Towards pultrusion process optimisation using artificial neural networks

• Published in: Journal of materials processing technology 1998-11, Vol.83 (1), p.131-141
• Main Authors: Wilcox, J.A.D., Wright, D.T.
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Optimisation; Process modelling; Pultrusion
• ISSN: 0924-0136
• DOI: 10.1016/S0924-0136(98)00052-1

The pultrusion process involves resin impregnated fibres and mats passed through a heated die, which cures the resin and produces the final product. When the process is examined in detail the number of parameters affecting the final product quality and process efficiency is very large. Parameters can range from the basic such as process speed, to the more complex such as the crosslinking reaction of the resin in the die. More important than the individual parameters are the relationships between these different parameters. Due to the complexity of the problem, the more usual compact mathematical descriptions of the whole process are not feasible in commercial manufacturing environments. This paper describes the use of artificial neural networks (ANNs) for pultrusion process modelling of real process data and their potential for intelligent machine control. It details how the use of ANNs can offer insights into the importance of the connections between the individual process parameters without having any ‘knowledge’ of the process. Such insights could lead to a greater understanding of the process, reduced product development time and increased manufacturing process capability and efficiency.