A material life prediction method based on microstructure analysis

Abstract

The present invention discloses a material life prediction method based on microstructure analysis, which relates to the field of material physical property measurement and analysis, including: irradiating the material to be tested with X-rays to collect a three-dimensional X-ray diffraction spectrum; spatially transforming the three-dimensional X-ray diffraction spectrum to obtain a three-dimensional microstructure map; analyzing the texture of the three-dimensional microstructure map by a multi-scale neural network to obtain an estimate of the average grain size; and calculating the life of the material to be tested by a stress fatigue life fitting function optimized by the microstructure. Starting from the microstructure, the present invention obtains the average microscopic grain size of the material to be tested through spatial transformation and texture recognition of the X-ray diffraction map, and uses this as an important microscopic parameter, combined with the macroscopic parameters of stress fatigue damage, to perform a more accurate material life prediction compared to traditional technologies.

Description

Technical Field

[0001] The invention relates to the field of material physical property measurement and analysis, and in particular to a material life prediction method based on microstructure analysis. Background Art

[0002] Material life prediction is an important topic in the engineering community. It is critical to the maintenance, replacement, and safe operation of equipment.

[0003] Traditional material life prediction methods mainly rely on macroscopic parameters such as material strength, hardness, strain, or empirical formulas such as exponential distribution expressions. However, these methods often fail to accurately predict the service life of materials because they ignore the influence of the internal microstructure of the material. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a material life prediction method based on microstructure analysis, which solves the problem of poor accuracy of traditional technology.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0006] A material life prediction method based on microstructure analysis comprises the following steps:

[0007] S1, irradiating the material to be tested with X-rays to obtain a three-dimensional X-ray diffraction pattern;

[0008] S2, spatially transforming the three-dimensional X-ray diffraction pattern to obtain a three-dimensional microstructure map;

[0009] S3, analyzing the texture of the three-dimensional microstructure map through a multi-scale neural network to obtain an estimate of the average grain size;

[0010] S4. Based on the estimated average grain size, the life of the material to be tested is calculated through the stress fatigue life fitting function of microstructure optimization.

[0011] The beneficial effects of the present invention are as follows: starting from the microstructure, the present invention obtains the average size of the microscopic grains of the material to be tested through spatial transformation and texture recognition of the X-ray diffraction pattern, and uses this as an important microscopic parameter, combined with the macroscopic parameters of stress fatigue damage, to make a more accurate material life prediction compared to traditional technologies.

[0012] Furthermore, the expression of the space conversion in S2 is:

[0013] ,

[0014] ,

[0015] ,

[0016] in, is the x-axis vector of the three-dimensional microstructure diagram, is the y-axis vector of the three-dimensional microstructure diagram, is the z-axis vector of the three-dimensional microstructure diagram, π is the circumference of a circle, ∙ is the vector dot product, and × is the vector cross product. is the x-axis vector of the three-dimensional X-ray diffraction pattern, is the y-axis vector of the three-dimensional X-ray diffraction pattern, is the z-axis vector of the three-dimensional X-ray diffraction pattern.

[0017] The beneficial effect of the above further scheme is that after X-ray irradiation of the material to be tested, a lattice diffraction image can be obtained. The image space is called reciprocal space, in which the particle lattice and the particles in the real space present a Fourier correspondence relationship. Through the spatial vector transformation of the above formula, the target object in the three-dimensional X-ray diffraction spectrum can be converted into the real space, which is more convenient for direct texture analysis and grain contour analysis.

[0018] Furthermore, the multi-scale neural network in S3 includes:

[0019] The multi-scale convolution sub-network is used to extract the edge texture of the material lattice in the three-dimensional microstructure map at different scales through multi-layer convolution, and perform weighted fusion to obtain a multi-scale texture feature map;

[0020] The fully connected operation sub-network is used to obtain an average grain size estimate based on a multi-scale texture feature map through a fully connected neural network operation.

[0021] The beneficial effect of the above further scheme is that the texture of the three-dimensional microstructure image is the microscopic grain outline of the material to be tested. Through multi-scale texture extraction and fully connected network calculation, the average grain size of the material to be tested can be obtained.

