A layer-wise melting defects mitigation method in laser powder bed fusion process based on machine learning and fuzzy inference.

Abstract

Melting defects in Laser Powder Bed Fusion (LPBF) processes, such as lack of fusion (LOF) or over-melting (OM), can cause significant deterioration in mechanical properties and surface roughness of printed parts, potentially leading to process failure. Previous attempts to utilize local melt pool-related information for LPBF process control have faced limitations due to the high requirements on sensors and data processing, as well as the lack of representativeness of local melt pool information. This study focuses on the surface quality of the parts and proposes an image based LPBF control to mitigate melting defect. A quality identification module utilizing convolutional neural networks (CNN) to perform layer-by-layer evaluation of the melting quality of the part. The CNN achieved an accuracy of up to 98.2% in identifying melting quality. Furthermore, based on the surface melting quality extracted by CNN, a fuzzy control strategy (FIC) integrated with a historical state consistency check mechanism (HSCCM) is introduced to determine the optimal control actions for subsequent layers. Experimental results affirm that the FIC integrated with HSCCM effectively alleviates surface melting defects, thereby enhancing surface roughness and manufacturing quality of components. This research offers a novel approach for online quality monitoring and improvement in LPBF processes.