Exploration of Production Data for Predictive Maintenance of Industrial Equipment: A Case Study

Abstract

Data-driven predictive maintenance is typically based on collected data from multiple sensors or industrial systems over a period of time, where historical and real-time data are combined as input to black-box machine learning models. In the current study we provide a case study of a major manufacturing company of large industrial equipment. We investigate the opportunity to utilize the manufacturing state of the equipment alone to predict future conditions. The production data contain information about the errors or defects in the equipment found in production. The defects are potentially repaired before the equipment is installed. We present a proactive approach based on interpretable machine learning models, where the production data are used to predict maintenance, which creates the opportunity to prevent future maintenance. The solution is easily translated into a simple set of rules that can be used to separate critical production errors from non-critical production errors. Identifying critical production errors potentially prevents future and more expensive repairs of errors detected in inspections after the equipment has been installed. Our paper contributes to the literature on predictive maintenance in two ways. Firstly, we show the viability of a more proactive approach utilizing production data to prevent future maintenance. Secondly, we demonstrate the applicability of interpretable machine learning models to understand the relationship between the features of the production errors and the later inspection errors.