A PILOT STUDY OF PREDICTIVE MAINTENANCE IN SEMICONDUCTOR MANUFACTURING

Abstract

Predictive maintenance (PdM) refers to the practice of regular monitoring of the operating condition of machines, estimating the future probability of failure, and/or projecting rate of machine degradation based on current and past equipment/process conditions, and scheduling maintenance actions just before they are required. With the successful implementation examples of PdM in other industries like chemical processing and aerospace to reduce the number and costs of unscheduled machine breakdown, this project aims to perform a pilot implement study of PdM in semiconductor industry. Based on the existing manufacturing system in the hosting company and the standard framework of PdM, this thesis proposed the approaches to implement PdM there, and developed models to extract features related to machine breakdown in the monitoring system, to estimate the probability of failure, and to project the time when machine failure happens. By model validation on the historical data under consideration, quite satisfying results were achieved.