Strength testing method and device for aluminum alloy components

By combining thermoelectric characteristic measurement and multimodal neural network model, the problems of insufficient accuracy and safety hazards in aluminum alloy component detection are solved, and lossless, fast, and full-field intensity detection is achieved, providing intuitive intensity distribution information, and improving Test efficiency and accuracy.

CN118655012BActive Publication Date: 2024-11-05DONGGUAN JSGS AUTOMATION EQUIP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411127173.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-11-05
Estimated Expiration
2044-08-16

AI Technical Summary

Technical Problem

The existing strength detection methods for aluminum alloy components have limitations, making it difficult to achieve lossless, accurate and applicable detection, especially in samples with uneven shapes and thicknesses. Traditional methods such as ultrasonic and X-ray diffraction have insufficient accuracy and safety risks. .

Method used

By obtaining a comprehensive database based on multiple sets of aluminum alloy samples, applying temperature gradients and mechanical stresses using the thermoelectric characteristic measurement system, combining the multimodal neural network model for intensity prediction, and generating visual results of intensity distribution.

Benefits of technology

It realizes lossless, fast, full-field and accurate detection of aluminum alloy components, overcomes the accuracy and safety risks of traditional methods, provides intuitive strength distribution information, and improves detection efficiency and accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118655012B_ABST
    Figure CN118655012B_ABST
Patent Text Reader

Abstract

The present invention discloses a strength detection method and device for aluminum alloy components, the method comprising: obtaining a comprehensive database based on multiple groups of aluminum alloy samples; optimizing the parameters of a thermoelectric characteristic measurement system according to the thermoelectric characteristic parameters in the comprehensive database; applying a temperature gradient and mechanical stress to the aluminum alloy component to be tested using the thermoelectric characteristic measurement system with optimized parameters, obtaining an original data set containing the temperature distribution and thermoelectric potential distribution of the aluminum alloy component to be tested; analyzing the original data set and extracting characteristic parameters to obtain a characteristic data set; constructing a multimodal neural network model based on the comprehensive database, optimizing the parameters of the multimodal neural network model, and obtaining a comprehensive prediction model; inputting the characteristic data set into the comprehensive prediction model, calculating the strength values ​​of each measurement point of the aluminum alloy component to be tested, and generating a visualization result of the strength distribution of the aluminum alloy component to be tested. The technical solution of the present invention can detect the strength of aluminum alloy components non-destructively.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of strength detection of aluminum alloy products, and in particular to a strength detection method and device for an aluminum alloy component. Background Art

[0002] Aluminum alloys are widely used in aerospace, automobile manufacturing, and construction engineering due to their excellent strength-to-weight ratio and good corrosion resistance. However, the strength of aluminum alloy components may change during service due to various factors, such as thermal cycles, mechanical stress, and environmental corrosion, thus affecting the performance and safety of the components. Therefore, accurate and non-destructive testing of the strength of aluminum alloy components is of great significance for ensuring the safety of the structure, extending its service life, and optimizing maintenance strategies.

[0003] At present, the commonly used aluminum alloy strength detection methods include hardness testing, ultrasonic testing and X-ray diffraction. However, these methods have some limitations. For example, hardness testing can only provide surface strength information, which is difficult to reflect the strength distribution of the entire component; although ultrasonic testing and X-ray diffraction can be used for non-destructive testing, they still face some challenges in application. Traditional ultrasonic and X-ray methods usually rely on fixed physical models, lack adaptive learning capabilities, and it is difficult to continuously improve the detection accuracy as data accumulates. In addition, ultrasonic testing may face the problem of complex sound wave propagation paths in complex-shaped components, affecting the accuracy of the test results. X-ray diffraction may produce diffraction patterns that are difficult to interpret in samples with uneven thickness, increasing the difficulty of data analysis. In addition, X-ray diffraction involves ionizing radiation, and special protective measures are required during operation, which may cause inconvenience in some application scenarios. Therefore, it is of great research significance and application value to develop a strength detection method for aluminum alloy components that can overcome these limitations and achieve safety, accuracy, and wide applicability. Summary of the invention

[0004] The main purpose of the present invention is to provide a method for non-destructively testing the strength of aluminum alloy components.

[0005] A first aspect of the present invention provides a strength detection method for an aluminum alloy component, the strength detection method for an aluminum alloy component comprising:

[0006] Acquire a comprehensive database based on multiple groups of aluminum alloy samples, wherein the comprehensive database contains microstructure parameters, mechanical property parameters, and thermoelectric property parameters of each group of aluminum alloy samples;

[0007] Optimizing the parameters of the thermoelectric characteristic measurement system according to the thermoelectric characteristic parameters in the comprehensive database;

[0008] Applying a temperature gradient and mechanical stress to the aluminum alloy component to be measured by using a thermoelectric characteristic measurement system with optimized parameters, and measuring the temperature and thermoelectric potential of each measuring point on the aluminum alloy component to be measured, so as to obtain an original data set including temperature distribution and thermoelectric potential distribution;

[0009] Analyzing the original data set and extracting characteristic parameters to obtain a characteristic data set;

[0010] Based on the comprehensive database, a multimodal neural network model is constructed, and parameters of the multimodal neural network model are optimized to obtain a comprehensive prediction model;

[0011] The characteristic data set is input into the comprehensive prediction model, the strength value of each measuring point of the aluminum alloy component to be tested is calculated, and a visualization result of the strength distribution of the aluminum alloy component to be tested is generated.

[0012] In some embodiments, constructing a multimodal neural network model based on the comprehensive database includes:

[0013] extracting thermoelectric characteristic data, microstructural data, and mechanical property data from the comprehensive database;

[0014] Performing a time series analysis on the thermoelectric characteristic data to determine the input dimension and time step of the long short-term memory network in the thermoelectric characteristic processing branch, performing an image feature analysis on the microstructure data to determine the convolution kernel size and network depth of the convolutional neural network in the microstructure processing branch, and performing a statistical analysis on the mechanical property data to determine the number of hidden layers and the number of neurons in each layer of the fully connected network in the mechanical property processing branch;

[0015] According to the above analysis results, the thermoelectric characteristics processing branch, microstructure processing branch and mechanical properties processing branch are constructed respectively;

[0016] Designing an attention mechanism module for fusing output features of the thermoelectric characteristic processing branch, the microstructure processing branch, and the mechanical property processing branch;

[0017] Constructing an intensity prediction module for mapping the features fused by the attention mechanism module into intensity values;

[0018] Connecting the thermoelectric property processing branch, the microstructure processing branch, the mechanical property processing branch, the attention mechanism module and the strength prediction module to form an initial multimodal neural network model;

[0019] Using the data in the comprehensive database to train the initial multimodal neural network model, and using a cross-validation method to evaluate the performance of the initial multimodal neural network model;

[0020] Based on the cross-validation results, adjusting the structure and parameters of the initial multimodal neural network model;

[0021] Performing adversarial training on the adjusted initial multimodal neural network model to generate adversarial samples and add them to the comprehensive database;

[0022] The initial multimodal neural network model is retrained using the comprehensive database after adding adversarial samples to obtain a multimodal neural network model.

[0023] In some embodiments, the step of optimizing the parameters of the multimodal neural network model to obtain a comprehensive prediction model includes:

[0024] Performing sensitivity analysis on the multimodal neural network model to determine the key parameters that have the greatest impact on model performance;

[0025] Based on the sensitivity analysis results, design parameter optimization strategies, including learning rate scheduling and regularization methods;

[0026] Using the parameter optimization strategy, adjusting the weights and biases of the multimodal neural network model;

[0027] Using data enhancement technology, transforming and expanding the samples in the comprehensive database to generate an enhanced data set;

[0028] Fine-tuning the adjusted multimodal neural network model using the enhanced dataset;

[0029] Performing performance tests on the fine-tuned multimodal neural network model under different working conditions to evaluate its prediction accuracy and stability;

[0030] According to the performance test results, the model with the best performance is selected as the comprehensive prediction model.

[0031] In some embodiments, the step of inputting the characteristic data set into the comprehensive prediction model, calculating the strength value of each measuring point of the aluminum alloy component to be tested, and generating a visualization result of the strength distribution of the aluminum alloy component to be tested includes:

[0032] Normalizing the feature data set to obtain a normalized feature data set;

[0033] Inputting the normalized characteristic data set into the comprehensive prediction model to obtain the initial strength prediction value of each measuring point of the aluminum alloy component to be tested;

[0034] Performing inverse normalization processing on the initial intensity prediction value to obtain an intensity prediction value of the original scale;

[0035] Using the Kriging interpolation method to spatially interpolate the intensity prediction values ​​of the original scale to obtain a continuous intensity distribution field;

[0036] Performing gradient calculation on the continuous intensity distribution field to obtain an intensity gradient field;

[0037] Refining the grid according to the intensity gradient field to obtain a refined prediction grid;

[0038] Performing intensity re-prediction and interpolation on the refined prediction grid to obtain an optimized intensity distribution field;

[0039] Dividing the optimized intensity distribution field into contour lines to obtain a plurality of intensity intervals;

[0040] Performing color mapping on the multiple intensity intervals to obtain a two-dimensional intensity distribution heat map;

[0041] A three-dimensional surface model is constructed based on the two-dimensional strength distribution heat map to obtain a visualization result of the three-dimensional strength distribution of the aluminum alloy component to be tested.

[0042] In some embodiments, the step of obtaining a comprehensive database based on multiple groups of aluminum alloy samples includes:

[0043] A group of aluminum alloy samples are observed by scanning electron microscopy, and the observation results are subjected to image segmentation and morphological analysis to extract the grain size distribution and the volume fraction of the second phase of the sample as microstructure parameters;

[0044] Performing a tensile test on the aluminum alloy sample, fitting the data obtained from the tensile test to obtain a stress-strain curve, and extracting yield strength, tensile strength and elongation as mechanical property parameters through the stress-strain curve analysis;

[0045] Applying a temperature gradient to the aluminum alloy sample and measuring the thermoelectric potential, calculating the Seebeck coefficient by the ratio of the thermoelectric potential to the temperature gradient, measuring the resistivity of the aluminum alloy sample, calculating the electrical conductivity by the inverse of the resistivity, measuring the thermal diffusion coefficient, specific heat capacity and density of the aluminum alloy sample, calculating the thermal conductivity by the product of the thermal diffusion coefficient, specific heat capacity and density, and using the Seebeck coefficient, the electrical conductivity and the thermal conductivity as thermoelectric characteristic parameters;

[0046] Recording microstructure parameters, mechanical property parameters and thermoelectric characteristic parameters obtained from the aluminum alloy sample;

[0047] Repeat the above steps for the remaining aluminum alloy samples until the tests of all groups of aluminum alloy samples are completed;

[0048] The recorded data of all groups were normalized by minimum-maximum method to obtain a comprehensive database.

