MANTIS · 22 ATLAS Atlas Copco Airpower Nv BEN BE 23 3E 3e Nv BEN BE 24 PCL Philips Consumer...

Cyber Physical System based Proactive Collaborative Maintenance

MANTIS

D1.2 Consolidated State-of-the-Art of Sensor-based Proactive Maintenance Appendix 4:

New sensor technology Work Package WP1 - Service platform architecture requirement definition. Scenarios

and use cases descriptions

Version 1.0

Contractual Date of Delivery 30/04/2016

Actual Date of Delivery 03/06/2016

Dissemination Level Public

Responsible Erkki Jantunen

Contributors Mikel Anasagasti (MONDRAGON), Lidia Godoy (ACCIONA), Rafael Socorro (ACCIONA), Raquel García (ACCIONA), Aitzol Iturrospe (MGEP), Mikel Viguera (FARR)

D1.2 Consolidated State-of-the-Art of Sensor-based Proactive Maintenance 662189 – MANTIS – ECSEL-2014-1

The MANTIS consortium consists of:

Num. Short Name Legal Name Role Country 1 MGEP Mondragon Goi Eskola Politeknikoa J.M.A. S.Coop. CO ES 2 MONDRAGON Mondragon Corporacion Cooperativa S.Coop. BEN ES 3 IKERLAN Ikerlan S.Coop. BEN ES 4 TEKNIKER Fundacion Tekniker BEN ES 5 FARR Fagor Arrasate S.Coop. BEN ES 5.1 KONIKER Koniker S.Coop. TP ES 6 GOIZPER Goizper S.Coop. BEN ES 7 ACCIONA Acciona Infraestructuras S.A. BEN ES 8 MSI Mondragon Sistemas De Informacion S.Coop. BEN ES 9 VTT Teknologian Tutkimuskeskus VTT Oy BEN FI 10 LUAS Lapin Ammattikorkeakoulu Oy BEN FI 11 NOME Nome Oy BEN FI 12 FORTUM Fortum Power And Heat Oy BEN FI 13 SQ Solteq Oyj BEN FI 14 WAPICE Wapice Oy BEN FI 15 AAU Aalborg Universitet BEN DK 16 DANFOSS Danfoss A/S BEN DK 17 UNIV Universal Foundation A/S BEN DK 18 HGE Hg Electric A/S BEN DK 19 VESTAS Vestas Wind Systems A/S BEN DK 20 SIRRIS Sirris Het Collectief Centrum Van De Technologische Industrie BEN BE 21 ILIAS Ilias Solutions Nv BEN BE 22 ATLAS Atlas Copco Airpower Nv BEN BE 23 3E 3e Nv BEN BE 24 PCL Philips Consumer Lifestyle B.V. BEN NL 25 PHC Philips Medical Systems Nederland B.V. BEN NL 26 PHILIPS Philips Electronics Nederland B.V. BEN NL 27 S&T Science and Technology B.V. BEN NL 28 TU/E Technische Universiteit Eindhoven BEN NL 29 RUG Rijksuniversiteit Groningen BEN NL 30 UNINOVA UNINOVA - Instituto de Desenvolvimento de Novas Tecnologias BEN PT 31 ISEP Instituto Superior de Engenharia do Porto BEN PT 32 INESC Instituto de Engenharia de Sistemas e Computadores do Porto BEN PT 33 ADIRA ADIRA - Metal Forming Solutions S.A. BEN PT 34 ASTS Ansaldo STS S.p.A. BEN IT 35 CINI Consorzio Interuniversitario Nazionale per l’Informatica BEN IT 36 AIT Austrial Institute of Technology GmbH BEN AT 37 HBM Hottinger Baldwni Messtechnik GmbH BEN AT 38 INNOTEC Innovative Technology and Science Limited BEN UK 39 AITIA AITIA International Inc. BEN HU 40 BME Budaperst University of Technology and Economics BEN HU 41 JSI Josef Stefan Institute BEN SI 42 XLAB XLAB d.o.o. BEN SI 43 FHG Fraunhofer Institute for Experimental Software Engineering IESE BEN DE 44 M2X M2Xpert GmbH & Co KG BEN DE 45 STILL STILL GMBH BEN DE 46 BOSCH Robert Bosch GMbH BEN DE 47 LIEBHERR Liebherr-Hydraulikbagger GmbH BEN DE

ii http://www.mantis-project.eu MANTIS

662189 – MANTIS – ECSEL-2014-1 D1.2 Consolidated State-of-the-Art of Sensor-based Proactive Maintenance

Document Revisions & Quality Assurance

Revisions:

Version Date By Overview

0.1 24.8.2015 Lidia Godoy (ACCIONA) First draft

0.2 4.9.2015 Aitzol Iturrospe (MGEP) Added Sensing strategies in manufacturing processes

0.3 8.9.2015 Lidia Godoy (ACCIONA) Added Abstract, Introduction and Conclusions

0.4 25.9.2015 Mikel Viguera (FARR) More sensor information in press forming

0.5 28.9.2015 Lidia Godoy (ACCIONA) Added Contributors

0.6 5.10.2015 Riku Salokangas Added contractual date of delivery etc.

1.0 02/06/2016 Mikel Muxika (MGEP) Format correction

Deliverable info update

http://www.mantis-project.eu iii MANTIS


Abstract

This appendix provides an overview of new sensor technology, highlighting the specific features of this technology and their advances in this field. Currently there exists a tendency to technology miniaturization, an example of these are Micro Electro Mechanical System sensors (MEMS) that consist of miniature devices which are able to sense, generate a signal and process it. Furthermore, it is essential to mention the wireless connection sensors, bidirectional data flow and the ability to contactless sensing of some devices for several applications. These are the main characteristics which new sensor technology is focused.

