One of the unique features of the PHM conferences is a full day of free technical tutorials on various topics in health management taught by industry experts. At the first European PHM conference, tutorials will take place on day one, Monday September 24. As educational events tutorials provide a comprehensive introduction to the state-of-the-art in the tutorial’s topic. Proposed tutorials address the interests of a varied audience: beginners, developers, designers, researchers, practitioners, and decision makers who wish to learn a given aspect of prognostic health management. Tutorials will focus both on theoretical aspects as well as industrial applications of prognostics. These tutorials reach a good balance between the topic coverage and its relevance to the community.
Monday Sep 24, 2012
Location: Nicollet D2/D3
Time: 8am –5pm
- Introduction to Improved Real-time Mechanical Systems Diagnostics
Presented by Carl Byington (Impact Technologies)
- Feature Extraction Methods
Presented by Tianyi Wang & Subrat Nanda (GE Global Research)
- An introduction to Prognosis, Uncertainty Representation, and Risk Measures
Presented by Marcos Orchard (University of Chile)
- Practical Data Mining
Presented by Ravi Patankar (Honeywell)
- System-wide Health Monitoring
Presented by Raj Bharadwaj (Honeywell)
PHM conference tutorials have been a popular event in the past and the PHM Society is proud to continue this service to the community. Tutorials from the past conferences can be freely accessed from respective conference pages.
Kai Goebel firstname.lastname@example.org
George Vachtsenavos. email@example.com
Tutorial Presenter: Carl Byington, Impact Technologies, a Sikorsky Innovations Company
Mechanical system diagnostics typically need to address structural fatigue and tribological-driven failure mechanisms. To provide more comprehensive coverage for these possible failure modes, consideration of both structural wave propagation (vibration, etc.) methods as well as fluid diagnostics technologies may be used to improve detection and diagnostic performance. Vibration-based diagnostics analysis has proven to provide some of the most quantitative and reliable indicators of rotating member fatigue detection and diagnosis. Methods to process such data from a diagnostic isolation perspective will be reviewed. Their usefulness however for certain wear-related or corrosion driven mechanisms may be less than desirable to achieve clear actionable information. In addition, there will be a discussion of the use of real-time fluid diagnostics for oil condition and debris characterization. Specifically candidate technologies that are being evaluated to reduce the dependency on oil sampling for these indicators will be presented and a correlation to addressable root cause issues will be covered. Lastly, the tutorial seeks to motivate the user to understand how to fuse or combine sources of evidence from each of these sources to improve the overall diagnostic performance. By integrating on-line oil quality/debris measurements with vibration measurements, gearbox failure mode indicators can be fused to produce improved fault classification, maintenance action recommendations, and component replacement recommendations.
Carl Byington is a registered Professional Engineer and prior Owner and Technical Director at Impact Technologies, now a Sikorsky Innovations Company. He possesses over 22 years of design, analysis, and testing experience with fluid power, thermal, mechanical, and electrical systems. He has performed as the Principal Investigator (PI) for innovative Prognostics and Health Management (PHM) and Condition-based Maintenance (CBM) technologies on over 60 programs for the military, NASA, and other customers. Carl is a member of ASME, IEEE, AIAA, SAE, STLE and AHS professional societies. He is also a past Technical Program Chairman of the MFPT Society and the PHM Society for the 2010 and 2011 meetings. For the past 10 years, he has been a Lead Instructor for the annual “PHM/CBM Design” Short Course offered by Impact to government and industry participants and has also been a guest lecturer at the Georgia Tech Fault Diagnostics course. In prior employments, Carl was the CBM Department Head at the Penn State Applied Research Lab and performed propulsion research at NASA Langley Research Center. He possesses degrees in Mechanical and Aeronautical Engineering and has produced well over 100 publications related to advanced sensing techniques; signal processing; vibration monitoring; diagnostics and control; prognostics; data fusion; and equipment health management. He is also coauthor on 5 US awarded or pending patents related to predictive monitoring technologies.
