Metal product mechanical property detection method and system based on artificial intelligence

Abstract

The invention relates to the technical field of metal product mechanical performance prediction, and provides a metal product mechanical performance detection method and system based on artificial intelligence, and the method comprises data fusion analysis, material constitutive performance prediction, mechanical performance quality prediction and performance detection evaluation. According to the method, a reinforcement learning model is embedded by introducing a structure evolution improvement factor and a causal graph, so that whole-process intelligent modeling and causal reasoning from process parameters to material constitutive performance to final mechanical performance are realized, and the method has triple capabilities of performance prediction, cause interpretation and optimization suggestion; the pre-service evaluation and quality control efficiency of the complex structural component is obviously improved; by constructing a four-module integrated linkage system of data fusion, material modeling, performance prediction and detection evaluation, full-chain closed-loop analysis from original data to performance evaluation is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of prediction of the mechanical properties of metal products, and in particular to an artificial intelligence-based method and system for detecting the mechanical properties of metal products. Background Art

[0002] Artificial intelligence-based methods and systems for testing the mechanical properties of metal products generally refer to the use of artificial intelligence technology to conduct a comprehensive analysis of data generated during the manufacturing and use of metal products (such as process parameters, material characteristics and test results) to predict and evaluate their mechanical properties (such as strength, hardness and ductility), thereby replacing or assisting traditional destructive tests, and achieving technical means and tool systems for rapid performance determination, quality anomaly traceability and manufacturing process optimization. They are widely used in manufacturing fields such as aviation, automobiles, and rail transit that have strict requirements on metal properties.

[0003] However, in the existing methods for testing the mechanical properties of metal products, there is a technical problem that the process information is disconnected from the final performance, which makes it difficult to trace the material causes and process paths in the test results; in the existing material bulk structure performance analysis process, there is a technical problem that the continuous influence of the process on the evolution of the material microstructure is ignored, resulting in poor generalization of the constitutive model; in the existing mechanical performance analysis process, there is a technical problem that the performance prediction results are not interpretable and lacks modeling support for the causal chain between process and performance; in the existing metal product performance analysis process, there are technical problems such as data fragmentation, isolated links, and the inability of prediction results to directly serve quality control decisions. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the existing technology, the present invention provides a method and system for detecting the mechanical properties of metal products based on artificial intelligence. In view of the technical problem that in the existing methods for detecting the mechanical properties of metal products, there is a disconnect between the process information and the final performance, which makes it difficult to trace the material causes and process paths of the detection results, this solution creatively proposes a dual-path analysis framework that combines material constitutive modeling and performance prediction. It can predict the performance range based on historical process data and material characteristics before detection, and combine the model to deduce and analyze the source of deviation after detection, realizing the shift from "result judgment" to "cause explanation" detection mode, greatly improving the detection of quality anomalies. Positioning and response efficiency; In view of the technical problem that the continuous influence of the process on the evolution of the material microstructure is ignored in the existing material structure performance analysis process, resulting in poor generalization of the constitutive model, this solution introduces a "structural evolution improvement factor" to explicitly incorporate dynamic processes such as phase change behavior and tissue morphology changes into the model parameters, and combines multi-factor training strategies to achieve accurate prediction of the stress-strain behavior of materials under different thermal paths. It is especially suitable for the performance analysis scenarios of metal materials such as steel and aluminum alloys that are manufactured in parallel with multiple specifications and multiple paths; In view of the fact that the performance prediction results are not interpretable and lack of process in the existing mechanical performance analysis process, this solution introduces a "structural evolution improvement factor" to explicitly incorporate dynamic processes such as phase change behavior and tissue morphology changes into the model parameters, and combines multi-factor training strategies to achieve accurate prediction of the stress-strain behavior of materials under different thermal paths. It is especially suitable for the performance analysis scenarios of metal materials such as steel and aluminum alloys that are manufactured in parallel with multiple specifications and multiple paths. —Technical issues in the modeling support of the causal chain between performances. Traditional methods often use black box prediction models. Although the error is controllable, it cannot answer the question of “why the strength is lower than expected” and it is difficult to guide the direction of process optimization. This solution constructs a “metal product mechanical causal diagram” based on process path and material state, and embeds a reinforcement learning strategy network to dynamically adjust the prediction strategy from the causal chain, and realize the “contribution factor weight” analysis behind the performance results, thus possessing the triple capabilities of “prediction, explanation and optimization”, significantly improving the performance assurance and design feedback efficiency of complex structural parts before service; in response to the existing problems of data fragmentation, link isolation, and prediction in the performance analysis process of metal products, this solution can effectively solve the problems of data fragmentation, link isolation, and prediction in the performance analysis process of metal products. To address the technical problem that the test results cannot directly serve the quality control decision-making, this solution realizes a closed-loop analysis of the entire chain from raw data to performance evaluation by constructing an integrated linkage system of four modules: "data fusion - material modeling - performance prediction - detection and evaluation". In the system structure designed by the present invention, each module can operate independently and achieve efficient collaboration through a unified data interface and scheduling logic. It supports the rapid injection of real-time collected process parameters (such as temperature field and stress state) into the model for online performance estimation, and combines actual detection data for deviation correction and feedback learning, thereby constructing an intelligent detection system with the self-circulation capability of "prediction-detection-correction".