[0022] Furthermore, the multi-scale convolutional subnetwork includes:

[0023] The first convolution layer is used to process the three-dimensional microstructure map through convolution operation to obtain a first-scale texture feature map;

[0024] A second convolution layer is used to process the first-scale texture feature map through a convolution operation to obtain a second-scale texture feature map;

[0025] A third convolutional layer, used for processing the second-scale texture feature map through a convolution operation to obtain a third-scale texture feature map;

[0026] A first maximum pooling layer is used to perform maximum pooling downsampling on the first-scale texture feature map to obtain a first-scale texture feature sampling map of the same size as the third-scale texture feature map;

[0027] The second maximum pooling layer is used to perform maximum pooling downsampling on the second-scale texture feature map to obtain a second-scale texture feature sampling map of the same size as the third-scale texture feature map;

[0028] A first multiplier, configured to weight the first scale texture feature sampling image according to a first weighting coefficient by performing a matrix multiplication operation;

[0029] a second multiplier, configured to weight the second scale texture feature sampling image according to a second weighting coefficient by performing a matrix multiplication operation;

[0030] a third multiplier, configured to weight the third-scale texture feature map according to a third weighting coefficient by performing a matrix multiplication operation;

[0031] The adder is used to perform matrix addition on the weighted first-scale texture feature sampling map, the second-scale texture feature sampling map and the third-scale texture feature map to obtain a multi-scale texture feature map.

[0032] Furthermore, the convolution kernels of the first convolution layer, the second convolution layer and the third convolution layer are:

[0033] ,

[0034] The first convolutional layer uses the following image convolution operation:

[0035] ,

[0036] Among them, fig out (x,y) is the pixel value of the x row and y column of the output image of the image convolution operation, c(i,j) is the value of the i row and j column of the convolution kernel of the image convolution operation, fig in (x+i, y+j) is the pixel value of the x+i row and y+j column of the image convolution operation input image, b ias is the bias coefficient of image convolution operation;

[0037] The second convolutional layer and the third convolutional layer use the following image hole convolution operation:

[0038] ,

[0039] in, The pixel value of the x row and y column of the output image of the hole convolution operation of the image, is the i-row and j-column value of the convolution kernel of the image hole convolution operation, is the pixel value of the x+i∙κ row and y+j∙κ column of the image atrous convolution operation input graph, where κ is the dilation rate. is the bias coefficient of the image dilated convolution operation.

[0040] The beneficial effects of the above further scheme are as follows: the convolution kernel designed and used in the present invention is a 3×3 matrix based on second-order differentials, which is suitable for detecting edge contours of high-frequency change areas in images and can effectively extract grain textures. In the second and third convolutional layers, dilated convolution operations are used to introduce expansion rates so that convolution kernels of the same size can obtain a larger receptive field. Multi-scale texture feature weighting takes into account both the receptive field and the resolution of details. The size matching before weighted summation of texture feature maps of each scale uses maximum pooling layer sampling, which can effectively protect texture information during downsampling and is more conducive to subsequent operations.

[0041] Furthermore, the fully connected operation subnetwork includes at least two fully connected layers, and uses a gradient descent algorithm and a mean square error function as a cost function to perform parameter training.

[0042] Furthermore, the expression of the stress fatigue life fitting function of the microstructure optimization in S4 is:

[0043] ,

[0044] Among them, N is the life of the material to be tested, γ is the environmental correction factor, σ0 is the fatigue strength coefficient, σ a is the stress amplitude, η is the fatigue strength coefficient correction factor, ξ is the grain size strengthening proportional factor, m is the grain size strengthening exponential factor, and b is the fatigue strength index.

[0045] Furthermore, the fitting method of the stress fatigue life fitting function of the microstructure optimization comprises the following steps:

[0046] A1. Determine the stress amplitude according to the working conditions of the material to be tested;

[0047] A2. Obtain fatigue strength coefficient and fatigue strength index according to the type of material to be tested;

[0048] A3. Set the initial values ​​of fatigue strength coefficient correction factor, grain size strengthening proportional factor and grain size strengthening exponential factor, use GA algorithm, based on material samples with known life, take the minimum life calculation error as the optimization goal, and optimize the parameters of fatigue strength coefficient correction factor, grain size strengthening proportional factor and grain size strengthening exponential factor.

