Layer Printing Time Prediction for Direct Metal Laser Sintering Based on Deep Learning

Precise prediction of printing duration in Direct Metal Laser Sintering (DMLS), a crucial process in highprecision metal additive manufacturing, is a persistent challenge due to the technology’s inherent dynamic multiphysics characteristics. Conventional approaches often struggle to deliver adequate accuracy due to limitations in capturing the influence of dynamic process parameters and exhibiting computational bottlenecks when dealing with intricate geometries. These limitations can lead to notable discrepancies between predicted and realized printing times, especially in the context of complex industrial parts subject to varying process conditions. To address this issue, we present a deep learning (DL)-based framework for DMLS printing time prediction. Our approach utilizes a comprehensive dataset of layer-wise printing images and corresponding process parameters acquired during DMLS fabrication. We trained and evaluated a suite of deep neural network architectures, including ResNeXt, ResNet, SqueezeNet, and Vision Transformer, to improve the fidelity of layer printing time predictions. Empirical evaluations indicate that the optimized ResNeXt model achieves a mean relative error of one point three percent in layer printing time prediction, a substantial improvement compared to a twenty-four points four eight percent relative error observed with a traditional prediction method. This work demonstrates the considerable promise of DL as a valuable tool for DMLS process monitoring and time estimation, providing a pathway to enhance process visibility without requiring simplified theoretical assumptions. This data-driven methodology has the potential to streamline DMLS process optimization and accelerate the adoption of additive manufacturing in industrial settings.