An Adaptive UKF Algorithm for Process Fault Prognostics

For standard unscented Kalman filters (UKF), the unknown covariance matrices of prior state estimate error, output prediction error and posterior state estimate error propagate recursively through fixed models, which does not consider the actual distribution information of errors. With respect to above problem, an adaptive UKF algorithm is proposed to improve the estimation of error covariance matrices. By introducing measurement innovation into the estimation of error covariance matrices, the proposed algorithm can compute the Kalman gain adaptively and make the future measurement innovation series uncorrelated. The adaptive UKF algorithm is then utilized for nonlinear process fault prognostics. Simulation results on a continuous stirred tank reactor demonstrate the effectiveness of the proposed algorithm.