In order to obtain more comprehensive understanding of this technology, the document presents an example of new sensor implemented in a real scenario such. This section shows the necessary elements which forming a sensor node (also named mote) and some parameters to measure with this technology in a production plant. Finally, sensing strategies in manufacturing processes are described, explaining how the sensors based on process monitoring systems work.

iv http://www.mantis-project.eu MANTIS


Table of Contents

1 Introduction ............................................................................................................................... 2 2 Miniature size ............................................................................................................................ 3

2.1 MEMS (Micro Electro Mechanical System) ....................................................................................... 4 3 Wireless connection sensors ........................................................................................................ 5

3.1 Wireless Sensor Network – WSN ...................................................................................................... 6 4 Contactless sensing .................................................................................................................... 7 5 Bidirectional data flow ................................................................................................................ 8 6 SoA New sensor technology in Pultrusion Line ............................................................................. 9 7 Sensing strategies in manufacturing processes ............................................................................ 17

7.1 Other sensors in Press forming machines ....................................................................................... 25

7.1.1 Electronic pressure sensor .......................................................................................................... 25

7.1.2 Bushing temperature measuring using sensors .......................................................................... 25 8 Conclusions .............................................................................................................................. 27 References ...................................................................................................................................... 28

http://www.mantis-project.eu 1 MANTIS


1 Introduction

The use of new solutions to utility problems on issues from proactive maintenance is focus on meet needs through the implementation of new sensors technology that continuously assess the conditions of aging assets and can provide decision support mechanism for both maintenance and replacement. There exist sensors advanced features to take into consideration when speaking about smart sensors and sensing technologies, i.e. contactless sensing, wireless sensors and network, miniature size, bidirectional data flowí As well as, the technological issues added that provides to an ordinary sensor of new and advances capabilities. This is the case of node sensors ímotesí which is a sensor network equipped with a radio transceiver, a microprocessor, some electronics and an energy source. It is capable of performing some processing, gathering sensory information and communicating with other connected nodes in the network.

In addition, the implementation of new sensors within a real use case is a significant point which must be taking into account in order to carry out sensing strategies in manufacturing process

2 http://www.mantis-project.eu MANTIS


2 Miniature size

Miniaturization has become a big success within the nowadays technologies. The fewer dimensions a sensor has, the shorter answer times it will provide. So, the speed of the generated signal and its processing will become much higher. Moreover, the small size feature provides higher reliability, lower power consumption and a higher integration rate.

The last developments in this area have promoted Micro Electro Mechanical System (MEMS) sensors that consist of miniature devices which are able to sense, generate a signal and process it. There is another similar but not so advanced area, which will be the following step to the micro sensors technology: nanotechnologies (NEMS). It is the same technology as the micro technology, but even in a more reduced space. This area is still under research.

What it is fully clear is that in the future sensors will contain the signal processing algorithms of the specific manufacturing process, of course, apart from the technological features.



2.1 MEMS (Micro Electro Mechanical System)

Micro Electro Mechanical System (MEMS) sensors have been a big challenge for the last two decades. A lot of effort has been invested on it and nowadays there are lots of different MEMS based sensors available in the market. Besides, a large increase of this kind of sensors is predicted for the next few years.

The most used type of sensor containing this top technology are acceleration and pressure sensors, which are used in automotive electronics, medical equipment, computer peripherals, wireless devices or smart portable electronics industry. MEMS accelerometers include capacitive, piezoresistive, electromagnetic, piezoelectric, ferroelectric, optical or tunneling systems. Anyway, the most successful accelerometer is based on capacitive transduction because of its simplicity. All these types of sensors are able to sense the required variable, process it and control the respective application.

These miniature electrical and mechanical components integrated on a single chip are produced in two steps: bulk micromachining and surface micromachining.

Figure 1. MEMS based sensor technology

Some of the most important benefits of MEMS based sensors are the following:

-Low Cost

-Low Power

-Miniaturization

-High performance

-High integration rate



3 Wireless connection sensors

Obviously, if the data is real time and reliable, the wireless data transmission fits the needs of any kind of application. However, especially in rotating applications the development of these kinds of sensors is a great step forward because its difficulty in working with cables.

After the last advantages of Micro Electro Mechanical Systems, two kinds of developments have emerged: Wireless communication and Wireless Sensor Network (WSN).



3.1 Wireless Sensor Network í WSN

It consists of a group of spatially dispersed, dedicated and small devices (nodes). The system can contain thousands of low cost and wireless sensor nodes. These devices are able to sense, process and store captured data. Each one of the nodes contains a radio transceiver with an antenna, a microcontroller, an interfacing electronic circuit and an energy source (battery). The required data in the sensor nodes is compressed and transmitted to a base station (gateway) by means of radio link. Not all the nodes transmit data directly to the gateway; some of them do it through other nodes as it can be visualized in the following picture.

Figure 2. WSN system layout

Usually this system is used to monitor and record physical conditions of the environment and organizing the collected data at a central location.

At the beginning of the WSNs creation, these systems were designed with a military aim, in order to improve military operations. However, recent advances in wireless and electronic technologies, has extended the use of WSN in different areas such as traffic surveillance, health monitoring, environment monitoring, industrial maintenance monitoringí

Focusing in the industry, it is true that wireless data transmission has been used for a long time. Nevertheless, the recent success of wireless sensor technology has provided reliable real time data to Industrial companies, fulfilling the needs and offering what industry is looking for.

WSN can monitor and optimize any production process, machine, component or even quality control in an industrial production facility. As mentioned, WSN has measures environmental conditions, such as temperature, pressure, flow level, speed... which also are captured in industrial processes.



4 Contactless sensing

Contactless measuring eliminates interference effects, wear and tear. It avoids vibrations, high temperatures, dirtiness, working in pressured environmentsí In few words, it works in better conditions, so it provides increased reliability.

There are several types of contactless sensors depending on the variable that should be captured. Each application condition determines which sensor technology is the proper one; nevertheless and as a global rule for contactless sensors, preferably magnetic sensor technologies are utilized:

• Semi-conductor sensors with integrated Hall.

• Semi-conductor sensors with magneto resistive elements.

• Magneto-inductive sensors.

• Temperature sensor devices.

