A steel hydrogen damage degree prediction method, system, medium and device

Abstract

This invention discloses a method, system, medium, and device for predicting the degree of hydrogen damage in steel, relating to the field of metal material damage detection technology. The method includes: constructing an experimental dataset of ultrasonic hydrogen damage in high-strength steel; constructing a numerical mechanism model of hydrogen damage and a numerical mechanism model of ultrasonic wave transmission to generate a simulation dataset of ultrasonic physics for hydrogen damage in high-strength steel; performing continuous wavelet transform on the experimental and simulation datasets to determine the corresponding time-frequency features, and constructing a feature alignment network based on an attention mechanism to fuse the time-frequency features and determine a hybrid feature vector; constructing and training a physical information neural network (PINN) with dual-channel input to obtain a prediction model for predicting the degree of hydrogen damage in steel.

Description

Technical Field

[0001] This invention relates to the field of damage detection of metallic materials, and in particular to a method, system, medium, and device for predicting the degree of hydrogen damage in steel. Background Technology

[0002] High-strength steel is widely used in critical fields such as aerospace, nuclear power, and oil and gas pipelines due to its excellent mechanical properties. However, in hydrogen-containing environments, high-strength steel is susceptible to hydrogen damage, leading to decreased material toughness, crack initiation, and sudden failure, seriously threatening structural safety. Traditional hydrogen damage assessment methods mainly rely on offline laboratory testing, such as slow strain rate testing or fracture morphology analysis. These methods are time-consuming, costly, and cannot achieve real-time on-site monitoring. Furthermore, the demand for intelligent monitoring of equipment health under the "Internet+" context has also brought higher requirements to the monitoring technology and life prediction theory research of high-end hydrogen-contaminated equipment.

[0003] In recent years, non-destructive testing technology has been used for hydrogen damage characterization due to its non-destructive characteristics. However, existing methods are mostly based on establishing empirical models based on single signal features. For example, based on extracted single signal features, empirical models are constructed using statistical methods or machine learning algorithms (such as regression analysis, support vector machines, etc.) to predict the degree of hydrogen damage. Hydrogen damage is a complex physical process that is affected by a variety of factors. Single signal features are difficult to fully reflect the intrinsic physical mechanisms of hydrogen diffusion and damage evolution, and are difficult to meet the needs of accurate detection of hydrogen-induced damage under complex working conditions. Summary of the Invention

[0004] This invention provides a method, system, medium, and device for predicting the degree of hydrogen damage in steel, to solve the aforementioned problems in the prior art, namely, how to achieve accurate prediction of the degree of hydrogen damage under complex environments. This invention provides a method for predicting the degree of hydrogen damage in steel, which includes: Tensile tests were performed on steel samples before and after hydrogen corrosion to determine the fracture elongation and reduction of area of ​​the steel samples before and after hydrogen corrosion. Based on the fracture elongation and reduction of area, the hydrogen embrittlement index, which is used to quantitatively characterize the degree of hydrogen damage of the samples, was determined. Based on the hydrogen embrittlement index, the degree of hydrogen damage in steel samples after hydrogenation corrosion is divided into different gradient levels, and the corresponding hydrogen damage degree gradient data is determined. Based on ultrasonic testing, ultrasonic echo signals of steel samples with different degrees of hydrogen damage are obtained. The hydrogen damage degree gradient data and ultrasonic echo signals are integrated to construct an experimental dataset of high-strength steel hydrogen damage ultrasound. Numerical mechanism models of hydrogen damage and ultrasonic transmission were constructed to simulate the theoretical tensile properties and ultrasonic time-frequency response of high-strength steel after hydrogenation corrosion, respectively, and to generate a simulation dataset of ultrasonic physics of hydrogen damage in high-strength steel. Continuous wavelet transform is performed on the experimental dataset and the simulation dataset to determine the wavelet entropy features corresponding to the experimental dataset and the dispersion coefficients corresponding to the simulation dataset. Then, the wavelet entropy features and dispersion coefficients are fused across domains through a feature alignment network based on an attention mechanism to determine the hybrid feature vector. A physical information neural network (PINN) with dual-channel input is constructed. The hybrid feature vector and the residual term of the hydrogen diffusion partial differential equation are respectively input into different channels of the physical information neural network PINN to train the physical information neural network PINN and obtain a prediction model for predicting the degree of hydrogen damage in steel.

