Prof. Mingjian Zuo, Canadian academician, University of Alberta, Canada
Machine Learning and its Applications in Prognosis and Health Management
Machine learning has great potential for reliability assurance through prognosis and health management (PHM) of engineering assets. It has been attracting attention from both academic and industrial sectors. Recent developments of machine learning, especially the evolving branches of deep learning, transfer learning, and reinforcement learning, bring new opportunities for effective PHM. This presentation will first introduce some general knowledge of machine learning and its applications in various disciplines. We will then introduce some fundamentals of deep learning, with emphasis on artificial neural networks. Two other branches of machine learning, i.e., transfer learning and reinforcement learning, will then be discussed. Finally, our recent research work on developing machine learning techniques for PHM will be described.
Dr. Mingjian Zuo received his Ph.D. degree in Industrial Engineering from Iowa State University, Ames, Iowa, U.S.A. He is currently Full Professor in the Department of Mechanical Engineering at the University of Alberta, Canada. His research interests include system reliability analysis, maintenance modeling and optimization, signal processing, and fault diagnosis. He is Department Editor of IISE Transactions, Associate Editor of IEEE Transactions on Reliability, Associate Editor of Journal of Risk and Reliability, Associate Editor of International Journal of Quality, Reliability and Safety Engineering, Regional Editor of International Journal of Strategic Engineering Asset Management, and Editorial Board Member of Reliability Engineering and System Safety, Journal of Traffic and Transportation Engineering, and International Journal of Performability Engineering. He is Fellow of Canadian Academy of Engineering (CAE), Fellow of Institute of Industrial and Systems Engineers (IISE), Fellow of Engineering Institute of Canada (EIC), Founding Fellow of International Society of Engineering Asset Management (ISEAM), and Senior Member of IEEE.
Prof. Stephan Heyns, University of Pretoria, South Africa
Useful concepts in the enhancement of vibration monitoring under noisy non-stationary conditions.
Vibration monitoring of rotating machinery under noisy non-stationary operating conditions remains difficult because of factors such as amplitude and frequency modulation and impulsive noise that impede the application of conventional condition indicators.
To improve the performance of these monitoring techniques various techniques that rely on traditional signal processing can be used. These range from the enhancement of weak damage components in the vibration signals, identifying informative frequency bands and to the use of new synchronous statistics for gearbox fault diagnostics.
Learning-based methods provide a complementary perspective on the signal processing problem, by potentially alleviating the shortcomings of traditional condition monitoring methods.
This paper highlights some of these concepts to enhance vibration monitoring under non-stationary conditions from signal processing and learning-based perspectives.
We emphasize the complementary nature of signal processing and learning-based approaches for improved condition monitoring.
Stephan Heyns is professor and director of the Centre for Asset Integrity Management in the Department of Mechanical and Aeronautical Engineering at the University of Pretoria in Pretoria, South Africa.
He was awarded his PhD in mechanical engineering from the University of Pretoria in 1988.
His personal research interests focus on rotating machinery diagnostics and prognostics with a special emphasis on gearbox, bearing and turbomachinery applications. He is interested in the application of signal processing as well as machine learning approaches and recently specifically in the complementary use of these approaches.
He is an accredited researcher with the National Research Foundation in South Africa, a registered professional engineer in South Africa, as well as a fellow of the South African Academy of Engineering, a fellow of the Royal Aeronautical Society, a fellow of the International Society of Engineering Asset Management and honorary fellow of the South African Institution of Mechanical Engineers.
Prof. Radoslaw Zimroz, Wroclaw University of Science and Technology, Poland
|Prof. Agnieszka Wyłomańska,Wroclaw University of Science and Technology, Poland|
Radoslaw Zimroz received an MSc degree in Acoustics from the Faculty of Electronics, WUST, in 1998, PhD in condition monitoring of mining machines in 2002 (Faculty of Mining, WUST). In 2004-2005 he was with the Applied Math and Computing Group (AMAC) @ Cranfield University working on the project supported by Caterpillar. In 2011 he received a habilitation degree based on the book on “adaptive approaches for condition monitoring in time varying load”. In 2012 he received a professor position @ WUST, in 2020 he is full professor.
In parallel he has started a service as professor in KGHM R&D Center, where he has created a department related to data analysis and predictive maintenance. He participated in several national and EU projects related to metrology, data analysis, condition monitoring, predictive maintenance, inspection robotics etc. His main interest is vibration signal processing and machine learning methods for fault detection (mostly gearboxes and bearings). He was (is) guest editor in Applied Acoustics, Shock and Vibration, Sensors, Electronics, Remote Sensing, International Journal of Mining Reclamation and Environment. He is member of the Editorial Board in Machines, Applied Science, Shock and Vibration, Frontiers in Mechanical Engineering. He is Topic Editor in Diagnostyka, Co-Editor of Applied Condition Monitoring Book Series (Springer). He is a member of IEEE, Polish Society of Technical Diagnostics (vice-president), Society of Mining Professors, and the Mining Committee of Polish Academy of Science. Now he is Dean of the Faculty of GeoEngineering, Mining and Geology and informal head of Digital Mining Center @ WUST.
