Log-based predictive maintenance in discrete parts manufacturing

The performance of discrete parts manufacturing systems is heavily influenced by unplanned machine breakdowns. Predictive maintenance allows for the conversation of unplanned machine breakdowns to scheduled corrective maintenance actions. We present a data-driven log-based approach for estimating the probability of machine breakdowns during specified time intervals in the future. Machine learning algorithms are utilized for a specific use-case which is based on real-world data-sets including machine log messages, event logs and operational information. We will present an overview on predictive maintenance strategies as well as applied data-preparation, feature-engineering and machine learning methods for estimating the remaining useful lifetime. The comparison of two different methods for RUL-estimation is indicating that machine failures can be predicted up to 168 hours in advance with promising precision and hit-rate.