[0005] Optionally, the step of determining the hydrogen embrittlement index, used to quantitatively characterize the degree of hydrogen damage to the sample, based on the fracture elongation and reduction of area, specifically includes: ; in, The hydrogen embrittlement index. Elongation at break of the uncharged hydrogen sample. The elongation at break of the sample after hydrogen charging for t hours is given. The section reduction rate is the value of the sample without hydrogen charging. The value represents the cross-sectional shrinkage rate of the sample after hydrogen charging for t hours.

[0006] Optionally, obtaining the residual terms of the hydrogen diffusion partial differential equation specifically includes: in, c This is an approximate solution for the hydrogen concentration. t For time, D is the diffusion coefficient.

[0007] Optionally, the physical information neural network PINN is trained by minimizing a composite loss function, wherein obtaining the composite loss function specifically includes: ; in, The mean square error term for the experimental data is used, and the Huber loss is employed to calculate the deviation between the predicted hydrogen damage level and the measured value. The residual of the hydrogen diffusion equation is... For the residuals of the damage evolution equation, The residual of the ultrasonic motion equation is denoted as .

[0008] This invention provides a system for predicting the degree of hydrogen damage in steel, comprising: The hydrogen embrittlement index acquisition module is used to perform tensile tests on steel samples before and after hydrogen corrosion to determine the fracture elongation and reduction of area of ​​the steel samples before and after hydrogen corrosion. Based on the fracture elongation and reduction of area, the hydrogen embrittlement index is determined to quantitatively characterize the degree of hydrogen damage of the sample. The experimental dataset acquisition module is used to classify the degree of hydrogen damage of steel samples after hydrogen corrosion into different gradient levels according to the hydrogen embrittlement index and determine the corresponding hydrogen damage degree gradient data; the ultrasonic echo signals of steel samples with different degrees of hydrogen damage are obtained based on ultrasonic testing, and the hydrogen damage degree gradient data and ultrasonic echo signals are integrated to construct an experimental dataset of hydrogen damage ultrasound of high-strength steel. The simulation dataset acquisition module is used to construct a numerical mechanism model of hydrogen damage and a numerical mechanism model of ultrasonic transmission to simulate the theoretical tensile properties and ultrasonic time-frequency response of high-strength steel after hydrogenation corrosion, and generate a simulation dataset of ultrasonic physics of hydrogen damage in high-strength steel. The fusion module is used to perform continuous wavelet transform on the experimental dataset and the simulation dataset to determine the wavelet entropy features corresponding to the experimental dataset and the dispersion coefficients corresponding to the simulation dataset. Then, through a feature alignment network based on an attention mechanism, the wavelet entropy features and dispersion coefficients are fused across domains to determine the hybrid feature vector. The model training and prediction module is used to construct a physical information neural network (PINN) with dual-channel input. The mixed feature vector and the residual term of the hydrogen diffusion partial differential equation are respectively input into different channels of the physical information neural network (PINN) to train the physical information neural network (PINN) and obtain a prediction model for predicting the degree of hydrogen damage to steel.

[0009] The present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting the degree of hydrogen damage to steel.

[0010] The present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-mentioned method for predicting the degree of hydrogen damage in steel.

