An application of time delay estimation by ANNs to frequency domain adaptive I-PD controller

We present an auto-tuning method of frequency domain adaptive I-PD controllers using online time delays estimation by ANNs (artificial neural networks). The ANN architecture used is a simple Adaline with the LMS training algorithm. The proposed control scheme applies estimated time delays to the Smith predictor and estimates the frequency response of the controlled object using the sliding DFT. The control gains are then determined so as to minimize the model error between the closed-loop frequency response of desired system and that of the actual system.