In-situ monitoring of layer-wise process quality and signal analysis for laser powder bed fusion using multi-source optical signal

Laser powder bed fusion is a key metal additive manufacturing technology capable of fabricating geometrically complex parts, yet its reliable industrial adoption is hindered by the inherent complexity and stochastic defect formation of the process. Current quality assessment is constrained by the inherent latency of offline methods and the diagnostic limitations of single-sensor monitoring. To address these challenges, this study developed a multi-source optical signal monitoring system integrating coaxial photodiodes and an off-axis industrial camera to achieve simultaneous powder spreading detection and radiation signal monitoring during LPBF layer-wise process quality monitoring. Based on the successful identification and analysis of typical detectable features, the YOLOv5s deep learning model was employed to achieve rapid and accurate detection of lack-of-powder defects during the printing process. The training results indicated that the model exhibited good performance metrics. The relationships between process parameters, typical defects, and multi-channel monitoring data were also investigated. The monitoring system achieved a spatial resolution of 300 μm for in-process monitoring and demonstrated high accuracy in detecting various defect types, including lack of powder, pores, warping, stitching seams, and printing failures. Furthermore, the algorithm-detected signal anomalies exhibited good spatial correlation with the actual surface defects. Simultaneously, wavelet time-frequency analysis was employed to evaluate molten pool dynamic stability under different process parameters and to analyze energy distribution for different defects. Furthermore, 3D model reconstruction from signals enabled effective correlation with actual part defects. Based on the signal-driven process optimization, complex conformal cooling molds were successfully fabricated with a grafting accuracy error of less than 0.12 mm on high-performance substrates, demonstrating the practical efficacy of the developed monitoring methodology. This study provides both a technological and a theoretical foundation for intelligent quality control in LPBF and its practical implementation in industry.