A Machine Learning Framework for Improving Diagnosability of a Reconfigurable Manufacturing System

Abstract

Reconfigurable Manufacturing Systems (RMS) effectively respond to fluctuating market needs and customer demands for finished product. Diagnosability is a supporting characteristic of RMS that has a say in the quality of finished product. Cost and time taken for manufacturing are also considerably affected if proper diagnosability measures are not taken. Previous studies on Diagnosability of RMS have been studied from Axiomatic System Theory as such Design For Diagnosability (DFD). Nevertheless Diagnosability remains to be the least studied characteristic of RMS. With the availability of digitized data, Machine Learning approaches to advance manufacturing have proven to be considerably effective. A research gap existed for the application of Machine Learning techniques in improving the Diagnosability of RMS. A framework of Machine Learning has been proposed to address this gap. The working of the framework has been illustrated by two demonstrations from the available datasets, one in identifying proper signals in semi-conductor manufacturing to predict excursions, and the second in predicting machine failures due to a variety of factors. The framework is rendered in a concurrent-engineering fashion. The framework is tested against two available manufacturing datasets. Increase in Diagnosability will decrease the cost and time taken to production.