CN-122021325-A - Electromagnetic ultrasonic simulation method and system based on physical information fusion neural network

CN122021325ACN 122021325 ACN122021325 ACN 122021325ACN-122021325-A

Abstract

The application discloses an electromagnetic ultrasonic simulation method and system based on a physical information fusion neural network, and relates to the technical field of electromagnetic ultrasonic simulation; the solid mechanics simplified model comprises a test piece to be tested and a semi-ellipse arranged at the skin depth of the upper surface of the test piece, the semi-ellipse replaces an eddy vibration source in an electromagnetic ultrasonic transducer, multiple groups of electromagnetic ultrasonic transducer parameters are obtained by random sampling in the parameter range of each electromagnetic ultrasonic transducer, simplified model parameters of the solid mechanics simplified model corresponding to each group of electromagnetic ultrasonic transducer parameters are obtained to form a data set, a deep neural network is trained based on the data set, input and output of the deep neural network meet the preset physical characteristic relation, and target electromagnetic ultrasonic transducer parameters are input into the trained deep neural network to obtain a predicted value of the simplified model parameters. The application can shorten the simulation time consumption and improve the output efficiency of simplified model parameters.

Inventors

SUN HONGYU
HUANG XINCHENG
FENG QIBO
LI JIAKUN
PENG LISHA
HUANG SONGLING

Assignees

北京交通大学
清华大学

Dates

Publication Date: 20260512
Application Date: 20260204

Claims (10)

