Fault prediction with combination of strong tracking square-root cubature Kalman filter and autoregressive model

To deal with the problem of prognosis of unmeasured parameters in nonlinear systems, we propose a fault prediction algorithm which is a combination of the strong tracking square-root cubature Kalman filter (STSCKF) with suboptimal fading factor and the autoregressive (AR) model. Future time values of measurement variables are forecasted by using the AR model time series prediction method; and then, the predicted values are used as STSCKF measurement variables. Thus, the prognostic problem is transformed into a filter estimation issue. The fading factor is introduced into the square root of the STSCKF prediction error covariance for adjusting the gain matrix in the filter process. As a result, STSCKF eliminates the disadvantage of slow tracking or even unable tracking of fault parameters in conventional SCKF when the time-varying functions of fault parameters are unknown, improving the capability for forecasting the varying trend of fault parameters. Simulation results on a continuous stirred tank reactor (CSTR) show that the predicting accuracy of STSCKF is higher than that of the conventional SCKF or the strong tracking unscented Kalman filter (STUKF), demonstrating the superiority of the performance capability of the proposed method.