A multi-parameter integrated testing system for magnesium-based solid hydrogen storage materials

Abstract

This invention relates to the field of solid-state hydrogen storage material testing technology, and discloses a multi-parameter comprehensive testing system for magnesium-based solid-state hydrogen storage materials. The system includes a sample feature extraction module, an in-situ data injection module, a dynamic test field module, a cross-domain mapping training module, and an instruction parsing and reporting module. The system applies multi-physical field composite excitations such as temperature, pressure, and stress to the material sample and simultaneously acquires multi-dimensional response signals; it generates a real-time environmental perturbation spectrum based on the signal sequence, uses its deviation from the standard model to drive adversarial training, dynamically generates test field control commands and material state evolution commands, and finally outputs a comprehensive test report. This system can realistically simulate multi-field coupled environments under complex working conditions, achieving adaptive optimization of test conditions and high-precision dynamic analysis of material performance evolution.

Description

Technical Field

[0001] This invention relates to the field of solid-state hydrogen storage material testing technology, specifically to a multi-parameter comprehensive testing system for magnesium-based solid-state hydrogen storage materials. Background Technology

[0002] In practical applications, the hydrogen storage performance of magnesium-based solid-state hydrogen storage materials is significantly affected by the complex coupling effects of multiple physical fields, including temperature, pressure, stress, and hydrogen concentration. Traditional testing methods often employ sequential or independent single-physical-field loading, such as thermogravimetric analysis, pressure-composition-isotherm testing, or mechanical property characterization. Data acquired in isolated environments using these methods struggles to accurately represent the dynamic response of materials under real-world multi-field synergistic effects, leading to discrepancies between performance assessments and actual conditions. The discreteness of data acquisition also makes real-time correlation analysis between different physical quantities difficult, failing to capture complete information on the transient behavior of materials under multi-field coupled excitation.

[0003] Existing testing systems typically rely on pre-set fixed procedures or operator experience for control strategies. Once test conditions are set, they remain constant in a single experiment or can only be adjusted manually in stages. This static or semi-static control mode lacks the ability to adaptively respond and optimize based on the material's real-time response. When material behavior deviates from expectations or is in a complex phase transition stage, the fixed procedure cannot dynamically adjust excitation parameters to accurately capture key state nodes, resulting in low intelligence in the testing process and limited ability to predict and analyze complex material evolution paths.

[0004] A testing technique is needed to achieve in-situ synchronous excitation of multi-physics fields and correlation acquisition of multi-dimensional signals to realistically simulate complex service environments. Simultaneously, a method needs to be developed that can dynamically optimize test conditions based on the material's real-time response and intelligently analyze the material's state evolution. This would overcome the shortcomings of isolated data and rigid control in traditional testing, enabling a more accurate and comprehensive evaluation of the performance of magnesium-based solid-state hydrogen storage materials. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-parameter comprehensive testing system for magnesium-based solid hydrogen storage materials to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a multi-parameter comprehensive testing system for magnesium-based solid hydrogen storage materials, the system comprising:

[0007] The sample feature extraction module is used to prepare magnesium-based solid hydrogen storage material samples with preset microstructure features and extract the microstructure feature information of the magnesium-based solid hydrogen storage material samples.

[0008] The in-situ data injection module is used to load the microstructure feature information and apply composite physical field excitation to the magnesium-based solid hydrogen storage material sample, and simultaneously acquire multidimensional response signal sequences.

[0009] The dynamic test field module is used to construct a dynamic test field, import the multidimensional response signal sequence into the dynamic test field, and drive the generation of a real-time environmental disturbance spectrum corresponding to the multidimensional response signal sequence.

[0010] The cross-domain mapping training module is used to establish a cross-domain mapping relationship based on the deviation between the real-time environmental perturbation spectrum and the preset standard perturbation model, and to perform synchronous adversarial training of the model based on the cross-domain mapping relationship, thereby generating test field control commands and material state evolution commands.

[0011] The instruction parsing and report module is used to parse the test field control instructions and material state evolution instructions to generate a comprehensive test report.

[0012] Preferably, the sample feature extraction module includes:

[0013] Based on the target hydrogen storage performance indicators, the grain size distribution, phase composition ratio and pore topology of magnesium-based solid hydrogen storage materials were designed in reverse.

[0014] The magnesium-based solid hydrogen storage material sample was prepared by physical vapor deposition and controlled sintering processes according to the grain size distribution, phase composition ratio and pore topology.

[0015] The magnesium-based solid hydrogen storage material sample was scanned using a micro-area composition analyzer and a three-dimensional morphology reconstruction instrument to obtain the actual micro-area composition distribution data and actual three-dimensional morphology data of the sample.

[0016] The actual micro-region component distribution data and the actual three-dimensional morphology data are combined to form the microstructure feature information.

[0017] Preferably, the in-situ data injection module includes:

[0018] The microstructure feature information is encoded into a machine-readable field control instruction set;

[0019] The field control instruction set is input to the multi-field coupling generator, which synchronously outputs thermal field fluctuation excitation, stress field gradient excitation and electromagnetic field pulse excitation according to the field control instruction set. The thermal field fluctuation excitation, stress field gradient excitation and electromagnetic field pulse excitation together constitute the composite physical field excitation.

[0020] By using a high-sensitivity sensor array surrounding the surface of the magnesium-based solid hydrogen storage material sample, the thermal radiation change signal, lattice strain propagation signal and surface potential fluctuation signal of the magnesium-based solid hydrogen storage material sample under the excitation of the composite physical field are simultaneously captured.

[0021] The thermal radiation change signal, lattice strain propagation signal, and surface potential fluctuation signal are aligned and packaged in time sequence to form the multidimensional response signal sequence.

[0022] Preferably, the dynamic test field module includes:

[0023] Initialize a digital twin space consisting of a computational fluid dynamics mesh and a discrete element mesh, as the basis of the dynamic test field;

[0024] The thermal radiation change signal, lattice strain propagation signal and surface potential fluctuation signal in the multidimensional response signal sequence are decoupled and mapped to the boundary condition input and volume force source input in the digital twin space, respectively.

[0025] Run the transient solver of the digital twin space to calculate the temperature fluctuation cloud map, stress distribution cloud map and hydrogen concentration diffusion flow field map of each grid node in the digital twin space;

[0026] Spatiotemporal evolution features are extracted from the temperature fluctuation cloud map, stress distribution cloud map, and hydrogen concentration diffusion flow field map, and compiled to generate the real-time environmental disturbance spectrum.

[0027] Preferably, the cross-domain mapping training module establishes a cross-domain mapping relationship based on the deviation between the real-time environmental perturbation spectrum and the preset standard perturbation model, including:

[0028] The preset standard perturbation model stored in the knowledge base is invoked, which describes the standard environmental perturbation response of magnesium-based solid hydrogen storage materials under ideal conditions;

[0029] By comparing the feature values ​​of the real-time environmental disturbance spectrum with those of the preset standard disturbance model at the same time and spatial location, the global deviation and local deviation distribution map are calculated.

[0030] Using the global deviation as the feedback gain coefficient and the local deviation distribution map as the weight distribution basis, a functional mapping network from the physical signal domain to the material performance parameter domain is constructed.

[0031] The internal connection weights of the functional mapping network are optimized by backpropagation algorithm, and the cross-domain mapping relationship is solidified.

[0032] Preferably, the cross-domain mapping training module performs synchronous adversarial training of the model based on the cross-domain mapping relationship, generating test field control instructions and material state evolution instructions, including:

[0033] The material performance prediction model and the test field control model are activated, and the cross-domain mapping relationship is used as a shared constraint condition between the material performance prediction model and the test field control model.

[0034] Under the shared constraints, the material performance prediction model attempts to predict the material performance degradation path under different environmental disturbances, while the test field control model attempts to generate a strengthening disturbance scheme that can maximize the exposure of the material performance degradation path.

