Quantitative prediction of additive manufacturing deposited layer offset based on passive visual imaging and deep residual network

Abstract

In the additive manufacturing process, the actual position of the welding torch will affect the shape of the weld pool, thereby affect the welding quality. So online monitoring of the deposited layer offset is essential. Based on the visual morphology of the molten pool obtained by the passive vision sensor, this paper proposes a quantitative prediction method for the deposition layer offset, during the arc additive manufacturing process. In order to prove the accuracy and effectiveness of this prediction method, this paper uses the deposited layer data to construct a model, and then predicts the offset. On this basis, the deviation direction of the initial deposited layer was changed, thus the generalization ability of the prediction method is verified. The experimental results show that the average prediction error of the deposited layer offset is less than 0.14 mm, while the network takes about 5 ms to process molten pool image. It proves that the proposed prediction method can be used to monitor the deposited layer offset real-time.