Recognition of unnatural variation patterns in metal-stamping process using artificial neural network and statistical features

Abstract

Unnatural process variation (UPV) is vital in quality problem of a metalstamping process. It is a major contributor to a poor quality product. The sources of UPV usually found from special causes. Recently, there is still debated among researchers in finding an effective technique for on-line monitoring-diagnosis the sources of UPV. Control charts pattern recognition (CCPR) is the most investigated technique. The existing CCPR schemes were mainly developed using raw data-based artificial neural network (ANN) recognizer, whereby the process samples were mainly generated artificially using mathematical equations. This is because the real process samples were commonly confidential or not economically available. In this research, the statistical features - ANN recognizer was utilized as the control chart pattern recognizer, whereby process sample was taken directly from an actual manufacturing process. Based on dynamic data training, the proposed recognizer has resulted in better monitoring-diagnosis performance (Normal = 100%, Unnatural = 100%) compared to the raw data- ANN (Normal = 66.67%, Unnatural = 26.97%).