Ferromagnetic metal material average grain size characterization method based on magnetic domain motion characteristics

Abstract

The invention discloses a ferromagnetic metal material average grain size characterization method based on magnetic domain motion characteristics, which comprises the following steps: firstly, preparing a sample for EBSD method material characterization and multi-magnetic detection equipment detection; secondly, carrying out quantitative characterization on the average grain size of the sample based on EBSD method material characterization; then, performing electromagnetic nondestructive testing on the ferromagnetic metal material based on a multi-magnetic testing device to obtain electromagnetic signals representing reversible and irreversible motion of a magnetic domain, and defining and extracting corresponding electromagnetic characteristic parameters of the ferromagnetic metal material at the same time; and finally, by analyzing the relationship between the electromagnetic characteristic parameters and the characteristic microstructure, screening out the electromagnetic characteristic parameters as input, and establishing and adjusting a deep neural network representation model of the characteristic microstructure based on the electromagnetic characteristic parameters of the magnetic domain reversible and irreversible motion. According to the invention, nondestructive testing of the average grain size of the ferromagnetic metal material is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the electromagnetic nondestructive testing technology of ferromagnetic materials, and more particularly to a method for representing the average grain size of ferromagnetic metal materials based on the motion characteristics of magnetic domains. BACKGROUND

[0002] At present, the microstructure of products obtained by domestic and foreign steel plants mainly relies on optical metallography, EBSD (Electron Back Scatter Diffraction, EBSD) and other means, and then supplemented by statistical analysis to evaluate the average grain size of the products. This experimental method cannot realize online detection, and has long detection cycle, is not intelligent enough, and wastes a lot of resources. Under the guidance of the concepts of green environmental protection and the like, the technical means for rapidly and nondestructively detecting the microstructure are becoming more and more important.

[0003] In this field, the motion characteristics of magnetic domains have two kinds of reversible motion of magnetic domains and irreversible motion of magnetic domains, and the incremental permeability technology based on the reversible motion of magnetic domains and the magnetic Barkhausen technology based on the irreversible motion of magnetic domains are feasible ways to realize the detection of the microstructure of materials. Under the excitation of an external magnetic field, different microstructures have different effects on the reversible and irreversible motion of magnetic domains, and also affect the structure of magnetic domains, so that the electromagnetic nondestructive detection signals representing the reversible and irreversible motion of magnetic domains are extracted, and the corresponding features are analyzed and extracted, which can reflect the microstructure of the measured material.

[0004] Most of the current technologies are focused on the detection method of the mechanical properties of ferromagnetic materials, and the method for the microstructure of ferromagnetic materials is also limited to the destructive test representation technology, so there is no nondestructive detection technology for the microstructure of ferromagnetic materials. SUMMARY

[0005] In view of the defects in the prior art, the purpose of the present application is to provide a method for representing the average grain size of ferromagnetic metal materials based on the motion characteristics of magnetic domains, so as to realize the nondestructive detection of the average grain size of ferromagnetic metal materials.

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0007] A method for representing the average grain size of ferromagnetic metal materials based on the motion characteristics of magnetic domains comprises the following steps:

[0008] First, a sample for EBSD method material representation and multi-magnetic detection equipment detection is prepared;

[0009] Secondly, the average grain size of the sample is quantitatively represented based on the EBSD method material representation;

[0010] Then, the ferromagnetic metal material is subjected to electromagnetic nondestructive testing based on the multi-magnetic detection device to obtain electromagnetic signals representing reversible and irreversible motion of magnetic domains, and to define and extract corresponding electromagnetic characteristic parameters of the ferromagnetic metal material;

[0011] Finally, by analyzing the relationship between the electromagnetic characteristic parameters and the characteristic microstructure, the electromagnetic characteristic parameters serving as inputs are screened, a characteristic microstructure deep neural network representation model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains is established, and the parameters of the characteristic microstructure feedforward deep neural network representation model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains are adjusted to obtain an optimal characteristic microstructure feedforward deep neural network representation model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains.

