Diagnosis of Speed Sensor Failure in InductionMotor Drive

F. Zidani', D. Diallo2,Senior Member IEEE, M.E.H. Benbouzid3, Senior Member IEEE, Eric Berthelot2'Laboratoire LSPIE, Electrical Engineering Department, University of Batna

2Laboratoire de Genie Electrique de Paris (LGEP)/ SPEE Labs, CNRS UMR 8507;Supelec, Univ Pierre et Marie Curie P6, Univ Paris Sud PI1, Plateau du Moulon, 91192 Gif-Sur-Yvette, France

E-mail: ddiallo ieee.or3Laboratoire d'Ingenierie Mecanique et Electrique (LIME), University of Western Brittany

IUT de Brest, Rue de Kergoat - BP 93169, 29231 Brest Cedex 3, France

Abstract - A large number of adjustable - speed drives in monitoring and fault/diagnosis schemes the incipient faults canindustry and emerging applications such as automotive (EV be detected in their early stages. Thus maintenances andor HEV) require high dynamic performances, robustness downtime expenses can be reduced and reliability can beagainst parameter variation and also reliability. parameter improved.detuning and mechanical speed sensor faults lead to adeterioration of the performances and even to instability. The indirect field oriented control (IFOC) is the most commontherefore condition monitoring is becoming mandatory in control method for implementing high performance inductionthose sensitive applications. motor drive [8-9]. In general in the IFOC technique the shaftthe objective of this contribution is to study the feasibility of speed, that is usually measured, and the slip speed calculateddetection and diagnosis of the mechanical speed sensor from the machine equations are added to find the rotor fluxfaults in an induction motor drive. using knowledge of the vector position which is a critical issue. Therefore a close lookmotor condition, the proposed technique based on fuzzy is required on parameter variations and speed sensor fault whichlogic is applied to discriminate load and parameter should lead to the controller detuning and even to the instabilityvariations from the speed sensor faults. both simulation and ofthe drive.experimental results are presented in terms of accuracy in Parameter variations are usually tracked by adaptivethe detection of speed sensor faults and knowledge mechanisms using additional sensors (stator temperature forextraction feasibility. example) or specific algorithms based on weak assumptions like

no saturation in the core. The drawbacks are an increase of theI. Introduction computational burden and a low accuracy without a significant

benefit in parameter adaptation.The induction machines are work horse of modem On the other hand the mechanical sensor defect leads to a

industries because of various technical and economical reasons. wrong computation of the rotating reference frame. As aBut these machines face various stresses during operating consequence, depending on the torque to rotor flux ratio theconditions which may lead to failures. A large number of torque control is detuned and this may lead to a instability if theadjustable speed drives (ASD) in industry and emerging deviation angle between the machine reference frame and theapplications such as automotive (EV or 1EV) require high controller one is above a certain limit.dynamic performances, robustness against parameter variations Hence one should pay attention to the deterioration of theand also reliability. However the ASD are very sensitive to mechanical speed sensor.faults which may occur within the power converter, theelectrical and mechanical components of the machine but also Fuzzy logic technology can be used to provide inexpensive butto the mechanical speed sensor. Hence the condition monitoring effective fault detection mechanism by an adequate analysis ofis becoming mandatory in sensitive applications like automotive the recorded data [16-19]. Fuzzy logic technique is easy tofor example [1-7]. Moreover on-line fault diagnosis technology extend and modify by the incorporation of new data orof incipient faults for induction drives is rapidly emerging to information. Based on measurement data this approach may aidavoid the unpredictable failure [1]. Different invasive and non in data management for fault diagnostics purpose. Usinginvasive approaches for motor incipient fault knowledge of the motor condition, the proposed technique isdetection/diagnosis have been reported [11-15]. Many of the applied to separate machine variations parameters effects andmotor incipient fault detection/diagnosis schemes can be sensor position faults.applied non invasively on-line without the need of expensive The paper is organized as follows. In section II, themonitoring equipment by using a microprocessor. With proper experimental bench under test is described and some results are

1-4244-0743-5/07/$20.00 ©2007 IEEE 1680

presented to validate the IFOC performance in normal operatingcondition, namely without parameter detuning or mechanical Asensor fault. In section 111, the data collected from numerousexperiments with parameter detuning and angle deviation are -analyzed to extract the knowledge which is introduced in thefuzzy detection block presented in section IV. The results of the 0I6detection block are presented in section V and eventually a 0,4conclusion is made. | M_0,2 _