[0011] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention provides a method for predicting the degree of hydrogen damage in steel. This method improves the generalization ability of the model when the sample size is insufficient by supplementing experimental data with physical simulation data. At the same time, it obtains multi-dimensional experimental data by combining ultrasonic testing and tensile testing, and generates simulation data by constructing a numerical mechanism model. It uses deep learning technology to achieve cross-domain feature fusion of experimental and simulation data, and finally constructs a dual-channel physical information neural network (PINN) model that can accurately predict the degree of hydrogen damage in high-strength steel. This model can accurately predict the degree of hydrogen damage under complex working conditions. In addition, based on ultrasonic non-destructive testing signals, this invention can achieve rapid inversion and dynamic tracking of the degree of hydrogen damage in high-strength steel, reduce equipment maintenance costs, and predict damage evolution trends, providing data support for equipment maintenance planning, avoiding sudden failure accidents, and improving the safety and reliability of equipment operation. Attached Figure Description

[0012] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0013] Figure 1 A flowchart of a method for predicting the degree of hydrogen damage in steel provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of a steel hydrogen damage prediction system provided in an embodiment of the present invention; Figure 3 A schematic diagram of a computer device for predicting the degree of hydrogen damage in steel provided in an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0015] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems are described in detail below with specific embodiments.

[0016] Figure 1 This is a flowchart of a method for predicting the degree of hydrogen damage in steel provided by an embodiment of the present invention, as shown below. Figure 1 As shown in this embodiment, a method for predicting the degree of hydrogen damage in steel includes: S1: Tensile tests were performed on steel samples before and after hydrogen corrosion to determine the fracture elongation and reduction of area of ​​the steel samples before and after hydrogen corrosion. Based on the fracture elongation and reduction of area, the hydrogen embrittlement index was determined to quantitatively characterize the degree of hydrogen damage to the samples.

[0017] For example, the electrochemical hydrogen charging test may include: placing a high-strength steel plate sample in an electrochemical hydrogen charging unit cell under constant temperature and pressure conditions. The unit cell contains 0.3 mol / L NaOH solution as the electrolyte solution for electrochemical hydrogen charging. The electrochemical hydrogen charging system consists of an electrochemical workstation, platinum sheet electrodes, high-strength steel sample, and electrolyte solution, forming a dual-electrode electrolytic hydrogen charging system. Electrolytic hydrogen charging is performed using a constant current polarization mode. The hydrogen charging current density is 50~200 mA / cm2, and the hydrogen charging time is 2~144 hours, thus preparing a high-strength steel hydrogen damage gradient sample. For example, a hydrogen damage calibration test may include: mounting a prepared high-strength steel plate sample on a universal tensile testing machine to perform tensile property testing, with the tensile rate set to 2~5×10⁻⁶. -4 s -1 By jointly analyzing the fracture elongation and reduction of area of ​​uncharged and hydrogen-corroded samples, the hydrogen embrittlement index, used to quantitatively characterize the degree of hydrogen damage in the samples, is determined using the following formula: In the formula, The hydrogen embrittlement index. Elongation at break of the uncharged hydrogen sample. The elongation at break of the sample after hydrogen charging for t hours is given. The section reduction rate is the value of the sample without hydrogen charging. The value represents the cross-sectional shrinkage rate of the sample after hydrogen charging for t hours.

[0018] S2: Based on the hydrogen embrittlement index, the degree of hydrogen damage in the steel samples after hydrogen corrosion is divided into different gradient levels, and the corresponding hydrogen damage degree gradient data is determined. Based on ultrasonic testing, ultrasonic echo signals of steel samples with different degrees of hydrogen damage are obtained. The hydrogen damage degree gradient data and ultrasonic echo signals are integrated to construct an experimental dataset of high-strength steel hydrogen damage ultrasound.

[0019] For example, ultrasonic echo signals can be obtained through ultrasonic testing. The specific process of obtaining ultrasonic echo signals may include: based on a high-frequency pulse reflection ultrasonic testing system, the high-strength steel plate sample after hydrogen corrosion is immediately placed on the test stage with the hydrogen-filled side of the sample facing upwards, the ultrasonic probe is placed at the center of the hydrogen-filled side of the sample, and a Gaussian pulse longitudinal wave is excited to collect the ultrasonic echo signal of the high-strength steel sample with gradient damage.

