Height Prediction Method Of Laser Cladding Thin Wall Component Based On Visual Monitoring

Laser cladding is a kind of additive manufacturing technology to manufacture or repair metal parts by powder melting and deposition.There are many kinds of physical quantities influencing each other in the processing process of this technology,which makes it difficult to guarantee the quality of the processed parts.When processing thin-walled parts,it is easy to have a high degree of instability.In order to ensure that the height of thin-walled parts increases steadily in the process of laser cladding,so as to obtain better formed thin-walled parts,the current height of thin-walled parts is obtained by using image processing algorithm of side axis molten pool image,which provides data support for height prediction of thin-walled parts;Based on the image of coaxial molten pool,the height prediction method of thin-walled parts based on LSTM neural network and cnn-lstm neural network is proposed.This paper is summarized as follows(1)The contour of the side axis weld pool is obtained by image processing algorithm,which is used to fit the height of thin-walled parts.Using high dynamic industrial camera and light compensation device,the processing image of side axis thin-walled parts is taken.The image contour of side axis molten pool is extracted by relevant image processing algorithm.The surface contour of thin-walled parts is taken as the surface contour of molten pool trajectory image.The height of thin-walled parts is calculated by camera calibration,and the error of image height calculation is analyzed.(2)A height prediction model of thin-walled parts based on feature extraction and LSTM neural network is proposed.Based on the coaxial molten pool image,three feature parameters which can represent the height change of thin-walled parts are extracted:the Laplacian variance of the image,the area of low frequency region after Fourier transform of the image and the area of molten pool transition texture,which are used as the input of LSTM neural network;Input to the trained LSTM neural network to realize the height prediction of thin-walled parts.(3)A cnn-lstm neural network height prediction model for thin-walled parts is proposed.The advantages of deep neural network are given full play.The model directly takes the image of coaxial molten pool as the input and inputs it into LSTM neural network to realize the height prediction of thin-walled parts.(4)Based on Lab VIEW and python,a real-time monitoring and prediction system for the height of laser cladding thin-walled parts is developed,which realizes the monitoring of the forming state of the thin-walled parts,and provides a reference for the subsequent manual adjustment of process parameters and the closed-loop control of the laser cladding system by using the control system.