Escolar Documentos
Profissional Documentos
Cultura Documentos
Proceedings of
IWAMA 2013
The Third International Workshop of Advanced Manufacturing and Automation 27 November 2013, Trondheim, Norway
Preface
IWAMA International Workshop of Advanced Manufacturing and Automation began in 2010 as a joint seminar between SFI Norman (SINTEF and NTNU) and Shanghai Key Laboratory of Manufacturing Automation and Robotics (Shanghai University). In 2013 IWAMA expanded to include academic and industrial experts in the fields of advanced manufacturing and automation worldwide when the University of Manchester and Tongji University joined as co-organizers. As IWAMA becomes an annual event, we expect more universities and industry sponsors will join the international workshop as co-organizers. Manufacturing and automation have assumed paramount importance and are vital factors for the economy of a nation and the quality of life. The field of manufacturing and automation is advancing at a rapid pace and new technologies are also emerging in the field. The challenges faced by todays engineers are forcing them to keep on top of the emerging trends through continuous research and development. IWAMA aims at providing a common platform for academics, researchers, practicing professionals and experts from industries to interact, discuss trends and advances, and share ideas and perspectives in the areas of manufacturing and automation. IWAMA 2013 takes place in Trondheim, Norway, 27 November 2013, organized by Norwegian University of Science and Technology, SFI Norman/SINTEF, Shanghai University, University of Manchester and Tongji University. The program is designed to improve manufacturing and automation technologies for the next generation through discussion of the most recent advances and future perspectives, and to engage the worldwide community in a collective effort to solve problems in manufacturing and automation. Participants from different countries, such as Norway, China, UK, Germany, Italy and India attend the workshop. Manufacturing research includes a focus on the transformation of present factories, towards reusable, flexible, modular, intelligent, digital, virtual, affordable, easy-to-adapt, easy-to-operate, easy-to-maintain and highly reliable smart factories. Therefore, IWAMA 2013 has mainly covered 3 sessions in manufacturing engineering: 1. 2. 3. Intelligent Manufacturing Technology Intelligent Logistics and Supply Chains Intelligent Diagnosis and Prognosis of Wind Turbines
The proceedings of IWAMA2013 provide up-to-date, comprehensive and worldwide state-ofthe-art knowledge of the manufacturing and automation science and engineering. All papers submitted to the workshop have been subjected to strict peer-review by at least 2 expert referees. Finally, 28 papers have been selected to be included in the proceedings after a revision process. We hope that the proceedings will not only give the readers a broad overview of the latest advances, and a summary of the event, but also provide researchers with a valuable reference in this field. On behalf of the organization committee and the international scientific committee of IWAMA 2013, I would like to take this opportunity to express my appreciation for all the kind support, from the contributors of high-quality keynotes and papers, and all the participants. My thanks are extended to all the workshop organizers and paper reviewers, NTNU and SFI Norman for the financial support, and co-sponsors for their generous contribution. Thanks are also given to Quan Yu, Yi Wang, Cecilia Haskins, Tonje Berg Hamnes, Line Holien, Jorurnn Auth, and Lilan Liu for
their hard editorial work of the proceedings and arrangement of the workshop. Thank also goes to Lasse Postmyr from Tapir Academic Press for producing the proceedings.
Kesheng Wang, Ph.D. Professor of NTNU Chairman of IWAMA 2013 Trondheim, Norway, 25th October, 2013
II
IWAMA 2013 & Akademika Publishing, 2013 Copyright is held by the respective authors. ISBN 978-82-321-0377-5 ISSN 1892-8110 This publication may not be reproduced, stored in a retrieval system or transmitted in any form or by any means; electronic, electrostatic, magnetic tape, mechanical, photo-copying, recording or otherwise, without permission.
Layout: The authors Cover layout: Akademika Publishing Printed and binded by: AiT Oslo AS
We only use environmentally certified printing houses. Akademika Publishing Oslo/Trondheim Norway www.akademikaforlag.no Publishing Editor: Lasse Postmyr (lasse.postmyr@akademika.no)
Honorary Chairs:
Prof. Minglun Fang
General Chairs:
Prof. Kesheng Wang, Prof. Jan Ola Strandhagen and Prof. Dawei Tu
Prof. Jan Ola Strandhagen, Prof. Kesheng Wang, Prof. Heidi Dreyer, Lars Tore Gellein, Quan Yu, Tonje Berg Hamnes, Associate Prof. Cecilia Haskins, Jorunn Auth
Secretariat
III
About Editors
Prof. Kesheng Wang holds a Ph.D. in production engineering from the Norwegian University of Science and Technology (NTNU), Norway. In 1993 he was appointed Professor at the Department of Production and Quality Engineering, NTNU. He is a director of the Knowledge Discovery Laboratory (KDL) at IPK, NTNU at present. He is also an active researcher and serves as a technical adviser in SINTEF. He was elected member of the Norwegian Academy of Technological Sciences in 2006. He has published 17 books, 10 book chapters and over 214 technical peer-reviewed papers in international journals and conferences. Prof. Wangs current areas of interest are intelligent manufacturing systems, applied computational intelligence, data mining and knowledge discovery, swarm intelligence, condition-based maintenance and structured light systems for 3D measurements and RFID. Prof. Jan Ola Strandhagen is a research director of the research center SFI NORMAN at SINTEF. He is also Professor at Department of Production and Quality Engineering, the Norwegian University of Science and Technology (NTNU). He holds a PhD in Production Engineering from NTNU (1994). His research has focused on production management and control, logistics, manufacturing economics and strategies. He has managed and performed R&D projects in close collaboration with a wide variety of Norwegian companies and participated as researcher and project manager in several European projects. Prof. Dawei Tu received his bachelors degree, masters degree and PhD from Zhejiang University in 1987, 1989 and 1993 respectively. Since 1993 he joined Shanghai University, and became a professor in 2000. Now he is the Executive Dean of the School of Mechatronics Engineering and Automation, Shanghai University. His research interests include optoelectronic detecting, precision mechanics and instrumentation, 3D vision and vision-based robot servo-control.