[0020] For example, based on the calculation results of the hydrogen embrittlement index, the degree of hydrogen damage to the sample can be divided into different gradient levels. The samples can be classified and organized according to the hydrogen damage degree gradient to ensure that there is a sufficient number of sample data under each gradient. Then, by integrating the hydrogen damage degree gradient data with the corresponding ultrasonic echo signal, an experimental dataset is formed. Each data point in the experimental dataset includes the hydrogen damage degree gradient label of the sample (hydrogen damage degree gradient data) and the corresponding ultrasonic echo signal.

[0021] S3: Construct numerical mechanism models for hydrogen damage and ultrasonic transmission to simulate the theoretical tensile properties and ultrasonic time-frequency response of high-strength steel after hydrogenation corrosion, respectively, and generate a simulation dataset of ultrasonic physics for hydrogen damage in high-strength steel.

[0022] For example, a numerical mechanism model for hydrogen damage may include: using the finite element method to simulate and obtain the theoretical tensile performance parameters of high-strength steel after hydrogenation corrosion; constructing a scaled simulation model of a high-strength steel tensile sample; combining the concentration distribution law of hydrogen in the material after hydrogen filling; defining the material properties of the simulation model in the thickness direction of the tensile sample; calculating the reduction of area and elongation at break of the simulation model through the stress-strain constitutive equation; and obtaining the theoretical hydrogen damage degree of high-strength steel with gradient damage degree by combining the expression for obtaining the hydrogen embrittlement index.

[0023] The numerical mechanism model of ultrasonic transmission includes: establishing a coupled numerical model based on the COMSOL Multiphysics platform, which includes a hydrogen diffusion model and an elastic wave propagation model containing Fick's second law; calculating the spatial distribution of hydrogen concentration and the theoretical ultrasonic time-frequency response by setting hydrogen charging current density and external load amplitude as boundary conditions; and generating a physical simulation dataset covering the entire parameter space.

[0024] S4: Perform continuous wavelet transform on the experimental dataset and the simulation dataset to determine the wavelet entropy features corresponding to the experimental dataset and the dispersion coefficients corresponding to the simulation dataset. Then, use a feature alignment network based on an attention mechanism to perform cross-domain feature fusion of the wavelet entropy features and the dispersion coefficients to determine the hybrid feature vector.

[0025] For example, the feature alignment network adopts a dual-stream architecture, specifically including: a feature encoder for experimental signals: using a 3-layer 1D convolutional network (kernel size: 3×1, stride 1, padding=1; first layer output channels 64, second layer 128, third layer 256; activation function is ReLU) to extract local features of wavelet time-frequency maps; a feature encoder for simulated signals: using a graph convolutional network (GCN, input is the physical field distribution of finite element mesh nodes, which may include hydrogen concentration, stress and displacement, hidden layer dimension 128, output dimension 256; activation function is ReLU) to process the physical field distribution of the finite element mesh; a cross-modal attention module: establishing a correlation mapping between experimental features and simulated physical field parameters through a multi-head attention mechanism, and achieving cross-domain feature fusion through weighted summation to generate a 256-dimensional hybrid feature vector with physical interpretability (the physical meaning of the fused features: includes both the actual damage information of the experimental signal and the physical field law associated with the simulated signal).

[0026] For example, the experimental and simulated ultrasonic signals can be continuously transformed using the Morlet wavelet basis (wavelet parameters: center frequency 2, bandwidth 0.5) to obtain the time-frequency matrix; singular value decomposition (SVD) is performed on the time-frequency matrix to extract the top 10 singular values ​​(cumulative contribution rate ≥95%) as basic features; at the same time, the wavelet entropy of the experimental signal (reflecting the signal complexity, the more severe the hydrogen damage, the greater the wavelet entropy) and the dispersion coefficient of the simulated signal (reflecting the change of sound wave propagation speed with frequency, hydrogen damage leads to a decrease in the dispersion coefficient) are calculated.

