Adaptive exponentially weighted moving average schemes using a Kalman filter

Two adaptive exponentially weighted moving average control schemes are proposed. The weighting coefficient is updated using a Kalman filter algorithm. The 2 test statistics incorporate an integral error term. Simulated average run lengths indicate that the proposed schemes are sensitive to small process shifts but that they do tend to ring false alarms when there is no process change. For medium and large process changes and trends, their performance is comparable to that of Lucas' combined Shewhart-CUSUM control scheme. Some application of the proposed schemes to correlated data indicate robust performance. It is concluded that the Kalman filter used to model a process, together with a detection mechanism applied to the residuals, closely resembles the work done in control theory.