The Research Of Identifying Molten Pool State Of Laser-based Direct Energy Deposition Based On Deep Learning

Metal laser melting deposition is a new processing technology that uses laser beam as focused thermal energy to synchronously melt and deposit materials.However,there is a complex multiphysics coupling effect during the processing,which makes the deposition quality difficult to guarantee.The quality of metal laser melting deposition is closely related to the state of the molten pool.Recognizing the state of the molten pool during processing according to the visual characteristics of the molten pool,and thus realizing the online prediction of the deposition quality is of great significance to the metal laser melting deposition technology.In this paper,the image of the molten pool of metal laser deposition is taken as the research object,and the image preprocessing and the recognition method of the molten pool based on the deep learning are proposed.The research content of the thesis is summarized as follows:(1)This paper proposes a PPCNN-based metal laser melting deposition pool state recognition method.As the molten pool area only occupies a small part of the original image,the method of threshold segmentation and edge detection is used to extract the region of interest in the original molten pool image as input.Aiming at the problem of high-level feature loss of ordinary convolutional neural networks,A pyramid pooled convolutional neural network model is constructed.For the problem of poor generalization of the model,an L2 penalty term is introduced to increase the generalization of the model.The influence of different network parameter settings on the model recognition effect is studied,and the validity of the model is verified through comparative experiments.(2)A state identification method for molten pool of metal laser deposition based on SFDCF and DenseNet is proposed.Aiming at the problem that the original molten pool image of interest contains more noise and interference information,a method of filtering and enhancing the original image by combining the spatial and frequency domain filtering methods is proposed;for the PPCNN to forward propagation In the process,there is still the problem of low transmission efficiency,and a densely connected convolutional neural network model is constructed;for the problem of large parameter calculations for featureextraction of standard convolution kernels,deep separable convolution is used to replace standard convolution.Through the verification of the molten pool test data,the proposed method can obtain a higher recognition rate and greatly reduce the training time of the model.A state identification method for molten pool of metal laser deposition based on SFDCF and DDenseNet is proposed.In view of the problem that the original molten pool of interest contains more noise and interference information,a method of filtering and image enhancement of the original molten pool of interest by combining the spatial and frequency domain filtering methods is proposed;for the standard convolutional neural network in In the forward propagation process,there is still a problem of low transmission efficiency,and a densely connected convolutional neural network model is constructed;for the problem of large parameter calculations for feature extraction of standard convolution kernels,deep separable convolutions are used to replace standard convolutions.Through the verification of the molten pool test data,the proposed method can obtain a higher recognition rate and greatly reduce the training time of the model.(3)Using LabVIEW and Python to realize the development and application of the proposed algorithm.Through the development of the metal laser melting deposition molten pool status recognition system,the actual application of metal laser melting deposition data collection and processing,online monitoring of the molten pool status,and deep learning-based molten pool status recognition are realized.Through the research of this project,it is possible to achieve accurate identification of the state of the metal laser melting deposition molten pool,and then provide theoretical and technical support for feedback control and improving the quality of the laser melting deposition.