[0049] In some embodiments, the step of optimizing parameters of a thermoelectric characteristic measurement system according to the thermoelectric characteristic parameters in the comprehensive database includes:

[0050] Performing principal component analysis on the thermoelectric characteristic parameters in the comprehensive database to extract the main influencing factors;

[0051] Constructing a parameter optimization objective function of a thermoelectric characteristic measurement system according to the main influencing factors;

[0052] Solving the objective function using an optimization algorithm to obtain an initial optimization parameter set;

[0053] Based on the initial optimized parameter set, construct a digital twin model of the thermoelectric characteristics measurement system;

[0054] simulating the digital twin model to generate simulation data;

[0055] Using the simulation data to train a machine learning model to establish a mapping relationship between system parameters and measurement accuracy;

[0056] Based on the mapping relationship, an optimization algorithm is used to fine-tune the parameters to obtain a final optimized parameter set;

[0057] The final optimized parameter set is applied to an actual thermoelectric characteristic measurement system to complete system parameter optimization.

[0058] In some embodiments, applying a temperature gradient to the aluminum alloy component to be measured using a thermoelectric property measurement system with optimized parameters includes:

[0059] Placing the aluminum alloy component to be tested horizontally on the thermal insulation test platform of the thermoelectric property measurement system;

[0060] Directly contact the heat source and the cold source of the thermoelectric property measurement system with the two opposite end surfaces of the aluminum alloy component to be measured;

[0061] Arrange N temperature sensors at equal intervals on the upper surface of the aluminum alloy component to be tested in the direction from the heat source to the cold source, where N≥5;

[0062] Controlling the temperature of the heat source to gradually increase to a preset high temperature value at a rate not exceeding 5°C per minute, while maintaining the temperature of the cold source at a preset low temperature value;

[0063] The following steps are executed cyclically until a stable state is reached: a. The temperature data of the surface of the aluminum alloy component to be tested is collected once every 10 seconds by the N temperature sensors; b. The temperature of the heat source and the cold source is fine-tuned according to the feedback data of the temperature sensor until the difference between the temperature data collected twice in succession by the same temperature sensor is less than 1°C, and this condition is simultaneously established for all the N temperature sensors;

[0064] When the above stable state is reached, continue to maintain this state for at least 10 minutes, and then record the temperature data of N temperature sensors at this time { , i=1, 2, ..., N};

[0065] The recorded temperature data are analyzed using the least squares method. } Perform linear fitting and obtain the fitting equation T=ax+b, where T is the temperature value, x is the distance of the temperature sensor from the heat source to the cold source, a is the temperature gradient, and b is the theoretical temperature value at x=0;

[0066] If the relative error between the fitted temperature gradient value |a| and the preset value is within the range of ±5%, and the goodness of fit R²>0.95, the temperature gradient application is confirmed to be successful; otherwise, the entire temperature gradient application process is restarted.

[0067] In some embodiments, it is characterized in that mechanical stress is applied to the aluminum alloy component to be tested, and the temperature and thermoelectric potential of each measuring point on the aluminum alloy component to be tested are measured simultaneously to obtain an original data set containing temperature distribution and thermoelectric potential distribution, including:

[0068] Installing a piezoelectric ceramic driver and a force sensor at both ends of the aluminum alloy component to be tested;

[0069] Applying an initial prestress to the aluminum alloy component to be tested, wherein the prestress value is 3% of the yield strength of the material;

[0070] Controlling the piezoelectric ceramic driver to increase stress at a rate of 0.5 MPa per second until a preset target stress is reached;

[0071] During the stress increase process, the temperature data of the surface of the aluminum alloy component to be tested is synchronously collected every 5 seconds through the N temperature sensors;

[0072] Synchronously measuring the thermoelectric potential between adjacent points of the N temperature sensors to obtain thermoelectric potential data;

[0073] Perform interpolation processing on the collected temperature data to obtain continuous temperature distribution data;

[0074] Performing interpolation processing on the collected thermoelectric potential data to obtain continuous thermoelectric potential distribution data;

[0075] The continuous temperature distribution data and thermoelectric potential distribution data are combined to form an original data set including temperature distribution and thermoelectric potential distribution.

[0076] In some embodiments, the analyzing the original data set and extracting characteristic parameters to obtain the characteristic data set includes:

[0077] Performing spatial gradient calculation on the temperature distribution data in the original data set to obtain temperature gradient characteristics;

[0078] Performing spatial gradient calculation on the thermoelectric potential distribution data in the original data set to obtain thermoelectric potential gradient characteristics;

[0079] Calculate the Seebeck coefficient distribution according to the temperature gradient characteristics and the thermoelectric potential gradient characteristics to obtain the Seebeck coefficient characteristics;

[0080] Performing time series analysis on the temperature distribution data to obtain temperature fluctuation characteristics;

[0081] Performing time series analysis on the thermoelectric potential distribution data to obtain thermoelectric potential fluctuation characteristics;

[0082] Calculate the local thermal conductivity according to the temperature distribution data and the thermoelectric potential distribution data to obtain a thermal conductivity characteristic;

[0083] The temperature gradient feature, the thermoelectric potential gradient feature, the Seebeck coefficient feature, the temperature fluctuation feature, the thermoelectric potential fluctuation feature and the thermal conductivity feature are combined to obtain a feature data set.

[0084] A second aspect of the present invention provides a strength detection device for an aluminum alloy component, comprising:

[0085] A data acquisition module, used for acquiring a comprehensive database based on multiple groups of aluminum alloy samples, wherein the comprehensive database contains microstructure parameters, mechanical property parameters and thermoelectric characteristic parameters of each group of aluminum alloy samples;

[0086] An optimization unit, used for optimizing parameters of a thermoelectric characteristic measurement system according to the thermoelectric characteristic parameters in the comprehensive database;

[0087] Thermoelectric characteristic measurement module, used for applying temperature gradient and mechanical stress to the aluminum alloy component to be measured by using the thermoelectric characteristic measurement system with optimized parameters, and measuring the temperature and thermoelectric potential of each measuring point on the aluminum alloy component to be measured, so as to obtain an original data set including temperature distribution and thermoelectric potential distribution;

[0088] A data processing unit, used for analyzing the original data set and extracting characteristic parameters to obtain a characteristic data set;

[0089] A model building module, used to build a multimodal neural network model based on the comprehensive database, optimize the parameters of the multimodal neural network model, and obtain a comprehensive prediction model;

[0090] The strength prediction unit is used to input the characteristic data set into the comprehensive prediction model, calculate the strength value of each measuring point of the aluminum alloy component to be tested, and generate a visualization result of the strength distribution of the aluminum alloy component to be tested.

[0091] The technical solution provided by the embodiments of the present application has at least the following advantages:

[0092] First, aluminum alloy has good thermal conductivity and electrical conductivity, which makes it efficient and feasible to apply temperature gradient and measure thermoelectric potential in the method of the present invention. The high thermal conductivity of aluminum alloy ensures that the temperature gradient can be established quickly and evenly in the component, while its good electrical conductivity ensures the accuracy of the generation and measurement of thermoelectric potential. Based on these characteristics, the present method obtains a high-quality original data set containing temperature distribution and thermoelectric potential distribution by simultaneously applying temperature gradient and mechanical stress, and measuring the temperature and thermoelectric potential at each point on the component. This multi-physics field coupled measurement method overcomes the limitations of traditional methods such as ultrasonic testing and X-ray diffraction, which are difficult to simultaneously obtain the thermal and mechanical properties of materials.

[0093] Secondly, the thermoelectric properties of aluminum alloys are sensitive to their microstructure and mechanical properties. The thermoelectric coefficient (such as Seebeck coefficient) and electrical conductivity of aluminum alloys are closely related to their microstructural characteristics such as grain size, second phase distribution, dislocation density, etc., and these microstructural characteristics also directly affect the mechanical properties of aluminum alloys, including strength. Therefore, by measuring the thermoelectric properties of aluminum alloys, its microstructural information and mechanical property changes can be indirectly obtained. This provides reliable input data and physical basis for the multimodal neural network model of the present invention.

[0094] Furthermore, aluminum alloys are prone to thermal-mechanical coupling effects during use. Due to the large thermal expansion coefficient of aluminum alloys, complex stress states will be generated inside them under the combined effects of temperature changes and mechanical loads, which will affect the thermoelectric properties of the material. The method of the present invention can effectively capture this thermal-mechanical coupling effect by applying temperature gradients and mechanical stresses at the same time, thereby more comprehensively evaluating the actual service status of aluminum alloy components. Compared with hardness tests that can only provide surface strength information, this method can reflect the strength distribution of the entire component through full-field measurement and data analysis.

[0095] Another significant advantage of the method of the present invention is that it uses a multimodal neural network model based on a comprehensive database for strength prediction. First, by analyzing the original data set and extracting characteristic parameters, a characteristic data set that can fully reflect the material properties is obtained. Then, based on a comprehensive database containing microstructure parameters, mechanical performance parameters, and thermoelectric characteristic parameters, a multimodal neural network model is constructed and optimized. This method overcomes the limitations of traditional methods that rely on fixed physical models and has strong adaptive learning capabilities. With the continuous accumulation of data, the model can be continuously optimized and the detection accuracy can be continuously improved, which is an advantage that traditional ultrasonic and X-ray methods do not have.

[0096] In addition, this method can effectively integrate multi-scale and multi-source information such as microstructure, thermoelectric properties and macroscopic mechanical properties, and achieve a comprehensive and accurate prediction of the strength of aluminum alloy components. This multi-modal and multi-scale information fusion capability greatly improves the accuracy and reliability of strength prediction and overcomes the limitations of traditional methods in processing and analyzing multi-source and heterogeneous data.

[0097] Finally, this method provides engineers and researchers with intuitive and comprehensive strength distribution information by generating visualization results of the strength distribution of components, which helps to better evaluate the performance and life of components. At the same time, this method is based on surface measurement and data analysis, and has relatively low requirements on the shape of components. It solves the problem of complex sound wave propagation paths faced by ultrasonic testing in complex-shaped components, and the problem of difficult-to-interpret diffraction patterns that may be produced by X-ray diffraction in samples with uneven thickness. In addition, this method does not involve ionizing radiation, is safer to operate, and does not require special protective measures, overcoming the safety hazards of X-ray diffraction methods in practical applications.

[0098] In general, the method of the present invention makes full use of the material properties of aluminum alloys, organically combines its thermoelectric properties, microstructure and mechanical properties, and realizes non-destructive, rapid, full-field and accurate detection of the strength of aluminum alloy components through data analysis and machine learning technology. This provides strong technical support for the performance evaluation, life prediction and maintenance decision-making of aluminum alloy components, significantly improves the efficiency and accuracy of aluminum alloy component strength detection, and has important practical application value. This method is not only applicable to aluminum alloys, but also has the potential to be extended to the strength detection of other metal materials, showing broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0100] Figure 1 A flow chart of an embodiment of a method for detecting strength of an aluminum alloy component in an embodiment of the present invention;

[0101] Figure 2 Schematic diagram of an embodiment of a strength detection device for an aluminum alloy component in an embodiment of the present invention.