IV
Table of Contents
Preface ............................................................................................................................... I Keynotes
Knowledge Dynamic in Software Applications for Process Controlling
Wladimir Bodrow......................................................................................................................... 1
Toward Intelligent Manufacturing: a Label Characters Marking and Recognition Method for Steel Products with Machine Vision The Development of Intelligent and Integrated RFID (II-RFID) System for Manufacturing Industries Responsible and Dynamic Logistics
Assessment of Geometric Tolerance Information as a Carrier of Design Intent to Manufacturing and Inspection Phase to Depth Conversion Algorithm by Using Epipolar Restraint in Objectadaptive Fringe Projection Technique Computer Vision Supported by 3D Geometric Modeling
Junzheng Peng, Yingjie Yu, Wenjing Zhou and Mingyi Chen .................................................. 71 Sven Fjeldaas ............................................................................................................................. 81
Studies on the Intelligent Manufacturing Technologies of the Fan Blade Virtual Simulation of Production Line for Ergonomics Evaluation
Wenhua Zhu, Baorui Li, Bin Bai, Hu Han and Minglun Fang ................................................... 93 Ming Chen and Jinfei Liu......................................................................................................... 113
A Point Cloud Registration Strategy Combining Particle Swarm Optimisation and Iterative Closest Point Method Graph Representation of N-Dimensional Space
Quan Yu and Kesheng Wang ................................................................................................... 123 Tomasz Kosicki ........................................................................................................................ 135
Yi Wang, Rishad Rayyaan and Richard Kennon ..................................................................... 171 Emrah Arica and Daryl J. Powell ............................................................................................. 185
Investigating the Fit of Planning Environments and Planning Methods, the Case of an Automotive Part Manufacturer Investigating the Applicability of Lean Principles in a Small, One-of-a-kind Production Company
Philipp Spenhoff, Marco Semini, Erlend Alfnes and Jan Ola Strandhagen ............................. 211
Incentive Regulation of Banks on Third Party Logistics Enterprises in Principalagent-based Inventory Financing
Automation in the ETO Production Situation: The Case of a Norwegian Supplier of Ship Equipment Competing in the Shell Eco-marathon. Challenges and Lessons Learned Tackling the Storage Problem through Genetic Algorithms
Brge Sjbakk, Maria Kollberg Thomassen and Erlend Alfnes .............................................. 259 Bassam A. Hussein and Fariborz Ali Haidrloo ........................................................................ 271 Lapo Chirici and Kesheng Wang ............................................................................................. 281
Session IDPWT: Intelligent Diagnosis and Prognosis of Wind Turbines (WINDSENSE Project)
Evaluation of Thermal Imaging System and Thermal Radiation Detector for Real-time Condition Monitoring of High Power Frequency Converters
Anders Eriksen, Dominik Osinski and Dag Roar Hjelme ........................................................ 295
VI
Wind Turbine Fault Detection Based on SCADA Data Analysis Using ANN
Models for Lifetime Estimation - An Overview with Focus on Applications to Wind Turbines
VII
List of Authors
Alfnes, E. 211, 259 Arica, E. 49, 185 Bai, B. 93 Belay, A. M. 157 Bodrow, W. 1 Cao, P. 9 Chen, M. 113 Chen, M. 71 Chirici, L. 281 Chu, X. 249 Dahir, L. 221 Eriksen, A. 295 Evanger, J. E. 53 Fang, M. 93 Fjeldaas, S. 81 Goborg, F. W. 53 Haidrloo, F. A. 271 Han, H. 93 Hjelme, D. R. 295 Hussein, B. A. 197, 271 Kennon, R. 171 Kosicki, T. 135 Lemu, H. G. 59, 147 Li, B 93 Liu, J. 113 Osinski, D. 295 Peng, J. 71 Pigagaite, D. 197 Powell, D. J. 185, 221 Rayyaan, R. 171 Semini, M. 211 Sharma, V. S. 307 Silva, P. P. 197 Sjbakk, B. 259 Spenhoff, P. 211 Strandhagen, J. O. 49, 211 Sun, X. 249 Thomassen, M. K. 259 Tu, D. 9 Ulonska, S. 235 Wang, K. 25, 123, 281, 307, 323, 337 Wang, Y. 171 Welo, T. 157, 235 Welte, T. M. 337 Yu, Q. 123 Yu, Y. 71 Zhang, Z. 307, 323 Zhao, Q. 9 Zhou, W. 71 Zhu, W. 93
VIII
Department of Production and Quality Engineering, Norwegian University and Science and Technology, Norway 2 Dept. of Industrial and Production Engineering, National Institute of Technology, India Abstract Most of the wind turbines have of some kind of CBM devices/systems, which provide the information about the device to the central data base i.e. SCADA database. These devices/systems make use some kind of data processing techniques/methods in order to locate faults. Finally all the information keeps on recoding in the SCADA data base. So the endeavour in this article is, to review the techniques/methods/algorithms developed, to carry out diagnosis and prognosis of the faults, based upon SCADA data obtained from wind turbines. The focus here is also on the Computational Intelligence methods developed. Keywords SCADA data, CI, CBM, WT Abbreviations
