Neural network technology for tuning process control systems

This paper presents an overview of neural network technology applications for tuning of industrial process control systems, particularly for paper machine control systems. The approaches described are well-suited for most conservative control practices because the neural network computations are essentially performed off-line, outside of the control loop. The neural network tuners provide significant improvement of the product quality by achieving highly optimized settings of the control system without directly relying on field personnel expertise. Two conceptually differing problem types illustrate the development and deployment of neural network tuning applications. The first example is a proportional integral (PI) controller tuner for processes with slow response. Tuning of such processes has been difficult for field personnel in the past. In the second example, the neural network performs identification and tuning of mapping and process response shape parameters in a cross-directional (CD) control system. For both problem types, simulation results are presented to illustrate the efficacy of the tuning solutions.