[0035] The prediction operation of the material property prediction model and the generation operation of the test field control model are executed iteratively until the material property prediction model can no longer accurately predict the disturbance effect generated by the test field control model.

[0036] The final enhanced perturbation scheme output by the test field control model at this time is recorded as the test field control command, and the final performance degradation path output by the material performance prediction model at this time is recorded as the material state evolution command.

[0037] Preferably, the instruction parsing and reporting module parses the test field control instructions and material state evolution instructions to generate a comprehensive test report, including:

[0038] By disassembling the test field control commands, we obtain temperature control sub-commands, pressure control sub-commands, and atmosphere control sub-commands for the dynamic test field.

[0039] Interpreting the material state evolution instructions yields the predicted hydrogen absorption rate curve, predicted hydrogen desorption plateau pressure curve, and predicted cycle life curve of the magnesium-based solid hydrogen storage material at different test stages.

[0040] The temperature control sub-instruction, pressure control sub-instruction, and atmosphere control sub-instruction are associated and labeled with the predicted hydrogen absorption rate curve, predicted hydrogen release plateau pressure curve, and predicted cycle life curve, and integrated into a structured document template to form the comprehensive test report.

[0041] Preferably, the initialization of a digital twin space consisting of a computational fluid dynamics mesh and a discrete element mesh includes:

[0042] A computational fluid dynamics mesh is created, with mesh nodes covering the geometric space of the magnesium-based solid hydrogen storage material sample. The mesh size is adaptively adjusted based on the pore topology in the microstructure feature information.

[0043] A discrete element mesh is created, which overlaps spatially with the computational fluid dynamics mesh. The particle parameters of the discrete element mesh are set based on the grain size distribution in the microstructure feature information.

[0044] The computational fluid dynamics mesh is coupled with the discrete element mesh, and coupling interface conditions are set. The coupling interface conditions are defined based on the correlation between the thermal radiation change signal and the lattice strain propagation signal in the multidimensional response signal sequence.

[0045] In the coupled mesh space, initial boundary conditions and initial field variables are defined. The initial boundary conditions are based on the parameter settings of the composite physical field excitation, and the initial field variables are initialized based on the microstructure feature information, thus completing the construction of the digital twin space.

[0046] Preferably, the construction of the preset standard perturbation model includes:

[0047] Standard magnesium-based solid hydrogen storage material samples were prepared under standard environmental conditions, and standard microstructure feature information was extracted.

[0048] Based on the standard microstructure feature information, a standard field control instruction set is generated, and a standard composite physical field excitation is applied to the standard magnesium-based solid hydrogen storage material sample. The standard composite physical field excitation includes standard thermal field fluctuation excitation, standard stress field gradient excitation, and standard electromagnetic field pulse excitation.

[0049] Standard response signals, including standard thermal radiation change signals, standard lattice strain propagation signals, and standard surface potential fluctuation signals, are synchronously acquired through a standard sensor array.

[0050] The standard response signals are aligned and processed according to time sequence to generate a standard environmental disturbance response curve, which is stored in the knowledge base as the preset standard disturbance model.

[0051] Preferably, optimizing the internal connection weights of the functional mapping network using the backpropagation algorithm includes:

[0052] Define the input layer, hidden layer, and output layer of the functional mapping network. The input layer nodes correspond to the eigenvalues ​​of the real-time environmental perturbation spectrum, and the output layer nodes correspond to the material performance parameters.

[0053] Initialize the inner connection weights to random values ​​and set the learning rate parameter;

[0054] The forward propagation computes the output value, which is calculated based on the input layer feature values ​​and the current internal connection weights;

[0055] Calculate the error between the output value and the target value, wherein the target value is derived based on the standard environmental disturbance response curve of the preset standard disturbance model;

[0056] The internal connection weights are adjusted based on the error using the backpropagation algorithm, with the adjustment amount calculated based on the learning rate and gradient descent.

[0057] Repeat the forward and backward propagation process until the error is lower than a preset threshold, and solidify the internal connection weights to obtain the cross-domain mapping relationship.

[0058] Compared with the prior art, the beneficial effects of the present invention are:

[0059] By integrating multi-physics excitation sources such as temperature, stress, and hydrogen pressure with corresponding sensor arrays, a composite field excitation is simultaneously applied to magnesium-based solid hydrogen storage material samples during testing, while synchronously acquiring multi-dimensional response signals in thermal, mechanical, and chemical dimensions, forming a time-strictly aligned multi-dimensional data sequence. This technical approach ensures that the material is always under the coupled field action of a near-real service environment, allowing its hydrogenation / dehydrogenation kinetics, phase transitions, stress evolution, and other processes to be observed in a realistic state of mutual influence. The synchronous acquisition of multi-dimensional signals establishes instantaneous and intrinsic correlations between different physical quantities, fully revealing the synergistic mechanism and transient characteristics of material behavior under multi-field coupling, and solving the problems of data fragmentation and state distortion caused by traditional sequential or single-field testing.

[0060] The real-time acquired multidimensional response signal sequence is imported into the constructed dynamic test field to generate a real-time environmental perturbation spectrum reflecting the dynamic changes of the external composite field. The deviation between this perturbation spectrum and the preset standard model is calculated, and this deviation is used to drive the machine learning model for cross-domain mapping and synchronous adversarial training. This training process enables the system to dynamically adjust the excitation parameters of the composite physical field, while simultaneously extrapolating the evolution path of the material's internal microstructure, phase composition, hydrogen content, and other states with high confidence. This approach transforms the testing process from a fixed, unidirectional execution of a fixed procedure into an intelligent system where "dynamic optimization of test conditions" and "real-time analysis of material states" interact and compete with each other. It can autonomously identify and focus on key test intervals where material properties undergo abrupt changes or phase transitions, adaptively adjusting the excitation intensity and method, thereby achieving more precise stimulation and capture of the material's complex nonlinear behavior and more intelligent prediction of evolution trends. Attached Figure Description

[0061] Figure 1 This is a timing diagram of the multi-parameter integrated testing system for magnesium-based solid hydrogen storage materials described in this invention.

[0062] Figure 2 A flowchart illustrating the operation of the sample feature extraction module;

[0063] Figure 3 A flowchart illustrating the operation of the dynamic test field module;

[0064] Figure 43D correlation analysis diagram of performance parameters of magnesium-based hydrogen storage materials;

[0065] Figure 5 This is a line graph showing the error evolution during training of a cross-domain mapping network. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] Please see Figure 1 This invention provides a multi-parameter comprehensive testing system for magnesium-based solid hydrogen storage materials. The system includes: a sample feature extraction module, an in-situ data injection module, a dynamic test field module, a cross-domain mapping training module, and an instruction parsing and reporting module. The sample feature extraction module is used to prepare magnesium-based solid hydrogen storage material samples with preset microstructure features and extract their microstructure feature information. The in-situ data injection module loads the microstructure feature information and applies a composite physical field excitation to the sample to synchronously acquire multi-dimensional response signal sequences. The dynamic test field module constructs a dynamic test field and imports the multi-dimensional response signal sequences into it to drive the generation of a real-time environmental perturbation spectrum. The cross-domain mapping training module establishes a cross-domain mapping relationship based on the deviation between the real-time environmental perturbation spectrum and a preset standard perturbation model, and performs synchronous adversarial training of the model based on this relationship to generate test field control instructions and material state evolution instructions. The instruction parsing and reporting module parses the test field control instructions and material state evolution instructions to generate a comprehensive test report.