[0102] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0103] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0104] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back...), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0105] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, "and / or" in the full text includes three solutions. Taking A and / or B as an example, it includes technical solution A, technical solution B, and technical solution that satisfies both A and B. In addition, the technical solutions between the various embodiments can be combined with each other, which must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0106] An embodiment of the present application provides a method for detecting the strength of an aluminum alloy component. Figure 1 A flow chart of a strength testing method for an aluminum alloy component provided in one embodiment of the present application.

[0107] refer to Figure 1 , obtaining a comprehensive database based on multiple groups of aluminum alloy samples, wherein the comprehensive database contains microstructure parameters, mechanical property parameters and thermoelectric characteristic parameters of each group of aluminum alloy samples;

[0108] In this embodiment, the acquisition of a comprehensive database based on multiple groups of aluminum alloy samples includes: performing scanning electron microscopy observation on a group of aluminum alloy samples, and performing image segmentation and morphological analysis on the observation results to extract the grain size distribution and the second phase volume fraction of the sample as microstructure parameters; performing a tensile test on the aluminum alloy samples, fitting the data obtained from the tensile test and obtaining a stress-strain curve, and extracting the yield strength, tensile strength and elongation as mechanical property parameters through the stress-strain curve analysis; applying a temperature gradient to the aluminum alloy samples and measuring the thermoelectric potential, and calculating the ratio of the thermoelectric potential to the temperature gradient to obtain the mechanical property parameters. The method comprises the following steps: measuring the Seebeck coefficient, measuring the resistivity of the aluminum alloy sample, calculating the electrical conductivity by the inverse of the resistivity, measuring the thermal diffusion coefficient, specific heat capacity and density of the aluminum alloy sample, calculating the thermal conductivity by the product of the thermal diffusion coefficient, specific heat capacity and density, and taking the Seebeck coefficient, the electrical conductivity and the thermal conductivity as thermoelectric characteristic parameters; recording the microstructure parameters, mechanical property parameters and thermoelectric characteristic parameters obtained for the aluminum alloy sample; repeating the above steps for the remaining aluminum alloy samples until the testing of all groups of aluminum alloy samples is completed; performing minimum-maximum normalization processing on the recorded data of all groups to obtain a comprehensive database.

[0109] Specifically, a scanning electron microscope is used to observe the aluminum alloy sample at high resolution to obtain clear images of its microstructure. These images are processed by professional image analysis software, and image segmentation technology is used to separate grain boundaries and second-phase particles from the matrix. For example, the threshold segmentation method is used to separate areas of different gray levels, or the edge detection algorithm is used to identify grain boundaries. Subsequently, morphological analysis is performed, such as using operations such as opening and closing operations to eliminate noise and optimize boundaries. Through these processes, the size of each grain can be accurately measured, and the grain size distribution can be statistically obtained. At the same time, the area ratio occupied by the second-phase particles is calculated to obtain the second-phase volume fraction. These data together constitute the microstructural parameters of the sample.

[0110] Next, the aluminum alloy samples were subjected to a standard tensile test. During the test, the sample was fixed on a tensile machine and stretched at a constant rate until it broke, while recording the load and displacement data. These raw data were converted into stress-strain data points by data processing software. Then, curve fitting techniques such as polynomial fitting or piecewise fitting were used to obtain a continuous stress-strain curve. From this curve, key mechanical performance parameters can be directly read or calculated. For example, the yield strength is determined by the 0.2% offset method, the tensile strength is the highest point on the curve, and the elongation can be calculated from the strain value corresponding to the fracture point. These parameters together constitute the mechanical performance indicators of the sample.

[0111] In order to obtain the thermoelectric characteristic parameters, a test device is designed. This device can apply a controllable temperature gradient at both ends of the sample, and accurately measure the temperature difference and the generated thermoelectric potential at both ends of the sample. The Seebeck coefficient is directly calculated by the ratio of the thermoelectric potential to the temperature gradient. For the measurement of conductivity, the four-probe method is used. This method applies a known current through two external probes and measures the voltage drop through two internal probes. Combined with the geometric dimensions of the sample, the resistivity can be calculated, and its reciprocal is the conductivity. The determination of thermal conductivity is relatively complicated, and it is necessary to measure the thermal diffusion coefficient, specific heat capacity and density separately. The thermal diffusion coefficient is measured by the laser flash method, the specific heat capacity is measured by differential scanning calorimetry, and the density is obtained by the Archimedean method. Multiplying these three parameters gives the thermal conductivity. These thermoelectric characteristic parameters jointly reflect the thermal-electric conversion ability and thermal conductivity characteristics of the material.

[0112] It should also be added that in addition to the methods mentioned above, a variety of complementary testing techniques can also be used to obtain the thermoelectric characteristic parameters of aluminum alloy samples. For the determination of the Seebeck coefficient, the steady-state method and the transient method can also be used. The steady-state method calculates the Seebeck coefficient by maintaining a constant temperature difference at both ends of the sample and measuring the generated thermoelectric potential, while the transient law calculates by rapidly heating one end of the sample and measuring the instantaneous temperature difference and thermoelectric potential. The combined use of these methods can improve the accuracy and reliability of the measurement. In addition to the four-probe method, the Van der Pauw Hall effect measurement method can also be used for the measurement of conductivity. This method determines the carrier concentration and mobility by applying a magnetic field and measuring the Hall voltage, and then calculates the conductivity. For large-size samples, the eddy current method is also an effective non-contact conductivity measurement method. This method infers the conductivity by measuring the intensity of the induced eddy current, which is suitable for rapid detection and large-area scanning. The determination methods of thermal conductivity also include the heat flow method and the laser flash method. The heat flow method calculates the thermal conductivity by applying a constant temperature difference at both ends of the sample and measuring the steady-state heat flow. This method is suitable for samples of various shapes. The laser flash method uses a pulsed laser to heat one side of the sample and measures the temperature response of the other side to calculate the thermal diffusion coefficient, and combines the density and specific heat capacity to obtain the thermal conductivity. This method has fast measurement speed and high accuracy. In addition to differential scanning calorimetry, adiabatic calorimetry and relaxation calorimetry can also be used to measure specific heat capacity. Adiabatic calorimetry calculates specific heat capacity by measuring the temperature change of the sample under completely adiabatic conditions. It has high accuracy but complex operation. The relaxation calorimetry method calculates specific heat capacity by measuring the temperature change rate of the sample during heating or cooling, which is suitable for small samples and fast measurements. In addition to the Archimedean method, the gas pycnometer method and X-ray diffraction method can also be used for density determination. The gas pycnometer method calculates density by measuring the volume of gas displaced by the sample, which is suitable for porous materials. The X-ray diffraction method can calculate the theoretical density by measuring the lattice parameters. These diverse test methods provide a wealth of options for obtaining accurate thermoelectric characteristic parameters. The combined use of different methods can verify each other and improve the reliability of the measurement results. At the same time, according to the characteristics of the sample, the measurement accuracy requirements and the available equipment, the most suitable test method can be flexibly selected. This comprehensive and flexible testing strategy ensures the accurate acquisition of thermoelectric characteristic parameters and provides a reliable data basis for subsequent strength prediction of aluminum alloy components.

[0113] Next, for each group of aluminum alloy samples, all the above test steps need to be repeated to ensure the integrity and consistency of the data. After completing the testing of all samples, an original data set containing microstructural parameters, mechanical property parameters, and thermoelectric property parameters can be obtained. In order to enable data of different dimensions and ranges to be compared and analyzed on the same scale, all parameters are subjected to minimum-maximum normalization. This processing method maps the value of each parameter to a range between 0 and 1, maintaining the relative distribution characteristics of the data while eliminating the influence of dimension. For example, for a certain parameter X, its normalized value X' can be obtained by the formula X'=(X- ) / ( - ) is calculated, where and are the minimum and maximum values ​​of the parameter in all samples respectively. By processing all parameters in this way, a standardized comprehensive database can be finally obtained, which provides high-quality basic data for subsequent model training and analysis. This comprehensive database not only contains rich material information, but also ensures the consistency and comparability of the data, laying a solid foundation for accurately predicting the strength distribution of aluminum alloy components.

[0114] Continue to refer Figure 1 , optimizing the parameters of the thermoelectric characteristic measurement system according to the thermoelectric characteristic parameters in the comprehensive database;

[0115] In this embodiment, the parameters of the thermoelectric characteristics measurement system are optimized according to the thermoelectric characteristics parameters in the comprehensive database, including: performing principal component analysis on the thermoelectric characteristics parameters in the comprehensive database to extract the main influencing factors; constructing a parameter optimization objective function of the thermoelectric characteristics measurement system according to the main influencing factors; solving the objective function by using an optimization algorithm to obtain an initial optimization parameter set; constructing a digital twin model of the thermoelectric characteristics measurement system based on the initial optimization parameter set; simulating the digital twin model to generate simulation data; using the simulation data to train a machine learning model to establish a mapping relationship between system parameters and measurement accuracy; based on the mapping relationship, using an optimization algorithm to fine-tune the parameters to obtain a final optimization parameter set; applying the final optimization parameter set to an actual thermoelectric characteristics measurement system to complete system parameter optimization.

[0116] Specifically, principal component analysis (PCA) is performed on the thermoelectric characteristic parameters in the comprehensive database. PCA is a powerful statistical method that can effectively compress multidimensional data into a low-dimensional space while retaining the main variation information of the data. In this process, thermoelectric characteristic parameters such as Seebeck coefficient, electrical conductivity and thermal conductivity can be used as input variables. By calculating the covariance matrix of these variables and performing eigenvalue decomposition on them, a series of orthogonal principal components can be obtained. Usually, the first few principal components with a cumulative explained variance of 85% or 90% are selected as the main influencing factors. The advantage of this method is that it can effectively reduce the data dimension and eliminate the correlation between variables, thereby simplifying the subsequent optimization process.

[0117] According to the extracted main influencing factors, the parameter optimization objective function of the thermoelectric characteristic measurement system can be constructed. This objective function is usually a multi-objective optimization problem, which needs to consider multiple aspects such as measurement accuracy, measurement speed and system stability at the same time. For example, the sum of squares of measurement errors can be used as the accuracy term, the measurement time as the speed term, and the degree of fluctuation of system parameters as the stability term, and then these sub-goals can be combined by weighted summation. The advantage of this comprehensive objective function is that it can comprehensively consider all aspects of system performance and avoid the problems that may be caused by the optimization of a single indicator.

[0118] Next, the constructed objective function is solved using an optimization algorithm to obtain the initial set of optimized parameters. Considering the complexity and nonlinear characteristics of the parameter optimization problem of the thermoelectric characteristic measurement system, heuristic algorithms such as particle swarm optimization (PSO) or genetic algorithm (GA) can be selected. Such algorithms can quickly find approximate optimal solutions in large-scale search spaces without easily falling into local optimality. During the optimization process, the algorithm will continuously adjust various parameters of the system, such as the size of the temperature gradient, the measurement time interval, the signal sampling frequency, etc., until the objective function reaches the optimal state or meets the preset termination conditions.

