Additive Manufacturing Morphological Feature Monitoring Based On Visual Feature

On-line metallurgical quality monitoring is one of the hot and important research contents in the field of fusion welding and additive manufacturing.Forming shape control of he deposited layer in additive manufacturing has become the focus and difficulty of additive manufacturing research.As one of the media of quality sensing in metallurgical process,light can best describe the quality characteristics of the process.In this paper,an online welding quality monitoring system based on visual information of weld pool in additive manufacturing process is studied.In the process of additive manufacturing,the actual position of welding torch will affect the weld pool shape,and thus affect the welding quality,so the online monitoring of the offset of he deposited layer is essential.Then the additive weld reinforcement is predicted by using convolutional neural network to extract the visual features of the molten pool during additive manufacturing.On the basis of quantitative prediction of the additive weld reinforcement,visual information of the additive weld reinforcement under another vision is introduced to achieve precision prediction of the additive weld reinforcement based on multiangle vision fusion.The main research contents of this paper are as follows:(1)In order to solve the problem that the existing monitoring technology cannot monitor the deposited layer offset based on passive vision,a prediction model of the deposited layer offset based on deep residual network and passive vision is proposed.The model uses molten pool visual information to monitor the offset of the deposited layer during arc additive manufacturing.In this paper,the data of two different offset starting directions are divided into training set and test set for training and testing respectively in a ratio of 4:1.The prediction results of offset network are 0.1054 mm and 0.08295 mm respectively.The validity of the method is verified.Two deposited layers with the same starting direction of offset were randomly selected,one for training and the other for testing.The multi-layer data with different offset starting directions are trained and tested jointly.The prediction results show that the precision of the prediction network is 0.1353 mm and 0.0983 mm,respectively.The accuracy and generalization ability of the prediction model of the deposited layer offset based on deep residual network and passive vision are verified.(2)In order to solve the problem that the existing monitoring technology cannot monitor the change of the additive weld reinforcement along the welding direction in the process of additive,a prediction model of the additive weld reinforcement based on multi-angle vision fusion was proposed.In this paper,the visual features of frontal molten pool image in additive manufacturing process are analyzed.It is found that compared with the square molten pool image,the lateral molten pool image provides more spatial location information.Based on the above findings,this paper constructed a multi-angle visual fusion additive weld reinforcement quantitative prediction system based on deep residual network.The experiment using only the frontal molten pool image as the training data and the multi-angle visual fusion information as the network training data were compared.Finally,the model prediction errors of the two experiments were 0.2564 mm and 0.1142 mm respectively,which verified the effectiveness of the additive weld reinforcement prediction model based on multi-angle visual fusion.