Predictive modelling of surface roughness and kerf widths in abrasive water jet cutting of Kevlar composites using neural network

Abstract

Abrasive water jet cutting (AWJC) is one of the important non-traditional machining processes used for cutting of difficult-to-cut materials and intricate profiles. Cutting of Kevlar fibre reinforced polymer composites is a complex process, making it difficult to model, predict and improve the cut surface quality. This paper presents a detailed approach of the usage and effectiveness of a back-propagation neural network (NN) for modelling and prediction of three cut surface characteristics namely top kerf width, bottom kerf width and surface roughness (Ra) in AWJC of aerospace grade Kevlar-epoxy composites. Statistically designed full factorial experiments based on three process parameters [water jet pressure (WJP), abrasive flow rate (AFR) and quality level (QL)] at three levels each were conducted to generate the NN training database. The results demonstrate that the NN model was able to successfully model and predict the two kerf widths and surface roughness closely matching the experimental results.