BT: Boosting Regression Tree OEM: Original equipment manufacturer CART: Standard classification & Regression tree FMECA: Failure Analysis, Failure Mode Effect and Criticality Analysis CBM: Condition Based Monitoring REP: Representative Tree CBMS: Condition Based Monitoring System DE: Differential evolution kNN: k Nearest Neighbor Network TPM: Total Productive Maintenance ANN: Artificial Neural Network WIP: Work in progress NNE: Neural Network Ensemble ES: Evolution Strategy RF: Random Forest RCM: Reliability Centred Maintenance SCADA: Supervisory Control & Data Acquisition PSO: Particle Swarm optimization SVM: Support Vector Machine
307
CMMS: Computerised Maintenance and Management Systems WT: Wind Turbine PCA: Principal Component analysis MLP : Multi-Layer Perceptron OEE: Overall Equipment Effectiveness CI: Computational Intelligence FFT: Fast-Fourier transform BT: Bragging Tree TSA: Time synchronous averaging FIS: Fuzzy inference system ANFIS: Adaptive neuro fuzzy inference system
Introduction
Renewable energy sources are playing an important role in the global energy mix, as a means of reducing the impact of energy production on climate change. Wind energy made its first actual step though simpler wind devices date back thousands of years ago with the vertical axis windmills found at the Persian-Afghan borders around 200 BC and the horizontal-axis windmills of the Netherlands and the Mediterranean following much later (1300-1875 AD). Further evolution and perfection of these systems was performed in the USA during the 19th century mainly for pumping water between 1850 and 1970. The first large wind machine to generate electricity of 12 kW was installed in Cleveland, Ohio, in 1888. Toward the end of World War I, use of 25 kW machines throughout Denmark was widespread. Further development of wind turbines in the USA was inspired by the aerospace developments, while subsequent efforts in Denmark, France, Germany, and the UK (between 1935 and 1970) showed that large-scale WTs could work. European developments continued after World War II and lot of development have been made till date(Kaldellis & Zafirakis, 2011). During the last thirty years, securities of energy supply and environmental issues have reheated the interest for wind energy applications. Also the challenge of wind energy applications is the target of 1000 GW of wind power by 2030. Further the trend in future is towards producing big and offshore turbines. The future forecast is indicated that the demand for wind power will increase many folds in the years to come. The rated capacity and size both will increase in time to come.
A wind turbine is a machine for converting the kinetic energy in wind into mechanical energy. Wind turbines are mainly classified into two general types: horizontal axis and vertical axis. A horizontal axis machine has its blades rotating on an axis parallel to the ground. A vertical axis machine has its blades rotating on an axis perpendicular to the ground. There are a number of available designs for both and each type has certain advantages and disadvantages. However, compared with the horizontal axis type, very few vertical axis machines are available commercially. Further horizontal wind turbines are also of two types upwind and downwind WTs. In case of upward WT the rotor is in the front of the unit. In order to keep it oriented into the wind, a yaw mechanism is needed. Also the extended nacelle is required to position the rotor far enough away from
308
the tower, in order to avoid any problems with a blade strike. In a downwind turbine the rotor is on the back side of the turbine. The nacelle is designed to seek the wind, thus there is no need to have a separate yaw mechanism. The flexible blade could be used but this advantage may also be a disadvantage, as the flexing may cause fatigue to the blades. Tower shadow is problem with downwind machines because the rotor blade is behind the tower. This may cause turbulence and lead to increased fatigue on the unit. Keeping in view the pros and cons of both the types, mostly upward WTs are used. The various parts of horizontal wind turbine (Figure 1) and their purpose are indicated: 1: blade (two or three blades, wind blowing over the blades causes the blades to "lift" and rotate); 2: rotor ( blades and the hub together); 3: pitch (blades are turned out of the wind to keep the rotor from turning in winds that are too high or too low to produce electricity); 4: brake (disc brake which can be applied mechanically, electrically, or hydraulically); 5: lowspeed shaft (turns at about 30 to 60 rpm); 6: gear box(connect the low-speed shaft to the high-speed shaft in order to increase speed from 30 or 60 rpm to about 1200 to 1500 rpm for producing electricity); 7: generator (induction generator that produces 60-cycle AC electricity). 8: controller ( starts or stops the WT at wind speeds of about 8 to 16 mph and 65 mph respectively; 9: anemometer (measures the wind speed and transmits wind speed data to the controller); 10: wind vane (measures wind direction and communicates with the yaw drive to orient the turbine); 11: nacelle (the rotor attaches to the nacelle, which sits atop the tower and includes the gear box, low and high-speed shafts, generator, controller, and brake, a cover protects the components inside the nacelle); 12: high-speed shaft (drives the generator); 13: yaw drive (upwind turbines face into the wind; the yaw drive is used to keep the rotor facing into the wind as the wind direction changes; downwind turbines don't require a yaw drive); 14: yaw motor (powers the yaw drive); 15: tower (made tubular steel or steel lattice, wind speed increases).
