Two-Day Short Course: PHM Fundamentals – From Monitoring/Sensing to Fault Diagnosis and Failure Prognosis
The Potential Customer Base:
- The Operations Manager/Field Commander: What is my confidence that I can deploy a particular asset for a specific mission/task?
- The Operators/Maintainers/Maintenance Technology Developers: Methods/tools for maintenance, repair and overhaul of critical systems
- The System Designer: How do I take advantage of CBM/PHM technologies to design high-confidence or fault-tolerant systems?
- The Educator: New course directions in emerging PHM technologies
- Small High-Tech Company Principals: Getting a share in the CBM/PHM market
What will they learn from this course?
- Basic understanding of state of the art of CBM/PHM technologies
- Sensors & Sensing Strategies — design, placement, operation and requirements
- Algorithm Development for Data Processing, Fault Diagnosis & Failure Prognosis
- Data Miming — methods to reduce raw censor data to usable information
- PHM Technologies demonstrated via actual case studies
|1. Introduction — State of PHM Technologies
Andrew J. Hess, The Hess PHM Group — President, PHM Society President,
|An overview of the technologies comprising PHM will be given with reference to historical evolutions in diverse sectors and domains. The technologies, sciences and skills that contribute to the broad multidisciplinary field of PHM will be defined with reference to the new PHM Society Capability Taxonomy. The organization of the course elements will be introduced with this basis.|
|2. The PHM Paradigm — From monitoring/sensing to fault diagnosis and failure prognosis
Dr. George Vachtsevanos, Professor Emeritus, The Georgia Institute of Technology
|This session will review basic concepts, tools and methods that comprise the Prognostics and Health Management technologies with emphasis on prognostics. We will introduce the constituent PHM topics from system monitoring and sensing strategies to data mining and diagnostics/prognostics. We distinguish between on-line, real time or health-based prognostics vs. usage-based prognostics employed in modern reliability analysis methods. These topics will covered in more detail in subsequent sessions. The intent is provide the attendees with an understanding of how these topics form a sequence of entities that start with requirements definition , failure analysis and understanding of the physics of failure mechanisms, monitoring and data processing and culminate with the development and implementation of robust and verifiable diagnostic and prognostic algorithms. We will show how these topics in sequence result in a holistic approach to health maintenance of critical engineered systems/processes.|
|3. CBM+ Technologies
Dr. George Vachtsevanos, Professor Emeritus, The Georgia Institute of Technology,
|A significant paradigm shift is clearly happening in complex engineered systems’ maintenance and support. The old-time coal miners used canaries to monitor the health of their mines. The “old” approach was to put a canary in a mine, watch it periodically, and if it died, you knew that the air in the mine was going bad. The “canary” paradigm has been adopted for component health monitoring where the “canary” now is a suitable that is prone to early failure due to environmental or other stresses providing a much earlier indication that the targeted component is starting to fail. The “new” approach would be to use current available technologies and capabilities to continuously monitor the health of critical systems and report as early and as accurately as possible on their health status. We will introduce fundamental concepts of CBM+ in this session that aim to increase operational availability and readiness throughout the system’s life cycle. It provides an overview of the current state of the art and details of a systems engineering process to health management of components/systems that builds upon advanced Failure Modes and Effects Criticality Analysis (FMECA), methods for fault monitoring, seeded fault testing, data management and data processing, Condition Indicator (CI) extraction and selection, incipient failure diagnosis and prognosis. We will also cover the extensions to traditional Condition Based Maintenance that constitute the + in CBM+. These include linkage between monitoring on assets to at-platform analysis and maintenance procedures and corporate business systems.|
|4. Sensors and Sensing Strategies
Dr. Karl Reichard, Research Associate, The Pennsylvania State University Applied Research Laboratory
|One of the cornerstones of prognostic health management is the acquisition of data related to the health and condition of the system. PHM systems acquire data and information from a variety of sources, but most of that data starts as the electrical output of some type of sensor. It is important to understand the mechanisms by which observable phenomena are converted into sensor signals, then into digital data. The digital data must then be converted into accessible formats used in diagnostic, prognostic, data fusion, and reasoning algorithms, modeling, or data analytics. This lecture will discuss the fundamental physics of the most common sensors used in PHM systems and the associated requirements for acquiring and converting the sensor outputs into digital data streams.|
|5. Data Mining — Feature selection and extraction
Dr. Neil Eklund, Chief Data Scientist, Schlumberger New Technology and Innovation Center
|Often a system will perform differently depending on the operating regime, making it helpful – if not essential – to model the normal operation of a system under different conditions. Even for well understood systems, it is often difficult to make a first-principles model that emulates system behavior well. Other systems are impossible to model. The capability to model a system from historical data can be a huge advantage in detecting faults and estimating remaining life. This session will outline how to perform data-driven system modeling, including feature selection, outlier accommodation, model selection, and model fusion.|