[0068] Example 1: See Figure 2The sample feature extraction module reverse-engineers the grain size distribution, phase composition ratio, and pore topology of magnesium-based solid hydrogen storage materials based on the target hydrogen storage performance indicators. This module uses physical vapor deposition and controlled sintering processes to prepare magnesium-based solid hydrogen storage material samples according to the grain size distribution, phase composition ratio, and pore topology. The sample is then scanned using a micro-area composition analyzer and a three-dimensional morphology reconstruction instrument to obtain the actual micro-area composition distribution data and actual three-dimensional morphology data. The actual micro-area composition distribution data and actual three-dimensional morphology data are then fused to form microstructure feature information. The in-situ data injection module encodes microstructure feature information into a machine-readable field control instruction set, which is then input to a multi-field coupling generator. The multi-field coupling generator synchronously outputs thermal field fluctuation excitation, stress field gradient excitation, and electromagnetic field pulse excitation according to the field control instruction set. These excitations together constitute a composite physical field excitation. A high-sensitivity sensor array surrounding the surface of the magnesium-based solid hydrogen storage material sample synchronously captures the thermal radiation change signal, lattice strain propagation signal, and surface potential fluctuation signal of the sample under the composite physical field excitation. The thermal radiation change signal, lattice strain propagation signal, and surface potential fluctuation signal are aligned in time and packaged to form a multidimensional response signal sequence.

[0069] In practical implementation, the sample feature extraction module reverse-engineers the grain size distribution, phase composition ratio, and pore topology of magnesium-based solid hydrogen storage materials based on the target hydrogen storage performance indicators. An exemplary target hydrogen storage performance indicator requires the magnesium-based solid hydrogen storage material to complete 90% of its hydrogen absorption capacity within 150 seconds at 300 degrees Celsius. The reverse design process outputs specific target values ​​for microstructure parameters. The target value for grain size distribution is between 20 nanometers and 200 nanometers and exhibits a bimodal distribution. The target value for the composition ratio of the main phase MgH2 is 85wt%, the target value for the composition ratio of the doped phase TiH2 is 15wt%, and the target value for pore topology is a porosity of 35% and a pore penetration rate of over 90%. These target values ​​constitute the preset microstructure features. The sample feature extraction module uses physical vapor deposition (PVD) and controlled sintering (CSS) to prepare magnesium-based solid hydrogen storage material samples according to grain size distribution, phase composition ratio, and pore topology. PVD is performed under an argon protective atmosphere with a co-sputtering power density of 8 watts per square centimeter for magnesium and titanium targets. The substrate temperature is controlled at 200 degrees Celsius, resulting in a magnesium-titanium composite thin film precursor with a composition conforming to the design. CSS is performed in a hot isostatic pressing (HIP) furnace with a heating rate of 5 degrees Celsius per minute, a final sintering temperature of 450 degrees Celsius, and a holding time of 60 minutes, while simultaneously applying an isostatic pressure of 150 MPa. This pressure-temperature-time parameter combination is determined by the preset pore topology target value through the following process relationship:

[0070]

[0071] in: Represents the applied isostatic pressure. Represents the sintering temperature. Indicates the heat preservation time. Represents the target porosity. and This is the process correlation coefficient, which controls the morphology and connectivity of the pores through this process relationship, ultimately obtaining a bulk magnesium-based solid hydrogen storage material sample with the desired microstructure. In specific implementation, the sample feature extraction module uses a micro-area composition analyzer and a three-dimensional morphology reconstruction instrument to scan the magnesium-based solid hydrogen storage material sample. The micro-area composition analyzer uses an energy dispersive spectroscopy (EDS) spectrometer equipped with a field emission electron microscope (FET). Five 100-micrometer by 100-micrometer regions are selected on the sample surface for surface scanning to obtain the mass percentage distribution data of magnesium, titanium, and oxygen in each region. The average value is calculated to obtain the actual phase composition ratio data. For example, one set of measurement data shows that the mass percentage of Mg is 85.3%, the mass percentage of Ti is 14.5%, and the mass percentage of O is 0.2%. The 3D topography reconstruction instrument uses X-ray tomography technology with a scanning voltage of 100 kV and a scanning resolution of 0.5 μm per voxel to reconstruct a 3D model of the internal pore structure of the sample. Porosity and pore connectivity are calculated using image analysis software. For example, a set of measurement data shows a porosity of 34.7% and a pore connectivity of 89.5%. The sample feature extraction module integrates the actual micro-area composition distribution data and the actual 3D topography data to form microstructure feature information. The microstructure feature information is stored in a structured data file, which includes a grain size statistical histogram, a phase composition ratio list, and a pore network model.

[0072] In practical implementation, the in-situ data injection module encodes microstructural feature information into a machine-readable field control instruction set. This instruction set is a text file following a specific protocol, containing a sequence of control parameters for multiple field coupling generators. For example, one instruction defines the temperature change curve function of thermal field fluctuation excitation over a time period of 0 to 300 seconds. The loading force function for stress field gradient excitation is: The magnetic field strength and frequency parameters of the electromagnetic pulse excitation are 0.5 Tesla and 10 Hz. These parameter values ​​are directly derived from the sensitivity data of the pore topology to the thermal stress field and the matching data of the grain size distribution to the electromagnetic field response frequency in the microstructure feature information. In specific implementation, the in-situ data injection module inputs the field control command set to the multi-field coupling generator. The multi-field coupling generator includes a resistance heating furnace, a servo hydraulic loader, and an electromagnetic coil array. The multi-field coupling generator parses the field control command set and synchronously outputs thermal field fluctuation excitation, stress field gradient excitation, and electromagnetic field pulse excitation. The thermal field fluctuation excitation is generated by the resistance heating furnace according to the command definition. The function accurately outputs the temperature field, and the stress field gradient excitation is applied spatially relative to the sample fixture by a servo-hydraulic loader. The distributed load and electromagnetic pulse excitation are generated by an electromagnetic coil array producing an alternating pulse magnetic field of 0.5 Tesla and 10 Hz. These three excitations work together in time and space to form a composite physical field excitation.

[0073] In practice, a high-sensitivity sensor array surrounding the surface of a magnesium-based solid hydrogen storage material sample synchronously captures the thermal radiation change signal, lattice strain propagation signal, and surface potential fluctuation signal of the sample under the excitation of a composite physical field. The high-sensitivity sensor array consists of 8 infrared temperature probes, 4 piezoelectric ceramic ultrasonic sensors, and 12 micro-contact potentiometers installed around the sample in a specific spatial configuration. The infrared temperature probes record a sequence of temperature distribution images on the sample surface at a rate of 1000 frames per second, and extract the data of radiation intensity change over time for each pixel as the thermal radiation change signal. The piezoelectric ceramic ultrasonic sensors receive ultrasonic waves passing through the sample at a sampling frequency of 20 MHz per second, and calculate the lattice strain propagation signal by analyzing the ultrasonic wave time difference and waveform distortion. The micro-contact potentiometers measure the potential values ​​of 12 fixed probe points on the sample surface at a frequency of 100,000 times per second to obtain the surface potential fluctuation signal. In practical implementation, the in-situ data injection module aligns and packages the thermal radiation change signal, lattice strain propagation signal, and surface potential fluctuation signal in time sequence to form a multidimensional response signal sequence. The time sequence alignment is based on a unified global positioning system timestamp, and the data streams collected by all sensors are synchronized with millisecond-level precision. The packaging process encapsulates all temperature distribution data, strain data of all ultrasonic channels, and potential data of all potentiometer channels under the same timestamp into a data package, which is arranged in time sequence to form a multidimensional response signal sequence. The data structure of the multidimensional response signal sequence includes a timestamp field, a signal type identifier field, and a binary signal data body.