(a)
(b)
Fig. 1 (a) Parts of Wind Turbine http://www.wwindea.org/technology/ch01/estructuraen.htm , (b): CBMS on WT (Kusiak et al, 2010)
A fault is a physical defect, imperfection or flaw that occurs within the system. This may cause a failure: the non-performance of some action that is due or expected. Fault
309
detection is determination of faults present in a system and time of detection. Fault isolation is determination of kind, location and time of detection of a fault; follows fault detection. Fault identification is determination of size and time-variant behavior of a fault; follows fault isolation and Fault diagnosis is determination of kind, size, location and time of a fault; follows fault detection, and includes fault isolation and identification [Donders, 2002]. Prognosis is an approach that combines information on each machines current condition with historical data from machines of the same class, models of the physics of failure of components and short term projected usage to predict the future probability of failure of that individual machine. That is, prognosis gives a probabilistic forecast that is specific to each machine, allowing a strategy that balances the risk of running a machine with damage indications against the lost revenue while waiting for maintenance [Hyers et al, 2006]. Data Mining is a process of extraction of useful information and patterns from huge data. It is also called as knowledge discovery process, knowledge mining from data, knowledge extraction or data /pattern analysis [Ramageri, 2010]. Various maintenance practices are have been described and classified, empirically, into the following approaches [Alsyouf, 2004]: reactive Approach; preventive approach; predictive approach, diagnostic (expert systems) approach ; autonomous approach ; lean approach ; proactive approach. The ultimate goal of Fault Diagnosis and Prognosis is to decide the appropriate maintenance strategy.
310
Table 2 Summary of sensory signals and signal processing methods [Garca et al, 2012]
Methods Statistical methods Description Root mean square, peak amplitude Maximum value, minimum value, mean, peak to peak, standard deviation, shape factor, crest factor, impulse factor, definite integral, energy ratio and kurtosis. Particular algorithms power output patterns from WT generators Applications Widely used the diagnosis of failures
Trend analysis
Least median squares, filtering with and without a classical statistical method (based on standard deviation) Variations in current signals and trends, vibration analysis, oil analysis, AE analysis
Monitoring pitch mechanisms, power output generators Vibration filtering Imbalances between the rotor and the stator phases, faults in the rotor windings of the generator. Gear vibrations Cracked gear tooth Identifies source of error Bearings Bearings Overall rotor condition including surface roughness, mass imbalance and aerodynamic asymmetry. Gears, bearings, mechanical systems Bearings All Systems
Cepstrum analysis Time synchronous averaging (TSA) Fast-Fourier transform (FFT) Amplitude demodulation Order analysis
Conversion of a digital signal from the time domain into the frequency domain, constant speed wind energy converters Low-amplitude and low-frequency, periodic signals that might be masked by other higher energy vibrations Out of phase signals can be separated and analysed, suitable for variable speed wind energy converters.
Time-frequency technique ,non-stationary signals Classification of patterns in trend analysis Computational Intelligent algorithms (Lei, Lin, He, & Kong, 2012),(Amjady & Hedayatshodeh, 2012),(Uraikul, Chan, & Tontiwachwuthikul, 2007)
311
Some existing CBMS are: Gram & Juhl TCM System; DMT WindSafe System; Siemens Monitoring & Safety System; PRFTECHNIK [Wiggelinkhuizen et al, 2008]
Supervisory Control and Data Acquisition (SCADA) is an application that collects data from a system and send them to a central computer for monitoring and control usages. Current CBM systems essentially provide the necessary sensor and data capture capability required for monitoring. This monitoring system consists of wireless sensors that communicate with an embedded microprocessor mounted on the devices. Its purpose is to allow diagnostic and fault detection algorithms to be deployed down at the sensor / hardware level. This will decrease the volume of data that must be transmitted and stored via the more traditional centralised SCADA system approach. The SCADA data resides on a server / PC in the form of a database, and the raw data resides on the hard-drive of the microprocessor / sensor system mounted on the turbine (Refer Figure 2). A typical SCADA data is 10 minute averaged data [Zaher and McArthur, 2007]. Thus the collected and stored SCADA data must then be examined in order to deduce the overall health of the turbine as well as its internal components [Uraikul et al, 2007]. An operational wind farm typically generates vast quantities of data. The SCADA data contain information about every aspect of a wind farm, from power output and wind speed to any errors registered within the system[Kim et al, 2011]. SCADA data may be effectively used to tune a wind farm, providing early warning of possible failures and optimising power output across many turbines in all conditions. Typical parameters recorded by SCADA on a WT could be broadly categorized into following types which could be used in fault diagnosis and prognosis activity: Wind parameters, such as wind speed and wind deviations; Performance parameters, such as power output, rotor speed, and blade pitch angle; Vibration parameters, such as tower acceleration and drive train acceleration; Temperature parameters, such as bearing temperature and gearbox temperature[Verma and Kusiak, 2012]. There are many success stories about using SCADA data for CBM. Initially Monostori et al (2009) discussed a framework for optimizing wind energy systems design, operation and maintenance. They emphasized, as wind turbines have several built-in sensors measuring various physical characteristics during the operation and SCADA systems serve with huge amount of data. There after Wiggelinkhuizen et al (2008) presented the framework of the EU funded Condition Monitoring for Offshore Wind Farms project. They studied a small wind farm of five turbines having been instrumented with several condition monitoring systems and also with the traditional measurement systems for measuring mechanical loads and power performance. Data from vibration and traditional measurements, together with data collected by the turbines system control and data acquisition (SCADA) systems, have been analyzed to assess. They could determine failures, detect early stage of failure and asses the components health. They developed, applied, and tested several data analysis methods and measurement configurations successfully and conclude that for all types of measurements SCADA data, time series, vibration monitoring could be used for CBM. Zaher and McArthur, 2007 also proposed an idea to use the combination of anomaly detection and data-trending techniques
312
encapsulated in a multi-agent framework for the development of a fault detection system for WTs.
