Process Mining for Reliability Modeling of Manufacturing Systems with Limited Data Availability

Accurate reliability modeling of manufacturing systems results in reduced maintenance costs and increased profits. The growing research field of data-driven reliability modeling seeks to harness the abundance of data from Smart Manufacturing Systems to improve and automate reliability modeling processes. Many manufacturing systems, however, still collect very limited amounts and types of data (typically operational state logs of production resources), which, we believe, can still be made useful to extract meaningful and useful reliability models. This motivates the work-in-progress that we present in this paper, a proof-of-concept method that extracts reliability models from state logs of production resources with deterministic durations of activities. First, the process model of a manufacturing system is extracted using a simple, yet efficient algorithm. Second, the reliability distributions for each production asset are fitted and useful reliability insights are integrated into the previously extracted process model. We conduct experiments of our proposed method using state logs from a drone manufacturing environment that is part of our university Industry 4.0 laboratory. The results show that a high-level overview of a system and its reliability can be provided with minimal data requirements and limited expert knowledge of the system. To this end, we also propose possible extensions and further developments of our method to make it more applicable to more challenging scenarios.