[0074] Example 2: See Figure 3The dynamic test field module initializes a digital twin space composed of a computational fluid dynamics grid and a discrete element grid as the basis of the dynamic test field. This module decouples and maps the thermal radiation change signal, lattice strain propagation signal and surface potential fluctuation signal in the multidimensional response signal sequence into boundary condition input and volume force source input in the digital twin space, respectively. The transient solver of the digital twin space is run to calculate the temperature fluctuation cloud map, stress distribution cloud map and hydrogen concentration diffusion flow field map of each grid node in the digital twin space. The spatiotemporal evolution features are extracted from the temperature fluctuation cloud map, stress distribution cloud map and hydrogen concentration diffusion flow field map and compiled to generate a real-time environmental perturbation spectrum. When initializing the digital twin space, a computational fluid dynamics (CFD) mesh is created, with mesh nodes covering the geometric space of the magnesium-based solid hydrogen storage material sample. The mesh size is adaptively adjusted based on the pore topology in the microstructure feature information. A discrete element mesh is then created, which overlaps spatially with the CFD mesh. The particle parameters of the discrete element mesh are set based on the grain size distribution in the microstructure feature information. The CFD mesh and the discrete element mesh are coupled, and coupling interface conditions are set. The coupling interface conditions are defined based on the correlation between the thermal radiation change signal and the lattice strain propagation signal in the multidimensional response signal sequence. Initial boundary conditions and initial field variables are defined in the coupled mesh space. The initial boundary conditions are based on the parameter settings of the composite physical field excitation, and the initial field variables are initialized based on the microstructure feature information, thus completing the construction of the digital twin space.

[0075] In practical implementation, the dynamic test field module initializes a digital twin space composed of a superimposed computational fluid dynamics (CFD) mesh and a discrete element mesh as the base of the dynamic test field. When creating the CFD mesh, the 3D scanning model of the magnesium-based solid hydrogen storage material sample is used as the geometric boundary. The nodes of the CFD mesh cover the entire geometric space of the magnesium-based solid hydrogen storage material sample. The basic size of the CFD mesh is adaptively adjusted based on the pore topology in the microstructure feature information. The adjustment process follows the matching relationship between the pore feature length and the mesh size. The pore feature length is extracted from the pore network model obtained by the 3D topography reconstruction instrument and characterizes the equivalent diameter of the pore channels. The formula for adjusting the size of the CFD mesh is:

[0076]

[0077] in: Represents the final size of the computational fluid dynamics mesh. This represents the average pore feature length statistically obtained from the pore topology. It is a size correlation coefficient, which enables computational fluid dynamics meshes to resolve hydrogen diffusion flow at the pore scale.

[0078] In some embodiments, when creating a discrete element mesh, the boundary of the discrete element mesh completely coincides with the geometric space of the computational fluid dynamics mesh. The discrete element mesh consists of a large number of spherical particles representing magnesium-based solid hydrogen storage material grains. The radius parameter of each particle in the discrete element mesh is set based on the grain size distribution in the microstructure feature information. For example, if the grain size distribution in the microstructure feature information shows that the grain diameter is mainly distributed between 20 nanometers and 200 nanometers, the discrete element mesh will generate a group of particles with a radius in the range of 10 nanometers to 100 nanometers. The radius distribution histogram of the particle group and the statistical histogram of the grain size in the microstructure feature information are compared. Figure 1 To.

[0079] In practical implementation, the dynamic test field module couples the computational fluid dynamics mesh with the discrete element mesh. The coupling process sets the coupling interface conditions for mass, momentum, and energy exchange between the computational fluid dynamics mesh and the discrete element mesh. The specific transmission coefficient of the coupling interface conditions is defined based on the correlation between the thermal radiation change signal and the lattice strain propagation signal in the multidimensional response signal sequence. For example, the thermal radiation change signal reflects the fluctuation of the surface temperature field, and the lattice strain propagation signal reflects the propagation of internal stress waves. By analyzing the phase difference and amplitude ratio of the two signals at the same time point, the internal thermo-mechanical coupling coefficient of the material can be calculated. This coupling coefficient is directly assigned as the thermoelastic coupling parameter on the coupling interface. In practical implementation, initial boundary conditions and initial field variables are defined in the coupled mesh space. The initial boundary conditions are based on the parameter settings of the composite physical field excitation. For example, the initial temperature of the thermal field fluctuation excitation is 300 degrees Celsius, the initial load of the stress field gradient excitation is 5 MPa, and the initial magnetic field strength of the electromagnetic field pulse excitation is 0 Tesla. These values ​​are set as the initial stress states of the temperature boundary, pressure boundary, and discrete element particle system boundary on the boundary of the computational fluid dynamics mesh. The initial field variables are initialized based on the microstructure feature information. The actual phase composition ratio data in the microstructure feature information is assigned to the hydride phase concentration field of each unit in the computational fluid dynamics mesh. The pore distribution extracted from the actual three-dimensional morphology data is mapped to the initial porosity field of the computational fluid dynamics mesh, thus completing the construction of the digital twin space.

[0080] It is understandable that the dynamic test field module decouples the thermal radiation change signal, lattice strain propagation signal, and surface potential fluctuation signal from the multidimensional response signal sequence. The thermal radiation change signal is a series of two-dimensional temperature distribution images that change over time. The time-temperature curve of each pixel position in the image is extracted as the transient temperature input condition at the corresponding spatial boundary in the digital twin space. The lattice strain propagation signal is ultrasonic waveform data recorded by multiple sensor channels. The data field of the three-dimensional strain tensor inside the sample changing over time is reconstructed through an inversion algorithm. The specific implementation of the inversion algorithm is as follows: using ultrasonic waveform data synchronously recorded by multiple piezoelectric ceramic ultrasonic sensor channels arranged around the surface of the magnesium-based solid hydrogen storage material sample, the signal correlation between each sensor channel is extracted by analyzing the time difference and waveform distortion characteristics of ultrasonic waves propagating in the material; the strain tensor estimates of each grid node inside the sample are gradually corrected until the matching error between the calculated waveform data and the measured data is minimized; finally, the complete data field of the evolution of the strain tensor components of each point in the three-dimensional space inside the sample over time is reconstructed. This data field is directly mapped to the volume force source input in the digital twin space. This data field is mapped to a volume force source input in the digital twin space. The volume force source simulates the internal stress field caused by lattice distortion. The surface potential fluctuation signal is a potential time series recorded by multiple potentiometers. The two-dimensional potential distribution evolution of the sample surface is reconstructed by spatial interpolation. The surface potential distribution is related to the surface adsorption energy of hydrogen atoms, and therefore is converted into the boundary condition input of the hydrogen atom coverage of the material surface in the digital twin space.

[0081] In practice, the transient solver running the digital twin space calculates the temperature fluctuation cloud map, stress distribution cloud map, and hydrogen concentration diffusion flow field map of each grid node in the digital twin space. The transient solver adopts a coupled solution strategy, simultaneously solving the mass, momentum, and energy conservation equations in the computational fluid dynamics grid and the particle motion equations and contact mechanics equations in the discrete element grid at each time step. The computational fluid dynamics grid and the discrete element grid exchange data in real time through the coupling interface. The solution process continues until the simulation time covers the entire acquisition time of the multidimensional response signal sequence. After the solution is completed, the temperature value, stress tensor components, and hydrogen molar concentration value of all grid nodes at each output time step are output. These values ​​are organized and rendered into a sequence of temperature fluctuation cloud maps, stress distribution cloud maps, and hydrogen concentration diffusion flow field maps according to spatial coordinates. Optionally, spatiotemporal evolution features are extracted from temperature fluctuation cloud maps, stress distribution cloud maps, and hydrogen concentration diffusion flow field maps and compiled to generate a real-time environmental disturbance spectrum. The extracted spatiotemporal evolution features include calculating the trajectory coordinate sequence of the highest temperature point in the temperature fluctuation cloud map as it moves over time, calculating the decay curve of the average principal stress over time in the stress distribution cloud map, and calculating the spatial location of the maximum concentration gradient and the gradient value change curve over time in the hydrogen concentration diffusion flow field map. These trajectories, curves, and change data are aligned along the time axis and encoded into a multidimensional array data structure, which is defined as the real-time environmental disturbance spectrum.