SCADA DATA RAW DATA Sensor 1 Micoprocessor
mounted on device remote location 10 mins average data
SCADA data
PRESENT DATA pre-processing SCADA DATA PAST/PRESENT DATA DATA MINING Fault diagnosis
Sensor 2
Micoprocessor
AOD data
Corroborate
Sensor 2
Micoprocessor
313
Power prediction
Power output
CART,BT,SVM, NN
Radial-basisfunction MLP
PSO, DE, ES
Inputs: wind, speed, wind direction, outside temperature; Outputs: turbine control Input: wind speed, blade pitch angle, generator torque; Outputs: rotor speed and turbine power Input : Torque value, blade pitch angle rate, wind deviation, blade pitch angle, wind speed; Outputs: drive train acceleration and tower acceleration Input: current wind speed, generator speed, wind direction, wind direction difference; Output: blade pitch/ yaw angle Wind speed difference tests
k-NN modelling best among studied Optimal control to improve the power output without large rotor speed changes. NN algorithm best among studied
Control approach generated optimized settings of the blade pitch and yaw angle Detected normal and abnormal states, PSO best among the studied Intelligent control system smoothed the power output, generator torque, and rotor speed Performance analytic power curve analysis separates out prefailure data from other normal operating data Power curve determines the health of WT, status/ fault, category A data-mining approach applied, Random forest algorithm best among studied NNE suitable drive train acceleration and the NN model is most suitable for predicting tower acceleration. Drive train acceleration, generator speed, torque and aerodynamic asymmetry. Boosting tree algorithm best among the studied
Extracts the process and the time series models from historical SCADA and wind data
Input: Wind conditions, electricity demand; Output: blade, pitch angle, generator torque. Power residual
NN,NNE,BT, SVM
Wind, performance Vibration and temperature parameters Power, wind speed, rotor speed, and generator speed
PCA
NN,NNE,BT, SVM, random forest,CART, kNN,NN Multilayered perceptron, boosting tree,k-NN Support vector machine
314
Approaches to analyse conditions of WT can be categorized into model-based and data driven ones. Model-based approaches attempts to incorporate physical models of system into estimation of the condition of WT. However, uncertainty due to the assumptions and simplifications of adopted model may pose significant limitations on this approach. Especially for SCADA data analysis we should use data-driven approaches. Data-driven
techniques utilize monitored operational data related to system health. They can be beneficial when understanding of first principles of system operation is not straightforward or when the system is so complex that developing an accurate model is prohibitively expensive. Data-driven approaches can often be deployed quickly and cheaply, and still provide wide coverage of system behavior. An added value of data-driven techniques is their ability to transform high dimensional noisy data into lower dimensional information useful for decisionmaking [Dragomir et al., 2007].
methods, ARMA models, etc.) and (2) Computational Intelligence (CI) techniques (Artificial Neural Networks, Fuzzy systems, etc.).We will focus on Computational Intelligence techniques in this paper. 5.1 Artificial Neural Networks
Pre-processing of SCADA is a must of extraction of useful information and patterns from huge data. The various data-driven methods being used for analysis of SCADA data from WTs can be divided into two categories: (1) statistical techniques (regression
ANNs can be used for a wide range of applications. They are inspired by the mechanism of the brain and can be classified by different categories as depending upon the learning mechanism or how they are trained (supervised \ unsupervised). ANNs accept input parameters, process them, and produce output parameters according to a nonlinear transfer function. Some of the key features for NN are their high processing speeds which are due to their massive parallelism, their proven ability to be trained, to produce instantaneous and correct responses from noisy or partially incomplete data, and their ability to generalize information over a wide range. These features make them a good choice for applying to WTs data analysis. Hofemann et al explored the possibilities to apply ANN not only within a wind park but on turbines located at different sites. Following the idea to develop a tool to forecast the particular loads of any wind turbine in the field without the need to install additional measuring systems, a model was developed that needed only signals contained in the SCADA data as input signals. They trained the network to predict the blade root moments of two wind turbines located at different sites. A supervised ANN was chosen, the load data was the target data of the network and was used to train the network as well as to verify the made predictions. Parameters as wind speed, wind direction, power output, pitch angle and yaw moment were used as input data.
315
An opportunity for increased transparency and openness of data-driven model is offered by fuzzy logic methods, which are increasingly proposed in modern diagnostic
and prognostic technologies. Based on the principles of Zadehs fuzzy set theory, fuzzy logic provides a formal mathematical framework for dealing with the vagueness of everyday reasoning [Zadeh, 1965]. As opposed to binary reasoning based on ordinary set theory, within the fuzzy logic framework measurement uncertainty and estimation imprecision can be properly accommodated.
Fuzzy systems are very useful in two general contexts like in situations involving highly complex systems whose behaviors are not well understood and in situations where an approximate, but fast, solution is desired. A further advantage of fuzzy systems is that existing expert knowledge can be implemented to improve the approximation by tuning, removing or adding of membership functions and rules.
