Nonlinear control charts for jump detection

This paper presents a new class of nonlinear control charts which respond quickly to small shifts and jump patterns in tme series. The underlying disturbance models for the control charts are nonlinear extensions of the IMA(1,1) model. The Kalman filtering algorithm generates Bayesian estimates of the process level for the control chart plotting. The single-parameter chart is identical to the EWMA, while the two- and three-parameter designs are much more effective in detecting small shifts mixed with local trends. The nonlinear control charting scheme is also capable of detecting a mean shift in independent observation.