Porosity Identification in Laser Powder Bed Fusion Through Feature-based Multi-Observation Hidden Markov Model

Laser Powder Bed Fusion (LPBF) offers significant advantages to fabricating parts in a net-shape or near-net-shape finish. However, the integrity of the as-fabricated part is often affected by defect formation during the printing process. In this study, a multi-variable observation Hidden Markov Model (HMM) is developed to predict porosity defect formation during the LPBF process using thermal signatures. This work is based on the nature of the LPBF’s layer-by-layer printing process, where each layer’s quality is influenced by the previous printed layers, which aligns with the general HMM assumption. The HMM is established to correlate the thermal signatures (observations) extracted from in-situ thermal images, such as the maximum temperature and cooling rate, with the porosity (hidden state) of each layer. X-ray CT scanning results are adopted as the ground truth of defect states. The HMM is evaluated in three case studies considering different thermal signatures as the observations: 1) maximum temperature, 2) cooling rate at a specific temperature, and 3) multi-observation combining maximum temperature and cooling rate at a specific temperature. Results from data analysis indicate that the multi-observation HMM achieves a prediction accuracy of 92.7%, surpassing that of both single-variable observation HMMs. This indicates the advantage of multiple variable observation HMM in improving the accuracy, robustness, and reliability for defect formation during LPBF.