Description:

ProgPy is an open-sourced python package, developed by NASA and several partners in the PHM Community, supporting research and development of prognostics and health management and predictive maintenance tools. It implements architectures and common functionality of prognostics, supporting researchers and practitioners. This tutorial is a hands-on overview of these tools, including how to use them and extend them.

This tutorial is a hands-on overview of these tools, including how to use and extend them. We will be building new ProgPy models and configuring existing models. We will also demonstrate how to use the ProgPy models for prognostics and understand the results. The tutorial will also include examples of how the tools are used at NASA.

Tutorial Resources and Links:

• https://nasa.github.io/progpy/
• https://github.com/nasa/progpy/blob/master/examples/2024PHMTutorial.ipynb

About the Presenters:

Katelyn Jarvis is an applied mathematician and research engineer at NASA Ames Research Center. Katy received her B.S. in Mathematics and M.S. and Ph.D. in Applied Mathematics from the University of California, Davis. At NASA, Katy is a researcher within the System-Wide-Safety and Data and Reasoning Fabric projects, as well as a member of the Diagnostics and Prognostics group. Her research interests include development of novel algorithmic approaches to prognostics, improvement of computational efficiency of these methods, and application of prognostics in human-health related fields.

Chris Teuber is the Diagnostics and Prognostics group lead at NASA Ames Research Center. Christopher received his B.S. in Aerospace Engineering from Iowa State University and his M.S. in Computer Science and Engineering from Santa Clara University. At NASA, Christopher performs research in software architectures for prognostics, is the lead of NASA’s Prognostics CoE, and is the lead developer for the Prognostics Python Packages (prog models, prog algs, and prog server).

Description:

This tutorial shows how to generate probabilistic distribution from a supervised deep learning model, with which users will have a measure of confidence for each inference. Combining independently trained deep neural networks is a convenient mechanism for generating uncertainty estimates but computationally expensive. Monte Carlo dropout offers a less costly alternative but is less reliable. Masksembles, the focus of this tutorial, strikes a balance between these two methods. The tutorial covers building, training, and deploying a model for classification and regression applications, with analogies to health state diagnosis and RUL estimation. We apply this technique to the 2024 PHM Data Challenge. Tutorial materials will be made available on GitHub.

Tutorial Resources and Links:

• https://github.com/globaltechinc/probablistic_deep_learning

About the Presenter:

Dr. Jesse Williams is GTC Analytics’s VP of Digital Engineering and CTO. He has a Ph.D. in materials engineering (UC Santa Barbara) and is a trained data scientist. Dr. Williams holds broad experience in mechanical, electronic, and material systems from his past experiences as CTO at LIM Innovation, lead scientist at the Nation Institute for Materials Science, International Center for Materials Nanoarchitectronics, and as a researcher at the Los Alamos National Laboratory. Dr. Williams also has experience in rapid prototyping of full-stack hardware and software solutions.

Tutorial Description:

The tutorial introduces large language models (LLMs) and multimodal models (e.g., language and vision) and their applications in Prognostics and Health Management (PHM). Foundational concepts like how transformers power LLMs will be discussed. Retrieval-augmented generation (RAG) methods will be explored, illustrating how domain-specific knowledge can be integrated into LLM outputs. Graph-enhanced RAG (GraphRAG) will also be discussed as an advanced approach that leverages knowledge graphs to improve reasoning and contextual awareness in PHM applications. These techniques can enhance predictive maintenance, failure detection, and diagnostics by providing more accurate insights derived from large-scale unstructured data.

The tutorial will focus on multimodal applications in PHM, combining language and vision to enhance the analysis of maintenance reports, technical documentation, and visual inspections. It will show how transformers process both textual and visual inputs simultaneously, with real-world examples demonstrating their use in identifying anomalies, interpreting diagrams, and improving maintenance decisions. Tutorial attendees will have a comprehensive view of how advanced generative AI techniques can be adapted and optimized for PHM applications.

About the Presenter:

Dr. Neil Eklund, FPHMS, is a leading expert in Asset Health Management and a seasoned technologist with over 25 years of experience in data science, industrial artificial intelligence, and machine learning. He has led high-impact projects across industries including aerospace, energy, healthcare, and oil & gas, resulting in 16 patents and over 70 technical publications. Dr. Eklund is one of the co-founders the Prognostics and Health Management Society and serves on its board of directors. His career includes significant roles at General Electric Research, Xerox PARC, and Schlumberger, where he was the Chief Data Scientist. At Schlumberger, he led the development of the first successful deployed Internet of Things (IoT) application in the oil industry, generating over $20 million in its first three months of operation.

Dr. Eklund’s extensive collaboration with organizations like DARPA, NASA, the DoD, Lockheed Martin, ExxonMobil, Ford Motor Company, and Boeing highlights his impact on both commercial and government sectors. His contributions extend beyond industry, having taught graduate-level courses in machine learning and classes through the PHM Society. He is also a former contributor to the International Standards Organization (ISO) in the area of diagnostics and prognostics for complex machinery.

Tutorial Description:

Diagnostics of an induction motors is a serious issue for improving plant reliability. Induction motors are the workhorse of the world today. In recent years application of axial flux (pancake) induction motors is happening in various industries, especially high-end electric vehicles. Motor current signature analysis (MCSA) and vibration data are commonly used for diagnosing induction motor problems. However, the fault signatures are complicated and lacks consensus. The problem is further complicated by the loading effects the defect signatures. This paper will present a comparison of vibration and motor current signals for induction motor subjected to different levels of torque, unbalance, and misalignment loadings. Experiments were performed on intentionally faulted motor with varying degrees of electro-mechanical defects. Different levels of mechanical torque, unbalance and misalignment loadings were applied to the rotor side. The data was analyzed using both vibration and motor current sensors. Results indicate motor current signature provide better indication of certain electrical faults such as air-gap eccentricity and broken rotos bars and vibration signature is better indicator of mechanical defects. The results suggest that both motor current and vibration measurements are required for more complete diagnostics of induction motors

About the Presenter:

Suri is the founder and president of Spectra Quest, Inc. Suri has expertise in machine condition monitoring and diagnostics, signal processing, modal analysis, finite element modelling, and (PhD thesis) constitutive modelling of amorphous (polymeric) materials. SpectraQuest’s flagship products Machinery Fault Simulators (MFS) and Drivetrain Diagnostics/Prognostics Simulators are his brainchild from concept to completion. These devices have been sold in over fifty countries around the world for research and teaching in industrial predictive maintenance.

He has over forty years of industrial and academic experience. Suri has authored over seventy-five papers, technical reports and articles in journals, magazines, and books. He has worked at Philip Morris, Fire Stone, and NASA’s Space Shuttle program at Martine Marietta. Suri is Chair of the MFPT society. He is fellow of the International Society of Condition Monitoring and Machinery Failure Prevention Technology. Suri obtained a Ph.D. in Mechanical Engineering from the University of Texas at Austin.