316
parametric model to monitor the wind farm performance. The k-NN model produced good performance for the wind farm operating in normal conditions. In their later study they [Kusiak et al., 2009] proposed the concept of anticipatory control applied to wind turbines. A modified evolutionary strategy algorithm was used to solve a nonlinear constrained optimization problem. The proposed approach has been tested on the data collected from a 1.5 MW wind turbine. In a subsequent study Kusiak et al (2010), proposed a framework for optimization of the power produced by wind turbines. They proposed control approach generated optimized settings of the blade pitch and yaw angle. They used integration of data mining and evolutionary computation in their approach. In a later study on SCADA data Kusiak and Zhang (2010) found that vibrations of a wind turbine have a negative impact on its performance. Data-mining algorithms were used to build models with turbine parameters of interest as inputs, and the vibrations of drive train and tower as outputs. The performance of each model was thoroughly evaluated based on metrics widely used in the wind industry. The neural network algorithm outperforms other classifiers and is considered to be the most promising approach to study wind turbine vibrations. Ye et al (2010), proposed that with design of wind speed difference tests to detect both hard failures (cause complete shutdown of the turbine e.g. gear failure) and soft failures (degrade the turbine performance but do not necessarily stop the turbine e.g. anemometer faults). They used PSO based approach to learn from historical data to decide the location and size of the boundary i.e. abnormal state from the normal state. Further Kusiak et al (2010) presented an intelligent wind turbine control system based on models integrating the following three approaches: data mining, model predictive control, and evolutionary computation. The results produced by the intelligent control system were found to be better than those of the current wind turbine control system. For turbulent wind, the intelligent control system smoothed the power output, generator torque, and rotor speed without compromising the electricity demand. Uluyol et al (2011) proved that the wind turbine power curve analytic is useful for assessing wind turbine performance and generating robust indicators for component diagnostics and prognostics. This approach makes use of SCADA information and provides easy configuration based on process control approaches for condition-based monitoring. They used operational regime-based condition indicators that prevented false alarms and increased the possibility of fault isolation. This approach could also detect slow performance degradation caused by component wear as well as degradation due to an impending failure. Kusiak and Li (2011) used fault data provided by SCADA system. They carried out fault prediction at three levels: (1) fault and no-fault prediction; (2) fault category (severity); and (3) the specific fault prediction. They used power curve to determine the health of a wind turbine. The model extraction at level 1 and 3 was performed using the Neural Network (NN), Neural Network Ensemble (NN Ensemble), the Boosting Tree Algorithm (BTA), and the Support Vector Machine (SVM). Whereas at level 2 they used Neural Network (NN), the Standard Classification and Regression Tree (CART), the Boosting Tree Algorithm (BTA), and the Support Vector Machine (SVM). Furthermore they could successfully predict faults 5-60 min before they occur at each level. Kusiak and Verma [2011] used an association rule mining algorithm to identify frequent status patterns of turbine components and systems to predicted using historical wind turbine data. They explored five data-mining algorithms namely bagging,
317
ripper, rotation forest, random forest, and k-nearest neighbor (k-NN, k= 10). They concluded that the best predication results were obtained with random forest algorithm. Verma and Kusiak [2012], proposed a methodology to predict the generator brush worn fault in wind turbines. They used SCADA systems information from 27 wind turbines. They used algorithms (Chi-square statistic, Boosting tree algorithm, Wrapper (genetic search) for parameter selection, by which the ten most important parameters were identified. They built the model using four well-known data-mining algorithms, namely, multilayered perceptron (MLP), boosting tree, k-NN (k10), and support vector machine (SVM). These algorithms were studied to evaluate the quality of the models for predicting generator brush faults and it was found that the boosting tree algorithm provided the best prediction results. Zhang and Kusiak [2012], proposed three monitoring models for detecting abnormal vibration of wind turbines in time domain based on SCADA data. The sampling interval of SCADA data was 10 s, it allows to detect abnormal statuses of wind turbines in the time domain. The data collected included training, test, and error data. They used a modified k-means clustering algorithm to first vibration monitoring model. The k-means algorithm grouped data into clusters by examining their similarity. The clusters were then labelled as normal or abnormal statuses of wind turbine vibration based on the error reports of wind turbines. They further incorporated the concept of control charts to develop models for monitoring of turbine vibration. They also addressed the detecting of abnormal drive train and tower vibration of a wind turbine. They compared seven different data-mining algorithms, namely, neural network ensemble (NNE) , neural network (NN) , boosting regression tree (BT) , support vector machine (SVM) , random forest with regression (RF) , standard classification and regression tree (CART) , and k nearest neighbor neural network (kNN) . Four metrics, the mean absolute error (MAE), standard deviation of absolute error (SD of AE), mean square error (MSE), and the standard deviation of square error (SD of SE), are utilized to evaluate the performance of data-mining algorithms in model extraction. They found that the NNE model is recognized as the most suitable for determining drive train acceleration and the NN model is considered to be the most suitable algorithm for developing the model for predicting tower acceleration. Yang et al [2013] in their work proposed a technique which would interpret the SCADA data collected from wind turbines. They developed an effective method for processing raw SCADA data, further propose an alternative condition monitoring technique based on investigating the correlations among relevant SCADA data; and realised the quantitative assessment of the health condition of a turbine under varying operational conditions. Both laboratory and site verification tests have been conducted and were found satisfactory.
