Domain Knowledge Injection Guidance for Predictive Maintenance

Abstract

AbstractWith the integration of Industry 4.0 technologies, overall maintenance costs of industrial machines can be reduced by applying predictive maintenance. Unique challenges that often occur in real-time manufacturing environments require the use of domain knowledge from different experts. However, there is hardly any guidance that suggests data scientists how to inject knowledge from predictive maintenance use cases in machine learning models. This paper addresses this lack and presents a guidance for the injection of domain knowledge in machine learning models for predictive maintenance by analyzing 50 use cases from the literature. The guidance is based on the informed machine learning framework by von Rueden et al. [1]. Finally, the guidance gives a recommendation to data scientists on how domain knowledge can be injected into different phases of model development and suggests promising machine learning models for specific use cases. The guidance is applied exemplarily to two predictive maintenance use cases.