofNozzle Geometry on Local ConvectiveHeat Transfer to a Confined ImpingingAir Jet”, Exp. Thermal Fluid Sci., 13, p.71-80. 1996.

[16] V P Schroeder and S V Garimella, “HeatTransfer form a Discrete Heat Source inConfined Air Jet Impingement”, Proc.11th Int. Heat Transfer Conf., Kyongju,Korea, 5, p. 23-28. 1998.

[17] V P Schroeder and S V Garimella, “HeatTransfer in the Confined Impingementof Multiple Air Jets”, Proc. ASME HeatTransfer Div. IMECE, Anaheim,California, HTD-1(1), p. 183-190.1998.

[18] L A Brignoni and S V Garimella,“Experimental Optimisation of ConfinedAir Jet Impingement on a Pin Fin HeatSink”, IEEE Trans. Components PackagingTech., 22(3), p. 399-404. 1999.

[19] H A El-Sheikh and S V Garimella, “HeatTransfer from Pin Fin Heat Sink underMultiple Impinging Jets”, IEEE Trans.Components Packaging Tech., 23(1), p.113-120. 2000.

[20] Y Kondo, M Behnia, W Nakayama andH Matsushima, “Optimisation of FinnedHeat Sinks for Impinging Cooling ofElectronic Packages”, J. ElectronicPackaging, ASME, 120, p. 259-266.1998.

NOTES FOR CONTRIBUTORS

Instructions to Authors

Authors of articles that fit the aims,scopes and policies of this journal areinvited to submit soft and hard copiesto the editor. Paper should be writtenin English. Authors are encouragedto obtain assistance in the writing andediting of their papers prior tosubmission. For papers presented orpublished elsewhere, also include thedetails of the conference or seminar.

Manuscript should be prepared inaccordance with the following:1. The text should be preceded by

a short abstract of 50-100 wordsand four or so keywords.

2. The manuscript must be typedon one side of the paper, double-spaced throughout with widemargins not exceeding 3,500words although exceptions willbe made.

3. Figures and tables have to belabelled and should be includedin the text. Authors are advisedto refer to recent issues of thejournals to obtain the format forreferences.

4. Footnotes should be kept to aminimum and be as brief aspossible; they must benumbered consecutively.

5. Special care should be given tothe preparation of the drawingsfor the figures and diagrams.Except for a reduction in size,they will appear in the finalprinting in exactly the sameform as submitted by the author.

6. Reference should be indicatedby the authors’ last names andyear of publications.

Publisher



31750 Tronoh

Perak Darul Ridzuan

MALAYSIA

Relay Feedback Auto-tuning Controller For Waste Water Treatment

V R Radhakrishnan



Ahmad Majdi Abdul Rani










Jariah Mohamad Juoi

New Radiation Grafted And Sulfonated Membranes

For PEM Fuel Cell

Mohamed Mahmoud Nasef, Hamdani Saidi and Hussin Mohd Nor

Neural Fuzzy Based 3d Anti-sway Modelling And

Control Design For Overhead Cranes

M Mahfouf

Scanning Electron Microscopy of Anisotropic Etching in Fabrication

of VMOS (Vertical Metal-Oxide Semiconductor) Transistor

Norani M Mohamed, Kamarulazizi Ibrahim & Leong Yew Wei


Mohd Shiraz Aris, I Rushyendran, G A Quadir & K N Seetharamu

P L A T F O R M

Volume 2 Number 2 Jul - Dec 2001

2

6

14

20

24

29

33

38

47

53

VO

LU

ME T

WO

N

UM

BER

TW

O J

ULY

- DEC

EM

BER

200

1

VOLUME TWO NUMBER TWO JULY - DECEMBER 2001

Volume 2 Number 2 Jul - Dec 2001press.utp.edu.my/wp-content/uploads/2018/12/Platform-v2n...NOTES FOR...

Documents

Transcript of Volume 2 Number 2 Jul - Dec 2001press.utp.edu.my/wp-content/uploads/2018/12/Platform-v2n...NOTES FOR...