[0012] Preferably, the average grain size rapid representation method of the ferromagnetic metal material specifically comprises the following steps:

[0013] S1, using a wire cutting method, the ferromagnetic metal material is cut into a sample with a size of 10mm*10mm for EBSD method material representation, and a sample with a size of 150mm*600mm for multi-magnetic detection device detection;

[0014] S2, using the EBSD method to quantitatively represent the microstructure- average grain size of the ferromagnetic metal material;

[0015] S3, repeating steps S1 and S2 to obtain quantitative representation results of different ferromagnetic metal materials;

[0016] S4, on the basis of step S1, using a multi-magnetic detection device to perform electromagnetic nondestructive testing on different ferromagnetic metal materials, and extracting electromagnetic characteristic parameters of reversible and irreversible motion of magnetic domains of the ferromagnetic metal material;

[0017] S5, according to the quantitative representation results in step S3, the characteristic microstructure of the ferromagnetic metal material is obtained, and the electromagnetic characteristic parameters of reversible and irreversible motion of magnetic domains of the ferromagnetic metal material obtained in step S4 are correlated;

[0018] S6, establishing a characteristic microstructure feedforward deep neural network representation model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains;

[0019] S7, using confidence to evaluate the accuracy of the characteristic microstructure feedforward deep neural network representation model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains;

[0020] S8, repeat the training of the characteristic microstructure structure feedforward deep neural network characterization model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains ten times according to the processes of step S6 and step S7, and take the average of the ten training results as the final detection value.

[0021] Preferably, the multi-magnetic detection device integrates the magnetic Barkhausen noise method, the incremental permeability method, the tangential magnetic field harmonic analysis method and the multi-frequency eddy current detection method.

[0022] Preferably, the multi-magnetic detection device adopts the magnetic Barkhausen noise method and the incremental permeability method, and analyzes 12 electromagnetic characteristic parameters in total.

[0023] Preferably, the 12 electromagnetic characteristic parameters specifically include:

[0024] MMAX, MMEAN, HCM, DH25M, DH50M and DH75M in MBN;

[0025] UMAX, UMEAN, HCU, DH25U, DH50U and DH75U in MIP.

[0026] Preferably, in step S6, the learning algorithm process of the characteristic microstructure structure feedforward deep neural network characterization model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains is as follows:

[0027] S61, give an input vector and an output vector;

[0028] S62, initialize the weights and thresholds;

[0029] S63, calculate the output of each node of the hidden layer and the output layer;

[0030] S64, calculate the error E between the expected output and the actual output;

[0031] S65, determine whether the error E meets the requirements, if yes, end, if not, go to step S66;

[0032] S66, calculate the error of each unit of the hidden layer and the output layer;

[0033] S67, calculate the error gradient;

[0034] S68, after updating the weights and thresholds of the hidden layer and the output layer, return to step S63.

[0035] Preferably, in the training process of the characteristic microstructure structure feedforward deep neural network representation model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains, the logsig, tansig, purelin activation functions and trainlm training functions are used, the error threshold is set to 0.000001, the initial learning rate is set to 0.01, and the iteration number is set to 1000.

[0036] Preferably, the confidence in the step S7 is defined as:

[0037]

[0038] In the formula, N 10 is the number of errors ρ of the model prediction value Y ci in the verification set less than 10% of the true value Y i , and N is the total number of the verification set, that is:

[0039]

[0040] The ferromagnetic metal material average grain size representation method provided by the application has the following advantages:

[0041] (1) The ferromagnetic metal material characteristic microstructure structure-average grain size is nondestructively detected, the detection accuracy is greater than 95%, the error is less than 10%, and the confidence rate is higher than 85%;

[0042] (2) A digital and rapid evaluation method is provided for the microstructure structure representation of the ferromagnetic metal material, so that the detection efficiency is improved and the detection cost is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 is a flowchart of the ferromagnetic metal material average grain size representation method of the application;

[0044] Figure 2 is a schematic diagram of the microstructure structure representation result of the step S2 in the rapid ferromagnetic metal material average grain size representation method of the application;

[0045] Figure 3 is a schematic diagram of the relationship between the average grain size and the yield strength of the step S3 in the rapid ferromagnetic metal material average grain size representation method of the application;

[0046] Figure 4 is a schematic diagram of the probe composition of the multi-magnetic detection device of the step S4 in the rapid ferromagnetic metal material average grain size representation method of the application;