[0082] Example 3: The cross-domain mapping training module calls the preset standard perturbation model stored in the knowledge base. The preset standard perturbation model describes the standard environmental perturbation response of magnesium-based solid hydrogen storage materials under ideal conditions. The global deviation degree and local deviation distribution map are calculated by comparing the feature values ​​of the real-time environmental perturbation spectrum and the preset standard perturbation model at the same time and spatial location. The global deviation degree is used as the feedback gain coefficient and the local deviation distribution map is used as the weight distribution basis to construct a functional mapping network from the physical signal domain to the material performance parameter domain. The internal connection weights of the functional mapping network are optimized by the backpropagation algorithm to solidify the cross-domain mapping relationship.

[0083] In practice, the cross-domain mapping training module calls the preset standard perturbation model stored in the knowledge base. The knowledge base is an associated database, and the preset standard perturbation model is stored as a data record. The preset standard perturbation model describes the standard environmental perturbation response of magnesium-based solid hydrogen storage materials under ideal conditions. Ideal conditions refer to the conditions where the ambient temperature is constant at 300 degrees Celsius, the ambient pressure is constant at 1 MPa in a hydrogen atmosphere, and there is no external stress or electromagnetic field interference. The standard environmental perturbation response is described by a multi-dimensional time series data. For example, this series data contains 3000 time points. Each time point records the standard surface temperature value, standard internal equivalent stress value, and standard body average hydrogen concentration value of the standard sample calculated under ideal conditions. These values ​​constitute the standard reference evolution trajectory.

[0084] In practice, the cross-domain mapping training module compares the feature values ​​of the real-time environmental perturbation spectrum with those of the preset standard perturbation model at the same time and spatial location. The real-time environmental perturbation spectrum is a data structure with the same time length and spatial resolution output from the dynamic test field module. The comparison process first performs time alignment and spatial coordinate registration. Then, for each registered data point, the difference between the real-time measured or simulated feature value of that point and the corresponding standard feature value in the preset standard perturbation model is calculated. An example feature value is the temperature of the center point of the sample's upper surface at 150 seconds. The standard temperature value here in the preset standard perturbation model is 573.15 Kelvin, while the corresponding real-time temperature value in the real-time environmental perturbation spectrum is 583.15 Kelvin, with a difference of 10 Kelvin. After performing this calculation on all registered data points, the squares of all differences are summed and averaged to obtain the global deviation. At the same time, the local difference at each spatial coordinate point is recorded to generate a local deviation distribution map indexed by spatial coordinates. It can be understood that the cross-domain mapping training module uses the global deviation as the feedback gain coefficient and the local deviation distribution map as the basis for weight distribution to construct a functional mapping network from the physical signal domain to the material performance parameter domain. The feedback gain coefficient is used to scale the learning step size during network training. The larger the global deviation, the larger the feedback gain coefficient, and the greater the adjustment range of the weights during training. The deviation value of each pixel position in the local deviation distribution map is normalized and used as the initial bias term for the connection weights from the corresponding input node to the first hidden node in the functional mapping network. Connections corresponding to regions with larger deviations are given higher attention in the early stage of training. In specific implementation, the constructed functional mapping network is a fully connected feedforward neural network with a three-layer structure. The nodes in the input layer correspond to the feature values ​​of the real-time environmental perturbation spectrum, including temperature, stress components, and hydrogen concentration gradient, totaling 256 nodes. The nodes in the output layer correspond to material performance parameters, including hydrogen absorption rate, hydrogen release plateau pressure, and cycle life index. The hidden layer contains 128 nodes. The network realizes the mathematical mapping from high-dimensional perturbation signals to key performance indicators, and the relationship can be expressed as:

[0085]

[0086] in: This represents the 256-dimensional feature vector of the input layer. This represents an 8-dimensional performance parameter vector of the output layer. and These represent the weight matrices from the input layer to the hidden layer and from the hidden layer to the output layer, respectively. and This represents the corresponding bias vector. and These represent the activation functions of the hidden and output layers.

[0087] In some embodiments, the internal connection weights of the functional mapping network are optimized using the backpropagation algorithm. The optimization process uses the standard performance parameters derived from a preset standard perturbation model as the target value. The backpropagation algorithm calculates the error between the network's predicted output and the target value, and uses gradient descent to adjust the weight matrix layer by layer from the output layer to the input layer. , and bias vector , The numerical values ​​are used to complete a full traversal of all training data and update the weights, which is called a training cycle. After a set number of training cycles, when the network's prediction error on the validation dataset is lower than a preset threshold, such as a root mean square error lower than 0.01, training stops. At this point, the set of weights and bias parameters fixed within the network constitutes the established cross-domain mapping relationship. It can be understood that the fixed cross-domain mapping relationship is stored as a configuration file containing all weight matrices, bias vectors, activation function types, and network structure parameters. This configuration file can be directly loaded and called by other sub-models in the cross-domain mapping training module to quickly map new real-time environmental perturbation spectra into predicted material performance parameters.

[0088] See Figure 4 This is a 3D correlation analysis diagram of the performance parameters of magnesium-based hydrogen storage materials, showing the comprehensive impact of temperature on hydrogen storage performance parameters (hydrogen absorption rate, hydrogen desorption plateau pressure, and cycle life). With increasing temperature, the hydrogen absorption rate and hydrogen desorption plateau pressure show an upward trend, while the cycle life shows a downward trend. The predicted values ​​closely match the standard values ​​(the broken line and the dashed line almost overlap), indicating that the performance prediction accuracy of the cross-domain mapping model is high. This diagram is used in the performance evaluation stage of magnesium-based hydrogen storage materials, intuitively reflecting the coupled effect of temperature on hydrogen storage performance, helping to verify the reliability of the cross-domain mapping model, and providing data support for the selection of practical application scenarios for the materials.

[0089] Example 4: The cross-domain mapping training module starts the material performance prediction model and the test field control model. The cross-domain mapping relationship is used as a shared constraint condition between the material performance prediction model and the test field control model. Under the shared constraint condition, the material performance prediction model attempts to predict the material performance degradation path under different environmental disturbances. At the same time, the test field control model attempts to generate a strengthening disturbance scheme that can maximize the exposure of the material performance degradation path. The prediction operation of the material performance prediction model and the generation operation of the test field control model are executed iteratively until the material performance prediction model can no longer accurately predict the disturbance effect generated by the test field control model. The final strengthening disturbance scheme output by the test field control model at this time is recorded as the test field control command, and the final performance degradation path output by the material performance prediction model at this time is recorded as the material state evolution command. The instruction parsing and reporting module decomposes the test field control instructions to obtain temperature control sub-instructions, pressure control sub-instructions, and atmosphere control sub-instructions for the dynamic test field. It interprets the material state evolution instructions to obtain the predicted hydrogen absorption rate curve, predicted hydrogen release plateau pressure curve, and predicted cycle life curve of the magnesium-based solid hydrogen storage material at different test stages. The temperature control sub-instructions, pressure control sub-instructions, atmosphere control instructions, and predicted hydrogen absorption rate curve, predicted hydrogen release plateau pressure curve, and predicted cycle life curve are associated, labeled, and integrated into a structured document template to form a comprehensive test report.