Figure 3. Hall sensor in an application



5 Bidirectional data flow

One of the last yearsí innovations in sensors has been the feedback input to the device: Bidirectional data bus. During years the normal working mode of a sensor was sending data of a captured variable (open loop). Nowadays, it is possible the sensor to receive (close loop) external information in order to be calibrated, tested, configuredí

Figure 4. Bidirectional bus example

This bidirectional data transmission bus is getting more importance nowadays, because it controls the machines variables in a close loop, what improves the system reliability and reduces maintenance costs.



6 SoA New sensor technology in Pultrusion Line

The pultrusion line from ACCIONA is a continuous automated process for manufacturing prototypes in carbon and glass fiber with constant cross-section, operating 24 hours a day in a challenging environment. Logistic operational and maintenance activities involve in the pultrusion line are based on manual procedures. Therefore there is a lack of process control (personnel, energy monitoring and automatized parameters control) and maintenance control of production tools.

Currently, only a few parameters involve in the pultrusion process has been measured through new sensor technology, in order to alert the production machine operators when important deviator appear in the pultrusion line, which can threaten the normal development of a proper production of the profiles. Each parameter can be measured by a sensor node (also named ímoteí) which is equipped with a radio transceiver, a microprocessor, some electronics and an energy source.

The following paragraph show several examples of parameters measured by these motes within pultrusion line and the technology used.

1. Environmental conditions: With the aim of monitoring the temperature and relative humidity of the factory building several wireless nodes with a suitable sensor for such measurements were integrated.

These sensors are installed on the factory building walls to an average height, determined by the

needs of each one of the measurement points.

The measurement points are:

o A sensor near the exit of the pultrusion machine and next to the entrance of the

installations.

o Four sensors distributed along the machine to know the temperature and humidity

conditions during the pultrusion process.

o A sensor next to the resin bath.

o Two sensors on the upper floor of the building.

The Sensor used is SHT11 sensor from Sensirion. It was chosen with the aim of obtaining the

best quality for the measurements.

SHT11 is a digital humidity and temperature sensor. It combines good accuracy and small size.

The capacitive humidity sensor is available up to high volumes and like another sensor type of

the SHTxx family, is fully calibrated and provides a digital output.



Figure 5. SHT11 Image sensor

Table 1 Main features of SHT11 Sensor

Main features of SHT11 Sensor

Energy consumption 80uW (at 12 bit, 3V, 1 measurement / s)

RH operating rate 0-100% RH

Tª operating range (-40, +125) °C

RH response time 8 sec (tau63%)

Output Digital (2 wire interface)

The following pictures show the box used for encapsulating the sensor:

Figure 6. Wireless sensor node developed in the R&D centre ACCIONA Infrastructure

The following image describes the different hardware elements within the box:



Figure 7. Hardware Scheme of ACCIONAís Temperature Sensor Node

2. Production speed: In order to measure the production speed of the pieces fabricated by the pultrusion machine as an indicator of the efficiency and productivity of the process. Also this parameter has implications related to maintenance due to impact in machine cycles. Therefore, a sensing node has been developed for determining production speed. Profiles are manufactured along a linear route in the machine which moves over a cylindrical tube rotating in proportion to the output linear motion of the work piece. It was decided to use the rotation of this tube in order to quantify production speed and store this information for a proper reproducibility of a profile in the future. Each time a magnet, mounted on the cylindrical part, passes through the sensor in the top of the piece, is considered a step. By controlling the time between two steps we know the final velocity.

Figure 8. Operating principles of the sensor



The following pictures show the sensor developed by ACCIONA.

Figure 9. ACCIONAís Magnetic Sensor of speed

Figure 10. Magnetic sensor encapsulated and magnet

3. Resin temperature and level: In order to know the depth of the resin bath and its temperature, a

sensing node with an optical measurement device (for resin level) together with a thermocouple (for resin temperature) has been integrated in the platform. The selection of a thermocouple is justified by the fact that it is cheap to replace in case this is needed, and it gives high precision and toughness. The optical sensor used to monitor the resin height is the Sharp GP2Y0A02YK



Figure 11. Features of the Optoelectronic Sharp Sensor

For the temperature measurement a thermocouple type T was selected. The following pictures show the structure of the sensing node and its final encapsulation.

Figure 12. Main features of the T-Thermocouple Sensor. Source: OMEGA Your One-Stop Source for Process Measurement & Control



Figure 13. Hardware Scheme of the sensors and the acquiring data node

Figure 14. Final encapsulated resin temperature and level sensors

4. Detection of broken fibre line: If during the pultrusion process any damage is detected in any of

the fiber threads which are being used, the operator has to act quickly and replace the thread. In order to support the operator in this surveillance task, ACCIONA has developed a sensor for detection of broken fibers, so that an alert is triggered and sent to the gateway whenever there is an incident, and then the sensing node sends a light alarm to the worker. The figure below shows the specific sensor developed by ACCIONA for detecting the broken fibers, based on photo-emitter diodes and photo-receptors. The sensor can be used to monitor up to 8 fiber threads.



Figure 15. Fibre line Broken Sensor

The following pictures show the structure of the sensing node and its final encapsulation.

Figure 16. Encapsulated Sensor and the acquiring data node

5. Fiber spools weight: By controlling the weight of the fiber spools during the pultrusion process it

can be detected when they should be replaced, achieving an increase in the productivity and efficiency of the process.

For this purpose ACCIONA developed its own sensor based on springs, which is shown in the

picture below.



Figure 17. Prototype of the Accionaís sensor to detect finishing spool (full)

The following figures show the structure of the sensing node which integrates this sensor and its

final encapsulation.

Figure 18. Fibre Spools Weight sensors encapsulated



7 Sensing strategies in manufacturing processes

The monitoring units used to monitor the performance of forming facilities are usually sensors based process monitoring equipment able to measure several channels, usually force and acoustic emission signals. The signals are usually coming from both the facility and the tool used in the process and the equipment evaluates, stroke by stroke, the correctness of those signals.

The sensors based process monitoring systems work in the following way [7].

After getting a correct process set up at the beginning of the production, a learning phase, during which the signals coming from the process are recorded, is carried out. The recorded signals correspond to a process behaving correctly and producing good quality parts.