[0119] Based on the initial set of optimized parameters obtained, the next step is to build a digital twin model of the thermoelectric characteristics measurement system. A digital twin is a high-fidelity virtual representation of a physical system in the digital world, which can reflect the state and behavior of the physical system in real time. During the construction process, professional simulation software, such as COMSOL Multiphysics or ANSYS, can be used to establish a complete numerical model containing geometric models, physical field equations, and boundary conditions. This digital twin model not only includes the physical structure of the thermoelectric characteristics measurement system, but also needs to consider multi-physical field coupling phenomena such as heat conduction, current distribution, and thermoelectric effect. The advantage of the digital twin model is that a large number of tests and optimizations can be performed in a virtual environment without the need for time-consuming and labor-intensive repeated experiments on the actual system.

[0120] After the construction is completed, the digital twin model can be fully simulated to generate a large amount of simulation data. These simulations can cover various possible working conditions and parameter combinations, including different temperature gradients, mechanical stress levels, material property changes, etc. Through simulation, the response characteristics of the system under these different conditions can be obtained, such as temperature distribution, thermoelectric potential distribution, measurement errors, etc. This digital twin-based simulation method can quickly and cost-effectively obtain a large amount of high-quality data, providing rich training samples for subsequent machine learning model training.

[0121] The generated simulation data can be used to train machine learning models and establish a mapping relationship between system parameters and measurement accuracy. Advanced machine learning algorithms such as random forests, support vector machines, or deep neural networks can be selected here. During the training process, system parameters are used as input features and measurement accuracy is used as output labels. Through training with a large amount of data, machine learning models can learn the complex nonlinear relationship of how system parameters affect measurement accuracy. This data-driven approach has stronger adaptability and generalization capabilities than traditional physical models, and can better handle uncertainty and nonlinear problems in complex systems.

[0122] Based on the established mapping relationship, a more sophisticated optimization algorithm can be used to fine-tune the parameters to obtain the final set of optimized parameters. This step can use efficient global optimization algorithms such as Bayesian optimization, which can find a solution close to the global optimal solution in a smaller number of iterations. The advantage of Bayesian optimization is that it can use the information of previous evaluation points to guide subsequent searches, thereby finding a better solution in a smaller number of function evaluations.

[0123] Finally, the final optimized parameter set is applied to the actual thermoelectric characteristics measurement system to complete the optimization of system parameters. Some actual testing and fine-tuning may be required in this process to ensure the effectiveness of the optimization results in the actual system. Through this series of optimization steps, the performance of the thermoelectric characteristics measurement system can be significantly improved, including improved measurement accuracy, shortened measurement time, and enhanced system stability. This data-driven and digital twin-based optimization method can not only make full use of existing data and knowledge, but also adapt to the dynamic changes of the system, providing a more reliable and efficient basic support for the strength detection of aluminum alloy components.

[0124] Continue to refer Figure 1 , applying a temperature gradient and mechanical stress to the aluminum alloy component to be measured using a thermoelectric characteristic measurement system with optimized parameters, and measuring the temperature and thermoelectric potential of each measuring point on the aluminum alloy component to be measured, to obtain an original data set including temperature distribution and thermoelectric potential distribution;

[0125] In this embodiment, the method of applying a temperature gradient to the aluminum alloy component to be tested by using the thermoelectric property measurement system with optimized parameters includes: placing the aluminum alloy component to be tested horizontally on an insulated test platform of the thermoelectric property measurement system; directly contacting the heat source and the cold source of the thermoelectric property measurement system with the opposite end surfaces of the aluminum alloy component to be tested respectively; arranging N temperature sensors at equal intervals on the upper surface of the aluminum alloy component to be tested in the direction from the heat source to the cold source, wherein N≥5; controlling the temperature of the heat source to gradually increase to a preset high temperature value at a rate not exceeding 5°C per minute, while maintaining the temperature of the cold source at a preset low temperature value; cyclically executing the following steps until a stable state is reached: a. collecting temperature data of the surface of the aluminum alloy component to be tested once every 10 seconds through the N temperature sensors; b. fine-tuning the temperature of the heat source and the cold source according to the feedback data of the temperature sensor until the difference between the temperature data of the same temperature sensor in two consecutive collections is less than 1°C, and this condition is simultaneously true for all N temperature sensors; when the above-mentioned stable state is reached, the state is maintained for at least 10 minutes, and then the temperature data of the N temperature sensors at this time are recorded. , i=1, 2, ..., N}; Use the least squares method to calculate the recorded temperature data { }Perform linear fitting and obtain the fitting equation T=ax+b, where T is the temperature value, x is the distance of the temperature sensor along the direction from the heat source to the cold source, a is the temperature gradient, and b is the theoretical temperature value at x=0; if the relative error between the fitted temperature gradient value |a| and the preset value is within the range of ±5%, and the goodness of fit R²>0.95, it is confirmed that the temperature gradient is applied successfully; otherwise, restart the entire temperature gradient application process.

[0126] Specifically, the design of this adiabatic test platform is crucial. It can effectively isolate the thermal interference of the external environment and ensure the accuracy of temperature control during the measurement process. Next, the heat source and cold source of the thermoelectric property measurement system are directly contacted with the two opposite ends of the aluminum alloy component to be tested. This direct contact method can minimize the contact thermal resistance during heat transfer and improve the efficiency and accuracy of temperature control.

[0127] On the upper surface of the aluminum alloy component to be tested, N temperature sensors are arranged at equal intervals from the heat source to the cold source, where N is not less than 5. This arrangement can fully capture the temperature distribution on the surface of the component and provide sufficient spatial resolution. For example, for a 50cm long aluminum alloy plate, a temperature sensor can be arranged every 10cm, for a total of 5 sensors, which can cover the entire test area without increasing the complexity of the system due to too many sensors.

[0128] Temperature control is a key link in the entire measurement process. The temperature of the heat source is controlled to gradually increase to a preset high temperature value at a rate of no more than 5°C per minute, while the temperature of the cold source is maintained at a preset low temperature value. This slow heating method helps to establish a stable temperature gradient inside the aluminum alloy component and avoid thermal stress or local temperature unevenness caused by rapid heating. For example, if the goal is to establish a temperature difference of 100°C, it may take about 20 minutes for the heat source end to rise from room temperature to 120°C. During this process, the cold source end can be maintained at 20°C.

[0129] During the temperature control process, the system will loop through the steps of temperature data collection and adjustment until a stable state is reached. Specifically, the temperature data of the surface of the aluminum alloy component to be tested is collected every 10 seconds through N temperature sensors. This frequent sampling interval can capture small changes in temperature in a timely manner, providing a basis for precise control. According to the feedback data from the temperature sensor, the system will fine-tune the temperature of the heat source and the cold source. The goal of the adjustment is to make the difference in temperature data of the same temperature sensor in two consecutive collections less than 1°C, and this condition needs to be true for all N temperature sensors at the same time. This strict stability requirement ensures that the temperature distribution on the entire component reaches a highly stable state.

[0130] When the system reaches the above stable state, it needs to maintain this state for at least 10 minutes. This additional stabilization time is very important, as it can ensure that the temperature distribution inside the aluminum alloy component is also stable and eliminate possible transient thermal effects. After the 10 minutes, the system records the temperature data of the N temperature sensors at this time { , i=1, 2, ..., N}. This set of data represents the final stable temperature distribution state.

[0131] In order to quantify the temperature gradient, the system uses the least squares method to perform a linear fit on the recorded temperature data {Ti}, and obtains the fitting equation T=ax+b. In this equation, T is the temperature value, x is the distance of the temperature sensor along the direction from the heat source to the cold source, a is the temperature gradient, b is the theoretical temperature value at x=0, and the absolute value of the coefficient a |a| is the temperature gradient value. The use of the least squares method can minimize the fitting error and provide the best linear approximation.

[0132] Finally, the system will evaluate the fitting results. If the relative error between the fitted temperature gradient value |a| and the preset value is within the range of ±5%, and the goodness of fit R² is greater than 0.95, the temperature gradient is confirmed to be applied successfully. The setting of these two conditions ensures both the accuracy of the temperature gradient and the linearity of the temperature distribution. R² greater than 0.95 means that at least 95% of the temperature changes can be explained by the linear model, indicating a high degree of linearity in the temperature distribution. If any of these two conditions is not met, the system will restart the entire temperature gradient application process.

[0133] For ease of understanding, let's take a specific experimental scenario as an example. Assume that there is an aluminum alloy plate with a length of 50 cm and a stable temperature gradient is established on it. This can be done as follows: Place the aluminum alloy plate horizontally on an insulated test platform, set a heat source (such as 120°C) at one end of the plate and a cold source (such as 20°C) at the other end, and place 5 temperature sensors (N=5) at equal intervals along the length of the plate, one every 10 cm. After the stabilization phase, the following temperature readings are obtained: T1 = 110 °C (x = 0 cm, close to the heat source), T2 = 90 °C (x = 10 cm), T3 = 70 °C (x = 20 cm), T4 = 50 °C (x = 30 cm), T5 = 30 °C (x = 40 cm, close to the cold source). The least squares method is used to linearly fit these data and the equation is obtained: T = -2x + 110, temperature gradient |a| = 2 °C / cm = 20 °C / 10 cm, assuming the preset temperature gradient value is 19 °C / 10 cm, relative error = |(20-19) / 19| × 100% = 5.26%, assuming the calculated R² = 0.98, although R²>0.95 meets the linearity requirement, the relative error is slightly higher than 5%. Therefore, it is necessary to readjust and repeat the temperature gradient application process.

[0134] This rigorous temperature gradient establishment and verification process has several advantages: first, it can establish accurate and stable temperature gradients in aluminum alloy components, providing a reliable basis for subsequent thermoelectric property measurements; second, through multi-point temperature measurement and data fitting, it is possible to fully understand the temperature distribution of the components, avoiding the errors that may be caused by single-point measurement; third, strict stability and linearity requirements ensure the consistency and repeatability of the measurement conditions, which is conducive to obtaining high-quality experimental data; finally, this process is automated and adaptive, and can be adjusted according to actual conditions, improving the efficiency and success rate of the experiment.

[0135] In this embodiment, mechanical stress is applied to the aluminum alloy component to be tested, and the temperature and thermoelectric potential of each measuring point on the aluminum alloy component to be tested are measured simultaneously to obtain an original data set containing temperature distribution and thermoelectric potential distribution, including: installing a piezoelectric ceramic driver and a force sensor at both ends of the aluminum alloy component to be tested; applying an initial prestress to the aluminum alloy component to be tested, and the prestress value is 3% of the material yield strength; controlling the piezoelectric ceramic driver to increase the stress at a rate of 0.5 MPa per second until a preset target stress is reached; during the stress increase process, the temperature data of the surface of the aluminum alloy component to be tested is synchronously collected every 5 seconds through the N temperature sensors; the thermoelectric potential between adjacent points of the N temperature sensors is synchronously measured to obtain thermoelectric potential data; interpolation processing is performed on the collected temperature data to obtain continuous temperature distribution data; interpolation processing is performed on the collected thermoelectric potential data to obtain continuous thermoelectric potential distribution data; the continuous temperature distribution data and thermoelectric potential distribution data are combined to form an original data set containing temperature distribution and thermoelectric potential distribution.

