An Automated Fault Diagnosis For Manufacturing Process Monitoring And Control

Abstract

A fault-tolerant monitoring system for manufacturing process control is defined with sensor fusion for process-feature preparation and neural networks for process-feature analysis. Considering the complexity of operation mechanisms and the variability of process parameters in typical manufacturing environments, the author develops an automated fault-diagnosis method for the fault-tolerant monitoring system itself, called adaptive feature scaling, which is a modified version of the input feature scaling algorithm. In conjunction with clustering-type neural networks, adaptive feature scaling performs not only feature evaluation for self-diagnosis of process features, but also feature adjustment for improvement of the monitoring performance. In experimental evaluation, adaptive feature scaling proves superior to conventional approaches from the implementation point of view.