As a result, two envelope curves are created, an upper envelope curve and a lower envelope curve. These envelope curves represent the limits that distinguish a nominal production (green area in curves of Figure 1) from a faulty production (red areas in curves of Figure 1). The operator sets up the distance between the envelope curves. This distance, named sensitivity depends principally on how stable the process is. For very stable processes, the sensitivity can be smaller while low stable processes will need a broader sensitivity. The drawback of setting up big sensitivities is that some process failures could be hidden inside the envelope curves and not being detected. Therefore, the ideal situation is to get a very stable signal (sensor located close to the point where the signal is generated) in a very stable process. In this case the operator can choose a small sensitivity value and even small process disturbances will be detected by the system.

Figure 12. Sensitivity definition

Once this is achieved, the monitoring system compares stroke by stroke the signals coming from the process with the envelope curves. Whenever at least one of the process signals goes beyond the envelope curves, it is interpreted as a disturbance of the process and a faulty production signal is immediately sent to the press control which can react by stopping the facility, issuing a warning or activating a sorter. As an example next figure shows a fault detection in a forming process. It is clearly seen that one of the signals has gone beyond the envelope curves.

Figure 13: Process fault detection by an acoustic emission signal



There exist several companies worldwide consecrated to the development of monitoring systems for metal forming applications. Among others, next ones can be highlighted:

• Brankamp (Germany) [6] • Schwer and kopka (Germany) [38] • Unidor (Germany) [44] • Siegfried (Czech Republic) [40] • Helm (US) [HELM09] • Imco (US) [20] • Artis (Germany) [2] more consecrated to cutting processes • Kistler (Germany) [KIS07] more consecrated to cutting processes • Montronix (USA) [33] more consecrated to cutting processes • Nordman (Germany) [35] more consecrated to cutting processes • Prometec (Germany) [36] more consecrated to cutting processes

SENSORS TYPE IN METAL FORMING PROCESSES

First monitoring systems in sheet metal forming processes were based on force measurement. The main purpose of those basic force-monitoring systems was to plot the total force that the press needed to form the parts, avoiding this way overloads in the facilities. Therefore, from a security point of view those basic force-monitoring systems gave very good results, but from a process monitoring point of view they did not give much information to the operators. Anyway, process control using force signals has for many years proven to be reliable and relatively low cost. Force signals have provided the following benefits: machine and tool protection, increased productivity and improved product quality. On the other hand, in the cold forming industry there has been and still there is a trend to boost machine output by increasing the running speed with additionally growing demands upon product quality. At the same time, the cold-formed parts are also increasing in their complexity and hence the probability of a higher failure rate is also increased. Thus, there is a need for monitoring devices with improved control accuracy.

Experience has shown that force-monitoring systems can be either too late in recognising, or unable to recognise, cracks in punches, dies, ejectors and spring elements. Experience has also shown that this problem can be overcome with the introduction of AE monitoring. The cracks, tears and breakages produce a short-term acoustic pulse that can be immediately recognised [43].

Figure 14: Comparison between force and AE signals during a crack growth and a punch breakage [43]

Therefore, the AE signals generated by the material during the deformation stage have become a promising technique to monitor and control sheet metal forming processes. It has been well known for centuries that wood and rocks emitted noises when they started cracking or breaking. Later, similar noise was identified during bending of tin bars, which is often known as ítin cryí. Joseph Kaiser, at the Technical University of Munich, made the first pioneering work on AE in 1950. Joseph Kaiser was able to examine the noise emitted by the deformation of materials by means of electronic equipment capable of detecting non-audible signals. One of the observations made was that irreversible processes were involved with this phenomenon, an effect later named the Kaiser effect [27].



AE monitoring techniques, combined with force monitoring techniques, have been and still are nowadays the most successful sensors based process monitoring techniques in metal forming processes. The higher sensitivity of AE allows this technique to detect future failures before producing catastrophic consequences [5]. Therefore, nowadays, sensors based process monitoring systems applied to sheet metal forming processes are based mainly on the measurement of two variables: forces and AE [10]. For example, Figure 2.8 shows the representation of two curves; one measuring the load of a press during a combined blanking-stamping operation and another one measuring the AE signals during a blanking operation.

Figure 15: Force and AE signals in a sheet metal forming process [10]

In fact, and depending the type of process failure to be measured, different approaches can be found in the literature where only type of signal or both signals, force and AE, are recorded. Next table summarizes different works found in bibliography that point out what type of signal is more recommended depending on the failure to be measured in deep drawing processes.

Table 1: Process failures and techniques for their detection in drawing processes

In order to measure the force and AE signals there are two main type of sensors used in industry: the resistive sensors and the piezoelectric sensors.

The resistive sensors, usually strain gages, are used for the measurement of deformations which are related with forces based on a calibration process. They can be attached directly to the facilities and/or attached to a pre-mounted device which is clamped to the facilities. The most common places for their position are the frame of the machine or the connection rods of the machine. This depends on the supplier company. As an example of this type of sensors next figure shows a sensor from the company ME Systeme mounted in a roll forming machine.



Figure 16: Example of resistive sensor in a roll forming facility

On the other hand, piezoelectric sensors can be used to measure both forces and acoustic emission signals in metal forming processes. The sensors are able to work up high frequencies of about 500 KHz although for force measurements the amplifier only works up to 1 kHz. It is usually clamped in the structure or connection rods of the facilities and used to measure the tensile and compressive forces of the process. The sensor is usually inserted in a borehole and preloaded. When intending to measure acoustic signals, the piezoelectric sensors work at higher frequencies and are usually installed closer to the process itself. In metal forming they are usually placed in the tooling instead of in the structure of the facility.

Figure 17: Example of piezoelectric sensor installation for force and acoustic signals measurement

DECISION MAKING APPROACHES

And finally, and once that the fault at the process is detected, next step is its identification and proposal of corrective solutions. In order to do so, a decision making algorithm, which is not usually included in the monitoring system itself, is necessary and different approaches can be found in the literature. The main aspect to be considered when selecting the solution for achieving the fault identification is the complex nature of forming processes. Metal forming processes show a high non-linear behaviour and are almost impossible to be defined with mathematical models (described as non-formalised problems in [37]). These processes are inherently quite unstable processes which main variables, like the material behaviour under deformation, the lubrication and the friction at the material/tool interface or the wear of the tools are highly non linear. Therefore, the application of traditional controllers to these processes has not offered good results yet.