From the literature reviewed it has been demonstrated successfully by researchers that by keeping track of wind speed and power output parameters, the overall health of the turbine can be supervised. Furthermore SCADA data could be usefully used for CBM of WTs and locating faults. There has been success in using SCADA data for power prediction, optimal control settings, performance evaluation, predict turbine faults (predicting drive train acceleration/tower acceleration/gear box failure) and also the vibrations on a WT. Many AI techniques have been applied including NN, Fuzzy,
318
ANFIS, GAs etc [Yang et al,2008, Garcia et al, 2006]. Use of AI model in CBM of wind turbines is justifiable as there are many variables involved and it not easy to establish accurate mathematical model for such kind of complicated systems. It is also proposed to use data fusion techniques for monitoring the health of WTs [Dempsey and Sheng, 2013]. Few challenges that need to overcome before SCADA data analysis becomes fully successful are: SCADA data can differ from turbine to turbine and SCADA data changes with the operational conditions. So because of this it becomes difficult to differentiate between a real fault and fake fault which is quite a challenge. More over WT SCADA data are usually 10 min average data, so some of the information is lost. Thus referring to the research of wind turbine condition monitoring, a frame work is proposed that takes SCADA data and also an add on high frequency data from the sensors (some) to diagnose and prognose with both conventional data and SCADA data. After comparison of the two results, the appropriate method can be chosen for maintenance decision making. A number of models have been proposed tried and tested[Wilkinson et al, 2013, Schlechtingen & Ferreira Santos, 2011], however the accuracy of the model depends upon the careful selection of variables and the quality of the data (free from noise). So pre-processing of the data plays an important role in the accuracy of the models. Performance of AI based data mining algorithms and CBM algorithms are showing quite promising results. Hence by using computational intelligence concepts more efficient models could be obtained thus enhancing the accuracy and robustness of the model. It is proposed that Data mining (AI based) and evolutionary computations could be integrated for building the models for prediction and monitoring.
Conclusions
The major outcomes from this review are listed below: A frame work is proposed for diagnosis and prognosis of CBM of WTs using a combinational approach. These make use of conventional SCADA data and add on high frequency data from the sensors. In addition to this it also analyses historical SCADA data. Few challenges that need to overcome in SCADA data analysis are: firstly SCADA can differ from turbine to turbine, secondly SCADA data changes with the operational conditions. So pre-processing the raw data can equally contribute to the success of the CBM algorithm itself. Some researchers have successfully demonstrated the use of AI algorithms for SCADA data analysis. Seeing their performance it is believed that AI models could enhance the accuracy and robustness of models. Though the comparison results of various models are mentioned but no one clear cut perfect modelling technique could emerge. So this area remains open for further explorations.
References
319
Alsyouf, I. (2004). Cost Effective Maintenance for Competitive Advantages. Thesis for the degree of Doctor of Philosophy (Terotechnology) School of Industrial Engg., Vxj University, Sweden. Vxj University Press, ISSN: 1404-4307. Amjady, N. and Hedayatshodeh, M. (2012). A New Power Transformer Fault Diagnosis System and its Application for Wind Farms. Journal of Basic and Applied Scientific Research, Vol 2, no. 5, pp 47584764. Dempsey, P. J. and Sheng, S. (2013). Investigation of data fusion applied to health monitoring of wind turbine drivetrain components. Wind Energy, pp 479489. Donders, S. (2002). Fault Detection and Identification for Wind Turbine Systems: a closed-loop analysis. Masters Thesis,University of Twente Faculty of Applied Physics Systems and Control Engineering. Dragomir, O.E., Gouriveau, R., Zerhouni, N., Dragomir, F. (2007). Framework for a Distributed and Hybrid Prognostic System, In 4th IFAC Conference on Management and Control of Production and Logistics. Entezami, M., Hillmansen, S., Weston, P. and Papaelias, M. P. (2012). Fault detection and diagnosis within a wind turbine mechanical braking system using condition monitoring. Renewable Energy, Vol. 47, pp. 175182. Fischer, K., Besnard, F. and Bertling, L. (2012). Reliability-Centered Maintenance for Wind Turbines Based on Statistical Analysis and Practical Experience. IEEE Transactions on Energy Conversion, Vol. 27, no.1, pp.184195. Garcia, M. C., Sanz-Bobi, M. A. and Del Pico, J. (2006). SIMAP: Intelligent System for Predictive Maintenance. Computers in Industry, Vol. 57, no. 6,pp. 552568. Garca Mrquez, F. P., Tobias, A. M., Pinar Prez, J. M. and Papaelias, M. (2012). Condition monitoring of wind turbines: Techniques and methods. Renewable Energy, Vol. 46, pp. 169 178. Hameed, Z., Ahn, S. H. and Cho, Y. M. (2010). Practical aspects of a condition monitoring system for a wind turbine with emphasis on its design , system architecture , testing and installation. Renewable Energy, Vol. 35, no. 5, pp. 879894. Hameed, Z., Hong, Y. S., Cho, Y. M., Ahn, S. H. and Song, C. K. (2009). Condition monitoring and fault detection of wind turbines and related algorithms: A review. Renewable and Sustainable Energy Reviews, Vol. 13, no. 1,pp. 139. Hyers, R. W., Mcgowan, J. G., Sullivan, K. L., Manwell, J. F. and Syrett, B. C. (2006). Condition monitoring and prognosis of utility scale wind turbines. Energy Materials: Materials Science and Engineering for Energy Systems, Vol. 1, no. 3, pp. 187203. Kaldellis, J. K. and Zafirakis, D. (2011). The wind energy (r)evolution: A short review of a long history. Renewable Energy, Vol. 36, no. 7,pp. 18871901. Kim, K., Parthasarathy, G., Uluyol, O., Foslien, W., Shuangwen, S. and Fleming, P. (2011). Use of SCADA Data for Failure Detection in Wind Turbines. Conference Paper NREL/CP-500051653. Kusiak, A. and Li, W. (2011). The prediction and diagnosis of wind turbine faults. Renewable Energy, Vol. 36, no. 1, pp. 1623. Kusiak, A., Li, W. and Song, Z. (2010). Dynamic control of wind turbines. Renewable Energy,Vol. 35, no. 2, pp. 456463. Kusiak, A., Song, Z. and Zheng, H. (2009). Anticipatory Control of Wind Turbines With DataDriven Predictive Models. IEEE Transactions On Energy Conversion, Vol. 24, no. 3, pp. 766 774. Kusiak, A. and Verma, A. (2011). Prediction of Status Patterns of Wind Turbines: A Data-Mining Approach. Journal of Solar Energy Engineering, Vol. 133, no. 1, pp. 011008101100810. Kusiak, A. and Zhang, Z. (2010). Analysis of Wind Turbine Vibrations Based on SCADA Data. Journal of Solar Energy Engineering,Vol. 132, no. 3, pp. 031008103100812.