1. The electromagnetic ultrasonic simulation method based on the physical information fusion neural network is characterized by comprising the following steps of: The method comprises the steps of obtaining an electromagnetic ultrasonic transducer parameter range of a test piece to be tested and a solid mechanics simplified model of an electromagnetic ultrasonic transducer model, wherein the electromagnetic ultrasonic transducer model is a model for carrying out electromagnetic ultrasonic detection on the test piece to be tested by adopting the electromagnetic ultrasonic transducer model, and the solid mechanics simplified model comprises the test piece to be tested and a semi-ellipse arranged at the skin depth of the upper surface of the test piece to be tested, wherein the semi-ellipse is used for replacing an eddy current vibration source in the electromagnetic ultrasonic transducer; Randomly sampling in the parameter range of each electromagnetic ultrasonic transducer to obtain a plurality of groups of electromagnetic ultrasonic transducer parameters under different electromagnetic ultrasonic transducer parameter combinations; Obtaining simplified model parameters of the solid mechanics simplified model corresponding to each group of electromagnetic ultrasonic transducer parameters, wherein each group of electromagnetic ultrasonic transducer parameters and the corresponding simplified model parameters are taken as one sample, and a plurality of samples form a data set; Based on the data set, taking a group of electromagnetic ultrasonic transducer parameters in each sample as input, and simplifying model parameters as output to train the deep neural network to obtain a trained deep neural network, wherein the input and the output of the deep neural network meet the preset physical characteristic relation; Inputting parameters of a target electromagnetic ultrasonic transducer into a trained deep neural network, and outputting a simplified model parameter predicted value; And determining an ultrasonic field distribution result of the test piece to be tested by adopting the parameter predictive value of the simplified model.
2. The electromagnetic ultrasonic simulation method based on the physical information fusion neural network according to claim 1, wherein the electromagnetic ultrasonic transducer parameters comprise a coil radius, a vertical distance from the bottom of a coil to the upper surface of a test piece to be tested, a time-varying current peak value, a permanent magnet residual magnetic flux density and an excitation frequency; the simplified model parameters include a semi-elliptical semi-major axis, a semi-elliptical semi-minor axis, and a time-varying force peak applied to a semi-elliptical vibration source.
3. The electromagnetic ultrasonic simulation method based on the physical information fusion neural network according to claim 2, wherein the preset physical characteristic relationship comprises two types, namely a first type relationship and a second type relationship; the first type of relationship is that a is in direct proportion to R, and a is in direct proportion to R A is a semi-elliptic semi-major axis, R is a coil radius, and H is a vertical distance from the bottom of the coil to the upper surface of a test piece to be tested; The second type of relation is F and And F is a time-varying force peak applied to the semi-elliptical vibration source, I is a time-varying current peak, and B is the residual magnetic flux density of the permanent magnet.
4. The electromagnetic ultrasonic simulation method based on the physical information fusion neural network according to claim 2, wherein the semi-elliptical semi-minor axis in the simplified model parameter predicted value is directly taken as a skin depth value of a test piece material to be tested, and the semi-elliptical semi-major axis and a time-varying force peak applied to a semi-elliptical vibration source are output by the trained deep neural network.
5. The electromagnetic ultrasonic simulation method based on the physical information fusion neural network, which is disclosed in claim 1, is characterized in that the geometric structure of the electromagnetic ultrasonic transducer model comprises a permanent magnet, a coil and a test piece to be tested, which are sequentially arranged from top to bottom, wherein the permanent magnet is used for providing a static bias magnetic field for the interior of the test piece to be tested, and the coil is used for generating eddy currents in the skin depth of the test piece to be tested after being electrified.
6. The electromagnetic ultrasonic simulation method based on the physical information fusion neural network according to claim 1, wherein obtaining the simplified model parameters of the simplified model of solid mechanics corresponding to each group of electromagnetic ultrasonic transducer parameters specifically comprises: And acquiring simplified model parameters of the solid mechanics simplified model corresponding to each group of electromagnetic ultrasonic transducer parameters by adopting a peak matching algorithm.
7. The electromagnetic ultrasonic simulation method based on the physical information fusion neural network according to claim 6, wherein a peak matching algorithm is adopted to obtain simplified model parameters of the simplified model of solid mechanics corresponding to each group of electromagnetic ultrasonic transducer parameters, and the method specifically comprises the following steps: For any group of electromagnetic ultrasonic transducer parameters, electromagnetic ultrasonic transducer simulation data of the group of electromagnetic ultrasonic transducer parameters are obtained, wherein the electromagnetic ultrasonic transducer simulation data are displacement field X components of particles at a plurality of specific positions on the test piece to be tested; And determining simplified model parameters of the simplified model of the solid mechanics corresponding to the group of electromagnetic ultrasonic transducer parameters by adopting a peak matching algorithm based on the X components of the displacement fields.
8. The electromagnetic ultrasonic simulation method based on the physical information fusion neural network according to claim 7, wherein the specific positions are a plurality of particles uniformly distributed on a vertical section of the test piece to be tested; The point taking modes of the specific positions are as follows: Taking one particle every h/4 vertically downwards from the central point of the upper surface of the test piece to be tested until 5 particles are firstly taken from the lower surface of the test piece to be tested; for each first-taken particle, taking two particles on each side every w/6 at the left and right sides of the particle, and taking 25 particles in total, wherein h is the height of the test piece to be tested, and w is the width of the test piece to be tested.
9. The electromagnetic ultrasonic simulation method based on the physical information fusion neural network according to claim 1, wherein the deep neural network adopts a multi-output full-connection layer architecture, the number range of neurons of each layer in the deep neural network is 64-256, and early-stop and dynamic learning rate adjustment strategies are adopted in the training process of the deep neural network.
10. An electromagnetic ultrasonic simulation system based on a physical information fusion neural network, which is characterized in that the electromagnetic ultrasonic simulation system based on the physical information fusion neural network applies the electromagnetic ultrasonic simulation method based on the physical information fusion neural network according to any one of claims 1 to 9, and the electromagnetic ultrasonic simulation system based on the physical information fusion neural network comprises: The electromagnetic ultrasonic transducer model is a model for carrying out electromagnetic ultrasonic detection on the test piece to be tested by adopting the electromagnetic ultrasonic transducer model, and the solid mechanical simplified model comprises the test piece to be tested and a semi-ellipse arranged at the skin depth of the upper surface of the test piece to be tested, wherein the semi-ellipse is used for replacing an eddy current vibration source in the electromagnetic ultrasonic transducer; the multi-group electromagnetic ultrasonic transducer parameter determining module is used for randomly sampling in the parameter range of each electromagnetic ultrasonic transducer to obtain multi-group electromagnetic ultrasonic transducer parameters under different electromagnetic ultrasonic transducer parameter combinations; the simplified model parameter determining module is used for obtaining simplified model parameters of the solid mechanics simplified model corresponding to each group of electromagnetic ultrasonic transducer parameters, wherein each group of electromagnetic ultrasonic transducer parameters and the corresponding simplified model parameters are taken as one sample, and a plurality of samples form a data set; The deep neural network training module is used for training the deep neural network based on the data set by taking a group of electromagnetic ultrasonic transducer parameters in each sample as input and simplifying model parameters as output, so as to obtain a trained deep neural network, wherein the input and the output of the deep neural network meet the preset physical characteristic relation; The simplified model parameter prediction module is used for inputting parameters of the target electromagnetic ultrasonic transducer into the trained deep neural network and outputting a simplified model parameter predicted value; and the simulation module is used for determining an ultrasonic field distribution result of the test piece to be tested by adopting the parameter predicted value of the simplified model.

