Hybrid data-driven maintenance policies with sequential pattern mining support

Abstract

The management of Operations and Maintenance (O&M) in industrial systems has evolved significantly with technological advancements, enabling real-time data collection through embedded sensors. These innovations provide opportunities for predicting failures and optimizing maintenance policies. However, challenges remain, particularly in interpreting discrete event data and addressing issues such as false negatives, defect induction, and maintenance impediments. This research introduces a novel framework that integrates Sequential Pattern Mining (SPM) with continuous improvement methodologies like Knowledge Discovery in Databases (KDD) and the Plan-Do-Check-Act (PDCA) cycle. The framework supports the development of hybrid maintenance policies for complex industrial systems by addressing both operational and managerial challenges. Key contributions include two innovative models tailored to distinct subsystems in a machining center: for the lubrication system, an opportunistic hybrid policy was designed to mitigate frequent interruptions and tool wear caused by lubrication failures, demonstrating cost reductions and operational

improvements; for the spindle subsystem, a hybrid maintenance policy incorporating a three- stage degradation model, external maintenance impediments, and defect induction scenarios

was developed, offering a comprehensive solution for maintenance optimization. This study advances the state of the art by integrating previously isolated maintenance concepts into cohesive hybrid policies, supported by numerical analyses that reveal significant cost optimization compared to traditional methods. Practical contributions include the identification of critical cost thresholds, guidelines for inspection frequency, and strategies to minimize defect induction. Additionally, the research highlights the economic and environmental benefits of proactive maintenance, aligning with sustainability goals and corporate social responsibility objectives. By bridging theoretical innovations with practical applications, this thesis provides robust tools for improving efficiency, reliability, and decision-making in industrial maintenance.