[0047] Figure 5is the MBN (MIP) butterfly curve schematic diagram of step S4 in the average grain size rapid characterization method of the ferromagnetic metal material of the application;

[0048] Figure 6 is the structural schematic diagram of the characteristic microstructure feedforward deep neural network characterization model of the electromagnetic characteristic parameter based on the reversible and irreversible motion of magnetic domains in the average grain size rapid characterization method of the ferromagnetic metal material of the application;

[0049] Figure 7 is the learning algorithm flowchart schematic diagram of the characteristic microstructure feedforward deep neural network characterization model of the electromagnetic characteristic parameter based on the reversible and irreversible motion of magnetic domains in the average grain size rapid characterization method of the ferromagnetic metal material of the application.

[0050] Figure 4 In the figure, 1 is a shell, 2 is an electronic board (pre-amplifier), 3 is a magnetic yoke, 4 is an electromagnetic coil, 5 is a connecting cable, 6 is a Hall sensor, 7 is a transmitter coil, 8 is a receiver coil, and 9 is an electromagnetic nondestructive testing sample. DETAILED DESCRIPTION

[0051] In order to better understand the above technical solutions of the application, the technical solutions of the application will be further described below in combination with the drawings and examples.

[0052] The average grain size characterization method of the ferromagnetic metal material based on the magnetic domain motion characteristics provided by the application quantitatively characterizes the average grain size of the ferromagnetic metal material based on the EBSD material characterization method. At the same time, the electromagnetic nondestructive testing of the ferromagnetic metal material is performed based on the multi-magnetic detection equipment, and the electromagnetic characteristic parameter representing the magnetic characteristics of the material is obtained. The electromagnetic characteristic parameter of the material is taken as the input, and the characteristic microstructure (average grain size) of the material is taken as the output, and a characteristic microstructure (average grain size) feedforward deep neural network (BP) characterization model based on the electromagnetic characteristic parameter of the reversible and irreversible motion of magnetic domains is established, so as to realize the nondestructive testing of the average grain size of the material.

[0053] Firstly, the ferromagnetic metal material is cut into small-size samples (10mm*10mm) (for material characterization by the EBSD method) and large-size samples (150mm*600mm) (for detection by the multi-magnetic detection equipment) in a wire cutting manner.

[0054] Secondly, the microstructure of the ferromagnetic metal material is quantitatively characterized based on the EBSD method, and the characteristic microstructure (average grain size) of the ferromagnetic metal material is obtained based on the correlation analysis of the microstructure of the ferromagnetic metal material and the mechanical properties thereof.

[0055] Then, the ferromagnetic metal material is subjected to electromagnetic nondestructive testing based on the multi-magnetic detection device to obtain electromagnetic signals representing reversible and irreversible motion of magnetic domains, and meanwhile, electromagnetic characteristic parameters of the ferromagnetic metal material are defined and extracted.

[0056] Finally, by analyzing the relationship between the electromagnetic characteristic parameters and the characteristic microstructure, the electromagnetic characteristic parameters serving as inputs are screened out, a characteristic microstructure deep neural network representation model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains is established, and parameters of the characteristic microstructure feedforward deep neural network representation model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains are adjusted to obtain an optimal characteristic microstructure feedforward deep neural network representation model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains, and finally, nondestructive testing of the characteristic microstructure- average grain size of the ferromagnetic metal material is realized.

[0057] In combination Figure 1 As shown in the figure, the method for rapidly representing the average grain size of the ferromagnetic metal material specifically comprises the following steps:

[0058] S1, the ferromagnetic metal material is cut into small-size samples (for EBSD method material representation) of 10mm*10mm and large-size samples (samples for detection by the multi-magnetic detection device) of 150mm*600mm in a wire cutting manner.

[0059] S2, the microstructure of the ferromagnetic metal material is quantitatively represented by the EBSD method, and the quantitative representation result is as shown in the figure. Figure 2

[0060] S3, the steps S1 and S2 are repeated to obtain the quantitative representation results of different ferromagnetic metal materials.

