Simultaneous identification of mean shift and correlation change in AR(1) processes

In this paper, we propose a neural-network-based identification system for both mean shift and correlation parameter change. The identifier is trained to detect mean shift, to recognize the presence of autocorrelation, and to identify shift and correlation magnitudes. Various magnitudes of process mean shift, under the presence of various levels of autocorrelation, are considered. Both in-control and out-of-control average run length are computed to measure the performance of the trained identifier. Additionally, we also measure the correction classification rate of shift and/or correlation magnitudes. The identifier is designed to work under two modes, i.e., with or without shift magnitude identification. When properly trained, the identifier is capable of simultaneously indicating whether the process change is due to mean shift, correlation change, or both. This approach is unique since all the statistical control charts developed so far can only detect mean (or variance) shift or parameter change when the deviation is beyond a certain specified control limit, but are incapable of distinguishing whether the shift is due to mean, correlation change, or both when they are concurrently taking place. The result is significant since it provides additional specific information about the process change and the graphical plot reveals the time and progression of the shift/change magnitude. Therefore, the result narrows down the scope of the assignable causes and speeds up the troubleshooting process.