Predicting and preventing layer-wise surface defects: Towards proactive quality control in laser powder bed fusion

The quality of as-built parts in laser powder bed fusion (LPBF) is significantly affected by the condition of each manufactured layer. This study introduces a proactive quality control approach that integrates predictive modeling and dynamic process control to improve layer-wise surface quality. Specifically, an encoder-convolutional long short-term memory (ConvLSTM)-decoder model is developed to predict the surface morphology of subsequent layers using sequential post-recoating images. These predictions enable a control strategy that dynamically adjusts laser power to maintain consistent surface quality. Experimental results demonstrate that this approach facilitates early detection of potential surface defects, allowing for timely process parameter adjustments and preventing defect progression. Parts manufactured with this proactive strategy exhibit significantly improved surface quality compared to those produced without such adjustments. This integration of predictive modeling and proactive control offers a promising solution to maintain high surface quality and enhance overall part quality in LPBF processes.