Neural network assisted autotuning of industrial PID controllers

A recent survey [1] has reported that the majority of industrial loops are controlled by PID-type controllers and many of the PID controllers in operation are poorly tuned. Poor PID tuning is due to the lack of a simple and practical tuning method for average users, and due to the tedious procedures involved in the tuning and retuning of PID controllers. Poor tuning as reported by the survey, suggests that commercially available self-tuning PID controllers and PID controllers with auto-tuning facilities have not been widely used. Realizing the economic and environmental advantages that will be gained from properly tuned PID controllers, the main focus of the work reported in this paper is the development of a scheme for automatic tuning of existing industrial PID controllers on-line. Recent success in exploiting the nonlinear function approximation properties of artificial neural networks [8], [9], [10], [4] for auto-tuning PID controllers proved that exploiting artificial neural networks for auto-tuning PID controllers is a viable approach. This approach is adopted in this paper to develop a scheme for automatic tuning of existing industrial PID controllers with the assistance of a trained neural network. The structure of the proposed neural network assisted PID auto-tuner (NNAT) is shown in Figure 1. There are two main components of the NNAT : transient analyser and neural network PID auto-tuner. The transient analyser samples, filters and analyses the closed-loop step response of the process when the controller is operating in proportional plus integral (PI) mode [6], and estimates either a first-order or a second-order plus dead-time model of the process. The estimated process model is then fed to the neural network PID auto-tuner which is trained to supply the optimal PID settings, according to some predefined criteria. In this scheme, the neural network PID auto-tuner acts as a nonlinear function approximator which approximates the mappings between the estimated process model parameters and the optimal PID settings. In this work, a radial basis function network [5], [7] was used in the neural network PID auto-tuner because of its fast training time.