Estimation of Energy and Time Usage in 3D Printing With Multimodal Neural Network

Estimation of resources (i.e. energy and time) consumed in the 3D printing process provides critical information for resource-efficient planning, designing and manufacturing. The existing approaches predict the resource usage by either making calculations based on numerical control (NC) codes or generating estimations by processing digital 3D models. On one hand, the former approach usually requires laborious experiments to calibrate the resources consumption of different printing actions by running human selected NC codes segments. On the other hand, the later approach is often less accurate due to the lack of printing process information (i.e. trajectory, velocity, temperature, etc.) varying significantly over different machines. In this paper, we present a deep learning-based approach that addresses those issues. Our model 1) directly learns from the NC code files accumulated in 3D printing, without human intervention, and 2) leverages the information from both 3D models and NC codes to yield improved estimations. Our approach achieves
$\mathbf{R}^{\mathbf{2}} > \mathbf{0}.\mathbf{93}$
on real-world 3D printing datasets.