PredNet and CompNet: Prediction and High-Precision Compensation of In-Plane Shape Deformation for Additive Manufacturing

The error compensation for printed objects in additive manufacturing (AM) has always been one of the most critical problems. The precision control of the AM is usually more difficult than the subtractive manufacturing system, whose precision can reach the micron level easily by using a servo system. For the AM, there usually exist shrinkage and curling effects which lead to deformation. In this paper, we focus on the in-plane shape deformation problem, and we build the PredNet and CompNet, using deep neural networks for the error prediction and compensation. We test our methods on dental crown models. We generate deformed models by simulation of the translation, scaling down and rotation deformation. The minimum F1 scores of error prediction and compensation can be up to 0.982.