Contents I. Introduction
As semiconductor and related industries continue to move from a reactive to predictive approach to operations [1], [2], one topic that has seen a noted rise in interest is Predictive Maintenance (PdM). PdM utilizes process and equipment state information to predict when a tool or a particular component in a tool might need maintenance. A good summary of the move from time-based or part-count based maintenance to conditioned based maintenance and ultimately PdM can be found in [3]. Utilizing PdM prediction information properly can lead to lower unscheduled downtimes, reduced mean-time-to-repair (MTTR), reduced scrap, and increased life of components and consumables [1]–[4]. It is logical to conclude that cost-effective PdM implementation requires: (1) existing automation infrastructure be leveraged to realize PdM systems, and (2) the maintenance predictions be of sufficient quality so that the cost of false positive occurrences is significantly outweighed by the benefits of early detection, unscheduled downtime avoidance, and fault classification (reducing mean-time-to-repair). In light of these objectives it is important that (1) existing infrastructure such as APC be leveraged and (2) data quality strengthening techniques be utilized to improve the viability of the PdM system.
