ONLINE PREDICTION OF FINAL PART DIMENSIONS IN BLOW MOLDING - A NEURAL-NETWORK COMPUTING APPROACH

Control over final part thickness distributions in extrusion blow mold ing would be very useful in resin optimization. An on-line measurement is essential for process monitoring and control of the part dimension s. Excessive resin usage results in material waste and increased cycle times because of increased cooling requirements. An inadequate thickn ess results in decreased mechanical strength, especially in regions al ong the part where large blow ratios or complex geometries exist. Neur al networks are investigated as a method for the on-line prediction of the final part distribution from the parison dimensions. The purpose of this work is to demonstrate the feasibility, for preliminary use, o f neural networks for this application. The network inputs include the initial parison thickness and temperature profiles, the bottle mold g eometry and a rheological parameter representative of the material. Va rying blow-up ratios are obtained from the bottle mold geometry. The n etwork accesses data from a pool of eighty data sets for the training sequence. The data sets are broadly distributed with regard to the ope rating conditions, so as to give the network a wide range of applicabi lity. The simulations are performed on data sets not present in the ac cess pool used for training.