CN-122021188-A - Inversion method of mechanical parameters of material based on composite indentation data and neural network

CN122021188ACN 122021188 ACN122021188 ACN 122021188ACN-122021188-A

Abstract

The invention relates to the field of intersection of material mechanical property testing technology and machine learning, in particular to a material mechanical parameter inversion method based on composite indentation data and a neural network. The method comprises the steps of carrying out finite element simulation based on a material constitutive model, establishing a mapping database, converting the residual indentation morphology and the maximum load in the mapping database into standardized characteristic quantities irrelevant to the size of a pressure head, updating the mapping database, constructing a reverse artificial neural network, training the reverse artificial neural network by utilizing the updated mapping database to obtain a reverse mapping model, obtaining a plurality of groups of residual indentation morphology and the maximum load in an indentation experiment of a material to be tested, converting the residual indentation morphology and the maximum load into standardized characteristic quantities irrelevant to the size of the pressure head, obtaining a first standardized measured value, inputting the first standardized measured value into the reverse mapping model, and outputting a material mechanical parameter inversion result of the material to be tested. In this way, the indentation response information can be fully fused, thereby realizing direct mapping from data to parameters.

Inventors

JIAO QUAN
LIU HAO
Yu Daohan
LI XIUYAN
LU KE

Assignees

辽宁材料实验室

Dates

Publication Date: 20260512
Application Date: 20260409

Claims (9)

1. The inversion method of the material mechanical parameter based on the composite indentation data and the neural network is characterized by comprising the following steps: performing finite element simulation based on a material constitutive model, and establishing a mapping database, wherein the mapping database comprises residual indentation morphology, maximum load, material mechanical parameters and mapping relations; Converting the residual indentation morphology and the maximum load in the mapping database into standardized characteristic quantities irrelevant to the size of the pressure head, and updating the mapping database; constructing a reverse artificial neural network, and training the reverse artificial neural network by using the updated mapping database to obtain a reverse mapping model; Obtaining a plurality of groups of residual indentation morphology and maximum load in an indentation experiment of a material to be measured, and converting the residual indentation morphology and the maximum load into standardized characteristic quantities irrelevant to the size of a pressure head to obtain a first standardized measured value; and inputting the first standardized measured value into a reverse mapping model, and outputting a material mechanical parameter inversion result of the material to be measured.
2. The method of claim 1, wherein converting the residual indentation morphology and maximum load in the mapping database into normalized feature quantities independent of indenter size, updating the mapping database comprises: Geometric standardization processing is carried out on the residual indentation morphology and the maximum load, and standardized residual indentation morphology and standardized maximum load are obtained; Replacing the residual indentation morphology and the maximum load in the mapping database with a normalized residual indentation morphology and a normalized maximum load.
3. The method of claim 2, wherein the inverse artificial neural network comprises an input layer, an output layer, and at least one hidden layer, wherein the hidden layer uses a ReLU activation function and the output layer uses a linear activation function; The input of the input layer is an updated mapping database, and the output of the output layer is a material mechanical parameter.
4. The method of claim 1, wherein training the inverse artificial neural network using the updated mapping database comprises: and dividing the updated data of the mapping database into a training set and a verification set, performing iterative optimization on the reverse artificial neural network by using the training set, and stopping iteration and completing training when the loss in the verification set is not reduced.
5. The method as recited in claim 1, further comprising: Carrying out finite element simulation on the inversion result of the material mechanical parameters of the material to be tested to obtain the predicted value of the residual indentation morphology and the maximum load of the material to be tested, and carrying out geometric standardization processing on the predicted value to obtain a first standardized predicted value; And constructing a composite error function related to the residual indentation morphology and the maximum load, calculating the error of the first standardized predicted value and the first standardized measured value by using the composite error function, and taking the inversion result of the mechanical parameters of the material to be measured as a final inversion result if the error is smaller than a preset error threshold value.
6. The method according to claim 2, wherein the geometric normalization of the residual indentation morphology comprises: Wherein, the Depth data of the residual indentation morphology; is the radius of the spherical pressure head; Depth data for the normalized residual indentation morphology.
7. The method of claim 2, wherein geometrically normalizing the maximum load comprises: Wherein, the Is the standardized maximum load; is the maximum load; Is the radius of the ball press head.
8. The method according to claim 1, wherein obtaining a plurality of sets of residual indentation morphologies in an indentation experiment of the material to be measured comprises: and obtaining surface elevation data of the three-dimensional morphology, uniformly extracting a plurality of radial contour lines along the circumferential direction by taking the center of the residual indentation as an origin, fusing the radial contour lines into a two-dimensional axisymmetric contour by a radial averaging method, and inhibiting measurement noise by utilizing a digital filtering technology to obtain the two-dimensional residual indentation morphology.
9. The utility model provides a material mechanics parameter inversion device based on compound indentation data and neural network which characterized in that includes: the simulation module is used for carrying out finite element simulation based on the material constitutive model and establishing a mapping database, wherein the mapping database comprises residual indentation morphology, maximum load, material mechanical parameters and mapping relation; The first conversion module is used for converting the residual indentation morphology and the maximum load in the mapping database into standardized characteristic quantities irrelevant to the size of the pressure head and updating the mapping database; The network construction module is used for constructing a reverse artificial neural network, and training the reverse artificial neural network by utilizing the updated mapping database to obtain a reverse mapping model; the second conversion module is used for obtaining a plurality of groups of residual indentation morphology and maximum load in an indentation experiment of the material to be tested, converting the residual indentation morphology and the maximum load into standardized characteristic quantities irrelevant to the size of the pressure head, and obtaining a first standardized measured value; And the output module is used for inputting the first standardized measured value into the reverse mapping model and outputting a material mechanical parameter inversion result of the material to be measured.