[0090] In practice, the cross-domain mapping training module activates the material performance prediction model and the test field control model. The cross-domain mapping relationship is loaded as a shared constraint between the two models. This constraint exists as a set of inequalities, such as limiting the predicted hydrogen desorption plateau pressure output by the material performance prediction model to within the physical range of 0.1 MPa to 10 MPa, and limiting the excitation temperature of the thermal field fluctuations generated by the test field control model to not exceed the melting point of the material. Under these shared constraints, the material performance prediction model attempts to predict the material performance degradation path under different environmental disturbances. The input to the model is the combination of disturbance parameters generated in real time by the test field control model, and the output is a sequence of predicted values ​​for the hydrogen absorption rate, hydrogen desorption plateau pressure, and cycle life index over the next five hydrogen absorption and desorption cycles. These predicted value sequences collectively constitute the material performance degradation path, which describes the trend of material performance parameters changing with the number of cycles. In some embodiments, the test field control model attempts to generate a reinforced perturbation scheme that maximizes the exposure of material performance degradation paths. The test field control model receives the performance degradation path output by the material performance prediction model and calculates an "exposure" score based on the path's shape. The exposure score quantifies the significance of material performance degradation under the current perturbation scheme. The test field control model adjusts its internal policy network parameters to generate new perturbation parameter combinations that result in a higher exposure score as the reinforced perturbation scheme. The prediction operation of the material performance prediction model and the generation operation of the test field control model are executed iteratively. Each iteration contains a complete "prediction-generation" loop. In each loop, the material performance prediction model predicts performance based on the perturbation scheme output by the test field control model in the previous round, while the test field control model generates the next round's perturbation scheme based on the degradation path predicted by the material performance prediction model in this round. The two models run alternately; see Table 1 for their input-output relationship.

[0091] Table 1: Iterative Process of Model Synchronous Adversarial Training

[0092] Iteration rounds Test field control model input (material performance degradation path) Test field control model output (enhanced perturbation scheme) Input to the material property prediction model (strengthening perturbation scheme) Output of material property prediction model (material property degradation path) 1 Initial path (default value) Option A: {Temperature: 300°C, Pressure: 3MPa, Frequency: 1Hz} Option A Path A: Predicted cycle life decreases from 100 to 95 2 Path A Option B: {Temperature: 320°C, Pressure: 4MPa, Frequency: 5Hz} Option B Path B: Predicted cycle life decreases from 100 to 88 3 Path B Option C: {Temperature: 350°C, Pressure: 5MPa, Frequency: 10Hz} Option C Path C: Predicted cycle life decreases from 100 to 80 ... ... ... ... ... N Path N-1 Option N: {Temperature: 380°C, Pressure: 6MPa, Frequency: 15Hz} Option N Path N: Predicted cycle lifetime decreases from 100 to 78

[0093] In practice, the prediction operation of the material performance prediction model and the generation operation of the test field control model are iteratively executed until the material performance prediction model can no longer accurately predict the perturbation effect generated by the test field control model. The condition for "no longer able to predict accurately" is that the similarity between the performance degradation paths predicted by the material performance prediction model for the perturbation schemes generated by the test field control model in three consecutive iterations exceeds 99%, indicating that the prediction of the material performance prediction model has become stable and can no longer distinguish the subtle differences of the newly generated perturbations by the test field control model. The final enhanced perturbation scheme output by the test field control model at this time is recorded as the test field control command. The final enhanced perturbation scheme is scheme N, which includes temperature, pressure, and electromagnetic field frequency. The final performance degradation path output by the material performance prediction model at this time is recorded as the material state evolution command. The final performance degradation path is path N, which includes hydrogen absorption rate, hydrogen desorption plateau pressure, and cycle life index. The instruction parsing and reporting module decomposes the test field control instructions to obtain temperature control sub-instructions, pressure control sub-instructions, and atmosphere control sub-instructions for the dynamic test field. The decomposition process is carried out according to the instruction encoding protocol. For example, the "temperature: 380°C" field is parsed from the final enhanced disturbance scheme "Scheme N" and formatted as a temperature control sub-instruction that can be executed by the resistance heating furnace in the dynamic test field. The "pressure: 6MPa" field is parsed and formatted as a pressure control sub-instruction that can be executed by the hydraulic loading system. The atmosphere control sub-instruction is fixed as a pure hydrogen atmosphere based on the characteristic that the material is a hydride.

[0094] In practical implementation, the instruction parsing and reporting module interprets the material state evolution instructions to obtain the predicted hydrogen absorption rate curve, predicted hydrogen release plateau pressure curve, and predicted cycle life curve of the magnesium-based solid hydrogen storage material at different test stages. The interpretation process reads the data sequence of the final performance degradation path "path N", separates the corresponding data points of hydrogen absorption rate, hydrogen release plateau pressure, and cycle life index with the number of cycles from the sequence, and uses an interpolation algorithm to generate smooth continuous curves. Optionally, the instruction parsing and reporting module associates and annotates the temperature control sub-instruction, pressure control sub-instruction, and atmosphere control sub-instruction with the predicted hydrogen absorption rate curve, predicted hydrogen release plateau pressure curve, and predicted cycle life curve. The association and annotation are achieved through a shared unique test identifier, such as the identifier "Test_N". In the structured document, the corresponding test conditions "at a temperature of 380°C, a pressure of 6MPa, and a frequency of 15Hz in a hydrogen atmosphere" are marked below each data curve. The instruction parsing and reporting module integrates temperature control sub-instructions, pressure control sub-instructions, atmosphere control sub-instructions, and predicted hydrogen absorption rate curves, predicted hydrogen release plateau pressure curves, and predicted cycle life curves into a structured document template to form a comprehensive test report. The structured document template is a predefined report format containing sections on test conditions, material performance evolution, and conclusions. The parsed instructions and curve data are filled into the corresponding section positions to generate a complete comprehensive test report document. The generation logic of the comprehensive test report can be summarized by the following formula:

[0095]

[0096] in: This represents the final generated comprehensive test report. The function process that represents report generation and integration. This represents a temperature control sub-command. This represents a pressure control sub-command. This represents the atmosphere control sub-instruction. This represents the predicted hydrogen absorption rate curve. This represents the predicted hydrogen desorption plateau pressure curve. This represents the predicted cycle life curve.

[0097] Example 5: Construction of a Preset Standard Perturbation Model. A standard magnesium-based solid hydrogen storage material sample was prepared under standard environmental conditions, and standard microstructure feature information was extracted. A standard field control instruction set was generated based on the standard microstructure feature information. A standard composite physical field excitation was applied to the standard magnesium-based solid hydrogen storage material sample. The standard composite physical field excitation included standard thermal field fluctuation excitation, standard stress field gradient excitation, and standard electromagnetic field pulse excitation. Standard response signals, including standard thermal radiation change signals, standard lattice strain propagation signals, and standard surface potential fluctuation signals, were synchronously acquired through a standard sensing array. The standard response signals were aligned in time sequence and processed to generate a standard environmental perturbation response curve. The standard environmental perturbation response curve was stored in the knowledge base as a preset standard perturbation model. When optimizing the internal connection weights of a functional mapping network using the backpropagation algorithm, the network is defined with an input layer, hidden layer, and output layer. The input layer nodes correspond to the feature values ​​of the real-time environmental perturbation spectrum, and the output layer nodes correspond to the material performance parameters. The internal connection weights are initialized with random values ​​and a learning rate parameter is set. The output value is calculated by forward propagation based on the input layer feature values ​​and the current internal connection weights. The error between the output value and the target value is calculated, and the target value is derived from the standard environmental perturbation response curve of a preset standard perturbation model. Based on the error, the internal connection weights are adjusted using the backpropagation algorithm. The adjustment amount is calculated based on the learning rate and gradient descent. The forward and backpropagation processes are repeated until the error is lower than a preset threshold, and the internal connection weights are solidified to obtain the cross-domain mapping relationship.

[0098] In the specific implementation, the construction of the pre-set standard perturbation model involves preparing standard magnesium-based solid hydrogen storage material samples under standard environmental conditions. The standard environmental conditions refer to a clean room environment with a temperature of 25 degrees Celsius, a relative humidity of 50%, and an atmospheric pressure of 101.325 kPa. Under this environment, the standard magnesium-based solid hydrogen storage material samples are prepared using the same physical vapor deposition and controllable sintering process as the sample feature extraction module. The formulation and process parameters of the standard magnesium-based solid hydrogen storage material samples adopt the magnesium-5wt% nickel system widely reported in the literature. After sintering, a bulk standard sample with uniform composition and dense structure is obtained. In practice, standard microstructure features of standard magnesium-based solid hydrogen storage material samples are extracted. The extraction process uses a micro-area composition analyzer and a three-dimensional morphology reconstruction instrument that are exactly the same as the sample feature extraction module to scan and obtain the actual micro-area composition distribution data and actual three-dimensional morphology data of the standard samples. For example, in the standard microstructure features, the mass percentage of magnesium is 95.0%, the mass percentage of nickel is 5.0%, the mass percentage of oxygen is less than 0.1%, the porosity is 0.5%, the grain size has a single-peak distribution, and the average grain size is 50 micrometers.

