Physics-Based Simulation-Assisted Machine Learning for Estimating Engineering System Failure Durations

EasyChair Preprint 14789

Abstract

In this paper, we explore the integration of physics-based simulations with machine learning (ML) to enhance the prediction of mean time to failure (MTTF) in engineering systems. Traditional failure prediction methods, while valuable, often fall short when dealing with complex systems due to limited historical data and an inability to model non-linear interactions. To address these challenges, we propose a hybrid approach that combines detailed simulations of physical systems with ML techniques to accurately predict system failure. By leveraging simulation-generated data alongside real-world sensor data, our method improves both the accuracy and generalization of failure predictions, particularly in scenarios with sparse or incomplete datasets. A case study involving wind turbine gearboxes illustrates the application of this method, demonstrating superior performance over purely data-driven models. The integration of physics-based features enables the model to generalize across a wider range of operating conditions, including rare or extreme events. We discuss the benefits, challenges, and future directions of this combined approach, highlighting its potential to improve the reliability and safety of critical engineering systems.

Keyphrases: Finite Element Analysis (FEA), Mean time to failure (MTTF), Predictive Maintenance, Simulation-assisted ML, Wind turbine gearbox, engineering systems, failure prediction, machine learning, physics-based simulations, system reliability

@booklet{EasyChair:14789,
  author    = {Kayode Sheriffdeen},
  title     = {Physics-Based Simulation-Assisted Machine Learning for Estimating Engineering System Failure Durations},
  howpublished = {EasyChair Preprint 14789},
  year      = {EasyChair, 2024}}