Online Identification and Alignment of MIMO Cross Directional Controlled Processes Using Second Order Statistics

Abstract

Among the factors influencing the performance of multi-input multi-output (MIMO) systems such as paper machines, accurate spatial alignment of the actuators with their corresponding measurement points is a key issue in cross directional (CD) control. This mapping is a stochastic, non-linear and time-varying problem. Most current methods of alignment require a manual open-loop bump test. Several actuators are excited to perform the bump test, then the observed peaks are assigned to the excited actuators. This paper uses the second-order statistical technique of the blind source separation methods to make the transition from a manual, open-loop bump test to closed-loop adaptive online mapping. This procedure estimates the mixing matrix by spatial decorrelation of the noise and source signal. The mixing matrix is a function of input autocorrelation input and output cross-correlation. The main advantage of this method comes from the fact that we have direct access to the source signals and outputs. The method's robustness in cases of spatially colored noise makes it applicable to the CD control of paper machines.