[0136] Specifically, the selection of piezoelectric ceramic drivers has the advantages of precise control and fast response, and can achieve micron-level displacement control, thereby accurately applying stress. Force sensors are used to monitor the applied stress in real time to ensure the accuracy of stress control. This configuration allows the system to accurately measure stress values ​​while applying stress, providing reliable mechanical parameters for subsequent data analysis.

[0137] Next, an initial prestress is applied to the aluminum alloy component to be tested, and the prestress value is set to 3% of the material's yield strength. There are two main considerations for selecting this prestress value: one is to ensure that the component is within the elastic deformation range to avoid permanent deformation; the other is to eliminate the tiny gaps or initial stress unevenness that may occur during the installation of the component, so that the subsequent stress application is more accurate and consistent. For example, if the yield strength of a certain aluminum alloy is 300MPa, the initial prestress will be set to 9MPa. This prestress application process needs to be carried out slowly to ensure that all parts of the component are evenly stressed.

[0138] After the initial prestress is applied, the piezoelectric ceramic actuator is controlled to increase the stress at a rate of 0.5 MPa per second until the preset target stress is reached. This slow loading rate design helps to avoid stress concentration and dynamic effects while giving the material sufficient time to respond to external loads. For example, if the target stress is 50 MPa, the loading process from the initial prestress to the target stress will last about 82 seconds. This relatively slow process is conducive to capturing the thermoelectric response characteristics of the material at different stress levels.

[0139] During the entire process of stress increase, N temperature sensors synchronously collect temperature data on the surface of the aluminum alloy component to be tested every 5 seconds. This high-frequency data acquisition can record in detail the dynamic process of temperature changes with stress. At the same time, the system also measures the thermoelectric potential between adjacent points of N temperature sensors to obtain thermoelectric potential data. This method of synchronously measuring temperature and thermoelectric potential can directly reflect the thermoelectric coupling effect of the material under different stress states, providing an important basis for subsequent strength analysis.

[0140] After obtaining the original data, the collected temperature data is interpolated to obtain continuous temperature distribution data. Common interpolation methods include linear interpolation, spline interpolation, etc. For example, the use of cubic spline interpolation can provide a smooth temperature distribution curve while ensuring data continuity. The advantage of this processing method is that it can infer the continuous temperature field of the entire component surface from discrete measurement points, thereby improving the spatial resolution of the data.

[0141] Similarly, the collected thermoelectric potential data is also interpolated to obtain continuous thermoelectric potential distribution data. The interpolation of thermoelectric potential data requires special attention to maintaining physical meaning, and a conformal interpolation method is usually used to ensure that the interpolation result conforms to the physical laws of the thermoelectric effect. This processed continuous thermoelectric potential distribution data can reflect the potential gradient inside the material and provide a basis for analyzing the thermoelectric effect.

[0142] Finally, the continuous temperature distribution data and thermoelectric potential distribution data are combined to form an original data set containing temperature distribution and thermoelectric potential distribution. This comprehensive data set contains the complete thermoelectric response information of the material under different stress levels. The data combination process needs to ensure the correspondence between temperature data and thermoelectric potential data in time and space, which is usually achieved by synchronizing timestamps and spatial coordinates.

[0143] This detailed data collection and processing method has multiple advantages: first, it provides the complete thermoelectric response process of the material during stress loading, which can capture transient effects and nonlinear behaviors; second, high-frequency data collection and fine interpolation processing ensure high temporal and spatial resolution of the data, which is conducive to discovering subtle changes and local features; third, the combination of temperature and thermoelectric potential data provides a basis for subsequent multi-physics field analysis, making it possible to fully consider the thermal-mechanical-electrical coupling effect; finally, the high-quality, multi-dimensional data set generated by this method provides a rich source of information for subsequent machine learning model training, which helps to improve the prediction accuracy and generalization ability of the model. Through this systematic data collection and processing method, a solid experimental data foundation is provided for the strength testing of aluminum alloy components.

[0144] Continue to refer Figure 1 , analyzing the original data set and extracting characteristic parameters to obtain a characteristic data set;

[0145] In this embodiment, the analysis and feature parameter extraction of the original data set to obtain a feature data set includes: performing spatial gradient calculation on the temperature distribution data in the original data set to obtain a temperature gradient feature; performing spatial gradient calculation on the thermoelectric potential distribution data in the original data set to obtain a thermoelectric potential gradient feature; calculating the Seebeck coefficient distribution according to the temperature gradient feature and the thermoelectric potential gradient feature to obtain a Seebeck coefficient feature; performing time series analysis on the temperature distribution data to obtain a temperature fluctuation feature; performing time series analysis on the thermoelectric potential distribution data to obtain a thermoelectric potential fluctuation feature; calculating the local thermal conductivity according to the temperature distribution data and the thermoelectric potential distribution data to obtain a thermal conductivity feature; and combining the temperature gradient feature, the thermoelectric potential gradient feature, the Seebeck coefficient feature, the temperature fluctuation feature, the thermoelectric potential fluctuation feature and the thermal conductivity feature to obtain a feature data set.

[0146] Specifically, first, the spatial gradient of the temperature distribution data in the original data set is calculated to obtain the temperature gradient characteristics. This process is performed using the finite difference method or spline interpolation method. For example, for a one-dimensional case, the central difference formula ∂T / ∂x≈(T(x+h)-T(xh)) / (2h) can be used to calculate the temperature gradient of each point, where h is the spatial step size. The temperature gradient characteristics reflect the direction and intensity of heat transfer in the material, which is crucial for understanding the thermal conduction behavior of the material.

[0147] Next, a similar spatial gradient calculation is performed on the thermoelectric potential distribution data in the original data set to obtain the thermoelectric potential gradient characteristics. The thermoelectric potential gradient characteristics represent the distribution of the electric field in the material and are key parameters for analyzing carrier motion. The advantage of these two gradient calculation steps is that they can convert the original distribution data into more physically meaningful gradient information, which helps to reveal the transport process inside the material.

[0148] The Seebeck coefficient distribution is calculated based on the obtained temperature gradient characteristics and thermoelectric potential gradient characteristics to obtain the Seebeck coefficient characteristics. The calculation formula of the Seebeck coefficient is S=-ΔV / ΔT, where ΔV is the thermoelectric potential difference and ΔT is the temperature difference. For example, for an aluminum alloy sample with a length of 10 cm, if the temperature difference between the two ends is 50K and the measured thermoelectric potential difference is 500μV, the average Seebeck coefficient of the sample is 10μV / K. The Seebeck coefficient characteristics directly reflect the thermoelectric conversion capability of the material and are an important indicator for evaluating the thermoelectric performance of the material.

[0149] Next, the temperature distribution data is analyzed in time series to obtain the temperature fluctuation characteristics. This step can use methods such as Fourier transform or wavelet analysis to decompose the temperature change over time into components of different frequencies. For example, the spectrum of temperature fluctuation can be obtained by fast Fourier transform (FFT), so as to identify the main fluctuation period and amplitude. Similarly, the thermoelectric potential distribution data is analyzed in time series to obtain the thermoelectric potential fluctuation characteristics. The advantage of these two time series analysis steps is that they can capture the response characteristics of the material under dynamic conditions, providing a basis for subsequent dynamic behavior prediction.

[0150] According to the temperature distribution data and the thermoelectric potential distribution data, the local thermal conductivity is calculated to obtain the thermal conductivity characteristics. The calculation of thermal conductivity is based on Fourier's heat conduction law, k=-q / ∇T, where q is the heat flux density and ∇T is the temperature gradient. In actual calculations, the finite element method or finite difference method can be used to solve the heat conduction equation to obtain the local thermal conductivity distribution. The thermal conductivity characteristics reflect the ability of the material to conduct heat and are an important parameter for evaluating the thermal performance of the material.

[0151] Finally, the temperature gradient feature, thermoelectric potential gradient feature, Seebeck coefficient feature, temperature fluctuation feature, thermoelectric potential fluctuation feature and thermal conductivity feature are combined to obtain a comprehensive feature data set. This combination process needs to consider the scale and unit of each feature, and is usually standardized, such as using the Z-score standardization method, to make different features comparable.

[0152] This comprehensive feature extraction and combination method has multiple advantages: first, it covers the thermal, electrical, and thermoelectric coupling properties of the material, providing a multi-angle description of the material properties; second, it includes static features (such as gradients) and dynamic features (such as fluctuations), which can fully reflect the behavior of the material under different conditions; third, by calculating high-order features (such as Seebeck coefficient and thermal conductivity), it provides more information directly related to the material properties; finally, this multi-dimensional feature set provides rich input for subsequent machine learning models, which helps to improve the model's prediction accuracy and generalization ability.

[0153] Through this systematic data analysis and feature extraction method, the original experimental data is converted into a set of features with rich physical meaning, laying a solid data foundation for accurately predicting the strength distribution of aluminum alloy components. This method is not only suitable for evaluating material properties under static conditions, but also provides the possibility to study the behavior of materials under dynamic loads, showing its broad prospects in material science research and engineering applications.

[0154] Continue to refer Figure 1 , based on the comprehensive database, construct a multimodal neural network model, optimize the parameters of the multimodal neural network model, and obtain a comprehensive prediction model;

[0155] In this embodiment, the multimodal neural network model is constructed based on the comprehensive database, including: extracting thermoelectric characteristic data, microstructure data and mechanical property data from the comprehensive database; performing time series analysis on the thermoelectric characteristic data to determine the input dimension and time step of the long short-term memory network in the thermoelectric characteristic processing branch, performing image feature analysis on the microstructure data to determine the convolution kernel size and network depth of the convolutional neural network in the microstructure processing branch, performing statistical analysis on the mechanical property data to determine the number of hidden layers and the number of neurons in each layer of the fully connected network in the mechanical property processing branch; according to the above analysis results, respectively constructing a thermoelectric characteristic processing branch, a microstructure processing branch and a mechanical property processing branch; designing an attention mechanism module for fusing the thermoelectric characteristic processing branch, the microstructure processing branch and the mechanical property processing branch. output features of the branch; construct an intensity prediction module for mapping the features fused by the attention mechanism module into intensity values; connect the thermoelectric characteristic processing branch, the microstructure processing branch, the mechanical property processing branch, the attention mechanism module and the intensity prediction module to form an initial multimodal neural network model; use the data in the comprehensive database to train the initial multimodal neural network model, and use a cross-validation method to evaluate the performance of the initial multimodal neural network model; based on the cross-validation results, adjust the structure and parameters of the initial multimodal neural network model; perform adversarial training on the adjusted initial multimodal neural network model, generate adversarial samples and add them to the comprehensive database; use the comprehensive database after adding the adversarial samples to retrain the initial multimodal neural network model to obtain a multimodal neural network model.

