Sensor fusion machine learning for surface quality prediction in laser powder bed fusion process

Abstract

In the laser powder bed fusion (LPBF) process, surface quality can impact the mechanical properties of the final part and may even result in process failure. Sensor monitoring is a promising approach, however, research on surface quality monitoring mainly rely on single-sensor approach, this approach cannot fully characterize surface quality. To address this limitation, this study introduces a novel sensor fusion machine learning (SFML) Framework for surface quality prediction. The SFML framework integrates sensors data from photodiodes and high-speed cameras, utilizing time–frequency and entropy features derived from both light intensity data and molten pool grayscale characteristics to establish robust correlations with surface quality. Then random forest model is employed to achieve over 96% accuracy, 100% in the area of receiver operating characteristic.This model classifies surface quality into 4 categories: acceptable melting, insufficient melting, slight bulge, and moderate bulge. Validation against experimental data demonstrates high precision of the framework in online surface quality prediction. By revealing the relationships between time–frequency/entropy features of molten pool light intensity/gray values and surface quality, the SFML framework exhibits robust performance across various LPBF process conditions, offering significant potential for real-time quality control, predictive maintenance, and defect detection in intelligent manufacturing.