[0061] S4, on the basis of the step S1, the ferromagnetic metal material is subjected to electromagnetic nondestructive testing by the multi-magnetic detection device, and electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains of the ferromagnetic metal material are extracted. The multi-magnetic detection device integrates multiple electromagnetic detection methods, and different electromagnetic principles are excited by applying different excitation magnetic fields from outside, such as the magnetic Barkhausen noise method, the incremental permeability method, the tangential magnetic field harmonic analysis method and the multi-frequency eddy current detection method. During detection, different alternating magnetic field excitations are applied to the ferromagnetic metal material, from zero magnetic field excitation to positive saturation and then back to zero magnetic field excitation, and then increased to negative saturation and then back to zero magnetic field excitation. During the entire repeated magnetization cycle, the magnetization processes of different microstructures are different. The cumulative effect of the motion of magnetic domains leads to changes in the permeability of the ferromagnetic metal material, the degree of magnetization deepens, and finally leads to changes in the entire magnetization curve. The signals obtained in this process can reflect the magnetic properties of different ferromagnetic metal materials and the microstructure characteristics of the ferromagnetic metal material. ​

[0062] The probe structure of the multi-magnetic detection device is shown in Figure 4 Four micro-magnetic non-destructive testing techniques are inherited in the probe, and 41 micro-magnetic characteristics are selected to represent the electromagnetic characteristic parameters of ferromagnetic metal materials. However, the stability of the electromagnetic characteristic parameters extracted by the tangential magnetic field strength detection technique is poor in the micro-magnetic detection experiment, so the tangential magnetic field strength detection technique is excluded. In the experimental analysis, the magnetic Barkhausen noise method (MBN) and the incremental permeability method (MIP) are used, and there are a total of 12 micro-magnetic characteristics.

[0063] In the MBN technique, a high-amplitude, low-frequency sinusoidal current is sent to the yoke coil 4 wound around the U-shaped yoke 3, and the applied magnetization amplitude is sufficient to excite the detected ferromagnetic metal material to reach the saturation level in order to ensure that the signal of irreversible domain motion is detected by the receiver coil 8.

[0064] The detected MBN signal is processed by a combination of a band-pass filter and a low-pass / high-pass filter, and is amplified, including post-amplification and signal smoothing rectification. The MBN butterfly curve is shown in Figure 5 As the vertical axis, the MBN amplitude time series signal after digital transformation and smoothing processing is taken as the horizontal axis, and the excitation magnetic field strength corresponding to the MBN amplitude signal is taken as the horizontal axis, to obtain the butterfly diagram (the curve shape is similar to the unfolded butterfly wings) about the MBN signal. From this, the maximum amplitude (the maximum value of the MBN signal, MMAX) can be obtained as a test statistic. Correspondingly, the magnetic field strength at MMAX is assigned to the test statistic (the horizontal coordinate value corresponding to the maximum value, HCM). The extension of the profile curve is evaluated at 25%, 50% and 75% of MMAX (referred to as width, the definition of width is "the interval (width) between the two intersection points of the profile curve at the vertical coordinate value of 25%, 50% and 75% of the maximum value", DH25M, DH50M and DH75M). The additional test statistic is MMEAN, which is the average value of the MBN signal amplitude in a certain period (the average value of the MBN signal amplitude in a butterfly diagram period).

[0065] During the reorganization of the magnetic domain, the displacement of the Bloch wall occurs in a discrete jump manner. The Bloch domain wall is affected by different microstructures, and therefore exhibits different motion characteristics. The changes in these microstructures can be reflected in the characteristics of MBN. The MMAX test statistic can be used to quantitatively obtain the finishing conditions, such as depth hardness and surface hardness. When the grain boundary represents the main obstacle to the displacement of the Bloch wall, HCM can be quantitatively related to the grain size. The relationship between the extension of the profile curve (DH25M, DH50M and DH75M) and internal stress or plastic deformation has been observed.

[0066] Unlike the MBN detection method, in the MIP technique, high and low frequency sinusoidal current is a necessary condition to obtain reversible domain motion information. Like the MBN method, the high amplitude low frequency (10-1000 Hz) excitation of the U-shaped yoke 3 generates a hysteresis cycle in the ferromagnetic metal material. At the same time, a low amplitude (milliampere level), high frequency (=10 kHz-1 MHz) sinusoidal current is required to be sent into the transmitter coil 7, just like the MFEC method, to generate a small asymmetric hysteresis loop superimposed on the main hysteresis curve.

