Neural networks for trajectory evaluation in direct laser writing

Material shrinkage commonly occurs in additive manufacturing and compromises the fabrication quality by causing unwanted distortions or residual stresses in fabricated parts. Even though it is known that the resulting deformations and stresses are highly dependent on the writing trajectory, no effective strategy for choosing suitable trajectories has been reported to date. Here, we present a path to achieve this goal in direct laser writing, an additive manufacturing method based on photopolymerization that commonly suffers from strong shrinkage-induced effects. First, we introduce a method for measuring the shrinkage of distinct direct laser written lines. We then introduce a semi-empirical numerical model to capture the interplay of sequentially polymerized material and the resulting macroscopic effects. Finally, we implement an artificial neural network to evaluate given laser trajectories in terms of the resulting part quality. The presented approach proves feasibility of using artificial neural networks to assess the quality of 3D printing trajectories and thereby demonstrates a potential route for reducing the impact of material shrinkage on 3D printed parts.