PROCESS BEHAVIOR DIAGNOSTICS AND ADAPTIVE PROCESS-CONTROL

The patterns exhibited in the recent history of the process input variable can indicate impending useful excitation data. A general method for detecting excitation based on an analysis of the process input is detailed, along with the use of this data to update an adaptive controller internal model. The method, applicable to a number of adaptive strategies, is demonstrated using an adaptive PI controller with Smith predictor. Because unmeasured disturbances can occur during excitation events and corrupt the resultant model, model gains are collected as they are computed and used in an automated procedure to generate a continuous gain schedule function. This function provides information for measuring the success of future model fits, even as operating regime changes for processes with nonlinear gain characteristics, and provides an adaptation mechanism for controlling such processes. To minimize the expertise required to implement these methods, an automated startup procedure is detailed. The methods are demonstrated on two simulated reactor processes that possess nonlinear and s. © 1990.