[0049] The beneficial effects of the above further scheme are as follows: the traditional material life fatigue calculation formula only considers the influence of alternating stress on the material, and only involves stress amplitude, fatigue strength coefficient and fatigue strength index, without considering that the microscopic grain size obtained in the actual production of the material is different, resulting in different point defects, surface defects and body defects, which makes the material fatigue resistance different. In addition, due to the influence of environmental temperature and humidity and corrosion, the coefficient index of the material will also shift. Therefore, while introducing the grain size, the present invention also sets multiple factors to fit the life and correct the details. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flow chart of a material life prediction method based on microstructure analysis provided in an embodiment of the present invention;

[0051] Figure 2 4 is a structural diagram of a multi-scale neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0053] like Figure 1 As shown, in one embodiment of the present invention, a material life prediction method based on microstructure analysis includes the following steps:

[0054] S1. Irradiate the material to be tested with X-rays to collect a three-dimensional X-ray diffraction pattern.

[0055] S2. The three-dimensional X-ray diffraction pattern is spatially transformed by the following formulas to obtain a three-dimensional microstructure map:

[0056] ,

[0057] ,

[0058] ,

[0059] in, is the x-axis vector of the three-dimensional microstructure diagram, is the y-axis vector of the three-dimensional microstructure diagram, is the z-axis vector of the three-dimensional microstructure diagram, π is the circumference of a circle, ∙ is the vector dot product, and × is the vector cross product. is the x-axis vector of the three-dimensional X-ray diffraction pattern, is the y-axis vector of the three-dimensional X-ray diffraction pattern, is the z-axis vector of the three-dimensional X-ray diffraction pattern.

[0060] After X-rays are irradiated on the material to be tested, a lattice diffraction image can be obtained. This image space is called reciprocal space, in which the particle lattice and the particles in the real space present a Fourier correspondence. Through the spatial vector transformation of the above formula, the target object in the three-dimensional X-ray diffraction spectrum can be converted to the real space, which is more convenient for direct texture analysis and grain contour analysis.

[0061] S3. Analyze the texture of the three-dimensional microstructure image through a multi-scale neural network to obtain an estimate of the average grain size.

[0062] like Figure 2 As shown, the multi-scale neural network includes:

[0063] The multi-scale convolution sub-network is used to extract the edge texture of the material lattice in the three-dimensional microstructure map at different scales through multi-layer convolution, and perform weighted fusion to obtain a multi-scale texture feature map;

[0064] The fully connected operation sub-network is used to obtain an average grain size estimate based on a multi-scale texture feature map through a fully connected neural network operation.

[0065] The texture of the three-dimensional microstructure image is the microscopic grain outline of the material to be tested. Through multi-scale texture extraction and fully connected network calculations, the average grain size of the material to be tested can be obtained.

[0066] The multi-scale convolutional subnetwork includes:

[0067] The first convolution layer is used to process the three-dimensional microstructure map through convolution operation to obtain a first-scale texture feature map;

[0068] A second convolution layer is used to process the first-scale texture feature map through a convolution operation to obtain a second-scale texture feature map;

[0069] A third convolutional layer, used for processing the second-scale texture feature map through a convolution operation to obtain a third-scale texture feature map;

[0070] A first maximum pooling layer is used to perform maximum pooling downsampling on the first-scale texture feature map to obtain a first-scale texture feature sampling map of the same size as the third-scale texture feature map;

[0071] The second maximum pooling layer is used to perform maximum pooling downsampling on the second-scale texture feature map to obtain a second-scale texture feature sampling map of the same size as the third-scale texture feature map;

[0072] A first multiplier, used for weighting the first scale texture feature sampling image according to a first weighting coefficient through a matrix multiplication operation, wherein the first weighting coefficient in this embodiment is 0.35;

[0073] a second multiplier, configured to weight the second scale texture feature sampling image according to a second weighting coefficient by performing a matrix multiplication operation, wherein the second weighting coefficient in this embodiment is 0.32;

[0074] a third multiplier, configured to weight the third-scale texture feature map according to a third weighting coefficient by performing a matrix multiplication operation, wherein the third weighting coefficient in this embodiment is 0.33;

[0075] The adder is used to perform matrix addition on the weighted first-scale texture feature sampling map, the second-scale texture feature sampling map and the third-scale texture feature map to obtain a multi-scale texture feature map.

[0076] The convolution kernels of the first, second, and third convolutional layers are:

[0077] ,

[0078] The first convolutional layer uses the following image convolution operation:

[0079] ,

[0080] Among them, fig out (x,y) is the pixel value of the x row and y column of the image convolution operation output image, c(i,j) is the value of the i row and j column of the convolution kernel of the image convolution operation, fig in (x+i, y+j) is the pixel value of the x+i row and y+j column of the image convolution operation input image, b ias is the bias coefficient of image convolution operation;

[0081] The second and third convolutional layers use the following image hole convolution operations:

[0082] ,

[0083] in, The pixel value of the x row and y column of the output image of the hole convolution operation of the image, is the i-row and j-column value of the convolution kernel of the image hole convolution operation, is the pixel value of the x+i∙κ row and y+j∙κ column of the image atrous convolution operation input graph, where κ is the dilation rate. is the bias coefficient of the image dilated convolution operation.