[0156] Specifically, building a multimodal neural network model is a core step in the strength testing method of aluminum alloy components. This process begins with extracting thermoelectric characteristic data, microstructure data, and mechanical property data from a comprehensive database. These three types of data represent different characteristics of the material and require different processing methods. First, the thermoelectric characteristic data is subjected to time series analysis. This step aims to determine the input dimension and time step of the long short-term memory network (LSTM) in the thermoelectric characteristic processing branch. For example, if the thermoelectric characteristic data contains time series of Seebeck coefficient, electrical conductivity, and thermal conductivity, and each characteristic is sampled 10 times per second for 100 seconds, then the input dimension will be 3 (corresponding to three characteristics) and the time step may be set to 10 (that is, 1 second of data is input each time). This setting can capture short-term fluctuations and long-term trends in thermoelectric properties, which helps the model learn the behavior of materials at different time scales.

[0157] For microstructure data, image feature analysis is performed to determine the convolution kernel size and network depth of the convolutional neural network (CNN) in the microstructure processing branch. For example, if the size of the microstructure image is 224×224 pixels, a 3×3 or 5×5 convolution kernel may be selected, and the network depth may be set to 5 to 10 layers. This configuration can effectively extract multi-scale features of the microstructure, from fine grain boundaries to large-scale phase distributions.

[0158] The mechanical properties data are statistically analyzed to determine the number of hidden layers and neurons in each layer of the fully connected network in the mechanical properties processing branch. For example, if the mechanical properties data contains five parameters such as yield strength, tensile strength, and elongation, a three-layer fully connected network may be designed, with 64, 32, and 16 neurons in each layer, respectively. This structure is capable of learning complex nonlinear relationships between mechanical property parameters.

[0159] According to the above analysis results, the thermoelectric characteristics processing branch, microstructure processing branch and mechanical properties processing branch are constructed respectively. The advantage of this branch structure is that it can use the most suitable neural network structure for different types of data and give full play to the advantages of various network structures.

[0160] Next, an attention mechanism module is designed to fuse the output features of the thermoelectric characteristic processing branch, the microstructure processing branch, and the mechanical property processing branch. The introduction of the attention mechanism allows the model to dynamically adjust the importance of different features. For example, in some cases, more attention may be paid to microstructure features, while in other cases, more attention may be paid to thermoelectric properties. This flexibility enables the model to adapt to different material states and environmental conditions.

[0161] Construct a strength prediction module to map the features fused by the attention mechanism module to strength values. This module is usually a fully connected network whose output is the predicted strength value. Connect the thermoelectric property processing branch, microstructure processing branch, mechanical property processing branch, attention mechanism module and strength prediction module to form an initial multimodal neural network model. The advantage of this multimodal structure is that it can comprehensively consider the multi-faceted characteristics of the material and provide a more comprehensive and accurate strength prediction.

[0162] The initial multimodal neural network model is trained using the data in the comprehensive database, and the performance of the initial multimodal neural network model is evaluated using the cross-validation method. For example, a 5-fold cross-validation can be used to divide the data set into 5 parts, using 4 parts as training sets and 1 part as validation set each time, and repeating 5 times. This method can effectively evaluate the generalization ability of the model and avoid overfitting.

[0163] Based on the cross-validation results, adjust the structure and parameters of the initial multimodal neural network model. This process may include changing the number of network layers, adjusting the learning rate, modifying the batch size, etc. The purpose of the adjustment is to improve the prediction accuracy and generalization ability of the model.

[0164] The adjusted initial multimodal neural network model is trained adversarially to generate adversarial samples and add them to the comprehensive database. The core idea of ​​adversarial training is to generate difficult-to-classify samples by adding small perturbations, thereby improving the robustness of the model. For example, small noise can be added to the microstructure image, or the thermoelectric characteristic data can be slightly deformed. These adversarial samples can help the model learn more robust feature representations.

[0165] Finally, the initial multimodal neural network model is retrained using the comprehensive database with adversarial samples added to obtain the final multimodal neural network model. This retraining process further improves the robustness and generalization ability of the model, enabling the model to better cope with various changes and interferences that may be encountered in practical applications.

[0166] This multi-step, iterative optimization method has multiple advantages: first, it makes full use of the characteristics of different types of data and adopts the most suitable network structure for processing; second, it realizes the dynamic fusion of different features through the attention mechanism, which improves the adaptability of the model; third, the introduction of cross-validation and adversarial training greatly enhances the generalization ability and robustness of the model; finally, the whole process is adaptive and can be continuously adjusted and optimized according to the characteristics of actual data, so as to obtain the neural network model that is most suitable for the specific aluminum alloy component strength detection task.

[0167] In this embodiment, the parameter optimization of the multimodal neural network model to obtain a comprehensive prediction model includes: performing a sensitivity analysis on the multimodal neural network model to determine the key parameters that have the greatest impact on model performance; designing a parameter optimization strategy based on the sensitivity analysis results, including a learning rate scheduling and a regularization method; using the parameter optimization strategy to adjust the weights and biases of the multimodal neural network model; using data enhancement technology to transform and expand samples in the comprehensive database to generate an enhanced data set; using the enhanced data set to fine-tune the adjusted multimodal neural network model; performing performance tests on the fine-tuned multimodal neural network model under different working conditions to evaluate its prediction accuracy and stability; and selecting the model with the best performance as the comprehensive prediction model based on the performance test results.

[0168] Specifically, global sensitivity analysis methods such as the Sobol index or the Morris screening method are used. For example, when using the Morris screening method, small perturbations are made to the various parameters of the model, and the changes in the output are observed to identify the parameters that have the most significant impact on the model output. In multimodal neural network models, these key parameters include the number of filters in the convolutional layer, the number of hidden units in the LSTM layer, the weights of the attention mechanism, etc. The advantage of sensitivity analysis is that it can help researchers focus on optimizing the most important parameters and greatly improve optimization efficiency.

[0169] Based on the results of the sensitivity analysis, design parameter optimization strategies, including learning rate scheduling and regularization methods. The learning rate scheduling strategy may adopt methods such as step decay or cosine annealing. For example, when using the cosine annealing strategy, the learning rate will change periodically between the preset maximum and minimum values ​​according to the cosine function. This method can converge quickly in the early stage of training and search for the optimal solution in detail in the later stage. Regularization methods may include L1, L2 regularization or Dropout. For complex models such as multimodal neural network models that may have the risk of overfitting, L2 regularization prevents the model from being too complex by penalizing large weights, while Dropout enhances the generalization ability of the model by randomly shutting down some neurons. The combined use of these optimization strategies can effectively improve the performance and generalization ability of the model.

[0170] The weights and biases of the multimodal neural network model are adjusted using a designed parameter optimization strategy. This process usually uses advanced optimization algorithms such as Adam or RMSprop, which can adaptively adjust the learning rate of each parameter to accelerate the convergence process. During the adjustment process, the performance of the model on the validation set is regularly evaluated to prevent overfitting.

[0171] Next, data enhancement technology is used to transform and expand the samples in the comprehensive database to generate an enhanced data set. For thermoelectric characteristic data, Gaussian noise can be added or the time series can be translated and scaled; for microstructure images, rotation, flipping or brightness adjustment can be performed; for mechanical property data, random perturbations can be added within a reasonable range. For example, a 90-degree rotation and horizontal flipping of a grain structure image can produce an enhanced sample that is 8 times the amount of the original data. The advantage of data enhancement is that it can significantly increase the diversity of training samples and improve the model's adaptability to various changes. It is also an effective means to alleviate overfitting.

[0172] The adjusted multimodal neural network model is fine-tuned using the generated augmented dataset. During fine-tuning, a smaller learning rate is usually used to retain the general features that the model has learned while adapting to the new data distribution. The purpose of this step is to enable the model to better adapt to various possible data changes and improve its robustness in practical applications.

[0173] The fine-tuned multimodal neural network model is tested for performance under different operating conditions to evaluate its predictive accuracy and stability. These operating conditions may include different temperature conditions, stress levels, and even simulated defects or abnormal conditions. For example, the model's predictive performance can be tested at room temperature, high temperature, and low temperature conditions, or the model's sensitivity to different degrees of material defects can be evaluated. Through these comprehensive tests, we can fully understand the performance of the model in various practical situations and ensure its reliability in practical applications.

[0174] Finally, based on the results of the performance test, the model with the best performance is selected as the comprehensive prediction model. The selection criteria include not only the average prediction accuracy, but also the stability and reliability of the model under various conditions. For example, a model that maintains high accuracy under all test conditions may be selected rather than a model that performs well only under certain specific conditions.

[0175] This systematic parameter optimization has multiple advantages: first, through sensitivity analysis and targeted optimization, the optimization efficiency is greatly improved; second, data enhancement and multi-condition testing significantly enhance the generalization ability and adaptability of the model; third, the whole process is data-driven and can be continuously adjusted and optimized according to the characteristics of actual data; finally, this method not only improves the prediction accuracy of the model, but also ensures the reliability of the model in various practical application scenarios. Through this series of steps, the final comprehensive prediction model can more accurately and stably predict the strength of aluminum alloy components, providing reliable decision support for engineering applications.

[0176] Continue to refer Figure 1 , input the characteristic data set into the comprehensive prediction model, calculate the strength value of each measuring point of the aluminum alloy component to be tested, and generate a visualization result of the strength distribution of the aluminum alloy component to be tested.

[0177] In this embodiment, the characteristic data set is input into the comprehensive prediction model, the strength value of each measuring point of the aluminum alloy component to be tested is calculated, and a visualization result of the strength distribution of the aluminum alloy component to be tested is generated, including: normalizing the characteristic data set to obtain a normalized characteristic data set; inputting the normalized characteristic data set into the comprehensive prediction model to obtain an initial strength prediction value of each measuring point of the aluminum alloy component to be tested; inverse normalizing the initial strength prediction value to obtain a strength prediction value of the original scale; and spatially interpolating the strength prediction value of the original scale using the Kriging interpolation method. Interpolation is performed to obtain a continuous intensity distribution field; gradient calculation is performed on the continuous intensity distribution field to obtain an intensity gradient field; the grid is refined according to the intensity gradient field to obtain a refined prediction grid; intensity re-prediction and interpolation are performed on the refined prediction grid to obtain an optimized intensity distribution field; contour line division is performed on the optimized intensity distribution field to obtain a plurality of intensity intervals; color mapping is performed on the plurality of intensity intervals to obtain a two-dimensional intensity distribution heat map; a three-dimensional surface model is constructed based on the two-dimensional intensity distribution heat map to obtain a three-dimensional strength distribution visualization result of the aluminum alloy component to be tested.