[0027] S5: Construct a physical information neural network (PINN) with dual-channel input; input the hybrid feature vector and the residual term of the hydrogen diffusion partial differential equation into different channels of the physical information neural network (PINN) respectively, train the physical information neural network (PINN), and obtain a prediction model for predicting the degree of hydrogen damage to steel.

[0028] For example, the Physical Information Neural Network (PINN) has a network architecture consisting of 8 hidden layers, each with 256 neurons, employing an adaptive tanh activation function; and a composite loss function. ,in, The mean square error term for the experimental data is used, and the Huber loss is employed to calculate the deviation between the predicted hydrogen damage level and the measured value. The residual of the hydrogen diffusion equation, For the residuals of the damage evolution equation, The residual of the ultrasonic motion equation is denoted as .

[0029] The joint training dataset uses a mixture of physical simulation data and experimental data in a certain proportion, and a dynamic weight adjustment strategy is used to balance the contributions of the two. The proportion is 40%-70% for physical simulation data. The training process adopts a course learning strategy, introducing physical constraints of different complexities in stages. In the initial stage, α:β:γ:δ=1:0.1:0.1:0.1, in the middle stage it is α:β:γ:δ=1:0.5:0.5:0.5, and in the final stage it is adjusted to α:β:γ:δ=1:1:1:1.

[0030] When training the Physical Information Neural Network (PINN), the Adam optimizer can be used (with an initial learning rate of 1×10⁻⁶). -4 The loss function value is reduced to 0.5 of its original value every 500 iterations. The batch size is set to 32, and the training iterations are 3000. During training, the composite loss function value is monitored in real time. When the loss function changes by ≤1×10⁻⁶ for 50 consecutive iterations, the loss function is considered lost. -5 When the time is up, stop training and obtain the trained model.

[0031] For example, a Monte Carlo Dropout layer (with a Dropout probability of 0.1) can be added to the hidden layer of the PINN network to make 50 predictions for the same input signal and calculate the 95% confidence interval of the predicted value (CI = [μ - 1.96σ, μ + 1.96σ], where μ is the mean of the 50 predictions and σ is the standard deviation). A confidence interval width threshold can be set (0.1-0.15 depending on the engineering accuracy requirements). When the interval width exceeds the threshold, the physical model is triggered to correct itself online.

[0032] For example, the online correction process may include: real-time acquisition of new experimental data and calculation of residual distribution; adjustment of the prior distribution of diffusion coefficient D and trap density parameters using a Bayesian update method; regeneration of the local simulation dataset after updating the material parameters of the finite element model; and incremental fine-tuning of the PINN network. Specifically, it includes the following steps:

[0033] (7) Real-time acquisition of newly added experimental data on site (including ultrasonic echo signals and corresponding hydrogen damage calibration values, with the sample size being 10%-20% of the initial experimental dataset each time), and calculation of the residual distribution between the model prediction value and the measured value (residual = prediction value - measured value). (8) Using the Bayesian update method, with the residual distribution as the likelihood function, the prior distributions of the diffusion coefficient D and the trap density parameter (N_trap) in the hydrogen diffusion model are adjusted (the initial prior is a normal distribution, and the updated one is a posterior distribution). (9) Based on the updated parameters, rerun the numerical mechanism model of hydrogen damage and the numerical mechanism model of ultrasonic transmission in the above steps to generate a local simulation dataset (the sample size is consistent with the newly added experimental data). (10) Mix the newly added experimental data with the local simulation dataset and perform incremental fine-tuning on the PINN network (training iterations of 300-500 times, learning rate of 1×10). -5 (Keep other parameters unchanged) Update the model parameters to ensure prediction accuracy.

[0034] For example, the ultrasonic echo signal of high-strength steel collected in real time on site is subjected to feature extraction and cross-domain fusion, and then input into the trained (or online corrected) PINN network to output a predicted value of hydrogen damage intensity (HEI). At the same time, the reliability of the prediction is evaluated by obtaining the 95% confidence interval of the predicted value. Based on the trend of HEI value change (e.g., HEI increases by 0.05-0.1 per week) and combined with the equipment safety threshold (e.g., shutdown and maintenance are required when HEI=0.6), dynamic tracking of the evolution of hydrogen damage intensity and life prediction can be realized.

