A novel extraction and evaluation method for melt pool images in laser powder bed fusion

Abstract

In-situ monitoring of the laser powder bed fusion (LPBF) process is essential for ensuring quality and stability. This study proposes a novel extraction and evaluation method for monitoring melt pool images during LPBF to improve the accuracy and efficiency of extracted data. The proposed method segments the melt pool into regions of interest (ROI) based on the intensity gradient of the melt pool boundary in the images, enabling the extraction of texture features from the Gray Level Co-Occurrence Matrix (GLCM) subsequently. The intensity-gradient image processing has improved robustness and accuracy. Based on the widths from the melt pool and single track, the mean square error (MSE) is 12.22 μm, outperforming the conventional fixed-threshold method by 25.3 %. In addition, a classification and regression tree (CART) model was used to categorize the melt pools into four types—balling, irregularity, normal, and over-melting—based on geometric, shape, temperature, and texture features, achieving a classification accuracy of 97.24 %. It is found that the texture feature, contributing to 58.4 % of the importance, and the temperature feature, contributing 37.6 %, emerged as vital features in the melt pool identification process. The texture feature is particularly effective in identifying melt pools in normal and over-melting modes, mainly due to the Correlation (COR) and Contrast (CON). Our research sheds light on improving the accuracy of melt pool segmentation significantly and offers crucial insights for feature extraction, thereby providing considerable assistance for online defect identification in LPBF.