Due to the presence of so many different process behaviours within the industry, there has been a tendency towards two schools of thought in the choice of a model structure for use in a control system. One school believes that the model should be based on known physical phenomena that characterise the process (model based control); that is, a first principle model (traditional control techniques). The other school tends towards a íblack boxí approach, which uses observed relations between the inputs and



the outputs of the processes to characterise a general, usually non-linear transformation (transfer function), which internal parameters are sometimes unknown [BAL88]. This second one is the most successful technique for metal forming processes. Next the two most successful approaches within this íblack boxí approach are briefly described.

One approach for the identification of faults in complex disciplines is Expert Systems (ES). Rule-based ES are computer programs that codify specific domain knowledge, as a set of IF-THEN rules in knowledge bases, and further use the codified rules to solve problems related to the specific domain. Rule-based ES use the codified rules along with information contained in the working memory (definition of the actual problem) to solve problems in the next way: when the "IF" portion of the rule matches the information contained in the working memory, the system performs the action specified in the "THEN" part of the rule [28].

Figure 18: Schematic representation of rule-based expert systems [14]

Therefore, while an ES is a computer program, it does not follow the traditional approach where the programmer specifies each step in solving the problem. Rather, the programmer (a knowledge engineer) codifies a large number of facts about the problem domain as rules. Rules specify one or more facts that may be inferred with some degree of certainty when some other facts are known to be true. A relatively simple control program called an inference engine can then be used to examine the rules in light of known facts. The ES methodology allows a system to draw useful conclusions even with incomplete or uncertain data. It is especially useful for solving problems where an algorithmic approach is either difficult or impossible to implement (traditional control strategies based on PID controllers) [32, 18].

Since the first ES named DENDRAL, created by Lederberg and Feigenbaum in 1965 at Stanford University, this AI methodology has been broadly applied both in the academic and in the industrial field. Next some previous research works focused on the identification of failures, the topic that the present research work covers, are briefly explained: 1. Chun Cheung Siu et al. developed a rule-based ES able to deal with fuzzy knowledge, and applied it

to the identification of vibration causes in rotating machines. The system was able to generate ranked fault hypothesis within an incrementally consultation and allowed for the revision of diagnosis results with respect to the revision of symptoms presented by the user [41].

2. R. Amyot et al. developed an operational ES prototype to help mill operators and engineers to troubleshoot and optimise the steam and condensate portion of paper machine dryer sections. A major output of the prototype was to quantify the thermodynamic performance of the machine in order to inform the user when the steam and condensate system was wasting energy. This way, when the machine was operating outside of specified thresholds, the ES entered into a diagnostic dialog with the user to obtain more information aimed at determining the possible cause(s) of the deteriorated performance. A three-month validation phase conducted simultaneously in two mills led to the overall conclusion that, despite some room for improvement in the system's usability and functionality, it is a fundamentally sound and useful tool for monitoring and recording the performance of a S&C system, and for helping to diagnose the causes of poor performance [17].



3. Warren R. Becraft et al. developed an operator advisory (INNATE/QUALMS) composed of a rule-based ES combined with artificial neural networks able to help operators at large-scale chemical process plants to diagnosis process failures. The developed diagnostic system exhibited good diagnostic performance under a variety of conditions including novel faults, and the presence of sensor noise [4].

4. D. Chester et al. developed FALCON, a rule-based ES able to identify probable causes of disturbances in a chemical process plant by interpreting data consisting of numerical values from gauges and the status of alarms and switches. The system interpreted the data by using knowledge of the effects induced by a fault in a given component and how disturbances in the input of a component lead to disturbances in the output [9].

5. William R. Nelson et al. developed REACTOR, an ES able to assist operators in the diagnosis and treatment of nuclear reactor accidents. The purpose of REACTOR was to monitor a nuclear reactor facility, detect deviations from normal operating conditions, determine the significance of the situation and recommend an appropriate response [34].

6. Massimo Gallanti et al. developed PROP, an ES for malfunction diagnosis and process surveillance concerning on-line monitoring of water pollution in a thermal power plant [15].

7. Peter Chan, during his PhD work, developed a prototype rule based ES for civil engineering applications in the knowledge domain of diagnosis of deterioration and other problems in reinforced concrete structures. The developed system performed satisfactorily with about a 70% rate of success in real cases. The confidence values provided were found to be reasonable and the system was shown to be adequate in providing diagnosis of common problems of reinforced concrete but it did not perform well in special cases outside its knowledge domain [8].

8. Agre et al. developed a rule-based ES intended to help the maintenance staff in search of faults in the personal computers of the family PRAVETS-8 (Apple-2 compatible) [AGR85]. At the same time Sgurev et al. also developed another rule-based ES intended to help the maintenance staff in the search of faults in disk subsystems, consisting of a controller and a hard disk drive module with 300 or 600 MB capacity [39].

Another approach for the fault identification is Case Based Reasoning (CBR) methodology. CBR is based on the idea of utilizing solutions to past problems to solve new problems. Thus, the solutions to ísimilarí problems are retrieved from a case memory of solutions, and applied to new problems. This way, when a CBR system is presented with a similar problem, it does not re-reason from an initial set of facts and rules. Instead, it uses the plan that embodies the reasoning already utilized in the retrieved solution [45].

CBR can be traced back to the work of Schankís dynamic memory model in 1982 [SCH82], but was Kolodner [22] who developed the first case-based reasoner, known as CYRUS. CYRUS was based on the abovementioned Schankís memory model and was a question and answer system that contained the knowledge, as cases, of the travels and meetings of ex-US Secretary-of-State Cyrus Vance [11].