[0084] The convolution kernel designed and used in the present invention is a 3×3 matrix based on second-order differentials, which is suitable for detecting edge contours of high-frequency change areas in images and can effectively extract grain textures. In the second and third convolutional layers, dilated convolution operations are used to introduce expansion rates so that convolution kernels of the same size can obtain a larger receptive field. Multi-scale texture feature weighting takes into account both the receptive field and the resolution of details. The size matching before weighted summation of texture feature maps of each scale uses maximum pooling layer sampling, which can effectively protect texture information during downsampling and is more conducive to subsequent operations.

[0085] In this embodiment, the fully connected operator subnetwork includes four fully connected layers, and the last fully connected layer is used as an output layer with only one neuron, which outputs an estimate of the average grain size.

[0086] The fully connected operator subnetwork uses a gradient descent algorithm and a mean square error function as a cost function to perform parameter training. The specific process of the gradient descent algorithm is well known to those skilled in the art and will not be described in detail in the present invention.

[0087] S4. Based on the estimated average grain size, the life of the material to be tested is calculated through the stress fatigue life fitting function of microstructure optimization.

[0088] The expression of the stress fatigue life fitting function of microstructure optimization is:

[0089] ,

[0090] Among them, N is the life of the material to be tested, γ is the environmental correction factor, σ0 is the fatigue strength coefficient, σ a is the stress amplitude, η is the fatigue strength coefficient correction factor, ξ is the grain size strengthening proportional factor, m is the grain size strengthening exponential factor, and b is the fatigue strength index.

[0091] The fitting method of the stress fatigue life fitting function for microstructure optimization includes the following steps:

[0092] A1. Determine the stress amplitude according to the working conditions of the material to be tested;

[0093] A2. Obtain fatigue strength coefficient and fatigue strength index according to the type of material to be tested;

[0094] A3. Set the initial values ​​of fatigue strength coefficient correction factor, grain size strengthening proportional factor and grain size strengthening exponential factor, use GA algorithm, based on material samples with known life, take the minimum life calculation error as the optimization goal, and optimize the parameters of fatigue strength coefficient correction factor, grain size strengthening proportional factor and grain size strengthening exponential factor.

[0095] In this embodiment, based on experience and to save time for iterative training, the initial value of the fatigue strength coefficient correction factor η is set to 1, dimensionless; the initial value of the grain size strengthening proportional factor ξ is set to 600, with the dimension of megapascals per micrometer to the power of m, and the initial value of the grain size strengthening exponential factor m is set to 2, dimensionless. The process of GA (Genetic Algorithm) is not repeated here.

[0096] The traditional material life fatigue calculation formula only considers the influence of alternating stress on the material, involving only stress amplitude, fatigue strength coefficient and fatigue strength index, without considering that the microscopic grain size obtained in the actual production of the material is different, resulting in different point defects, surface defects and body defects, which makes the material fatigue resistance different. In addition, due to the influence of environmental temperature and humidity and corrosion, the coefficient indicators of the material will also shift. Therefore, while introducing the grain size, the present invention also sets multiple factors to fit the life and correct the details.

[0097] In summary, the present invention starts from the microstructure and obtains the average microscopic grain size of the material to be tested through spatial transformation and texture recognition of the X-ray diffraction pattern, and uses this as an important microscopic parameter. Combined with the macroscopic parameters of stress fatigue damage, a more accurate material life prediction is performed compared to traditional technologies.