|6. Fault Diagnosis
Dr. José R. Celaya, Research Scientist / SGT Inc., NASA Ames Research Center
|Traditionally, fault detection and isolation has been at the core of PHM technologies. Fault diagnostics techniques are varied depending on the application domain and in the PHM context, these techniques are heavily dependent on data along with models that represent the physical behavior of the system. A complete fault diagnosis solution will consist of mathematical formalisms (models) which could be conceived purely based on data (observations on the system) or could be constructed starting from the physical knowledge of the system. This session will focus mainly on model-based methods.|
|7. Failure Prognosis
Dr. Marcos Orchard, Associate Professor, University of Chile
|Particle filters (PF) have been established as the de facto state-of-the-art in failure prognosis, and particularly in the representation of the uncertainty that is associated to long-term predictions. This talk explores some of the most important aspects associated to the problem of failure prognosis, with special emphasis on the utilization of Particle Filter (PF) algorithms and outer feedback correction loops for uncertainty representation and management. Application examples will include problems associated to battery management systems, mechanical systems, and volatility prediction in finance.|
|8. Analytical Foundations and Numerical Solutions for PHM
Mark A. Powell, Consultant, Adjunct Professor, Stevens Institute of Technology
|Failure and fault prognosis via PHM is performed to enable good operational decisions. Some of these decisions concern optimal structuring of a maintenance and logistics program; some involve real time operational decisions to preclude a fault or failure. Starting with first principles of physics, math, statistics, decision theory, and information theory, we will in this session derive the equations the solutions to which are necessary to make these good RAM and PHM decisions. These derivations will be assumption free, taking advantage of information theoretic models providing maximum objectivity. We will then discover that these equations necessary for good decision making in RAM and PHM are neither analytically tractable nor solvable via ordinary numerical methods packages. All is not lost however; modern numerical methods will be presented enabling quick, easy and accurate solutions to these equations.|
|9. Performance Metrics
Dr. Abhinav Saxena, Research Scientist / SGT Inc., NASA Ames Research Center
|Metrics are required to determine the performance of prognostics methods. This session will examine the unique methods needed to determine requirements and to rate the quality of the prognostic tools. More importantly, characteristics of various different prediction methods will be highlighted that require different treatments performance evaluation. Special attention will be given to performance evaluation of condition-based prediction methods. State-of-the-art performance metrics will be presented and illustrated through their use in a variety of applications. It will be discussed how application context and end user requirements call for different performance requirements and how such requirements can be derived for prognostic algorithms. At the end of this session participants will be able to understand what are key attributes of prognostic performance that should be measured and potentially identify the performance metrics that are relevant to their applications.|
|10. PHM Case Studies
|a. PHM Technology Transition – From Aerospace Engines to Surface Vehicles:
Marvin Zaluski, National Research Council Canada.
This case study presents new analytical and statistical tools/techniques developed for the CF-18 aircraft exploiting: operational flight data from sensors and Built-In Test Equipment (BITE) and maintenance activities recorded by personnel. This case study first investigates the utility of readily available data to develop data mining-based models for PHM systems. Then lessons learned in the development of data mining models/tools are discussed. Next some synergies are identified in the surface transportation industry and are linked to expected outcomes of the National Research Council of Canada new Fleet Forward 2020 program’s Vehicle Diagnostic and Prognostic activities.
b. Prognosis in Structural Health Monitoring – Case Studies on Metallic and Composite Structures:
Carl Byington, Sikorsky Aircraft Corporation and Dr. Abhinav Saxena, SGT Inc., NASA Ames Research Center.
Majority of structural health monitoring literature tends to focus on damage detection and diagnosis and not so much on prognostics. This session will illustrate some recent developments in predicting damage growth to assess Remaining Useful Life (RUL) of structures through several case studies. These case studies will briefly touch upon various different methods used for prognostics in structures. A more detailed discussion will focus on prognostics for composites. While modeling fault growth in metallic structures is well understood, it has been very challenging to develop damage growth models for composites due to their non-homogenous structure. Results from a case study will be presented where experimental data collection and analysis was augmented by Bayesian model selection for identifying the most suitable damage growth model and making predictions. Overall, this session will provide an appreciation of what is needed for making condition based life-predictions for structures beyond existing SHM technologies and what approaches may be taken to accomplish this.
c. Battery Health Management and Reliability Analysis:
Dr. Marcos Orchard University of Chile and Dr. George Vachtsevanos, The Georgia Institute of Technology.