[0178] Specifically, first, the feature data set is normalized to obtain a normalized feature data set. This step uses the minimum-maximum normalization or Z-score normalization method. For example, when using minimum-maximum normalization, each feature value x will be converted to ( ) / (max-min), where min and max are the minimum and maximum values ​​of the feature respectively. The advantage of normalization is that it can map features of different scales to the same range, which helps to improve the training efficiency and prediction accuracy of the model.

[0179] The normalized feature data set is input into the comprehensive prediction model to obtain the initial strength prediction value of each measurement point of the aluminum alloy component to be tested. In this process, the comprehensive prediction model will use the previously trained multimodal neural network structure to process and predict the input features. Subsequently, the initial strength prediction value is inversely normalized to obtain the strength prediction value of the original scale. The inverse normalization process is the inverse operation of normalization, which converts the prediction result back to the original strength unit, such as MPa. This step ensures that the prediction result is comparable with the actual physical quantity.

[0180] Next, the Kriging interpolation method is used to spatially interpolate the strength prediction values ​​of the original scale to obtain a continuous strength distribution field. Kriging interpolation is a spatial interpolation method based on statistical principles, which takes into account the spatial correlation between data points. For example, for a 100mm×100mm aluminum alloy plate, if there are strength prediction values ​​at 10 discrete points, Kriging interpolation can generate a continuous strength distribution on the entire plate surface. The advantage of this method is that it can provide the best linear unbiased estimate and take into account the spatial structure of the data.

[0181] The gradient of the continuous intensity distribution field is calculated to obtain the intensity gradient field. The gradient calculation usually uses the finite difference method or the finite element method. For example, using the central difference method, the gradient can be expressed as ∇f≈(f(x+h)-f(xh)) / (2h), where h is the spatial step size. The intensity gradient field reflects the direction and rate of intensity change, which helps to identify stress concentration areas.

[0182] The mesh is refined according to the intensity gradient field to obtain a refined prediction mesh. In areas with large intensity changes, the mesh will be encrypted to improve local prediction accuracy. For example, in areas of stress concentration, the mesh size may be reduced to 1 / 4 of the original size. This adaptive mesh refinement technology can reduce the consumption of computing resources while ensuring accuracy.

[0183] Intensity re-prediction and interpolation are performed on the refined prediction grid to obtain the optimized intensity distribution field. This step utilizes a finer grid to capture more local details and improve the spatial resolution of the prediction.

[0184] The optimized intensity distribution field is divided into contour lines to obtain multiple intensity intervals. Contour line division usually adopts equal interval or percentile-based methods. For example, the intensity range can be evenly divided into 10 intervals, or an appropriate division method can be selected according to the distribution characteristics of the intensity value. Contour line division helps to intuitively display the spatial distribution characteristics of the intensity.

[0185] Subsequently, multiple intensity intervals are color mapped to obtain a two-dimensional intensity distribution heat map. Color mapping usually uses a gradient color spectrum from cold to warm, such as from blue (low intensity) to red (high intensity). This visualization method can intuitively display the spatial distribution of intensity, making it easier for engineers to quickly identify high stress areas and potential failure points.

[0186] Finally, a three-dimensional surface model is constructed based on the two-dimensional strength distribution heat map to obtain the three-dimensional strength distribution visualization results of the aluminum alloy component to be tested. This step usually uses three-dimensional rendering technology to map the intensity value to the height in three-dimensional space. For example, for a flat component, the intensity value can be expressed as the height of the z-axis to form an undulating three-dimensional surface. The advantage of this three-dimensional visualization technology is that it can simultaneously display the numerical value and spatial distribution of the strength, providing engineers with more intuitive and comprehensive information.

[0187] This series of data processing and visualization steps has multiple advantages: first, the process from normalization to inverse normalization ensures the consistency and interpretability of data processing; second, Kriging interpolation and adaptive grid refinement techniques improve the spatial resolution and accuracy of prediction results; third, the combination of multiple visualization techniques (contour lines, heat maps, 3D models) provides users with different needs with rich information display methods; finally, the whole process is automated, which can quickly process large amounts of data and generate intuitive visualization results. This systematic approach not only improves the accuracy and efficiency of strength testing of aluminum alloy components, but also provides a powerful decision support tool for engineering design and failure analysis.

[0188] A second aspect of the present invention provides a strength detection device for an aluminum alloy component, comprising:

[0189] A data acquisition module 101 is used to obtain a comprehensive database based on multiple groups of aluminum alloy samples, wherein the comprehensive database contains microstructure parameters, mechanical property parameters and thermoelectric characteristic parameters of each group of aluminum alloy samples;

[0190] An optimization unit 102, configured to optimize parameters of a thermoelectric characteristic measurement system according to the thermoelectric characteristic parameters in the comprehensive database;

[0191] Thermoelectric characteristic measurement module 103, used to apply temperature gradient and mechanical stress to the aluminum alloy component to be measured by using the thermoelectric characteristic measurement system with optimized parameters, and simultaneously measure the temperature and thermoelectric potential of each measurement point on the aluminum alloy component to be measured, so as to obtain an original data set including temperature distribution and thermoelectric potential distribution;

[0192] The data processing unit 104 is used to analyze the original data set and extract characteristic parameters to obtain a characteristic data set;

[0193] A model building module 105 is used to build a multimodal neural network model based on the comprehensive database, and optimize the parameters of the multimodal neural network model to obtain a comprehensive prediction model;

[0194] The strength prediction unit 106 is used to input the characteristic data set into the comprehensive prediction model, calculate the strength value of each measuring point of the aluminum alloy component to be tested, and generate a visualization result of the strength distribution of the aluminum alloy component to be tested.

[0195] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A strength testing method for an aluminum alloy component, characterized in that: include: Acquire a comprehensive database based on multiple groups of aluminum alloy samples, wherein the comprehensive database contains microstructure parameters, mechanical property parameters, and thermoelectric property parameters of each group of aluminum alloy samples; The method of obtaining a comprehensive database based on multiple groups of aluminum alloy samples includes: performing scanning electron microscope observation on a group of aluminum alloy samples, and performing image segmentation and morphological analysis on the observation results to extract the grain size distribution and the second phase volume fraction of the sample as microstructure parameters; performing a tensile test on the aluminum alloy samples, fitting the data obtained from the tensile test and obtaining a stress-strain curve, and extracting the yield strength, tensile strength and elongation as mechanical property parameters through the stress-strain curve analysis; applying a temperature gradient to the aluminum alloy samples and measuring the thermoelectric potential, and calculating the Seebeck coefficient by the ratio of the thermoelectric potential to the temperature gradient. Gram coefficient, measure the resistivity of the aluminum alloy sample, calculate the electrical conductivity by the inverse of the resistivity, measure the thermal diffusion coefficient, specific heat capacity and density of the aluminum alloy sample, calculate the thermal conductivity by the product of the thermal diffusion coefficient, specific heat capacity and density, and use the Seebeck coefficient, the electrical conductivity and the thermal conductivity as thermoelectric characteristic parameters; record the microstructure parameters, mechanical property parameters and thermoelectric characteristic parameters obtained for the aluminum alloy sample; repeat the above steps for the remaining aluminum alloy samples until the test of all groups of aluminum alloy samples is completed; perform minimum-maximum normalization processing on the recorded data of all groups to obtain a comprehensive database; Optimizing the parameters of the thermoelectric characteristic measurement system according to the thermoelectric characteristic parameters in the comprehensive database; Applying a temperature gradient and mechanical stress to the aluminum alloy component to be measured by using a thermoelectric characteristic measurement system with optimized parameters, and measuring the temperature and thermoelectric potential of each measuring point on the aluminum alloy component to be measured, so as to obtain an original data set including temperature distribution and thermoelectric potential distribution; Analyzing the original data set and extracting characteristic parameters to obtain a characteristic data set; analyzing the original data set and extracting characteristic parameters to obtain a characteristic data set includes: performing spatial gradient calculation on the temperature distribution data in the original data set to obtain a temperature gradient feature; performing spatial gradient calculation on the thermoelectric potential distribution data in the original data set to obtain a thermoelectric potential gradient feature; calculating the Seebeck coefficient distribution according to the temperature gradient feature and the thermoelectric potential gradient feature to obtain a Seebeck coefficient feature; performing time series analysis on the temperature distribution data to obtain a temperature fluctuation feature; performing time series analysis on the thermoelectric potential distribution data to obtain a thermoelectric potential fluctuation feature; calculating local thermal conductivity according to the temperature distribution data and the thermoelectric potential distribution data to obtain a thermal conductivity feature; combining the temperature gradient feature, the thermoelectric potential gradient feature, the Seebeck coefficient feature, the temperature fluctuation feature, the thermoelectric potential fluctuation feature and the thermal conductivity feature to obtain a characteristic data set; Based on the comprehensive database, a multimodal neural network model is constructed, and the parameters of the multimodal neural network model are optimized to obtain a comprehensive prediction model; the multimodal neural network model is constructed based on the comprehensive database, including: extracting thermoelectric characteristic data, microstructure data and mechanical property data from the comprehensive database; performing time series analysis on the thermoelectric characteristic data to determine the input dimension and time step of the long short-term memory network in the thermoelectric characteristic processing branch, performing image feature analysis on the microstructure data to determine the convolution kernel size and network depth of the convolutional neural network in the microstructure processing branch, performing statistical analysis on the mechanical property data to determine the number of hidden layers and the number of neurons in each layer of the fully connected network in the mechanical property processing branch; according to the above analysis results, a thermoelectric characteristic processing branch, a microstructure processing branch and a mechanical property processing branch are respectively constructed; and an attention mechanism module is designed to fuse the thermoelectric characteristic processing branch , the output features of the microstructure processing branch and the mechanical property processing branch; constructing an intensity prediction module for mapping the features fused by the attention mechanism module into intensity values; connecting the thermoelectric characteristic processing branch, the microstructure processing branch, the mechanical property processing branch, the attention mechanism module and the intensity prediction module to form an initial multimodal neural network model; using the data in the comprehensive database to train the initial multimodal neural network model, and using a cross-validation method to evaluate the performance of the initial multimodal neural network model; based on the cross-validation results, adjusting the structure and parameters of the initial multimodal neural network model; performing adversarial training on the adjusted initial multimodal neural network model, generating adversarial samples and adding them to the comprehensive database; using the comprehensive database after adding the adversarial samples to retrain the initial multimodal neural network model to obtain a multimodal neural network model; The characteristic data set is input into the comprehensive prediction model, the strength value of each measuring point of the aluminum alloy component to be tested is calculated, and a visualization result of the strength distribution of the aluminum alloy component to be tested is generated.

