Defect Characterization and Prediction for Femtosecond Laser Drilling of Carbon Fiber‐Reinforced Polymer Based on Image Processing and Machine Learning

ABSTRACT

Heat‐affected zone (HAZ) and taper are critical defects affecting the femtosecond laser drilling quality of carbon fiber‐reinforced polymer. Owing to the discontinuity of the unilateral edges of HAZ images, the missegmentation rate of the single threshold image segmentation method can be as high as 20.31%. A novel hybrid image segmentation method integrating fuzzy C‐means clustering and iterative threshold segmentation was proposed in this study, which reduced the missegmentation rate of HAZ images to 1.55% and improved the characterization accuracy of the HAZ. Based on a high‐quality dataset constructed from high‐precision characterization data of defects in 125 laser drilling experiments, a systematic comparison was conducted on the predictive performance of six machine learning regression models—linear regression, decision tree, random forest, support vector regression, k‐nearest neighbor, and artificial neural network for HAZ and taper. The results demonstrated that the random forest regression model exhibits the optimal performance in both prediction tasks: the coefficient of determination reaches 0.9882 for the average HAZ width prediction and 0.9905 for the taper angle prediction; the corresponding root mean square errors are 1.1706 and 0.4358, and the mean absolute errors are 0.9392 and 0.3720. Feature importance analysis based on SHapley Additive exPlanations indicates that scanning speed is the primary factor influencing the average HAZ width, while average power has the most significant impact on the taper angle.