Machine Learning Based Predictive Maintenance Strategy: A Super Learning Approach with Deep Neural Networks

Abstract

Effective predictive maintenance (PdM) strategy is needed in microelectronic manufacturing to reduce cost and loss of cycle time associated with unplanned maintenance events in fab. Adoption of recent advances in big data analytics methods will help in developing an effective PdM strategy. This paper discusses advanced machine learning based PdM strategy. Various approaches to formulate the PdM problem and solve it are discussed. The performance of respective machine learning algorithms on equipment data is evaluated. To benefit from the strengths of various algorithms a super learning based approach with deep neural networks is proposed. This approach is robust compared to single method selection and handles changing production environment well. This cross validation based stacked model ensemble predictive maintenance strategy is also easy to automate.