2. The strength testing method of an aluminum alloy component according to claim 1, characterized in that: The step of optimizing the parameters of the multimodal neural network model to obtain a comprehensive prediction model includes: Performing sensitivity analysis on the multimodal neural network model to determine the key parameters that have the greatest impact on model performance; Based on the sensitivity analysis results, design parameter optimization strategies, including learning rate scheduling and regularization methods; Using the parameter optimization strategy, adjusting the weights and biases of the multimodal neural network model; Using data enhancement technology, transforming and expanding the samples in the comprehensive database to generate an enhanced data set; Fine-tuning the adjusted multimodal neural network model using the enhanced dataset; Performing performance tests on the fine-tuned multimodal neural network model under different working conditions to evaluate its prediction accuracy and stability; According to the performance test results, the model with the best performance is selected as the comprehensive prediction model.

3. The strength testing method of an aluminum alloy component according to claim 1, characterized in that: The step of inputting the characteristic data set into the comprehensive prediction model, calculating the strength value of each measuring point of the aluminum alloy component to be tested, and generating a visualization result of the strength distribution of the aluminum alloy component to be tested includes: Normalizing the feature data set to obtain a normalized feature data set; Inputting the normalized characteristic data set into the comprehensive prediction model to obtain the initial strength prediction value of each measuring point of the aluminum alloy component to be tested; Performing inverse normalization processing on the initial intensity prediction value to obtain an intensity prediction value of the original scale; Using the Kriging interpolation method to spatially interpolate the intensity prediction values ​​of the original scale to obtain a continuous intensity distribution field; Performing gradient calculation on the continuous intensity distribution field to obtain an intensity gradient field; Refining the grid according to the intensity gradient field to obtain a refined prediction grid; Performing intensity re-prediction and interpolation on the refined prediction grid to obtain an optimized intensity distribution field; Dividing the optimized intensity distribution field into contour lines to obtain a plurality of intensity intervals; Performing color mapping on the multiple intensity intervals to obtain a two-dimensional intensity distribution heat map; A three-dimensional surface model is constructed based on the two-dimensional strength distribution heat map to obtain a visualization result of the three-dimensional strength distribution of the aluminum alloy component to be tested.

4. The strength testing method of an aluminum alloy component according to claim 1, characterized in that: Optimizing the parameters of the thermoelectric characteristic measurement system according to the thermoelectric characteristic parameters in the comprehensive database includes: Performing principal component analysis on the thermoelectric characteristic parameters in the comprehensive database to extract the main influencing factors; Constructing a parameter optimization objective function of a thermoelectric characteristic measurement system according to the main influencing factors; Solving the objective function using an optimization algorithm to obtain an initial optimization parameter set; Based on the initial optimized parameter set, construct a digital twin model of the thermoelectric characteristics measurement system; simulating the digital twin model to generate simulation data; Using the simulation data to train a machine learning model to establish a mapping relationship between system parameters and measurement accuracy; Based on the mapping relationship, an optimization algorithm is used to fine-tune the parameters to obtain a final optimized parameter set; The final optimized parameter set is applied to an actual thermoelectric characteristic measurement system to complete system parameter optimization.

5. The strength testing method of an aluminum alloy component according to claim 1, characterized in that: The method of applying a temperature gradient to the aluminum alloy component to be measured by using a thermoelectric property measurement system with optimized parameters includes: Placing the aluminum alloy component to be tested horizontally on the thermal insulation test platform of the thermoelectric property measurement system; Directly contact the heat source and the cold source of the thermoelectric property measurement system with the two opposite end surfaces of the aluminum alloy component to be measured; Arrange N temperature sensors at equal intervals on the upper surface of the aluminum alloy component to be tested in the direction from the heat source to the cold source, where N≥5; Controlling the temperature of the heat source to gradually increase to a preset high temperature value at a rate not exceeding 5°C per minute, while maintaining the temperature of the cold source at a preset low temperature value; The following steps are executed cyclically until a stable state is reached: a. The temperature data of the surface of the aluminum alloy component to be tested is collected once every 10 seconds by the N temperature sensors; b. The temperature of the heat source and the cold source is fine-tuned according to the feedback data of the temperature sensor until the difference between the temperature data collected twice in succession by the same temperature sensor is less than 1°C, and this condition is simultaneously established for all the N temperature sensors; When the above stable state is reached, continue to maintain this state for at least 10 minutes, and then record the temperature data of N temperature sensors at this time { , i=1, 2, ..., N}; The recorded temperature data are analyzed using the least squares method. } Perform linear fitting and obtain the fitting equation T=ax+b, where T is the temperature value, x is the distance of the temperature sensor from the heat source to the cold source, a is the temperature gradient, and b is the theoretical temperature value at x=0; If the relative error between the fitted temperature gradient value |a| and the preset value is within the range of ±5%, and the goodness of fit R²>0.95, the temperature gradient application is confirmed to be successful; otherwise, the entire temperature gradient application process is restarted.

6. The strength testing method of an aluminum alloy component according to claim 5, characterized in that: Applying mechanical stress to the aluminum alloy component to be tested, and measuring the temperature and thermoelectric potential of each measuring point on the aluminum alloy component to be tested, to obtain an original data set including temperature distribution and thermoelectric potential distribution, including: Installing a piezoelectric ceramic driver and a force sensor at both ends of the aluminum alloy component to be tested; Applying an initial prestress to the aluminum alloy component to be tested, wherein the prestress value is 3% of the yield strength of the material; Controlling the piezoelectric ceramic driver to increase stress at a rate of 0.5 MPa per second until a preset target stress is reached; During the stress increase process, the temperature data of the surface of the aluminum alloy component to be tested is synchronously collected every 5 seconds through the N temperature sensors; Synchronously measuring the thermoelectric potential between adjacent points of the N temperature sensors to obtain thermoelectric potential data; Perform interpolation processing on the collected temperature data to obtain continuous temperature distribution data; Performing interpolation processing on the collected thermoelectric potential data to obtain continuous thermoelectric potential distribution data; The continuous temperature distribution data and thermoelectric potential distribution data are combined to form an original data set including temperature distribution and thermoelectric potential distribution.

7. A strength detection device for an aluminum alloy component, characterized in that: include: A data acquisition module is used to obtain a comprehensive database based on multiple groups of aluminum alloy samples, wherein the comprehensive database contains microstructure parameters, mechanical property parameters, and thermoelectric property parameters of each group of aluminum alloy samples; the method of obtaining the comprehensive database based on multiple groups of aluminum alloy samples includes: performing scanning electron microscope observation on a group of aluminum alloy samples, and performing image segmentation and morphological analysis on the observation results to extract the grain size distribution and the second phase volume fraction of the sample as microstructure parameters; performing a tensile test on the aluminum alloy samples, fitting the data obtained from the tensile test and obtaining a stress-strain curve, and extracting the yield strength, tensile strength, and elongation as mechanical property parameters through the stress-strain curve analysis; and .... Apply a temperature gradient to the sample and measure the thermoelectric potential, calculate the Seebeck coefficient by the ratio of the thermoelectric potential to the temperature gradient, measure the resistivity of the aluminum alloy sample, calculate the electrical conductivity by the inverse of the resistivity, measure the thermal diffusivity, specific heat capacity and density of the aluminum alloy sample, calculate the thermal conductivity by the product of the thermal diffusivity, specific heat capacity and density, and use the Seebeck coefficient, the electrical conductivity and the thermal conductivity as thermoelectric characteristic parameters; record the microstructure parameters, mechanical property parameters and thermoelectric characteristic parameters obtained for the aluminum alloy sample; repeat the above steps for the remaining aluminum alloy samples until the testing of all groups of aluminum alloy samples is completed; perform minimum-maximum normalization processing on the recorded data of all groups to obtain a comprehensive database; An optimization unit, used for optimizing parameters of a thermoelectric characteristic measurement system according to the thermoelectric characteristic parameters in the comprehensive database; Thermoelectric characteristic measurement module, used for applying temperature gradient and mechanical stress to the aluminum alloy component to be measured by using the thermoelectric characteristic measurement system with optimized parameters, and measuring the temperature and thermoelectric potential of each measuring point on the aluminum alloy component to be measured, so as to obtain an original data set including temperature distribution and thermoelectric potential distribution; a data processing unit, configured to analyze the original data set and extract characteristic parameters to obtain a characteristic data set; the analyzing and extracting characteristic parameters of the original data set to obtain a characteristic data set comprises: performing spatial gradient calculation on the temperature distribution data in the original data set to obtain a temperature gradient feature; performing spatial gradient calculation on the thermoelectric potential distribution data in the original data set to obtain a thermoelectric potential gradient feature; calculating the Seebeck coefficient distribution according to the temperature gradient feature and the thermoelectric potential gradient feature to obtain a Seebeck coefficient feature; performing time series analysis on the temperature distribution data to obtain a temperature fluctuation feature; performing time series analysis on the thermoelectric potential distribution data to obtain a thermoelectric potential fluctuation feature; calculating local thermal conductivity according to the temperature distribution data and the thermoelectric potential distribution data to obtain a thermal conductivity feature; and combining the temperature gradient feature, the thermoelectric potential gradient feature, the Seebeck coefficient feature, the temperature fluctuation feature, the thermoelectric potential fluctuation feature and the thermal conductivity feature to obtain a characteristic data set; A model building module is used to build a multimodal neural network model based on the comprehensive database, optimize the parameters of the multimodal neural network model, and obtain a comprehensive prediction model; the building of the multimodal neural network model based on the comprehensive database includes: extracting thermoelectric characteristic data, microstructure data and mechanical property data from the comprehensive database; performing time series analysis on the thermoelectric characteristic data to determine the input dimension and time step of the long short-term memory network in the thermoelectric characteristic processing branch, performing image feature analysis on the microstructure data to determine the convolution kernel size and network depth of the convolution neural network in the microstructure processing branch, performing statistical analysis on the mechanical property data to determine the number of hidden layers and the number of neurons in each layer of the fully connected network in the mechanical property processing branch; according to the above analysis results, respectively constructing a thermoelectric characteristic processing branch, a microstructure processing branch and a mechanical property processing branch; designing an attention mechanism module for fusing the thermoelectric characteristic data. The method comprises the steps of: calculating the output features of the property processing branch, the microstructure processing branch and the mechanical property processing branch; constructing an intensity prediction module for mapping the features fused by the attention mechanism module into intensity values; connecting the thermoelectric property processing branch, the microstructure processing branch, the mechanical property processing branch, the attention mechanism module and the intensity prediction module to form an initial multimodal neural network model; training the initial multimodal neural network model using the data in the comprehensive database, and evaluating the performance of the initial multimodal neural network model using a cross-validation method; adjusting the structure and parameters of the initial multimodal neural network model based on the cross-validation results; performing adversarial training on the adjusted initial multimodal neural network model, generating adversarial samples and adding them to the comprehensive database; retraining the initial multimodal neural network model using the comprehensive database after adding the adversarial samples to obtain a multimodal neural network model; The strength prediction unit is used to input the characteristic data set into the comprehensive prediction model, calculate the strength value of each measuring point of the aluminum alloy component to be tested, and generate a visualization result of the strength distribution of the aluminum alloy component to be tested.