320
Kusiak, A., Zheng, H. and Song, Z. (2009). Models for monitoring wind farm power. Renewable Energy, Vol. 34, no. 3, pp. 583590. Kusiak, A., Zheng, H. and Song, Z. (2010). Power optimization of wind turbines with data mining and evolutionary computation. Renewable Energy, Vol. 35, no. 3, pp. 695702. Kusiak, B. A. and Verma, A. (2010). The Future of Wind Turbine diagnostics. windsystemsmag.com, pp. 6671. Lei, Y., Lin, J., He, Z. and Kong, D. (2012). A method based on multi-sensor data fusion for fault detection of planetary gearboxes. Sensors,Vol. 12, no. 2, pp 200517. Liu, W. Y., Zhang, W. H., Han, J. G. and Wang, G. F. (2012). A new wind turbine fault diagnosis method based on the local mean decomposition. Renewable Energy, Vol. 48,pp. 411415. Lu, B. (2009). A Review of Recent Advances in Wind Turbine Condition Monitoring and Fault Diagnosis. Power Electronics and Machines in Wind Applications, IEEE, pp. 1 7. Nadakatti, M., Ramachandra, A. and Kumar, A. N. S. (2008). Artificial intelligence-based condition monitoring for plant maintenance. Assembly Automation, Vol. 28, no. 2, pp. 143150. Ramageri, B. M. (2010). Data mining techniques and applications. Indian Journal of Computer Science and Engineering,Vol. 1, no. 4, pp. 301305. Schlechtingen, M. and Ferreira Santos, I. (2011). Comparative analysis of neural network and regression based condition monitoring approaches for wind turbine fault detection. Mechanical Systems and Signal Processing,Vol. 25, no. 5, pp. 18491875. Uluyol, O., Parthasarathy, G., Foslien, W. and Kim, K. (2011). Power Curve Analytic for Wind Turbine Performance Monitoring and Prognostics. Annual Conference of the Prognostics and Health Management Society, pp. 18. Uraikul, V., Chan, C. W., and Tontiwachwuthikul, P. (2007). Artificial intelligence for monitoring and supervisory control of process systems. Engineering Applications of Artificial Intelligence,Vol. 20, no.2,pp. 115131. Verma, A. and Kusiak, A. (2012). Fault Monitoring of Wind Turbine Generator Brushes: A DataMining Approach. Journal of Solar Energy Engineering, Vol. 134, no. 2, pp. 0210011 0210019. Wiggelinkhuizen, E., Verbruggen, T., Braam, H., Rademakers, L., Xiang, J. and Watson, S. (2008). Assessment of Condition Monitoring Techniques for Offshore Wind Farms. Journal of Solar Energy Engineering, Vol. 130, no. 3, pp. 03100410310049. Wilkinson, M., Darnell, B. and Harman, K. (2013). Annual Comparison of Methods for Wind Turbine Condition Monitoring with SCADA Data. EWEA 2013 Vienna, 4-7 February 2013, pp. 47. Yang, S., Li, W. and Wang, C. (2008). The intelligent fault diagnosis of wind turbine gearbox based on artificial neural network. International Conference on Condition Monitoring and Diagnosis, pp. 13271330. Ye, X., Yan, Y. and Osadciw, L. A. (2010). Learning Decision Rules by Particle Swarm Optimization ( PSO ) for Wind Turbine Fault Diagnosis. Annual Conference of the Prognostics and Health Management Society. Zadeh, L.(1965). Fuzzy sets, Inform. Control, Vol. 8, pp. 338353. Zaher, A. S. and McArthur, S. D. J. (2007). A Multi-Agent Fault Detection System for Wind Turbine Defect Recognition and Diagnosis. IEEE Lausanne Power Tech, pp. 2227. Zhang, Z. and Kusiak, A. (2012). Monitoring Wind Turbine Vibration Based on SCADA Data. Journal of Solar Energy Engineering, Vol. 134, no. 2, pp. 021004-1-021004-12.
321