Agnieszka Wyłomańska received the M.Sc. degree in Financial and Insurance Mathematics from Institute of Mathematics and Computer Science at the Wroclaw University of Technology (WUT), Poland in 2002 and the Ph.D. degree in Mathematics from WUT in 2006. In years 2007-2016 she was an Assistant Professor with the Faculty of Pure and Applied Mathematics (previously Faculty of Fundamental Problems of Technology), WUT. Currently she is a Professor of WUST (Wroclaw University of Science and Technology) and a member of the Hugo Steinhaus Center for Stochastic Processes. In 2015 she received a D.Sc. degree in Mining and Geology from the Faculty of Geoengineering, Mining and Geology, WUST. Her area of interest relates to time series analysis, stochastic modelling and statistical analysis of real data (especially technical data related to the mining industry, indoor air quality and financial time series). She is an author of more than 100 research papers. She cooperates with industrial companies, especially from the mining industry.
|Speech Title: |
Signal processing techniques for local damage detection in gears and bearings in presence of non-Gaussian noise - a stochastic perspective
During the Keynote we will present recent developments related to signal processing for local damage detection in bearings and gearboxes with focus on heavy duty mining mechanical systems. We will address following issues：Why local damage? - still hot topic for condition monitoring community. State of the art - focus on heavy duty industry perspective. Challenges - T-V Load/speed conditions, non-Gaussian noise, multiple damage, mixture of various sources, poor SNR. Recent solutions - cyclostationary analysis in presence of nonGaussian noise, optimal filter design for SOI extraction, source separation, denoising. Methods - alternative dependence measures, robust statistics, statistical modelling, stochastic processes, NonNegative Matrix Factorisation. Future - inspection robots, acoustic signals. Each approach presented in the talk will touch objects, their specific design and operational factors, math background for proposed solutions and finally results obtained for real data from industry.
Prof. Luc Thévenaz, IEEE Fellow, École Polytechnique Fédérale de Lausanne(EPFL), Switzerland
Distributed optical fibre sensing - a powerful tool for structural monitoring and for a safer society
Distributed fibre sensing offers the only opportunity to monitor systems by collecting simultaneously a large number of data using a single instrument and a very miniature and seamlessly integrated sensing element: the optical fibre. After explaining the principles, the presentation will be illustrated by some representative applications.
Prof. Zhike Peng, Shanghai Jiao Tong University, China
To be added soon.
Prof. Ruqiang Yan, Xi'an Jiaotong University, China
Machine Doctor in the Era of Artificial Intelligence
In our daily life, when we are sick, we usually go to the hospital and ask experienced doctors for diagnosis and treatment. The long-term operation of machines will also produce "disease", that is the so-called fault. We need to seek the help of "doctor" of the machine to diagnose the occurrence of the fault and predict its development trend, and then provide guidance for its operation and maintenance. The new generation of artificial intelligence technology represented by deep learning provides a new way of intelligent diagnosis for machine doctors. On the basis of introducing the development history of artificial neural network, this talk introduces the concept and characteristics of deep learning, and then discusses several typical deep network models and their application in intelligent diagnosis of machines, as well as the development trend of deep learning in the future.
Prof. Ruqiang Yan is Director of International Machinery Center, School of Mechanical Engineering, Xi’an Jiaotong University. He received his Ph.D. degree in Mechanical Engineering from the University of Massachusetts Amherst, USA, in 2007. Prof. Yan is a Fellow of American Society of Mechanical Engineers (ASME), and received the Millions of Leading Engineering Talents Award and the IEEE Instrumentation and Measurement Society Technical Award in 2019. His research interests include advanced data analytics, artificial intelligence, and energy-efficient sensing and sensor networks for the condition monitoring and fault diagnosis of complex engineering systems.
Prof. Qingbo He, Shanghai Jiao Tong University, China
Parameterized Time-Frequency Analysis: Methods and Applications
It is of increasing importance to utilize collected vibration signals for real-timely machine condition monitoring and effectively diagnosing possible faults to make an efficient machine maintenance strategy. Non-stationary signals exhibit the time-varying frequency characteristics that generally appears in a wide range of engineering applications, including vibration measurements in machinery condition monitoring. Time-frequency analysis (TFA) has become a powerful tool to analyze non-stationary signals, as it is well-known to be able to characterize signals in a joint time-frequency domain. Almost all classical TFA methods are non-parameterized methods, such as short-time Fourier transform, continuous wavelet transform, and Wigner-Ville distribution. Theoretically, traditional non-parameterized TFA can analyze any signal, but it is unable to provide the best representation for complex signals. Another emerging research direction in this area is parameterized TFA. Parameterized TFA provides a better representation of signal by parameterizing kernel functions using additional parameters. Recently, parameterized TFA has become a new trend to analyze signals with complex time-frequency structures. This lecture first briefly revisits non-parameterized TFAs, then the lecture reviews three new parameterized TFA methods developed. After that, the nonlinear chirp mode decomposition methods are described for separating the non-stationary signal components. Finally, the applications of parameterized TFA in several cases are provided. The lecture demonstrates the significance of parameterized TFA for advancing non-stationary signal analysis in condition monitoring of machinery.