[0099] In some embodiments, a standard field control instruction set is generated based on standard microstructure feature information. The generation logic of the standard field control instruction set is consistent with that of the non-standard sample, but the input microstructure feature information comes from the standard sample. For example, the thermal field fluctuation excitation defined by the standard field control instruction set is a linear heating from 25 degrees Celsius to 350 degrees Celsius at a heating rate of 5 degrees Celsius per minute, the stress field gradient excitation is kept at 0 MPa, and the electromagnetic field pulse excitation is kept at 0 Tesla. A standard composite physical field excitation is applied to a standard magnesium-based solid hydrogen storage material sample. The standard composite physical field excitation includes standard thermal field fluctuation excitation, standard stress field gradient excitation, and standard electromagnetic field pulse excitation. The application process is carried out in a multi-field coupling generator to ensure that, except for the thermal field changing according to the command, the stress field and electromagnetic field remain in a zero excitation state. It is understandable that standard response signals are synchronously acquired through a standard sensor array. The model, specifications, and layout of the standard sensor array are exactly the same as those used in the in-situ data injection module. The acquired signals include standard thermal radiation change signals, standard lattice strain propagation signals, and standard surface potential fluctuation signals. The standard thermal radiation change signal records the temperature change curve of the standard sample surface under standard thermal field fluctuation excitation. The standard lattice strain propagation signal records the weak strain signal caused by thermal expansion of the standard sample in a near-stress-free state. The standard surface potential fluctuation signal records the change of the surface potential of the standard sample as the temperature increases.

[0100] In practice, the standard response signals are time-series aligned and processed to generate standard environmental disturbance response curves. Time-series alignment uses the start time of the standard thermal field fluctuation excitation as the zero point. The data streams of the standard thermal radiation change signal, the standard lattice strain propagation signal, and the standard surface potential fluctuation signal are synchronized. The processing includes filtering and noise reduction, normalization, and feature extraction of the original signals. For example, the temperature-time curve of the sample center point is extracted from the standard thermal radiation change signal; the longitudinal wave velocity versus time curve is extracted from the standard lattice strain propagation signal; and the average potential versus time curve is extracted from the standard surface potential fluctuation signal. These three curves are merged into a multi-channel time series data set, which is defined as the standard environmental disturbance response curve. The standard environmental disturbance response curve is stored as a preset standard disturbance model in a knowledge base. The knowledge base assigns a unique identifier to each preset standard disturbance model and associates it with the corresponding standard magnesium-based solid hydrogen storage material system formulation, standard microstructure feature information summary, and the original data file of the standard environmental disturbance response curve.

[0101] In some embodiments, when optimizing the internal connection weights of the functional mapping network using the backpropagation algorithm, the input layer, hidden layer, and output layer of the functional mapping network are defined. The input layer nodes correspond to the feature values ​​of the real-time environmental perturbation spectrum. For example, 256 nodes correspond to the time series statistical features of physical quantities such as temperature, stress, and hydrogen concentration, respectively. The output layer nodes correspond to material performance parameters. For example, 8 nodes correspond to hydrogen absorption capacity, hydrogen release plateau pressure, and cycle stability index, respectively. A single hidden layer is set, containing 128 nodes, each of which uses the Sigmoid activation function. The internal connection weights are initialized with random values, and a learning rate parameter is set. The internal connection weights include the weight matrix from the input layer to the hidden layer and the weight matrix from the hidden layer to the output layer. Each element of the weight matrix is ​​randomly assigned within the interval [-0.1, 0.1]. The learning rate parameter η is set to 0.01. Forward propagation calculates the output value, which is based on the input layer feature values ​​and the current internal connection weights. The calculation process involves left-multiplying the input feature vector by the weight matrix from the input layer to the hidden layer plus a bias, transforming it through the hidden layer activation function, left-multiplying it again by the weight matrix from the hidden layer to the output layer plus a bias to obtain the original value of the output layer, and finally obtaining the final output value through the linear activation function of the output layer.

[0102] It is understandable that the error between the output value and the target value is calculated. The target value is derived from the standard environmental perturbation response curve of the preset standard perturbation model. The derivation process uses an independent calibration function, which takes the standard environmental perturbation response curve as input and outputs the corresponding standard material performance parameter value. For example, for the standard magnesium-5wt% nickel system, its standard hydrogen absorption capacity at 350 degrees Celsius and 3 MPa hydrogen pressure is 6.5wt%, which is used as a target value. The error is calculated using the mean square error function. Based on the error, the internal connection weights are adjusted using the backpropagation algorithm. The adjustment amount is calculated based on the learning rate and gradient descent method. The weight update follows the following formula:

[0103]

[0104] in: Represents the updated weights. This represents the weight before the update. Represents the learning rate parameter. This represents the mean square error between the output value and the target value. The gradient representing the error relative to the weights is calculated layer by layer from the output layer to the input layer using the chain rule. The forward and backward propagation processes are repeated until the error falls below a preset threshold, which is set to a total mean squared error below 0.001. The training process terminates when the preset threshold is reached or the maximum number of iterations (10,000) is reached. The internal connection weights are solidified to obtain the cross-domain mapping relationship. The final stable weight matrix from the input layer to the hidden layer, the weight matrix from the hidden layer to the output layer, the bias vectors of each layer, and the network structure parameters are saved as an immutable data file; this data file represents the solidified cross-domain mapping relationship.

[0105] See Figure 5 This is a line graph showing the error evolution during the training of the cross-domain mapping network, illustrating the change in mean squared error (MSE) with the number of iterations. The error decreases extremely rapidly: only in the early stages of iteration (approximately the first 100 iterations), the MSE drops rapidly from 250,000 to near 0. The errors on the training and validation sets highly overlap, indicating that the model has not overfitted and has strong generalization ability. The error is far below the threshold: after training, the error stabilizes near 0, meeting the preset error threshold (0.001). This graph was used to evaluate the training effect of the cross-domain mapping model for magnesium-based hydrogen storage materials, visually reflecting the model's convergence speed and generalization ability, and verifying the training effectiveness of the cross-domain mapping network.

