Applying ICA monitoring and profile monitoring to statistical process control of manufacturing variability at multiple locations within the same unit

Abstract

The assessment of in-process observations can provide useful information on potential sources of process variability. In this research, each source of variation was assumed to generate specific patterns in the spatial and temporal variations observed in data related to the measurement of quality characteristics. The spatial variation pattern and temporal pattern caused by a variation source may turn into the observed within-part and between-part variations in the monitoring of product measurements from multi-location. The traditional chart is a useful tool to deal with this type of process monitoring. To further improve the monitoring performance, this paper proposes two new process control methods based on independent component analysis (ICA) monitoring and profile monitoring. The ICA monitoring uses various monitoring statistics obtained from ICA to construct a control procedure. In the profile monitoring approach, a set of distance-based and statistical features is used as the input of a support vector regression (SVR)-based decision function to create a process monitoring method. Average run length (ARL) is used as a performance criterion to evaluate the capability of various control methods in detecting abnormalities. Simulation results show that profile monitoring approach has the best overall performance in terms of ARL, followed by ICA monitoring, and chart. This paper makes an important contribution to the monitoring of within-part and between-part variations.