DeepFMEA -- A Scalable Framework Harmonizing Process Expertise and Data-Driven PHM
Authors:
Christoph Netsch,
Till Schöpe,
Benedikt Schindele,
Joyam Jayakumar
Abstract:
Machine Learning (ML) based prognostics and health monitoring (PHM) tools provide new opportunities for manufacturers to operate and maintain their equipment in a risk-optimized manner and utilize it more sustainably along its lifecycle. Yet, in most industrial settings, data is often limited in quantity, and its quality can be inconsistent - both critical for developing and operating reliable ML…
▽ More
Machine Learning (ML) based prognostics and health monitoring (PHM) tools provide new opportunities for manufacturers to operate and maintain their equipment in a risk-optimized manner and utilize it more sustainably along its lifecycle. Yet, in most industrial settings, data is often limited in quantity, and its quality can be inconsistent - both critical for developing and operating reliable ML models. To bridge this gap in practice, successfully industrialized PHM tools rely on the introduction of domain expertise as a prior, to enable sufficiently accurate predictions, while enhancing their interpretability.
Thus, a key challenge while developing data-driven PHM tools involves translating the experience and process knowledge of maintenance personnel, development, and service engineers into a data structure. This structure must not only capture the diversity and variability of the expertise but also render this knowledge accessible for various data-driven algorithms. This results in data models that are heavily tailored towards a specific application and the failure modes the development team aims to detect or predict. The lack of a standardized approach limits developments' extensibility to new failure modes, their transferability to new applications, and it inhibits the utilization of standard data management and MLOps tools, increasing the burden on the development team.
DeepFMEA draws inspiration from the Failure Mode and Effects Analysis (FMEA) in its structured approach to the analysis of any technical system and the resulting standardized data model, while considering aspects that are crucial to capturing process and maintenance expertise in a way that is both intuitive to domain experts and the resulting information can be introduced as priors to ML algorithms.
△ Less
Submitted 13 May, 2024;
originally announced May 2024.
