Maintenance Strategies for Manufacturing Systems using Markov Models

Abstract

Deteriorated equipment condition has a significant impact on the product quality and maintenance policies. It is well known that online diagnosis systems and intelligent maintenance strategy play an important role to support maintenance and production in the modern manufacturing industry. Among numerous issues related to manufacturing maintenance, we address three major challenging problems: 1) an adaptive anomaly detection algorithm for condition-based maintenance, 2) more accurate stochastic models for preventive maintenance, and 3) joint maintenance and production scheduling for a multiple product and multiple station system. Adaptive anomaly detection allows us to conduct not only the online degradation assessment but also anomaly diagnosis in the presence of unknown faults. This algorithm is realized by using the hidden Markov model with reinforcement learning techniques. Online machine health information can further be investigated for the relationship on the product quality and equipment deterioration. Based on impact of machine condition to the product quality, we develop an integrated maintenance and dynamic product sequencing policy that can be applied to a multiple product and multiple station system. For preventive maintenance, the traditional degradation models only focus on a single machine system and ignore maintenance durations. We perform analytical and numerical examination of production lines with the Markov process framework, focusing on the more accurate dynamic behavior modeling and multiple maintenance tasks. Non-exponential holding time distributions in Markov chain are approximated by inserting multiple intermediate states based on a phase-type distribution. By having an adequate model representing both deterioration and maintenance processes, we can find different optimal maintenance policies to maximize the availability or productivity for different configurations of components.