Predictive maintenance enabled by machine learning: Use cases and challenges in the automotive industry

TLDR

Machine‑learning–driven predictive maintenance is increasingly important for ensuring automotive functional safety while controlling costs, yet no comprehensive survey of ML‑based PdM has been published. This paper surveys and categorizes existing ML‑based PdM studies, identifies open challenges, and proposes future research directions. The authors systematically reviewed the literature, classified papers by application and ML technique, and analyzed them from both application and methodological perspectives. They conclude that publicly available data would accelerate research, most studies use supervised methods needing labeled data, multimodal data fusion improves accuracy, and deep learning will grow but demands efficient, interpretable models and large labeled datasets.

Abstract

Recent developments in maintenance modelling fuelled by data-based approaches such as machine learning (ML), have enabled a broad range of applications. In the automotive industry, ensuring the functional safety over the product life cycle while limiting maintenance costs has become a major challenge. One crucial approach to achieve this, is predictive maintenance (PdM). Since modern vehicles come with an enormous amount of operating data, ML is an ideal candidate for PdM. While PdM and ML for automotive systems have both been covered in numerous review papers, there is no current survey on ML-based PdM for automotive systems. The number of publications in this field is increasing — underlining the need for such a survey. Consequently, we survey and categorize papers and analyse them from an application and ML perspective. Following that, we identify open challenges and discuss possible research directions. We conclude that (a) publicly available data would lead to a boost in research activities, (b) the majority of papers rely on supervised methods requiring labelled data, (c) combining multiple data sources can improve accuracies, (d) the use of deep learning methods will further increase but requires efficient and interpretable methods and the availability of large amounts of (labelled) data.