-0,6 -0,4 -0,2 0 0,2 0,4 0,6

II. DESCRIPTION OF THEEXPERIMENTAL Time(s)SETUP

Fig. 1 shows the experimental setup. The squirrel-cageinduction motor and inverter data are given in the Appendix. Af_The currents flowing in the stator windings are measured withtwo Hall Effect current sensors and a 4098 points pulseincremental encoder is used as position sensor. The IFOC isimplemented on a dSpaceg 1103 board using the Matlab-Simulinkg software package. Fig. 2 shows the speed reversal -C 5 -0,3 -0 1 10 A 0.3 ° 5tracking and the rotor flux module at rated load. This -20 \establishes the performances of the drive. 30

-40

Time(s)|~~~~~~~~~~~~~~~~~~~~~~ v W W,= ^- ~~~~~~~~~~~~~~~~~~~~~~~~~~T ime(s)

Fig.2. Rotor flux, rotor speed and stator current at rated load.

III. DATAANAL YSIS

In this section the rotor speed and stator currentevolutions are analyzed when a parameter variation or aspeed sensor fault occur at different operating points. Theissue is to be able to discriminate those two effects.Therefore, the sensitivity of the IM has been studied to the

vaiutationg the trandtesientoanlbeh viro spevaiations

cniusreding stheadstratsent Tblehvo ofsumrzspehedvresutsoFig. 1: Experimental setuptran.sient = Qref - tran.sient and current error

1681

AIs-transient = isref -istransient for four operating points.The transient analysis examines the error AQtransient and

AIs-transient in response to the parameter variation. Thereference value corresponds to the nominal case in steadystate.

TABLE 1

Experimental Data: Parameter sensitivity" Rotor resistance variation"

A Rr (Oo) =re5rd/s T Qfre=lO rd/s Qrrf=40 rd/s Qrrf=60 rd/s

APansi( %s-as rAnsi( N A~4r~ LY % AQ-trans[ r/\fgan.sie% ^E fransi% /AOO s-fransi $ansie% ransl AmansM%20% 32 13.24 37 11.5 4.62 1.48 0.016 0.21

40% 49 27.02 42.5 15 10.87 12.97 0.016 0.21

60% 53.7 25 13.42 22.57 0.016 0.21

80% 78.7 35.5 54.67 31.48 0.016 0.21

100% 98.7 50 69.7 46.59 13.55 10.86

B) Speed sensorfailure analysis: sensitivity to stator The rotor flux and the electromagnetic torque are kept constant.

angle variation Table 2 summarizes the collected experimental data AQtransientThe mechanical sensor fault is emulated by introducimg and Ms-transient , the speed and current variations and the

in the IFOC algorithm different angle variations as an error deviation A0s for four different speeds.AO0S in the determination of the rotor flux vector at steady state.

TABLE 2

Experimental data: Speed sensor fault analysis

Qref=5rd/s Qref=O rd/s Qref=40 rd/s Qref=6O rd/sAO (0) o AX%

-2 144.2 65.96 91.3 66.66 11.12 162.25 11.41 17.98

-1 108.4 46.98 66 27.38 7.07 50 7.25 15.02

-05 108.6 19.57 56.8 26.78 5.55 23.75 5.61 8.86

0~~5 133.2 13.55 68.7 16.07 4.4 11.25 2.88 37.9

1 t )6.2 18.67 r 70. 23c 5.22 A425 3.75 5394

L2 136.2 28.31 100.5 5.11 7.2 141.25 4.16 58.12

It is found that the largest values of AQtransient and As-transient

are due to stator angle variation. This analysis and the data IV. FUZZYBLOCKDESCRIPTION OFSPEEDSENSOR FAULTS

collected will be used in the design of the diagnosis procedure Fzyssesrl nasto ue.Teerls hl

which is described in the following paragraph. superficially similar, allow the input to be fuzzy (i.e., more likethe way humans tend to express their way of thinking). Thus, a

1682

power engineer might refer to an electrical machine as "some- If (el is LIM/H and e2 is ZE) then FS is NF. This casewhat secure" or a "little overloaded." This linguistic input can ponds in oube expressed directly by a fuzzy system. Therefore, the natural c es L

r eriment toeload iaiNefcs.format greatly eases the interface between the engineer's If (e1 is LMH and e2is L) then ES iS NE. This caseknowledge and the domain expert. Furthermore, infinite corresponds in our experiment and according to data analysis tograduations of truth are allowed, a characteristic that accurately parameter variation effects.mirrors the real world, where decisions are seldom "crisp"

