Gain-Scheduled PID Control of Nonlinear Plant via Artificial Neural Networks

The high-performance control of nonlinear industrial plants in a wide operation range requires intelligent techniques. The aim of the present research is to develop an engineering approach for adaptation of the gains of the well-mastered and widely applied linear PID controller based on an offline-trained backpropagation artificial neural network (BANN) that assesses the plant parameters for the current operation point. The controller’s gains are online-computed from the empirical relationship with the plant parameters. Robust stability and robust performance conditions are derived for the gain-scheduled BANN-PID system. Their fulfilment ensures system feasibility in an industrial environment. The approach is demonstrated for the control of temperature in a laboratory dryer for fruits. The BANN training is based on data derived and validated from experiments using the Takagi–Sugeno–Kang nonlinear plant model. Simulations show that the BANN-PID system outperforms both the gain-scheduled fuzzy logic PID control system, designed in previous research, and the PID real-time control system by reducing overshoot six times and settling time 1.8 times and improving robustness 1.3 times.