[0106] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0107] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-parameter comprehensive testing system for a magnesium-based solid-state hydrogen storage material, characterized in that, The system includes: The sample feature extraction module is used to prepare magnesium-based solid hydrogen storage material samples with preset microstructure features and extract the microstructure feature information of the magnesium-based solid hydrogen storage material samples. The in-situ data injection module is used to load the microstructure feature information and apply composite physical field excitation to the magnesium-based solid hydrogen storage material sample, and simultaneously acquire multidimensional response signal sequences. The dynamic test field module is used to construct a dynamic test field, import the multidimensional response signal sequence into the dynamic test field, and drive the generation of a real-time environmental disturbance spectrum corresponding to the multidimensional response signal sequence. The cross-domain mapping training module is used to establish a cross-domain mapping relationship based on the deviation between the real-time environmental perturbation spectrum and the preset standard perturbation model, and to perform synchronous adversarial training of the model based on the cross-domain mapping relationship, thereby generating test field control commands and material state evolution commands. The cross-domain mapping training module establishes a cross-domain mapping relationship based on the deviation between the real-time environmental perturbation spectrum and the preset standard perturbation model, including: The preset standard perturbation model stored in the knowledge base is invoked, which describes the standard environmental perturbation response of magnesium-based solid hydrogen storage materials under ideal conditions; By comparing the feature values ​​of the real-time environmental disturbance spectrum with those of the preset standard disturbance model at the same time and spatial location, the global deviation and local deviation distribution map are calculated. Using the global deviation as the feedback gain coefficient and the local deviation distribution map as the weight distribution basis, a functional mapping network from the physical signal domain to the material performance parameter domain is constructed. The internal connection weights of the functional mapping network are optimized by backpropagation algorithm, and the cross-domain mapping relationship is then solidified. The cross-domain mapping training module performs synchronous adversarial training of the model based on the cross-domain mapping relationship, generating test field control commands and material state evolution commands, including: The material performance prediction model and the test field control model are activated, and the cross-domain mapping relationship is used as a shared constraint condition between the material performance prediction model and the test field control model. Under the shared constraints, the material performance prediction model attempts to predict the material performance degradation path under different environmental disturbances, while the test field control model attempts to generate a strengthening disturbance scheme that can maximize the exposure of the material performance degradation path. The prediction operation of the material property prediction model and the generation operation of the test field control model are executed iteratively until the material property prediction model can no longer accurately predict the disturbance effect generated by the test field control model. The final enhanced perturbation scheme output by the test field control model at this time is recorded as the test field control command, and the final performance degradation path output by the material performance prediction model at this time is recorded as the material state evolution command. The instruction parsing and report module is used to parse the test field control instructions and material state evolution instructions to generate a comprehensive test report.

2. The multi-parameter comprehensive testing system of a magnesium-based solid-state hydrogen storage material according to claim 1, characterized in that, The sample feature extraction module includes: Based on the target hydrogen storage performance indicators, the grain size distribution, phase composition ratio and pore topology of magnesium-based solid hydrogen storage materials were designed in reverse. The magnesium-based solid hydrogen storage material sample was prepared by physical vapor deposition and controlled sintering processes according to the grain size distribution, phase composition ratio and pore topology. The magnesium-based solid hydrogen storage material sample was scanned using a micro-area composition analyzer and a three-dimensional morphology reconstruction instrument to obtain the actual micro-area composition distribution data and actual three-dimensional morphology data of the sample. The actual micro-region component distribution data and the actual three-dimensional morphology data are combined to form the microstructure feature information.

3. The multi-parameter comprehensive testing system of a magnesium-based solid hydrogen storage material according to claim 2, characterized in that, The in-situ data injection module includes: The microstructure feature information is encoded into a machine-readable field control instruction set; The field control instruction set is input to the multi-field coupling generator, which synchronously outputs thermal field fluctuation excitation, stress field gradient excitation and electromagnetic field pulse excitation according to the field control instruction set. The thermal field fluctuation excitation, stress field gradient excitation and electromagnetic field pulse excitation together constitute the composite physical field excitation. By using a high-sensitivity sensor array surrounding the surface of the magnesium-based solid hydrogen storage material sample, the thermal radiation change signal, lattice strain propagation signal and surface potential fluctuation signal of the magnesium-based solid hydrogen storage material sample under the excitation of the composite physical field are simultaneously captured. The thermal radiation change signal, lattice strain propagation signal, and surface potential fluctuation signal are aligned and packaged in time sequence to form the multidimensional response signal sequence.

4. The multi-parameter comprehensive testing system for magnesium-based solid hydrogen storage materials as described in claim 3, characterized in that, The dynamic test field module includes: Initialize a digital twin space consisting of a computational fluid dynamics mesh and a discrete element mesh, as the basis of the dynamic test field; The thermal radiation change signal, lattice strain propagation signal, and surface potential fluctuation signal in the multidimensional response signal sequence are decoupled and mapped to the boundary condition input and volume force source input in the digital twin space, respectively. Run the transient solver of the digital twin space to calculate the temperature fluctuation cloud map, stress distribution cloud map and hydrogen concentration diffusion flow field map of each grid node in the digital twin space; Spatiotemporal evolution features are extracted from the temperature fluctuation cloud map, stress distribution cloud map, and hydrogen concentration diffusion flow field map, and compiled to generate the real-time environmental disturbance spectrum.

5. The multi-parameter comprehensive testing system for magnesium-based solid hydrogen storage materials as described in claim 1, characterized in that, The instruction parsing and reporting module parses the test field control instructions and material state evolution instructions to generate a comprehensive test report, including: By disassembling the test field control commands, we obtain temperature control sub-commands, pressure control sub-commands, and atmosphere control sub-commands for the dynamic test field. Interpreting the material state evolution instructions yields the predicted hydrogen absorption rate curve, predicted hydrogen desorption plateau pressure curve, and predicted cycle life curve of the magnesium-based solid hydrogen storage material at different test stages. The temperature control sub-instruction, pressure control sub-instruction, and atmosphere control sub-instruction are associated and labeled with the predicted hydrogen absorption rate curve, predicted hydrogen release plateau pressure curve, and predicted cycle life curve, and integrated into a structured document template to form the comprehensive test report.

6. The multi-parameter comprehensive testing system for magnesium-based solid hydrogen storage materials as described in claim 4, characterized in that, The initialization of a digital twin space consisting of a computational fluid dynamics mesh and a discrete element mesh includes: A computational fluid dynamics mesh is created, with mesh nodes covering the geometric space of the magnesium-based solid hydrogen storage material sample. The mesh size is adaptively adjusted based on the pore topology in the microstructure feature information. A discrete element mesh is created, which overlaps spatially with the computational fluid dynamics mesh. The particle parameters of the discrete element mesh are set based on the grain size distribution in the microstructure feature information. The computational fluid dynamics mesh is coupled with the discrete element mesh, and coupling interface conditions are set. The coupling interface conditions are defined based on the correlation between the thermal radiation change signal and the lattice strain propagation signal in the multidimensional response signal sequence. In the coupled mesh space, initial boundary conditions and initial field variables are defined. The initial boundary conditions are based on the parameter settings of the composite physical field excitation, and the initial field variables are initialized based on the microstructure feature information, thus completing the construction of the digital twin space.

7. The multi-parameter comprehensive testing system for magnesium-based solid hydrogen storage materials as described in claim 1, characterized in that, The construction of the preset standard perturbation model includes: Standard magnesium-based solid hydrogen storage material samples were prepared under standard environmental conditions, and standard microstructure feature information was extracted. Based on the standard microstructure feature information, a standard field control instruction set is generated, and a standard composite physical field excitation is applied to the standard magnesium-based solid hydrogen storage material sample. The standard composite physical field excitation includes standard thermal field fluctuation excitation, standard stress field gradient excitation, and standard electromagnetic field pulse excitation. Standard response signals, including standard thermal radiation change signals, standard lattice strain propagation signals, and standard surface potential fluctuation signals, are synchronously acquired through a standard sensor array. The standard response signals are aligned and processed according to time sequence to generate a standard environmental disturbance response curve, which is stored in the knowledge base as the preset standard disturbance model.

8. The multi-parameter comprehensive testing system for magnesium-based solid hydrogen storage materials as described in claim 1, characterized in that, The optimization of the internal connection weights of the functional mapping network using the backpropagation algorithm includes: Define the input layer, hidden layer, and output layer of the functional mapping network. The input layer nodes correspond to the eigenvalues ​​of the real-time environmental perturbation spectrum, and the output layer nodes correspond to the material performance parameters. Initialize the inner connection weights to random values ​​and set the learning rate parameter; The forward propagation computes the output value, which is calculated based on the input layer feature values ​​and the current internal connection weights; Calculate the error between the output value and the target value, wherein the target value is derived based on the standard environmental disturbance response curve of the preset standard disturbance model; The internal connection weights are adjusted based on the error using the backpropagation algorithm, with the adjustment amount calculated based on the learning rate and gradient descent. Repeat the forward and backward propagation process until the error is lower than a preset threshold, and solidify the internal connection weights to obtain the cross-domain mapping relationship.

Images