Generally, a case-based reasoner will receive a problem presented by either a user or another program or system. The case-based reasoner then searches its memory of past cases (the case base) and attempts to find a case that has the same problem specification as the current case. If the reasoner cannot find an identical case in its case base, it will attempt to find the case or cases in the case base that most closely match the current query case.



Figure 19: Schematic representation of a CBR system [24]

The CBR methodology is known to be well suited to those domains where formalized and widely recognized background knowledge is not available [29]. In these scenarios, the acquisition of cases becomes a natural mechanism for knowledge acquisition and avoids the need to extract the principles underlying a domain. Many works in the literature suggest that this significantly alleviates the knowledge acquisition problem [23].

As mentioned before, first CBR system, CYRUS, was developed in 1982. Since then, CBR methodology has been broadly applied both in the academic and in the industrial field. Some previous research works focused on the identification of failures, the topic that the present research work covers, are briefly explained next: 1. Stefania Montani et al. applied CBR for failure diagnosis and remediation in software systems with

the purpose of developing distributed software systems with self-healing capabilities. The suitability of the approach was demonstrated by some tests conducted on the Moodle application, running on a distributed architecture. Moreover, it was also demonstrated the no necessity of structured knowledge, such as models of the system behaviour, thus easing its applicability to large-scale, complex software systems [MON06].

2. Erik Olsson et al. demonstrated the successful performance of CBR methodology to the identification and diagnosis of faults during the assembly of robots based on the recording of abnormal acoustic signals [OLS04].

3. Paal Skalle et al. also applied CBR in the oil extraction industry to the identification and solution of lost circulation during oil well drilling. The author demonstrated the capacity of CBR to provide useful knowledge such as cause factors and remedial actions when new problems arrived to the system [42].

4. T. Warren Liao et al. implemented CBR methodology to the correct detection and identification of welding flaws in automated weld inspection systems. The system developed used radiographic weld images of the welding line and compared them to previous already classified welding flaws to perform a correct identification of the flaw at the part. The results obtained in the study indicated that better performance in terms of higher accuracy rate and lower false positive rate can be achieved than that of the fuzzy clustering methods employed before [26].

5. Mark Devaney et al. developed a log identification system able to provide the data necessary to characterize operating cycles, maintenance schedules, periodic breakdowns, and most importantly, to identify and address abnormal failure rates in big industrial facilities before critical problems arise. All the knowledge was implemented into a database using a CBR methodology and the identification of new operating failures was provided to experienced maintenance engineers and managers who assessed the utility of the system [12].

6. Mark Devaney et al. also developed a case-based reasoner for gas turbine diagnosis at the monitoring and diagnosis centre of General Electric in Atlanta. The main purpose of the developed system was to improve turbine and system reliability, reduced turbine operating/maintenance costs, and produce the greatest possible sustained availability from the power generation equipment. The case-based reasoner worked in next way: right after any gas turbine shut down, the monitoring and diagnosis centre of General Electric in Atlanta automatically received the operating data of the turbine. Then the data were analysed and the reasons and solutions to be apply were identified



based on the data recorded from previous experiences. Finally the solutions to be applied were communicated to the maintenance personnel of the corresponding gas turbine [13].

The industry demands monitoring systems for machine-tools increasingly efficient that reduce maintenance costs and increasing machine availability. Nowadays, widely employed monitoring systems are mostly based in additional sensor technology which is prone to errors and must be carefully maintained. Several failure risks that must be prevented by means of advanced and robust monitoring system. On the other hand, health assessment capability allows the change from the classical íscheduled maintenanceí strategy to a more ícondition basedí one. This evolution increases the operational life of the components and improves the overall machine availability reducing the unscheduled corrective actions.

Process monitoring involves knowledge several process states and machine parameters, such as dynamic, during machine tool operational lifetime. Measuring forces is error-prone, expensive and often intrusive process. Furthermore, it occurs regularly that force measurements at the desired locations are prohibited due to space limitations or too harsh circumstances.

A promising approach is indirect measurement technology by adopting a virtual sensing or soft sensing strategy. This involves the evaluation and development of observers and estimation techniques that combine high-fidelity physical models, both for the machine and the manufacturing process, with affordable non-intrusive sensors or control signals to estimate unknown states in a fast, accurate and on-line way. The main scientific challenges are modelling the process and machine, model order reduction and maximizing the observability for a given signal set. The developed models should also include the electro-mechanical subsystems.

These techniques are used in many industry applications as well as in several metalwork processes, such as machining processes and forming processes. Evidences of this are described in some papers that show how a large list of industrial processes and their variables estimation are controlled and enhanced by means of soft sensing. Some of those monitored processed and variable estimation techniques are proposed in the literature:

1. Zeng yi-hui et al. developed a soft sensing model for the roughness of machining surface based on the support vector machines using rotate speed n, feed peed vf, and depth of cutting as independent parameters, taking groups of actual machining experiment data as samples. The allowable error ε and the pos an adaptive genetic algorithm [48].

2. J. P. Lei et al. designed a soft sensing model to effectively realize fast and high accurate measurements of flatness error on the surface of machining workpiece. Thereby, the results of training, testing and practical application show, after the optimization of 200 steps, the soft measurement value of flatness error on the surface of machining workpiece decreases [25].

3. Victor M. Zavala et al. proposed a moving horizon estimation (MHE) as an efficient optimization-based strategy for state estimation. Despite its difficulties to be applied in industrial settings due to real-time dynamic optimization, the author demonstrates a fast MHE algorithm able to solve those problems based on advances in nonlinear programming algorithms and sensitivity concepts [46].

Regarding metal forming processes, there are some researches in this field:

4. Meixing Ji et al. has investigated the application of interdisciplinary approaches to optimization of multi-stage metal forming to different aspects of the metal forming processes using soft sensing tools. In process control systems of the forging process, as a result of the limitation incurred by process technology or measurement techniques, some important process variables and material properties of the initial billet are very difficult or impossible to monitor and also adjust promptly. Thanks to soft sensing those properties and variables can be estimated [31].