Tutorial Title: Feature Extraction Methods [download]
Tutorial Presenters: Tianyi Wang and Subrat Nanda, GE Global Research
Feature extraction transforms raw signals into more informative, and in many cases more compact, signatures or fingerprints of a system. It plays a significant role in data driven diagnosis and prognosis modeling. Selection of a good feature set, in most cases, has a greater impact to the performance of the whole analytic solution than the selection of the downstream analytic models. However, feature extraction is challenging because it usually requires a certain level of understanding to the subject matter and somehow relies on the experience of the model developer. In this tutorial, we will summarize a variety of frequently used feature extraction algorithms for different signal types and applications, introduce the approaches to semi-automate the feature extraction process, and discuss the various issues and their handling methods regarding to feature extraction in real world PHM applications.
Tianyi Wang is an Information Scientist in the Machine Learning Laboratory at GE Global Research since 2010. His research interests include computational intelligence, data mining, equipment diagnostics and prognostics, signal processing, etc. Previously, he worked in the NSF I/UCR Center for Intelligent Maintenance Systems (IMS) at University of Cincinnati (2005-2010) on Prognostics and Health Management (PHM). He was the overall winner of the 2008 PHM data challenge, and the second winner in industry category of the IEEE 2012 PHM challenge. Tianyi Wang received a B.E degree in Mechanical Engineering from Tsinghua University, Beijing, China, in 2002, an M.S. degree in Mechatronics from University of Siegen, Siegen, Germany, in 2005, and a Ph.D. degree in Industrial Engineering from University of Cincinnati, Cincinnati, OH, in 2010. He is a member of IEEE and member of IEEE Computer Society.
Subrat Nanda was born in New Delhi, India and is a Lead Engineer at GE Global Research Center. He earned his M.S degree in Autonomous Systems from University of Exeter, United Kingdom in 2003 and a B.Engg degree in Production Engineering from Nagpur University in India in 2001 and. His research interests are mainly in applying machine learning and pattern recognition methods to industrial problems such as prognostics & health management, condition monitoring and risk assessment, reliability engineering, bio-inspired optimization and fusion of data driven methods with physics. He is a member of IEEE and IEEE Reliability Society and has lead and directly contributed to several research projects in the area of PHM and data driven modeling.
Tutorial Presenter: Marcos Orchard, Universidad de Chile
Uncertainty and risk management holds the key for a successful penetration of health management strategies in industrial applications. While methods to estimate and handle uncertainty have received a reasonable amount of attention in the diagnostics domain, uncertainty management for prognostics is an area which awaits major advances. This tutorial explores some of the most important aspects associated to the problem of failure prognosis and risk evaluation, with special emphasis on the utilization of Particle Filter (PF) algorithms and outer feedback correction loops for uncertainty representation and management in long-term prediction.
Dr. Marcos Orchard is Assistant Professor at the Department of Electrical Engineering, Universidad de Chile and was part of the Intelligent Control Systems Laboratory at The Georgia Institute of Technology. His current research interest is the design, implementation and testing of real-time frameworks for fault diagnosis and failure prognosis, with applications to battery management systems, mining industry, and finance. For nearly ten years, Dr. Orchard has been exposed to the theoretical aspects of nonlinear state estimation and worked extensively in research projects regarding system identification, fault detection and failure prognosis. His field of expertise includes the implementation of diverse statistical monitoring systems (wavelets, filtering techniques,) and parametric/non-parametric modeling techniques such as fuzzy expert systems, and neural networks. His research work at the Georgia Institute of Technology was the foundation of novel real-time fault diagnosis, failure prognosis approaches based on Particle Filtering (Sequential Monte Carlo methods, SMC) and other Bayesian estimation techniques.
Tutorial Title: Practical Data Mining [download]
Tutorial Presenter: Ravi Patankar, Honeywell Aerospace
The tutorial will discuss practical issues in data mining for PHM. Different sources of data that can contribute to PHM knowledge will be introduced. Traditional maintenance data collection processes and the needs of data mining will be assessed.