Description

Electromagnetic ultrasonic simulation method and system based on physical information fusion neural network Technical Field The application relates to the technical field of electromagnetic ultrasonic simulation, in particular to an electromagnetic ultrasonic simulation method and system based on a physical information fusion neural network. Background Electromagnetic ultrasonic transducers (Electromagnetic Acoustic Transducer, EMAT) are an important technology in the field of nondestructive testing, and generate Lorentz force or magnetostriction effect to excite ultrasonic waves in conductors through electromagnetic induction, so that the ultrasonic transducer has the outstanding advantages of non-contact, no need of coupling agent, suitability for high-temperature detection and the like. The numerical simulation is a core means for analyzing an interaction mechanism of ultrasonic waves and a test piece and optimizing geometric parameters (such as coil size and lift-off distance) and electrical parameters (such as current, frequency and magnetic field strength) of the EMAT, and is important for shortening the research and development period and reducing the test cost. At present, a finite element method is mainly adopted to directly carry out numerical simulation calculation on an electromagnetic model of a target parameter EMAT. However, since the EMAT simulation involves large-scale calculation and strong multi-physical field coupling, a fine grid needs to be constructed, and a smaller time step is used for time domain solution, so that a full-scale numerical simulation is performed on a test piece, a large amount of calculation resources are required, and the calculation of a complex model can take several hours to days. The efficiency defect severely restricts the application of the EMAT in scenes such as large-scale parameter optimization and the like. In recent years, a plurality of scholars aim to apply the deep learning technology to the EMAT nondestructive testing, and a new thought and method are provided for improving the detection efficiency. However, few researches on EMAT simulation optimization and acceleration are performed, one proposed method is to firstly generate sample data through a finite element model, then train various neural networks to learn the input-output relationship of a simulation model, and finally call the trained neural network with extremely high calculation speed in an optimization algorithm to replace the original simulation. The goal of this approach is to quickly optimize the results of the pre-set ultrasonic signal amplitude ratio and signal strength, rather than to accelerate the physical field solving process. Therefore, the metamodel method based on pure data driving has obvious effect in specific optimization tasks, but cannot further analyze the distribution characteristics of the whole ultrasonic field in the test piece, so that the application of the metamodel method in wider scenes is limited. Therefore, it is needed to develop an electromagnetic ultrasonic numerical simulation intelligent acceleration algorithm to quickly and accurately obtain the calculation result of the EMAT finite element simulation, and improve the efficiency of optimizing the EMAT parameters. Disclosure of Invention The application aims to provide an electromagnetic ultrasonic simulation method and system based on a physical information fusion neural network, which improve the output efficiency of simplified model parameters and shorten the simulation time. In order to achieve the above object, the present application provides the following solutions: in a first aspect, the application provides an electromagnetic ultrasonic simulation method based on a physical information fusion neural network, which comprises the following steps: The method comprises the steps of obtaining an electromagnetic ultrasonic transducer parameter range of a test piece to be tested and a solid mechanics simplified model of an electromagnetic ultrasonic transducer model, wherein the electromagnetic ultrasonic transducer model is a model for carrying out electromagnetic ultrasonic detection on the test piece to be tested by adopting the electromagnetic ultrasonic transducer model, and the solid mechanics simplified model comprises the test piece to be tested and a semi-ellipse arranged at the skin depth of the upper surface of the test piece to be tested, wherein the semi-ellipse is used for replacing an eddy current vibration source in the electromagnetic ultrasonic transducer; Randomly sampling in the parameter range of each electromagnetic ultrasonic transducer to obtain a plurality of groups of electromagnetic ultrasonic transducer parameters under different electromagnetic ultrasonic transducer parameter combinations; Obtaining simplified model parameters of the solid mechanics simplified model corresponding to each group of electromagnetic ultrasonic transducer parameters, wherein each group of electr