The processes of speed sensor defects detection can be divided~If ( el is H and e2 is VH) then FS is NF. This case correspondsinto four blocks: in our experiment and according to data analysis to machine

o The data acquisition. acceleration/deceleration.o The signal conditioning as applied to measured

quantities to provide the input quantities to the If ( el is L and e2 is VH) then FS is HF. This case correspondsevaluation method. in our experiment and according to data analysis as a severe

o The evaluation method is the fault detection fault of the speed sensor.technique based on fuzzy logic.

o The last block is fault assessment which provides V. FUZZYAPPROACH VALIDATIONan indication of the fault and its severity.

In this work a fuzzy system is used to strictly tie the operatingconditions of the motor with the diagnosis that can be given by To evaluate the diagnosis method a deviation ofexamination of the data analysis given in §111. A60 = +20 or A6, =-1 is considered. Fig. 3 shows the

The inputs of the fuzzy evaluation block are chosen and are fuzzy output (fault severity FS (/o)). The diagnosis gives a high

defined asel = Q2ref -_Q(k) and e2 = I - (HF) fault severity (FS 800%) for the first case and a medium1 ref 2sref 's *severity index in the second case (FS 50_ O). It can be noticed

The internal structure of this block is chosen similar to that of a on Fig 3.c that when a rotor resistance variationtR =75O% Rro is considee,tefutsvrydceasfuzzy logic controller (fuzzification, inference engine and r mom ered, the fault severity decreasessignificantly (FS 20%) and can be taken as NF.

defuzzification). Both errors are normalized by dividing with

the respective gains factors. For the purpose of this study, the 09

fuzzy evaluation block output is the fault severity "FS" of the 0.8 - -_0.7 LLLLLLLL

speed sensor. The selected membership functions are based on 0.6 _ _ _

the previously collected experimental data. The amplitudes of o4 I I I I I I I

the input variables have been defined as: ZE: zero, L: light, M: 0.3 - r T T T T

Medium, H: high and VH: Very high. 0.1 I I IXo 500 000 1500 2000 22500 33000 3500 4000 4500 5000

Fuzzy sets associated to the output have been defined as: NF:

no fault (fault severity is considered as insignificant/negligible), a) AOs + 20

LF: light fault, MF: Medium fault and HF: High fault.09 -- --1---1-- -- --1--

TABLE 3 0.8Rule base of fuzzy detection block

0I I I I I I

Le2~~~~~~~ ~ ~~ ~~~~~~~~~~ I IL I I I05 I11e2 ~~L M H 0.5 . 4--p

ZE NF NF NF 0.4-_-.

L NF NF NF 0.3 -T - -T-_ -- -.-- -T--

M NF NF NF 0.2 ---

H MF NF HF 0 -<-----_I L I I I

VHHFTLF NE 0T500 1000 -1500 2000 2500 3000 3500 4000 4500 5000

As illustration: b)AO\s =- l1°

1683

1 -I I I

_ 171 Consoli, F. Gennaro, A. Raciti and A. Testa, "Fuzzy logicI application to pre-fault diagnosis of induction motors, " In

I III I I I I I --Proc. IEEE-L4S Annual Meeting, pp. 249-254, 1998.0.7--- 1T - - - V. Youk, " Hybrids: then and now, " in Proc. IEEE0.6--_8 V._Youk_____ _____ ___

o--2I-- k ---- - 44- Spectrum, Vol.32, pp.16-21, July 1995.0.4 ________ I------------- I-- ___ [9] J.G.W. West, "DC, induction, reluctance and PM motors

I I I I I I I TI s, C0.3- for electric vehicles, " Power Eng. Journal, pp. 77-88,0.2 --I-- -- l --t --e---I---I-- e - - - --

0.1 ------11

April 1994.0 5 00 1000 15002000 2500 3000 3500 4000 4500 5000 1101 C.D.E. Schaeffer, T. Couraud and M Zaim, " Simulation

and detection of faults in induction ...' Internationalc) ARr = 75% Rrnom Symposium on Diagnostics for electrical Machines, Power

Fig. 3. Fuzzy output of the detection block (FS). electronics and Drives, pp.3-8, 1997.111 M.Y. Chow, Methodologies of using artificial neural

