In-Situ Sensing Framework for Machine Learning Based Defect Prediction in Laser Powder Bed Fusion

Abstract

Laser powder bed fusion (LPBF) enables the fabrication of complex metal components through additive manufacturing, but component quality remains limited by the formation of internal defects such as porosity, lack of fusion, and keyhole induced voids. These defects are typically identified using techniques such as X-ray computed tomography (XCT), which provide high resolution characterization but do not allow for detection during fabrication. This work investigates whether in-situ photoemission and acoustic measurements can be used to characterize and predict defect formation in LPBF. Samples of 316L stainless steel were fabricated across a range of laser power and scan speed conditions to generate varying defect states. During fabrication, a photodiode and microphone recorded continuously to capture optical emission and acoustic activity. Following fabrication, XCT was used to quantify internal defects and establish a reference dataset. A data processing pipeline was developed to extract descriptive features from both XCT volumes and in-situ signals, enabling direct comparison between process behavior and defect characteristics. Machine learning models were then applied to evaluate the predictive relationship between sensor derived features and XCT characterized defect metrics. The results demonstrate that in-situ measurements contain measurable predictive information for key defect characteristics, with model performance indicating meaningful relationships for several targets including pore size related metrics and the overall porosity fraction of samples. While the models are not optimized for deployment, the findings establish the feasibility of using photodiode and acoustic measurements to characterize and predict defect behavior during LPBF, providing a foundation for future development of in-situ monitoring and control strategies.