7.1 Other sensors in Press forming machines

7.1.1 Electronic pressure sensor

Nowadays, the PN 3070 electronic sensor is used to measure the pressure of the overload chambers in FAGOR ARRASATE presses. The disadvantage of this kind of pressure transducers is that they don´t measure the negative force of the press, so the cutting shock effect is not possible to measure.

Figure 19: PN 3070 Electronic pressure sensor

The automotive sector is replacing the conventional sheet metal steel by high-strength sheet metal steel. Tis steel, called Third generation steel causes the increase of the loads applied to the metal forming machines in the processing. Comparing to conventional sheet metals a higher cutting-shock effect appears after cutting this type of metals. The components of the press start to oscillate and there is the possibility of an early crack initiation consequence. It is necessary to investigate the specific demands while cutting modern sheet metals and take them into account. That is why, in the MANTIS project, one of the activity will be to monitor the effects of the cutting shock.

7.1.2 Bushing temperature measuring using sensors

Figure 20 PT-100 temperature probe

FAGOR ARRASATE press machines are equipped with a PT-100 temperature probes to measure the temperature of the bushings. These probes are installed on the surface of these elements. It can be assumed that a PT-100 temperature probe with direct output to the cable can operate between -50ºC and 250ºC. In our case, the limit is fixed to 100ºC because a higher temperature can damage the properties of the oil and consequently the quality of the bushing. Nowadays is no easy to distinguish the

6



temperature between the surface and the core of the bushing. This problem together with the improvement of temperature control is going to be a goal for research during MANTIS project.



8 Conclusions

Sensors can be found in many electronics systems and devices. However, most of them have not the

ability to process and analyze the detected data, limited to function as a transducer performing the

measurement of parameters and sends this information to a processor central. This situation is

changing, new sensor generation can be found, it is equipped with its own intelligence, able to organize

themselves and to interface wirelessly with other similar.

Depending on the application within the use case, several new sensors technologies will be evaluated

according on the use case requirements, i.e. vibrations, high temperatures, dirtiness, precisioní



References

http://www.ukessays.com/essays/information-technology/future-trends-in-wireless-sensor-networks-information-technology-essay.php

http://www.techopedia.com/definition/25651/wireless-sensor-network-wsn

http://en.wikipedia.org/wiki/Wireless_sensor_network

http://www.researchgate.net/publication/3090944_Sensor_technology_advances_and_future_trends

http://www.freescale.com/webapp/sps/site/overview.jsp?code=SNSMEMSOVERVIEW

http://www.sensorland.com/HowPage023.html#anchor203959

https://en.wikipedia.org/wiki/Microelectromechanical_systems

SHT11 Sensor Datasheet:

http://www.sensirion.com/fileadmin/user_upload/customers/sensirion/Dokumente/Humidity/Sensirion_Humidity_SHT1x_Datasheet_V5.pdf

GP2Y0A21YK Optoelectronic Device Datasheet:

http://www.sharpsma.com/webfm_send/1208

Digital Temperature Controller CB103 Datasheet:

http://www.rkcinst.co.jp/english/pdf/cb3_03e.pdf

[1] Almansa A., Melchor M. A. Controlador del proceso de estampaciín basado en redes neuronales y lígica difusa. XIII Congreso de Míquina-herramienta y Tecnologías de Fabricaciín. pp 263-282. Donostia (Spain). October 2000.

[2] http://www.artis-systems.com/

[3] Balchen G. J. íProcess Controlí. Blackie A & P. Glasgow (Great Britain), 1988, ISBN: 0-7514-0298-2.

[4] Becraft W. R., Lee. P. L., Newell R. B. íIntegration of Neural Networks and Expert Systems for Process Fault Diagnosisí. Proceedings of the Twelfth International Conference. on Artificial Intelligence IJCAI 1991: 832-837.

[5] Process monitoring: Both tool and press protected. Brankamp Journal, 1994.

[6] www.brankamp.com

[7] Brankamp PK550 User Manual. www.brankamp.com

[8] Chan P. F. íAn Expert System For Diagnosis Of Problems In Reinforced Concrete Structuresí Royal Melbourne Institute of Technology. December 1996.

[9] Chester D., Lamb D. Dhurjati P. íRule-based computer alarm analysis in chemical process plants. In Proceedings of the Seventh Annual Conference on Computer Technology, pages 22-29. IEEE, March 1984.

[10] Cowper.S. Acoustic-Emission monitoring making waves in press shops. Metalforming, 49-52, 2000.

[11] Cobb C. L., Zhang Y., Agogino A.M. íMEMS Design Synthesis: Integrating Case-based Reasoning and Multi-objective Genetic Algorithmsí. Smart Structures, Devices, and




http://www.techopedia.com/definition/25651/wireless-sensor-network-wsn

http://en.wikipedia.org/wiki/Wireless_sensor_network

http://www.researchgate.net/publication/3090944_Sensor_technology_advances_and_future_trends

http://www.freescale.com/webapp/sps/site/overview.jsp?code=SNSMEMSOVERVIEW

http://www.sensorland.com/HowPage023.html%23anchor203959

https://en.wikipedia.org/wiki/Microelectromechanical_systems



http://www.sharpsma.com/webfm_send/1208

http://www.rkcinst.co.jp/english/pdf/cb3_03e.pdf


Systems III, edited by Said F. Al-Sarawi, Proc. of SPIE, Vol. 6414, 641419, (2007) 0277-786X/07/$15 doi: 10.1117/12.695609.

[12] Devaney M., Ram A., Qiu H., Lee J. íPreventing Failures By Mining Maintenance Logs With Case-Based Reasoningí. 59th Meeting of the Society for Machinery Failure Prevention Technology (MFPT-59), 2005, 369-381.

[13] Devaney M., Cheetham B. íCase-Based Reasoning for Gas Turbine Diagnosticsí. The 18th International FLAIRS Conference, Clearwater Beach, Florida, (2005).

[14] http://faculty.ac.id

[15] Gallanti M., Guida G., Spampinato L., Stefanini A. íRepresenting procedural knowledge in expert systems: An application to process controlí. Proceedings of the Ninth International Conference on Artificial Intelligence IJCAI (1985).

