Automated Machine Learning for Remaining Useful Life Predictions

TL;DR

The paper tackles prognostics by predicting remaining useful life (RUL) using AutoML. It introduces AutoRUL, an end-to-end AutoML pipeline that cleans data, engineers features, and regresses RUL using standard models, optimized via Bayesian search to minimize $L_{\mathrm{RMSE}}$ on validation data. Evaluations on eight real-world and synthetic datasets show AutoRUL achieves state-of-the-art or competitive performance compared with hand-crafted models and AutoCoevoRUL, demonstrating strong generalization across diverse domains. The work demonstrates that domain experts can obtain data-driven RUL predictions without ML expertise, though high data quality and a fixed search space are important considerations; the authors also provide open-source code to support reproducibility.

Abstract

Being able to predict the remaining useful life (RUL) of an engineering system is an important task in prognostics and health management. Recently, data-driven approaches to RUL predictions are becoming prevalent over model-based approaches since no underlying physical knowledge of the engineering system is required. Yet, this just replaces required expertise of the underlying physics with machine learning (ML) expertise, which is often also not available. Automated machine learning (AutoML) promises to build end-to-end ML pipelines automatically enabling domain experts without ML expertise to create their own models. This paper introduces AutoRUL, an AutoML-driven end-to-end approach for automatic RUL predictions. AutoRUL combines fine-tuned standard regression methods to an ensemble with high predictive power. By evaluating the proposed method on eight real-world and synthetic datasets against state-of-the-art hand-crafted models, we show that AutoML provides a viable alternative to hand-crafted data-driven RUL predictions. Consequently, creating RUL predictions can be made more accessible for domain experts using AutoML by eliminating ML expertise from data-driven model construction.

Figures (2)

• Figure 1: Schematic representation of pipelines constructed by AutoRUL to map sensor data to RUL predictions. The pipeline is structured into three major phases: data cleaning, feature engineering and regression. Each phase contains multiple steps, displayed below the ML pipeline, configured by various hyperparameters. Steps with solid borders indicate mandatory steps included in every pipeline while dashed borders represent optional pipeline steps.
• Figure 2: Visualization of the mean performance, aggregated over all datasets and iterations, with standard deviations plotted over time. Displayed is the immediate regret, i.e., the performance difference to the best solution, of the best so-far found configuration as a function over wall clock time.