Intelligent process control: application of neural network to glass melting furnace control

The authors propose a control scheme for a color TV glass-bulb melting process using neural network (NN). For modeling and synthesizing the input-output mapping of the process, the scheme has a hierarchically unified neural network (HUNN) architecture which integrates recurrent NN and feedforward NN. Both multilayer NNs are trained to predict the states and performance of the melting operations. Then, the sensitivity analysis is applied to the trained NN for identifying the key control input variables which strongly affect the process performance. The optimal control for these key variables are calculated by partial inversion algorithm so that the performance of the controlled process can maintain at the maximum target level. Experimental results show the effectiveness of the proposed control scheme