[16] Garcia C. Artificial intelligence applied to automatic supervision, diagnosis and control in sheet metal stamping processes. Journal of Materials Processing ¨Technology, 164-165, 1351-1357, 2005.

[17] Gowing J., Wylie R., Henzell R., Futcher J., Reinsborough J., Coderre A., Henzell P., Vadas O. íSteam and Condensate Diagnostic Expert Systemí. Source: Preprints of the 81st Annual Meeting, Canadian Pulp and Paper Association (Technical Section). pp. A41-44. NRC Publication Number: NRC 38335, 2001.

[18] Harrod.R.A., Tietz.L. On-line expert system and their role in the control systems of the future. ISA transactions, 29: 41-46, 1990.

[19] http://www.naatinc.com/index.php

[20] http://www.loadmonitors.com/aboutus.htm

[21] Kirchheim.A., Schaffner.G., Wolfer.P. Piezoelectric sensors for the combined measurement of forces and acoustic emission for in-process monitoring. Monitoring of machining and forming processes, March 30-31st, Dísseldorf 1995.

[22] Kolodner J. íCase-Based Reasoningí. Morgan Kaufmann Publisher, San Mateo, CA.

[23] Leake D. B. íBecoming An Expert Case-Based Reasoner: Learning To Adapt Prior Casesí. Proceedings of the Eighth Annual Florida Artificial Intelligence Research Symposium, 1995, pp. 112-116.

[24] Leake D. B. íCBR in Context: The Present and Futureí. In Leake, D., ed., 1996, Case-Based Reasoning: Experiences, Lessons, and Future Directions. Menlo Park: AAAI Press/MIT Press, 1996. Chapter 1.

[25] J. P. Lei, J. M. Chen, "Soft-Sensing Model of Flatness Error on the Surface of Machining Workpiece and its Application", Applied Mechanics and Materials, Vol. 628, pp. 436-441, Sep. 2014

[26] Liao T. W., Triantaphyllou E., Chang P.C. íDetection of Welding Flaws with MLP Neural Network and Case Based Reasoningí. Intelligent Automation and Soft Computing, Vol. 9, No. 4, pp. 259-267, 2003, Copyright © 2003, TSI® Press, Printed in the USA. All rights reserved.

[27] Licht.T, Acoustic Emission. Brueel & Kjaer Technical Review, 2, 3-36, 1979.

[28] Lu S. P. í Control and Supervision of the Aluminium Electrolysis Process with Expert Systemí. PhD work presented in University of Quebec September 2002.

[29] Main J., Dillon T., Shiu S. íA tutorial on case-based reasoningí. Proceedings of the Conference Soft Computing in Case Based Reasoning, 2000, 1-18.



[30] Martínez F., García C., Ortubay R. Monitorizaciín y diagnístico del proceso de embuticiín. XIV Congreso de Míquina-herramienta y Tecnologías de Fabricaciín. pp 363-378. Donostia (Spain). 14-16 October 2002.

[31] Meixing Ji, M.S., FEM based interdisciplinary approaches to optimization of multi-stage metal forming processes.

[32] Miller R. K. íArtificial intelligence: a new tool for manufacturingí. Manufacturing Engineering, 94: 56-62, 1985.

[33] http://www.montronix.com/

[34] Nelson W. R. íReactor: An expert system for diagnosis and treatment of nuclear reactor accidentsí. In proceedings of AAAI-82, August 1982.

[35] http://www.nordman.org

[36] http://www.prometec.com

[37] Rybina G., Rybin V. íStatic And Dynamic Integrated Expert Systems: State Of The Art, Problems And Trendsí. International Journal "Information Theories & Applications", Vol.12, 2005, p203-211.

[38] http://www.sk-gmbh.de/english/index.htm

[39] Sgurev V., Dochev D., Agre G., Dichev C., Markov Z. íDiagnostic expert system for digital electronics, industrial applications of artificial intelligenceí. Alty, J.L., and Mikulich, L.i (Eds), North-Holland, Amsterdan 1991, pp 220-224.

[40] http://www.siegfried-ag.ch/processors_e.htm

[41] Siu C. C., Shen Q., Milne R. íA Fuzzy Expert System for Turbomachinery Diagnosisí. Proceedings of the International Conference on Fuzzy Systems. 1997. p555-560.

[42] Skalle P., Sveen J., Aamodt A. íImproved Efficiency Of Oil Well Drilling Through Case Based Reasoningí. 6th Pacific Rim International Conference on Artificial Intelligence Melbourne, Australia, August 28 - September 1, 2000, 712-722.

[43] Terzyk.T., Oppel.F. Force and acoustic monitoring in cold forming. Wire Industry, 63: 622-623, 1996.

[44] http://www.unidor.de/unidor/engl/home.htm

[45] Veerakamolmal P. íAutomating Multiple Products Disassembly Process Planning with Case- Based Reasoningí. The Second International Conference on Operations and Quantitative Management (ICOQM-II) January 3-6, 1999, Ahmedabad, India.

[46] Victor M. Zavala *, Carl D. Laird 1, Lorenz T. Biegler. íA fast moving horizon estimation algorithm based on nonlinear programming sensitivityí 2008

[47] Yoshihara.S., Manabe.K., Nishimura.H. Adaptability to frictional change of fuzzy adaptive blank holder control for deep drawing. Proceedings of the Second International Conference on Intelligent Processing and Manufacturing of Materials, 1: 601-606, 1999.

[48] Y. H. Zeng, J. Q. E, X. P. Yang, H. M. Li, "Soft-Sensing Model on the Roughness of Machining Surface under the Numerical Control and its Application", Applied Mechanics and Materials, Vols. 48-49, pp. 1077-1085, Feb. 2011


MANTIS · 22 ATLAS Atlas Copco Airpower Nv BEN BE 23 3E 3e Nv BEN BE 24 PCL Philips Consumer...

Documents

Transcript of MANTIS · 22 ATLAS Atlas Copco Airpower Nv BEN BE 23 3E 3e Nv BEN BE 24 PCL Philips Consumer...