The 3D Feature Prediction of the Molten Metal Direct Writing Result based on Neural Network

Abstract

The three-dimensional features of the single track forming result are one of the key factors for metal 3D printing, while few studies are carried out to predict the three-dimensional features in liquid metal jetting printing area. In this paper, the results of the single track forming are scanned by 3D displacement sensor to obtain accurate 3D model. Based on the 3D model, the average width and average height are logically selected to demonstrate the three-dimensional features. To predict the type and size of the single track forming result, 2 neural networks are designed, which both contain 2 hidden layers. After training and testing, the Classification model of the two Multi-Input-Multi-Output neural networks has more than 90% accuracy in the classification prediction of the single channel forming results, and the average prediction errors of the average width and average thickness in the size prediction model are limited to 6%.