VI. CONCLUSION network and fuzzy logic technologies for motor incipient

This paper has demonstrated that it is feasible to detect a fault detection. Singapore. World Scientific, 1997.speed sensor failure and moreover to discriminate it from [121A.K. Sood, A.A. Fahs, and N.A. Henan " Engine faultparameter detuning effects on the drive performances. The analysis, Part I: statistical methods, " IEEE Transactions onproposed technique uses available system variable signals such Industry Electron., vol. 32, pp. 294-300, Nov. 1985.as stator current and motor speed. The experimental results are [131 A.K. Sood, A.A. Fahs, and N.A. Henan Engine faultvery promising. On the other hand, the fuzzy validation analysis, Part II: Parameter estimation approach, " IEEEapproach is simple and easy to implement. Transactions on Industry Electronics, vol. 32, pp. 294-300,

Nov. 1985.The proposed scheme is recommended for application when 1141R. Isermann "Process fault diagnosis based on modelingdetection and isolation of speed sensor fault is preferable sinceit keeps the drive partially operative. This is the case of fault pp. 387-404e 1984.tolerant controllers in sensitive applications where a high level pp .387-404,f1984.

is madaor like transportaionsystem. [151]R. Isermann and B. Freyermuth, " Process fault diagnosisof reliability is mandatory like transportation systems. based on process model knowledge - Part I, " Journal ofREFERENCES Dynam. Syst., Meas., Contr., Vol. 113, pp. 620-626, 1991.

[16] P.V. Goode and M.Y. Chow, " Using a neural/Fuzzy[11 Seddique, G.S. Yadava and B. Singh, "Applications of system to extract heuristic knowledge of incipient faults in

artificial intelligence techniques for induction machine induction motors: Part 1- Methodology, " IEEE Trans. Ind.stator diagnosis, " in Proc. IEEE SDEMPED 2003 Electron., Vol. 42, pp. 131-138, Apr. 1995.Conference-symposium on Diagnosis for Electric 117] C.T. Lin and C.G. LEE, " Neural network- based fuzzyMachines, Power Electronics and Drives, pp. 29-34, 24626 logic control and decision system, " IEEE Trans. Comut.,,August 2003. Vol. l 13, pp. 627-633, 1991.

[2] G.B. Kliman and J. Stein, 'Induction motor fault detection [18] R. Isermann and B. Freyermuth, " Process fault diagnosisvia passive current monitoring- A brief survey, " in Proc. based on process model knowledge - Part II, " Journal of44th Meeting of the Mechanical Failures Prevention Group, Dynam. Syst., Meas., Contr., Vol. 113, pp. 627-633, 1991.pp. 49-65, April 1990. 119] S. Altug, M.Y. Chow and H.J. Trussell, " Fuzzy inference

[3] W.T. Thomson, " Research and development of on-line systems implemented on neural architectures for motordiagnostic monitoring systems for electrical machines, " fault detection and diagnosis, " IEEE Trans. Ind Electron.,IEE Coll. Instrumentation ofRotating Electrical Machines, Vol. 46, N06, pp. 1069-1079, Dec. 1999.pp. 5-7, 1991.

141 M.E.H. Benbouzid, " Bibliography on induction motors APPENDIXfaults detection and diagnosis, " IEEE Trans. Energy INDUCTIONMOTOR PARAMETERSConversion, Vol. 14, pp. 1065-1073, Dec. 1999.

[5] S. Nandi and H.A. Toliyat, " Condition monitoring and Rated values: Power 1.1kW, Voltage (A/Y) 220/380V, Currentfault diagnosis on electrical machines - a review, " In Proc. (A/Y)4.7/2.7 A, Speed 1400 tr/minIEEE-L4S Annual Meeting 1999, Vol.1, pp. 197-204, 1999.

[6] F. Zidani, M.E.H. Benbouzid, D. Diallo and M.S. Nait said, fs =50 Hz, osnom= 1.22 Wb, Cmom =7 Nm" Induction motor stator faults diagnosis by a current Parameters: Rs= 8 Q, Rr=4 Q, Ls =0.47 H, v= 0.075, J=Concordia pattern-based fuzzy decision system, " IEEE 0 06 Kg.m2,f= 0.04 Nm.sTrans. On Energy Conversion, Vol.18, N°4, pp. 469-475,'Dec. 2003.

1684