Neural Network Approach to Density Control of Rigid PVC Foam in Extrusion Process

Abstract

Abstract In rigid PVC foam extrusion, foam density is very crucial in determining the mechanical characteristics of the extruded material and the profitability of the manufacturing process. This paper presents a new application of artificial neural computing in the control of PVC foam density in profile extrusion process. A 3-layer multi-layer perceptron (MLP) artificial neural network was developed to estimate the foam density (weight) based on the known processing conditions; mainly heating zones temperatures and screw speed. The network was developed and tested on a specific formula that is used in wood-like products. A two factorial design of experiment was carried out to determine the significant process variables before training the neural networks using a back propagation algorithm. Finally, a comparison between the true weights, and the estimated weights using artificial neural networks is presented.