Neural network based diagnosis system for looper height controller of hot strip mills

In this study, auto tuning of PID control gains in hot strip looper controller is made based on a recurrent neural network (RNN) model. Further, a neural network emulator (Neuro emulator) is employed to model the characteristics of looper dynamics. Combining Neuro emulator and the RNN model, an auto tuning system of PID control gains is constructed. As the inputs to RNN, plural evaluation functions which reflect individual preference of human experts. Self learning mechanism is embedded in the RNN model which enables adaptation both of rolling characteristics. However, over time, deterioration of mechanical characteristics and the control system will be induced. To overcome the problem, it is necessary to diagnose the true cause of failure and to compensate for it. For this purpose, the hierarchical neural network (HNN) is built into the auto tuning of PID control gains. The HNN model, which enables compensation to the deterioration of mill system, can estimate current system parameters such as control gains and mill constants. In developing the advanced control system combining HNN, RNN and Neuro Emulator, we aim for the realization of the simultaneous operations of tuning, diagnosis and estimation of the plant like skillful human experts. Through numerical experiments, the effect of the proposed method is ascertained.