A Novel Dead-Time Adaptive Controller

Abstract

Most adaptive control techniques which have seen industrial applications require that the model structure be fixed in terms of the process delay or dead time. This is a severe limitation in processes where significant throughput changes cause marked variations in transport delays. Typical approach has been to specify the model dead time securely longer than the actual process dead time to avoid closed-loop instability should the process delay lengthen. We present an adaptive control algorithm which addresses this problem in a more direct and less conservative manner. The algorithm is patterned after the classical self-tuning regulator structure; however, there are several added features. The prediction step of the controller is computed on-line using a digital filter dedicated specifically to this purpose. When a change is required, the adaptive controller structure is shifted to reflect this. Concurrently, an on-line estimation scheme produces model parameters which are converted into controller coefficients using the indirect method with the polynomial identity. The extent of the polynomial identity must be changed as the estimated dead time shifts. Results of experimental tests on a distillation feed preheater are presented which demonstrate stable performance under circumstances where fixed-delay schemes fail.