Improved ICA-based mixture control chart patterns recognition using shape related features

Abstract

Quality control and improvement tools and techniques can add value to the supply chain. Quality management practices improve not only product quality, but also supply chain performance, through their impact on variance reduction. Statistical process control (SPC) uses control charts to achieve process stability and improve quality by reducing variability. Various techniques have been applied to identify the presence of unnatural control chart patterns (CCPs); however, most studies have focused on recognizing basic CCPs from a single type of unnatural assignable cause. Where more than one type of unnatural variation exists simultaneously within the manufacturing process, a mixture of CCPs result and these might be incorrectly classified. The Independent Component Analysis (ICA) technique is one of the techniques that have been used to estimate the independent components of a mixture of two basic CCPs. However, the separation performance of an ICA-based approach is relatively poor for basic CCP pairs that are highly correlated. This paper will investigate using shaped-related features to improve the overall performance for mixture CCP recognition.