[0067] Like the MBN, the maximum amplitude of the MIP (the maximum value of the signal, UMAX) is extracted as an important feature. The magnetic field strength at UMAX (the abscissa value at the maximum, HCU) is also derived as a statistical parameter. In addition, the 25%, 50%, and 75% curve extensions (defined above, DH25U, DH50U, and DH75U) and the average UMEAN of the time period are also used as MIP features. MIP can be used to characterize the material properties near the surface (surface hardening). The shell depth information comes from the amplitude of the UMAX signal received from the core structure, and the hardness information can be obtained from the related forced field intensity HCU. The stress state information is quantitatively described by the curve extension (DH25U, DH50U, and DH75U).

[0068] Through electromagnetic nondestructive testing experiments on ferromagnetic metal material samples by a multi-magnetic detection device, 12 electromagnetic characteristic parameters of MBN and MIP shown above can be obtained. Each sample is detected for 120 seconds, and more than 100 groups of electromagnetic characteristic parameter data are obtained. After cleaning of abnormal values, 100 groups of data are retained for subsequent analysis.

[0069] S5, according to the quantified ferromagnetic metal material characteristic microstructure obtained in step S3, and the electromagnetic characteristic parameters of the reversible and irreversible motion of the magnetic domain of the ferromagnetic metal material obtained in step S4, a correlation relationship is established, taking the MBN and MIP electromagnetic characteristic parameters of the ferromagnetic steel material as the ordinate and the characteristic microstructure of the ferromagnetic steel material as the abscissa.

[0070] It can be found that the electromagnetic characteristics representing the reversible and irreversible motion of the magnetic domain and the characteristic microstructure show a monotonic trend, so the electromagnetic characteristics representing the reversible and irreversible motion of the magnetic domain can be used to represent the characteristic microstructure- average grain size of the ferromagnetic metal material.

[0071] S6, a characteristic microstructure feedforward deep neural network representation model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of the magnetic domain is established.

[0072] A multi-layer perceptron (MLP) is a feed-forward artificial neural network ANN that maps a set of input vectors to a set of output vectors. An MLP can be viewed as a directed graph consisting of multiple layers of nodes, with each layer of nodes being fully connected to the next. Each node, except the input nodes, is a neuron with a non-linear activation function. Supervised learning methods using BP backpropagation are used to train an MLP, overcoming the shortcoming of perceptron that cannot identify linearly inseparable data, and generalizing the perceptron.

[0073] Learning based on backpropagation is a typical feedforward network, and information is processed layer by layer from the input layer to each hidden layer and then to the output layer. The hidden layer realizes a nonlinear mapping of the input space, and the output layer realizes linear classification. The nonlinear mapping method and linear discriminant function can be learned simultaneously.

[0074] Before training an MLP neural network, the mathematical relationship between the data does not need to be explicitly defined. By giving the input and output of the training set, the training algorithm adjusts the weights and biases of the MLP neural network and iteratively updates them to reduce the loss function of the model, thereby completing the training. The structure of the MLP neural network is shown in Figure 6 .

[0075] The above derivation process is for a single training sample and a BP neural network with a single hidden layer. The basic idea is also applicable to a BP neural network with multiple hidden layers. When calculating the adjustment amount of the weight and threshold, the output layer is recursively propagated forward to the first hidden layer through the intermediate hidden layers. The error calculation method for multiple training samples is the cumulative sum of the error of each single sample.

[0076] In summary, the execution process of the BP neural network learning algorithm can be summarized as two main parts: forward propagation of signals and backward propagation of errors. For each training sample, during forward propagation, the input vector is propagated from the input layer to the output layer layer by layer; during backward propagation, the error is propagated from the hidden layer to the input layer in some form. These two processes are repeated until the set termination condition is met. The algorithm flowchart is shown in Figure 7 .

