On-line estimation of first order plus dead time process parameters based on under or over-parameterized models

Abstract

Simple on-line process identification methods that can be used in tuning PID controllers or any other controllers have been studied in this work. The underlying continuous-time first order plus dead time FOPDT process model parameters ('k', [tau], [tau]'d') have been estimated effectively at every sampling period from the frequency response of an under or over-parameterized model assigned for the process and enhanced with recursive least squares. It is shown that for any under or overparameterized model structure and with any SNR as low as 0dB, the parameters of the FOPDT model are estimated reliably. This has been accomplished by using either the Nelder-Mead optimization approach or the line fitting approach. Line fitting approach provides better estimation results and faster parameter convergence than the Nelder-Mead optimization approach especially when small sampling periods are used. Line fitting approach is applied experimentally to a distillation column and evaluated successfully in this thesis.