Model-based PID autotuning enhanced by neural structural identification

This paper presents an autotuning method for industrial PID controllers in the 1-d.o.f. ISA form. The major feature of the method is that the model structure employed for the process is selected on-line based on a step response record, by means of a multilayer perceptron neural network. Thanks to the exclusive use of normalized I/O data, the network can be trained off-line with simulated data, therefore simplifying the method’s implementation. Once the model structure is selected and its parameters are identified, the IMC approach is used for synthesizing a regulator that is then approximated with a PID. Simulation and experimental results are reported to show the effectiveness of the proposed tuning method and its advantages with respect to IMC-based PID tuning with the model structure fixed a priori.