HOW USEFUL ARE RECURRENT NEURAL NETWORKS FOR REAL-TIME CALCULATION OFTHE AVERAGE CHAIN-LENGTH OF POLYMETHYLMETHACRYLATE

This paper presents a method for an accurate estimation of the average chain length of polymers based on neural networks. A simulation of a continuous solution polymerisation reactor, with varying setpoints, is used to train a real-time recurrent neural network. The inherently dy namic structure of the real-time recurrent network and its capability to model highly non-linear systems makes this type of network an ideal tool for the real-time determination of average chain lengths. The ma jor advantages of this technique are that it gives a fast and accurate on-line estimation of the polymers chain length and that it can repla ce the more elaborate and time consuming analytical methods. Pretraine d networks are used to process sets of untrained input patterns.