[0035] The above are one or more embodiments of the steel hydrogen damage prediction method provided in this specification. Based on the same idea, this specification also provides a corresponding steel hydrogen damage prediction system, such as... Figure 2 As shown, it includes: The hydrogen embrittlement index acquisition module is used to perform tensile tests on steel samples before and after hydrogen corrosion to determine the fracture elongation and reduction of area of ​​the steel samples before and after hydrogen corrosion. Based on the fracture elongation and reduction of area, the hydrogen embrittlement index is determined to quantitatively characterize the degree of hydrogen damage of the sample. The experimental dataset acquisition module is used to classify the degree of hydrogen damage of steel samples after hydrogen corrosion into different gradient levels according to the hydrogen embrittlement index and determine the corresponding hydrogen damage degree gradient data; the ultrasonic echo signals of steel samples with different degrees of hydrogen damage are obtained based on ultrasonic testing, and the hydrogen damage degree gradient data and ultrasonic echo signals are integrated to construct an experimental dataset of hydrogen damage ultrasound of high-strength steel. The simulation dataset acquisition module is used to construct a numerical mechanism model of hydrogen damage and a numerical mechanism model of ultrasonic transmission to simulate the theoretical tensile properties and ultrasonic time-frequency response of high-strength steel after hydrogenation corrosion, and generate a simulation dataset of ultrasonic physics of hydrogen damage in high-strength steel. The fusion module is used to perform continuous wavelet transform on the experimental dataset and the simulation dataset to determine the wavelet entropy features corresponding to the experimental dataset and the dispersion coefficients corresponding to the simulation dataset. Then, through a feature alignment network based on an attention mechanism, the wavelet entropy features and dispersion coefficients are fused across domains to determine the hybrid feature vector. The model training and prediction module is used to construct a physical information neural network (PINN) with dual-channel input. The mixed feature vector and the residual term of the hydrogen diffusion partial differential equation are respectively input into different channels of the physical information neural network (PINN) to train the physical information neural network (PINN) and obtain a prediction model for predicting the degree of hydrogen damage to steel.

[0036] Specific limitations regarding the steel hydrogen damage prediction system can be found in the limitations of the steel hydrogen damage prediction method described above, and will not be repeated here. Each module in the aforementioned steel hydrogen damage prediction system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0037] The present invention also provides a computer-readable storage medium storing a computer program that can be used to perform the above-described prediction of the degree of hydrogen damage to steel.

[0038] The present invention also provides Figure 3 The schematic diagram of the computer device shown is as follows: Figure 3 As shown, at the hardware level, the computer device includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to implement the steel hydrogen damage prediction method provided in the above embodiment.

[0039] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0040] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this invention.

Claims

1. A method for predicting the degree of hydrogen damage in steel, characterized in that, include: Tensile tests were performed on steel samples before and after hydrogen corrosion to determine the fracture elongation and reduction of area of ​​the steel samples before and after hydrogen corrosion. Based on the fracture elongation and reduction of area, the hydrogen embrittlement index, which is used to quantitatively characterize the degree of hydrogen damage of the samples, was determined. Based on the hydrogen embrittlement index, the degree of hydrogen damage in steel samples after hydrogenation corrosion is divided into different gradient levels, and the corresponding hydrogen damage degree gradient data is determined. Based on ultrasonic testing, ultrasonic echo signals of steel samples with different degrees of hydrogen damage are obtained. The hydrogen damage degree gradient data and ultrasonic echo signals are integrated to construct an experimental dataset of high-strength steel hydrogen damage ultrasound. Numerical mechanism models of hydrogen damage and ultrasonic transmission were constructed to simulate the theoretical tensile properties and ultrasonic time-frequency response of high-strength steel after hydrogenation corrosion, respectively, and to generate a simulation dataset of ultrasonic physics of hydrogen damage in high-strength steel. Continuous wavelet transform is performed on the experimental dataset and the simulation dataset to determine the wavelet entropy features corresponding to the experimental dataset and the dispersion coefficients corresponding to the simulation dataset. Then, the wavelet entropy features and dispersion coefficients are fused across domains through a feature alignment network based on an attention mechanism to determine the hybrid feature vector. A physical information neural network (PINN) with dual-channel input is constructed. The hybrid feature vector and the residual term of the hydrogen diffusion partial differential equation are respectively input into different channels of the physical information neural network PINN to train the physical information neural network PINN and obtain a prediction model for predicting the degree of hydrogen damage in steel.