[0098] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0099] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A material life prediction method based on microstructure analysis, characterized in that: The following steps are involved: S1, irradiating the material to be tested with X-rays to obtain a three-dimensional X-ray diffraction pattern; S2, spatially transforming the three-dimensional X-ray diffraction pattern to obtain a three-dimensional microstructure map; S3, analyzing the texture of the three-dimensional microstructure map through a multi-scale neural network to obtain an estimate of the average grain size; S4. According to the estimated average grain size, the life of the material to be tested is calculated by using a stress fatigue life fitting function optimized by microstructure; The expression of the space conversion in S2 is: ， ， ， in, is the x-axis vector of the three-dimensional microstructure diagram, is the y-axis vector of the three-dimensional microstructure diagram, is the z-axis vector of the three-dimensional microstructure diagram, π is the circumference of a circle, ∙ is the vector dot product, and × is the vector cross product. is the x-axis vector of the three-dimensional X-ray diffraction pattern, is the y-axis vector of the three-dimensional X-ray diffraction pattern, is the z-axis vector of the three-dimensional X-ray diffraction pattern; The multi-scale neural network in S3 includes: The multi-scale convolution sub-network is used to extract the edge texture of the material lattice in the three-dimensional microstructure map at different scales through multi-layer convolution, and perform weighted fusion to obtain a multi-scale texture feature map; A fully connected operation sub-network, used for obtaining an average grain size estimate according to a multi-scale texture feature map through a fully connected neural network operation; The multi-scale convolutional sub-network includes: The first convolution layer is used to process the three-dimensional microstructure map through convolution operation to obtain a first-scale texture feature map; A second convolution layer is used to process the first-scale texture feature map through a convolution operation to obtain a second-scale texture feature map; A third convolutional layer, used for processing the second-scale texture feature map through a convolution operation to obtain a third-scale texture feature map; A first maximum pooling layer is used to perform maximum pooling downsampling on the first-scale texture feature map to obtain a first-scale texture feature sampling map of the same size as the third-scale texture feature map; The second maximum pooling layer is used to perform maximum pooling downsampling on the second-scale texture feature map to obtain a second-scale texture feature sampling map of the same size as the third-scale texture feature map; A first multiplier, configured to weight the first scale texture feature sampling image according to a first weighting coefficient by performing a matrix multiplication operation; a second multiplier, configured to weight the second scale texture feature sampling image according to a second weighting coefficient by performing a matrix multiplication operation; a third multiplier, configured to weight the third-scale texture feature map according to a third weighting coefficient by performing a matrix multiplication operation; An adder, used for performing matrix addition on the weighted first-scale texture feature sampling map, the second-scale texture feature sampling map and the third-scale texture feature map to obtain a multi-scale texture feature map; The expression of the stress fatigue life fitting function of the microstructure optimization in S4 is: ， Among them, N is the life of the material to be tested, γ is the environmental correction factor, σ0 is the fatigue strength coefficient, σ a is the stress amplitude, η is the fatigue strength coefficient correction factor, ξ is the grain size strengthening proportional factor, m is the grain size strengthening exponential factor, and b is the fatigue strength index.

2. The material life prediction method based on microstructure analysis according to claim 1, characterized in that: The convolution kernels of the first convolution layer, the second convolution layer and the third convolution layer are: ， The first convolutional layer uses the following image convolution operation: ， Among them, fig out (x,y) is the pixel value of the x row and y column of the image convolution operation output image, c(i,j) is the value of the i row and j column of the convolution kernel of the image convolution operation, fig in (x+i, y+j) is the pixel value of the x+i row and y+j column of the image convolution operation input image, b ias is the bias coefficient of image convolution operation; The second convolutional layer and the third convolutional layer use the following image hole convolution operation: ， in, The pixel value of the x row and y column of the output image of the hole convolution operation of the image, is the i-row and j-column value of the convolution kernel of the image hole convolution operation, is the pixel value of the x+i∙κ row and y+j∙κ column of the image atrous convolution operation input graph, where κ is the dilation rate. is the bias coefficient of the image dilated convolution operation.

3. The material life prediction method based on microstructure analysis according to claim 1 is characterized in that: The fully connected operator network includes at least two fully connected layers, and uses a gradient descent algorithm and a mean square error function as a cost function to perform parameter training.

4. The material life prediction method based on microstructure analysis according to claim 1, characterized in that: The fitting method of the stress fatigue life fitting function of the microstructure optimization comprises the following steps: A1. Determine the stress amplitude according to the working conditions of the material to be tested; A2. Obtain fatigue strength coefficient and fatigue strength index according to the type of material to be tested; A3. Set the initial values ​​of fatigue strength coefficient correction factor, grain size strengthening proportional factor and grain size strengthening exponential factor, use GA algorithm, based on material samples with known life, take the minimum life calculation error as the optimization goal, and optimize the parameters of fatigue strength coefficient correction factor, grain size strengthening proportional factor and grain size strengthening exponential factor.

Images