[0077] and the electromagnetic characteristic parameters representing the reversible and irreversible domain motion of the ferromagnetic steel material are used as input, and the average grain size (true value) of the ferromagnetic steel material is used as output. In the training process of the DNN model, the "logsig" activation function and the "trainlm" training function are used, the error threshold is set to 0.000001, the initial learning rate is set to 0.01, and the number of iterations is set to 1000.

[0078] S7, the accuracy of the characteristic microstructure structure feedforward deep neural network representation model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains is evaluated using the confidence. The confidence is defined as:

[0079]

[0080] where N 10 is the number of errors ρ between the predicted value Y ci of the model in the verification set and the true value Y i is less than 10%, and N is the total number of the verification set, i.e.:

[0081]

[0082] S8, according to the processes of steps S6 and S7, the characteristic microstructure structure feedforward deep neural network representation model based on the electromagnetic characteristic parameters of the reversible and irreversible motion of magnetic domains is repeatedly trained ten times, and the average of the ten training results is taken as the final detection value. The evaluation accuracy is that the confidence rate is greater than 85% under the condition that the error ρ is less than 10%.

[0083] Embodiment

[0084] This embodiment takes the pickled steel of a certain steel plant cold rolling production line as an example, and proposes a ferromagnetic metal material average grain size representation method based on the magnetic domain motion characteristics, and the specific process is as shown in Figure 1 The microstructure quantitative representation result is as shown in Figure 2 The relationship between the microstructure structure and the yield strength is as shown in Figure 3 The recrystallization rate of the pickled steel can be determined as the characteristic microstructure structure. The probe structure of the multi-magnetic detection device is as shown in Figure 4 The reversible and irreversible magnetic domain motion signals are as shown in Figure 5 The relationship between the electromagnetic characteristic values of the reversible and irreversible magnetic domains of the ferromagnetic steel material and the characteristic microstructure structure is obtained, and the specific steps include:

[0085] 3 The hidden layer DNN model (each layer node is 9, 10, 21) is established, and the pickled steel MBN electromagnetic characteristic parameters are taken as the input and the pickled steel characteristic microstructure structure is taken as the output. The training set data is 145, and the verification set data is 44. In the training process of the DNN model, the “logsig”, “tansig”, “purelin” activation function and “trainlm” training function are used, the error threshold is set to 0.000001, the initial learning rate is set to 0.01, and the iteration number is set to 1000. According to the above modeling and training process, the training is repeated ten times, and the average of the ten training results is taken as the final detection value. The evaluation accuracy of the average grain size is greater than 95%: the confidence rate is greater than 85% under the condition that the error is less than 10%.

[0086] Those skilled in the art will recognize that the above-described embodiments are merely illustrative of the application and should not be construed as limiting the scope of the application. Variations and modifications to the embodiments disclosed herein can be made based on the description set forth herein, without departing from the scope and spirit of the application as recited in the following claims.

Claims

1. A method for characterizing the average grain size of ferromagnetic metallic materials based on magnetic domain motion characteristics, characterized in that: First, samples were prepared for material characterization using the EBSD method and for detection using a multimagnetic detection device; Secondly, the average grain size of the sample was quantitatively characterized based on the EBSD method for material characterization. Then, electromagnetic non-destructive testing of ferromagnetic metal materials is carried out based on multi-magnetic detection equipment to obtain electromagnetic signals characterizing the reversible and irreversible motion of magnetic domains, and at the same time, the corresponding electromagnetic characteristic parameters of ferromagnetic metal materials are defined and extracted. Finally, by analyzing the relationship between electromagnetic characteristic parameters and characteristic microstructures, electromagnetic characteristic parameters as inputs are selected, and a deep neural network representation model of characteristic microstructures based on electromagnetic characteristic parameters of reversible and irreversible magnetic domain motion is established. The parameters of the feedforward deep neural network representation model of characteristic microstructures based on electromagnetic characteristic parameters of reversible and irreversible magnetic domain motion are adjusted to obtain the optimal feedforward deep neural network representation model of characteristic microstructures based on electromagnetic characteristic parameters of reversible and irreversible magnetic domain motion.