Examples of data mining will be shown with the problems and solutions in
1) Parameter optimization,
2) empirical modeling ,
3) supervised learning, and ,
4) unsupervised learning.
The focus will be on supervised and unsupervised learning examples to generate PHM knowledge. The examples will use RapidMiner, an open source, user friendly graphical tool for data mining and visualization. Attendees are encouraged to bring their computers and to install RapidMiner to follow the tutorial with some hands-on data mining exercises.
Ravi Patankar has been employed at Honeywell Aerospace since 2007 as a Principal Systems Engineer. He worked as a Program Manager at Intelligent Automation, Inc from 2004-2007 and as an Assistant Professor of Mechanical Engineering at Michigan Technological University from 2001-2004 and as a Project Engineer at Delphi Corp. from 1999-2001. He received a PhD in Mechanical Engineering from the Pennsylvania State University in 1999. He received a Bachelor’s and Master’s degree in Mechanical Engineering from the Indian Institute of Technology, Mumbai in 1994.
His areas of research and contribution are control systems, fatigue modeling, diagnostic and prognostic algorithms, maintenance reasoners and data mining.
Tutorial Title: System-wide Health Monitoring [download]
Tutorial Presenter: Raj Mohan Bharadwaj, Honeywell
A systems view is necessary to monitor, detect, diagnose, predict, and mitigate adverse events to maintain asset operation and avoid damage or accidents. The design of the monitoring system is mainly driven by the system health monitoring objectives; such as safety, maintenance, operational costs and the type of information exchanged by the monitored subsystems. For example, on many aircrafts, different subsystems are provided by different vendors. Those vendors often compete on platforms and do not want to share subsystem design information. Therefore, a system-wide health monitoring solution must work with heterogeneous data types that provide protection for proprietary information. The first half of this tutorial will cover the high level design choices and present a primer on techniques used to accomplish the stated objectives. In the second half of the tutorial we will discuss system-wide health monitoring with specific case studies from aircraft and wind farm monitoring systems.
The first case study describes the latest developments on the vehicle level reasoning within a model-based framework used by the central maintenance system on Boeing 777 and 787. This tutorial covers a number of approaches that utilize the progress made in developing prognostic indicators to integrate the heterogeneous (both continuous and discrete signals), asynchronous data streams for detection of potential adverse events. The tutorial will show the methodology of developing reference models, reasoning and use of data mining to develop and support the health monitoring system.
The second case study explores the challenges of monitoring a modern wind farm that have several tens or even hundreds of wind turbines at a site, located in remote locations, and operate under severe environments. It is no surprise that operations and maintenance costs of wind farms are high. We will discuss the challenges in distributed monitoring of wind turbines. Finally, this cases study will show how to apply the associative model methodology to capture the relationship amongst several wind turbines on a farm to detect anomalous conditions.
Raj Bhardwaj received Ph.D. in Electrical Engineering from Texas A&M University. He is currently Principal R&D Scientist in the Vehicle Health Management group at Honeywell Aerospace Advance Technologies. His work at Honeywell is centered on prognostics health management, algorithms and system design. He is the Principal Investigator/Program Manager for several PHM programs including AATD Drive Train prognostics, NASA Vehicle Level Reasoning and FAA Sensory Prognostics Management Systems projects. Prior to joining Honeywell he was with the General Electric Global Research Center, Niskayuna, NY. At GE, he worked on diagnostics for a variety of power system equipment and locomotives. He was also the newsletter editor and a board member for the Society for Machinery Failure Prevention Technology (MFPT) from 2005 to 2011. He is the recipient of IEEE Industry Application Society 2006 First Prize Paper and is member of the VLRS team that received “2011 Associate Administrator Award for Technology Innovation” from NASA Aeronautics Research Mission Directorate. He has co-authored over 25 publications and has 7 assigned patents.