A Bayesian network based approach to improve the effectiveness of maintenance actions in Semiconductor Industry

The Semiconductor Industry (SI) is facing the challenge of high-mix low-volume production due to increasing diversity in customer demands. This has increased unscheduled equipment breakdowns followed by delays in diagnosis and ineffective maintenance actions that reduce the production capacities. At present, these challenges are addressed with mathematical approaches to optimize maintenance actions and their times of intervention. However, few studies take into account the ineffectiveness of maintenance actions, which is the key source for subsequent breakdowns. Hence, in this paper, we present a methodology to detect poorly executed maintenance actions and predict their consequences on the product quality and/or equipment as the feedback for technicians. It is based on the definition of maintenance objectives and criteria by experts to capture information on the extent to which the objective is fulfilled. Data collected from maintenance actions is then used to formulate Bayesian Network (BN) to model the causality between defined criteria and effectiveness of maintenance actions. This is further used in the respective FMECA defined for each equipment, to unify the maintenance knowledge. The key advantages from the proposed approach are (i) dynamic FMECA with unified and updated maintenance knowledge and (ii) real time feedback for technicians on poor maintenance actions. The SI is characterized by fastest change in smallest period of time and has become a 300+ B $ industry in less than 6