2. The method for characterizing the average grain size of ferromagnetic metallic materials based on magnetic domain motion characteristics according to claim 1, characterized in that, The rapid characterization method for the average grain size of ferromagnetic metallic materials specifically includes the following steps: S1, using wire cutting, cuts ferromagnetic metal materials into samples with a size of 10mm*10mm for material characterization using the EBSD method, and samples with a size of 150mm*600mm for detection by multi-magnetic detection equipment; S2. The EBSD method was used to quantitatively characterize the microstructure and average grain size of ferromagnetic metal materials. S3. Repeat steps S1 and S2 to obtain the quantitative characterization results of different ferromagnetic metal materials. S4. Based on step S1, use a multi-magnetic detection device to perform electromagnetic non-destructive testing on different ferromagnetic metal materials, and extract the electromagnetic characteristic parameters of the reversible and irreversible motion of the magnetic domains of the ferromagnetic metal materials. S5. Based on the quantitative characterization results in step S3, the characteristic microstructure of the ferromagnetic metal material is obtained and a correlation is established with the electromagnetic characteristic parameters of the reversible and irreversible motion of the magnetic domains of the ferromagnetic metal material obtained in step S4. S6. Establish a feedforward deep neural network representation model for the characteristic microstructure based on electromagnetic characteristic parameters of reversible and irreversible magnetic domain motion. S7 uses confidence level to evaluate the accuracy of the feedforward deep neural network representation model of the characteristic microstructure based on the electromagnetic characteristic parameters of reversible and irreversible magnetic domain motion. S8. Following the process of steps S6 and S7, the feedforward deep neural network representation model of the characteristic microstructure based on the electromagnetic characteristic parameters of reversible and irreversible magnetic domain motion is repeatedly trained ten times, and the average of the ten training results is taken as the final detection value.

3. The method for characterizing the average grain size of ferromagnetic metallic materials based on magnetic domain motion characteristics according to claim 2, characterized in that: The multi-magnetic detection device integrates the Barkhausen noise method, incremental permeability method, tangential magnetic field harmonic analysis method, and multi-frequency eddy current detection method.

4. The method for characterizing the average grain size of ferromagnetic metallic materials based on magnetic domain motion characteristics according to claim 3, characterized in that: The multi-magnetic detection device uses the magnetic Barkhausen noise method and incremental permeability method to analyze a total of 12 electromagnetic characteristic parameters.

5. The method for characterizing the average grain size of ferromagnetic metallic materials based on magnetic domain motion characteristics according to claim 4, characterized in that, The 12 electromagnetic characteristic parameters specifically include: MBN includes MMAX, MMEAN, HCM, DH25M, DH50M, and DH75M; The MIP series includes UMAX, UMEAN, HCU, DH25U, DH50U, and DH75U.

6. The method for characterizing the average grain size of ferromagnetic metallic materials based on magnetic domain motion characteristics according to claim 2, characterized in that, In step S6, the learning algorithm flow of the feedforward deep neural network representation model of the characteristic microstructure based on the electromagnetic characteristic parameters of reversible and irreversible magnetic domain motion is as follows: S61, given the input vector and the output vector; S62, Initialize weights and thresholds; S63, calculate the output of each node in the hidden layer and the output layer; S64, calculate the error E between the expected output and the actual output; S65, determine whether the error E meets the requirements. If yes, end; otherwise, proceed to step S66. S66, calculate the error of each unit in the hidden layer and the output layer; S67, Calculate the error gradient; S68, after updating the weights and thresholds of the hidden layer and the output layer, return to step S63.

7. The method for characterizing the average grain size of ferromagnetic metallic materials based on magnetic domain motion characteristics according to claim 6, characterized in that: During the training process of the feedforward deep neural network representation model of the characteristic microstructure based on the electromagnetic characteristic parameters of reversible and irreversible magnetic domain motion, the logsig, tansig, purelin activation functions and trainlm training function are used, the error threshold is set to 0.000001, the initial learning rate is set to 0.01, and the number of iterations is set to 1000.

8. The method for characterizing the average grain size of ferromagnetic metallic materials based on magnetic domain motion characteristics according to claim 6, characterized in that, In step S7, the confidence level is defined as: In the formula, N 10 To verify the predicted value Y of the ensemble model ci Compared with the true value Y i The number of validation sets with an error ρ less than 10%, where N is the total number of validation sets, i.e.:

Images