Intelligent Geometry Compensation for Laser Powder Bed Fusion by Machine Learning Derived from Optical Scanning Data

Recently, laser powder bed fusion (LPBF) has shown great potential in advanced manufacturing. However, the rapid melting and re-solidification of the molten pool in LPBF leads to the distortion of parts, especially thin-walled parts. The traditional geometric compensation method, which is used to overcome this problem, is simply based on mapping compensation, with the general effect of distortion reduction. In this study, we used a genetic algorithm (GA) and backpropagation (BP) network to optimize the geometric compensation of Ti6Al4V thin-walled parts fabricated by LPBF. The GA-BP network method can generate free-form thin-walled structures with enhanced geometric freedom for compensation. For the GA-BP network training, an arc thin-walled structure was designed and printed by LBPF and measured via 3D scanning measurements, resulting in point cloud data sets of more than 470,000 data points. The final distortion of the compensated arc thin-walled part based on GA-BP was reduced by 72.7%, which is better than the compensation method based on BP and mapping. The effectiveness of this GA-BP compensation method is further evaluated in an application case using new data points, and the result shows that the final distortion of the oral maxillary stent was reduced by 71%. In summary, the GA-BP-based geometric compensation proposed in this study can better reduce the distortion of thin-walled parts with higher time and cost efficiencies.