Control Chart Pattern Recognition in Metal-Stamping Process Using Statistical Features-Ann

Identification for the sources of unnatural variation (SOV) in manufacturing process is vital in quality control. In case of metal stamping process, the SOV based on special causes has become a major contributor to poor quality product. In recent years, researchers are still debating to find an effective technique for on-line monitoring-diagnosis the SOV. Control chart pattern recognition (CCPR) method has been reported as applicable for this purpose, whereby the existing CCPR schemes were trained using the artificially statistical process control (SPC) samples. Inversely, the trained scheme using real SPC samples have not been reported since the data are limited or not economically available. In this paper, the SPC samples were taken directly from an actual metal stamping process to be used as the dynamic training patterns. The proposed features-based method has resulted in higher diagnosis accuracy (normal patterns = 100%, unnatural patterns = 100%) compared to the raw data-based method (normal patterns = 66.67%, unnatural patterns = 26.97%).