This talk describes the implementation of particle-filtering-based prognostic frameworks for state-of-charge (SOC) and state-of-health (SOH) monitoring in energy storage devices (ESDs), and more specifically in Li-ion batteries. A probabilistic characterization of battery use profiles is used to incorporate the uncertainty associated with the future operation of the system, thus helping to generate a more realistic estimate of the Time-of-Failure probability density function. Analyzed data include voltage, current, and temperature measurements.
d. Corrosion Detection, Prediction and Mitigation Strategies:
Carl Byington, Sikorsky Aircraft Corporation and Dr. George Vachtsevanos, The Georgia Institute of Technology.
Corrosion of critical components, structures and a variety of complex systems/ processes has attracted the attention of the research community over the past years due to the severe consequences of corrosion damage. Corrosion detection, prediction and mitigation technologies are taking central stage in government and industrial enterprises.
In this case study, we introduce an integrated methodology to sensing, detection and prediction and exploit a typical aircraft surface to illustrate the efficacy of the approach. We highlight those innovative aspects of the application domain and point out areas of need for improvement.
e. PHM Implementation Strategies on Legacy Tactical Trucks:
Karl Reichard, The Pennsylvania State University Applied Research Laboratory
Case study will share lessons learned from the implementation of technologies to meet Army CBM+ requirements on tactical wheeled vehicles. The presentation will briefly describe the results of cost benefit analyses for the introduction of health monitoring technology on the vehicles, and discuss the infrastructure improvements required to implement prognostic health management on these vehicles.
f. Maintenance Optimization for Wind Turbines in Wind Farms:
Peter Sandborn and Xin Lie, CALCE, University of Maryland.
This case study describes the use of remaining useful life (RUL) predictions for wind turbine subsystems to assess and optimize the value of maintenance. Results for individual turbines and multiple turbines within a farm will be presented. The case study uses a simulation-based real options analysis to determine the value of waiting after obtaining a RUL indication from prognostics and health management (PHM) structures. This methodology accounts for the anticipated future availability of maintenance resources, uncertainties in future wind resources, and market uncertainties (e.g., anticipated future price and demand for energy).
g. Uncertainty Representation and Management in PHM:
Shankar Sankararaman, NASA Ames.
This case study focuses on understanding the significance, interpretation, and quantification of uncertainty in prognostics and health management, with an emphasis on remaining useful life prediction. In order to facilitate meaningful decision-making, it is important to analyze the various sources of uncertainty that affect prognostics and quantify their combined effect on the remaining useful life prediction through rigorous statistical methods. First, a conceptual example will be discussed to illustrate the importance of uncertainty quantification, and this will be followed by practical example where the goal is to predict the end-of-discharge of a lithium-ion battery. It will be demonstrated that the quantification of uncertainty in the remaining useful life prediction is non-trivial even in simple problems involving linear models and Gaussian variables. A few computational methods for uncertainty quantification in prognostics will be discussed in detail and the challenges related to uncertainty quantification in prognostics will be outlined. Finally, some aspects of uncertainty management will be discussed with numerical illustrations.
|11. PHM Lessons Learned/Where do we go from here
Andrew J. Hess, The Hess PHM Group — President, PHM Society President,
|With audience participation, the course leaders will attempt to capture key needs, gaps and opportunities in PHM. Ways and means of continuing educational and professional development will be presented with reference to the PHM Society’s new Continuing Professional Development Guidelines.|
- For conference non-participants: $700
- For conference participants: $350
- A discount of $150 is offered to the first 20 students registering for the course.
If you are interested in attending the short course, please register here.
|Day 1 – October 3, 2014|
|8:00 – 8:30||Registration|
|8:30 – 9:00||Introduction — State of PHM Technologies|
|9:00 – 9:30||The PHM Paradigm — From monitoring/sensing to fault diagnosis and failure prognosis|
|9:30 – 10:00||CBM+ Technologies|
|10:00 – 10:30||Sensors and Sensing Strategies|
|10:30 – 10:45||Break|
|10:45 – 12:00||Data Mining — Feature selection and extraction|
|12:00 – 1:00||Lunch|
|1:00 – 2:30||Fault Diagnosis|
|2:30 – 2:45||Break|
|2:45 – 5:00||Failure Prognosis|
|Day 2 – October 4, 2014|
|8:30 – 9:30||Performance Metrics|
|9:30 – 10:30||Analytical Foundations and Numerical Solutions for PHM|
|10:30 – 10:45||Break|
|10:45 – 11:30||Performance Metrics|
|11:30 – 12:30||PHM Case Studies|
|12:30 – 1:30||Lunch|
|1:30 – 3:30||PHM Case Studies|
|3:30 – 4:30||PHM Lessons Learned/Where do we go from here; Q&A|