Research On Digital Image Correlation Deformation Measurement Method Based On Convolution Neural Network

The deformation measurement method based on digital speckle image is an important method of object deformation characterization.Because of its non-contact and full field measurement advantages,it is widely used in deformation measurement of materials.With the development of deep learning image processing technology based on convolutional neural network(CNN),this technology has been widely used in various image feature recognition and classification tasks.In this paper,the two-dimensional digital speckle image correlation deformation measurement method based on convolutional neural network is studied by using the powerful image processing learning ability of convolutional neural network.The accuracy and calculation efficiency of deformation measurement method based on digital image correlation are improved,and the means of digital image correlation deformation measurement are extended.Firstly,based on the CNN depth learning technology,the learable depth CNN model for the displacement field calculation of digital images is studied.Based on the depth learning framework,pytorch,a CNN model for the recognition and calculation of displacement field of digital speckle image was proposed.The speckle image before and after deformation is used as the input of the network and the displacement field is the output.A method of data set construction method based on computer simulation speckle image is proposed.The digital speckle image before and after deformation with accurate displacement label is generated.This data set including random displacement mode,axial uniform deformation and shear deformation,and Gaussian noise is added to the training data.The model of network is trained by supervised learning method,and the depth neural network(DNN)is trained to realize the calculation ability of displacement field of digital speckle image.Secondly,the effectiveness of CNN method proposed in this paper is studied by means of simulated speckle experiment and uniaxial tensile test of silica gel phantom sheet.Firstly,the relative error between the displacement field and the real value calculated by this method is less than 10% on the simulated test data set.The traditional digital correlation non iterative gray gradient method and inverse combination Gauss Newton iteration method are compared with the algorithm proposed in this paper.It shows that the method proposed in this paper is reliable and effective,has high calculation efficiency and has the displacement calculation ability of digital speckle image.At the same time,the mapping ability of the proposed method is stronger in the calculation of the random displacement field constructed in this paper.Secondly,the uniaxial tensile test of the silica gel phantom sheet with surface spray spot was used to verify the method.The tensile process made the specimen extend by 0.0050 mm.The average value of the clamp head displacement calculated by CNN method is 0.0044 mm.This demonstrated The validity and accuracy of the method proposed in this paper.Finally,based on the CNN digital image correlation method of displacement field measurement,two experiments are carried out.Firstly,the micro surface thermal deformation of ferroalloy composite is studied.It is found that different components of the material show different surface deformation response rules under the change of temperature.Secondly,based on OCT image,the distribution of displacement field of the internal cross section of silicone biomimetic hose material under different loading conditions is studied.It is found that the displacement of the cross section increases with the increase of water pressure,and the displacement mainly occurs in the inside of the hose,which is in line with the actual deformation.