Digital parameter-adaptive control of processes with unknown dead time

Maximum Likelihood Estimation of Time Delay for First Order Linear System

“…In time-delay approximation methods, the time delay is not an explicit parameter in the model. In [3], Kurz and Goedecke estimated time delay by measuring the time delay to the start (the beginning of the nonzero part) of an estimated impulse response of the system. Carter [4] found the maximum of the cross-correlation between input and output, which was a common method.…”

Identification of a time-delayed process model using an overparameterisation method

“…Many variations of the overparameterisation methods have been defined to calculate the numerator (and denominator) parameters, and subsequently the time delay, for processes that may be modelled in SISO form. Kurz [1] and Kurz and Goedecke [2], for example, define a robust method for estimating the SISO model parameters that is equivalent to determining the best match between the impulse response of the overparameterised model and the impulse response of a non-overparameterised model with a pure time delay; the method suffers from the disadvantage of having a heavy computational load. Other methods offer various trade-offs between robustness and computational load .…”

“…In interesting papers, Keviczky and Banyasz [32], [33] identify the time delay index using overparameterisation in the delta domain. Other authors identify MIMO process models (with time delays) using the method of overparameterisation; Gurubasavaraj and Brogan [34], for instance, extend the method of Kurz and Goedecke [2] to estimate the time de lay for each input-output pair of a MIMO process model. Simulation results presented by the authors show that the time delays may be estimated in 20 sample periods, for a 2x2 MIMO process with a maximum time delay index of 4; the process order is, however, assumed known a priori.…”

A review of time-delay estimation techniques

“…The time-delay is the delay for the impulse response to start [2,3,19,21]. Finding the peak of the cross correlation between input and output, which is a common method [13], is in principle the same thing.…”

“…For low SNR the estimated impulse and step responses are very noisy [2]. (See also [21].) In an attempt to mitigate this, the methods ICT and SCT uses CUSUM (cumulative sum) thresholding, which is a nonlinear averaging operation [12].…”