Description

Inversion method of mechanical parameters of material based on composite indentation data and neural network Technical Field The present invention relates generally to the field of intersection of material mechanical property testing techniques and machine learning, and more particularly to a method for inverting material mechanical parameters based on composite indentation data and neural networks. Background The instrumented indentation technology has become one of the important methods for the characterization of the mechanical properties of materials because of the characteristics of small sample requirement, convenient operation and the like. Conventional indentation inversion analysis relies primarily on load-displacement curves, but this approach suffers from inherent non-uniqueness in that different combinations of material parameters may produce similar load-displacement responses. To overcome this limitation, some studies have introduced residual indentation morphology information. However, the existing method for introducing the residual indentation morphology information often depends on manually extracting morphology features, and has the limitations of strong subjectivity and insufficient information utilization. In the inversion strategy, the existing mainstream method mostly adopts an optimization route based on iterative finite element simulation, and continuously adjusts material parameters and runs the finite element simulation until the simulation response is matched with experimental data. In recent years, a machine learning model is introduced to replace finite element simulation, a forward proxy model is constructed, and inversion is performed by combining an optimization algorithm, so that the problems of optimization cycle, initial value sensitivity and the like are not solved although the simulation times are reduced. Most of the presently disclosed methods train neural networks by using indentation morphology data or load displacement data alone, which can lead to loss of key mechanical information of the material, thereby affecting inversion accuracy of mechanical property parameters of the material. Disclosure of Invention According to the invention, a material mechanics parameter inversion scheme based on composite indentation data and a neural network is provided. The scheme can fully fuse indentation response information, and further realize direct mapping from data to parameters. In a first aspect of the invention, a method for inverting mechanical parameters of a material based on composite indentation data and a neural network is provided. The method comprises the following steps: performing finite element simulation based on a material constitutive model, and establishing a mapping database, wherein the mapping database comprises residual indentation morphology, maximum load, material mechanical parameters and mapping relations; Converting the residual indentation morphology and the maximum load in the mapping database into standardized characteristic quantities irrelevant to the size of the pressure head, and updating the mapping database; constructing a reverse artificial neural network, and training the reverse artificial neural network by using the updated mapping database to obtain a reverse mapping model; Obtaining a plurality of groups of residual indentation morphology and maximum load in an indentation experiment of a material to be measured, and converting the residual indentation morphology and the maximum load into standardized characteristic quantities irrelevant to the size of a pressure head to obtain a first standardized measured value; and inputting the first standardized measured value into a reverse mapping model, and outputting a material mechanical parameter inversion result of the material to be measured. Further, converting the residual indentation morphology and the maximum load in the mapping database into standardized feature quantities independent of the size of the indenter, and updating the mapping database, wherein the method comprises the following steps: Geometric standardization processing is carried out on the residual indentation morphology and the maximum load, and standardized residual indentation morphology and standardized maximum load are obtained; Replacing the residual indentation morphology and the maximum load in the mapping database with a normalized residual indentation morphology and a normalized maximum load. Further, the geometric normalization processing is carried out on the residual indentation morphology, which comprises the following steps: Wherein, the Depth data of the residual indentation morphology; is the radius of the spherical pressure head; Depth data for the normalized residual indentation morphology. Further, the geometric normalization processing is performed on the maximum load, including: Wherein, the Is the standardized maximum load; is the maximum load; Is the radius of the ball press head. Further,