A Novel Measurement Scheme for Periodic Fouling in Recipe Alternation Based on Hybrid Fuzzy Neural Netwok

For on-line measuring the periodic fouling resistance in recipe alternation, a novel soft-sensing approach based on FNN (fuzzy neural network), FCM (fuzzy c-means clustering) and OED (orthogonal experimental design) was investigated. The periodic fouling resistance is modeled by two parts: the increase of short-term fouling in one batch and the long-term irreversible fouling at the beginning of the batch. FCM algorithm was used to cluster all the recipes into classes, and give each recipe a fuzzy membership belongs to one class. Based on the membership matrix, a FCM neuron was designed and accordingly a hybrid FNN that is robust to recipe alternation was proposed. OED were used to select out the key parameters influencing the long-term fouling and short-term fouling respectively. Two hybrid FNN networks are trained to learn the formation trends of the short-term fouling and long-term fouling trend. The new measurement scheme combines the advantages of FCM, OED and FNN, have a good performance to predict the fouling in recipe alternation. The experimental results and the comparison to other methods show the new scheme is effective