High-accuracy predictive model for carbon fiber reinforced polymer laser machining quality using neural networks

Abstract

In order to address the issue of thermal damage induced by laser processing of carbon fiber reinforced polymer (CFRP), researchers have conducted an optimization study of process parameters in the laser processing of CFRP. Their aim is to elucidate the relationship between process parameters and processing quality to minimize thermal damage. However, during laser processing, there exists a complex nonlinear relationship between process parameters and processing quality, making it challenging to establish high-precision predictive models, while the intrinsic connection between these two aspects remains incompletely revealed. In light of this, this study proposes utilization of machine learning techniques to explore the inherent relationship between process parameters and processing quality and establishes a 5-13-5 type back-propagation (BP) neural network predictive model. Subsequently, genetic algorithms are employed to optimize the weights and thresholds of the BP neural network, and the model is then subjected to validation. The results indicate that the BP neural network predictive model yields average errors of 5% for surface heat-affected zone (HAZ), 2.9% for groove width, 5.9% for cross-sectional HAZ, 1.8% for groove depth, and 4.5% for aspect ratio, demonstrating a relatively high level of accuracy but with notable fluctuations. The GA-BP model, when predicting the surface HAZ and the groove width, achieves errors of 4.5% and 2.7%, respectively, which are lower when compared to the BP model, indicating a higher predictive accuracy. The GA-BP model established in this study unveils the intrinsic connection between process parameters and processing quality, providing a novel means for an effective quality prediction in the processing of CFRP.