2. The method for predicting the degree of hydrogen damage in steel as described in claim 1, characterized in that, The hydrogen embrittlement index, used to quantitatively characterize the degree of hydrogen damage in a sample, is determined based on fracture elongation and reduction of area, specifically including: ； in, The hydrogen embrittlement index. Elongation at break of the uncharged hydrogen sample. The elongation at break of the sample after hydrogen charging for t hours is given. The section reduction rate is the value of the sample without hydrogen charging. The value represents the cross-sectional shrinkage rate of the sample after hydrogen charging for t hours.

3. The method for predicting the degree of hydrogen damage to steel as described in claim 1, characterized in that, The acquisition of the residual terms of the hydrogen diffusion partial differential equation specifically includes: in, c This is an approximate solution for the hydrogen concentration. t For time, D is the diffusion coefficient.

4. The method for predicting the degree of hydrogen damage in steel as described in claim 1, characterized in that, The physical information neural network PINN is trained by minimizing a composite loss function. The acquisition of the composite loss function specifically includes: ； in, The mean square error term for the experimental data is used, and the Huber loss is employed to calculate the deviation between the predicted hydrogen damage level and the measured value. The residual of the hydrogen diffusion equation is... For the residuals of the damage evolution equation, The residual of the ultrasonic motion equation is denoted as .

5. A system for predicting the degree of hydrogen damage to steel, characterized in that, include: The hydrogen embrittlement index acquisition module is used to perform tensile tests on steel samples before and after hydrogen corrosion to determine the fracture elongation and reduction of area of ​​the steel samples before and after hydrogen corrosion. Based on the fracture elongation and reduction of area, the hydrogen embrittlement index is determined to quantitatively characterize the degree of hydrogen damage of the sample. The experimental dataset acquisition module is used to classify the degree of hydrogen damage of steel samples after hydrogen corrosion into different gradient levels according to the hydrogen embrittlement index and determine the corresponding hydrogen damage degree gradient data; the ultrasonic echo signals of steel samples with different degrees of hydrogen damage are obtained based on ultrasonic testing, and the hydrogen damage degree gradient data and ultrasonic echo signals are integrated to construct an experimental dataset of hydrogen damage ultrasound of high-strength steel. The simulation dataset acquisition module is used to construct a numerical mechanism model of hydrogen damage and a numerical mechanism model of ultrasonic transmission to simulate the theoretical tensile properties and ultrasonic time-frequency response of high-strength steel after hydrogenation corrosion, and generate a simulation dataset of ultrasonic physics of hydrogen damage in high-strength steel. The fusion module is used to perform continuous wavelet transform on the experimental dataset and the simulation dataset to determine the wavelet entropy features corresponding to the experimental dataset and the dispersion coefficients corresponding to the simulation dataset. Then, through a feature alignment network based on an attention mechanism, the wavelet entropy features and dispersion coefficients are fused across domains to determine the hybrid feature vector. The model training and prediction module is used to construct a physical information neural network (PINN) with dual-channel input. The mixed feature vector and the residual term of the hydrogen diffusion partial differential equation are respectively input into different channels of the physical information neural network (PINN) to train the physical information neural network (PINN) and obtain a prediction model for predicting the degree of hydrogen damage to steel.

6. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steel hydrogen damage prediction method according to any one of claims 1-4.

7. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steel hydrogen damage prediction method according to any one of claims 1-4.

Images