[0005] The technical solution adopted by the present invention is as follows: The present invention provides a method for detecting the mechanical properties of metal products based on artificial intelligence, the method comprising the following steps:

[0006] Step S1: data fusion analysis;

[0007] Step S2: prediction of material constitutive properties;

[0008] Step S3: mechanical property quality prediction;

[0009] Step S4: Testing the mechanical properties of metal products.

[0010] Furthermore, in step S1, the data fusion analysis is used to integrate multi-source structured data of metal products during the manufacturing, processing and inspection process and perform data fusion and optimization. Specifically, the original data set of the metal product manufacturing and inspection process is obtained through multi-source data collection, and the metal product feature optimization data is obtained by performing cleaning, standardization and association modeling operations;

[0011] The original data set of the metal product manufacturing and inspection process, specifically including material composition data, process parameter data, processing data, historical actual inspection data and environmental condition data;

[0012] The metal product feature optimization data specifically includes structured material composition feature vectors, structured process parameter feature vectors, structured machining process feature vectors, structured environmental condition feature vectors and process material coding feature data.

[0013] Furthermore, in step S2, the material constitutive property prediction is used to intelligently predict the stress-strain relationship parameters of the metal material under different process conditions. Specifically, based on the metal product feature optimization data, a multi-factor metal material constitutive modeling method combined with structural evolution improvement is used to predict the material constitutive property and obtain the material constitutive prediction parameters, including the following steps:

[0014] Step S21: input feature processing, specifically extracting the structured material composition feature vector and the structured process parameter feature vector from the metal product feature optimization data, and performing normalization processing to obtain normalized material constitutive feature data;

[0015] Step S22: defining constitutive targets, specifically defining target parameters for predicting the structural properties of the material by defining a constitutive relationship target prediction vector, and obtaining a constitutive target function; the constitutive target function specifically includes an elastic modulus target, a yield strength target, a strength coefficient target, and a strain hardening exponent target;

[0016] Step S23: constructing a basic prediction model, specifically constructing a standard XGBoost model as the basic model of the regression prediction model, and optimizing the model prediction by minimizing the square error loss function;

[0017] Step S24: constructing a structural evolution improvement factor, specifically by defining a process trend function and constructing a trend deviation index function to represent the trend deviation of the material structure body affected by the process parameters, obtaining a trend deviation loss term, and using the trend deviation loss term to construct a structural evolution improvement factor, and performing a weighted loss fusion by weighting the trend deviation loss term and the minimization of square error loss to obtain a comprehensive loss function;

[0018] The process trend function specifically represents the assumption that the influence of the process parameters on the constitutive objective function is a single-peak trend, and defines the trend characteristic representation;

[0019] The trend deviation loss term is specifically defined by taking the difference of the trend change direction;

[0020] Step S25: Material constitutive prediction, specifically, training a material constitutive prediction model based on the comprehensive loss function to obtain a material constitutive prediction model, and using the material constitutive prediction model to perform material constitutive prediction to obtain material constitutive prediction parameters;

[0021] The material constitutive prediction parameters specifically include material constitutive stress-strain relationship characteristics and a combination of predicted constitutive parameters.

[0022] Furthermore, in step S3, the mechanical property quality prediction is used to predict the final mechanical properties of the product under actual use or testing conditions. Specifically, based on the material constitutive prediction parameters and the metal product feature optimization data, a reinforcement learning performance prediction method combined with causal graph improved embedding is used to perform mechanical property quality prediction to obtain mechanical property prediction parameters, including the following steps:

[0023] Step S31: inputting feature coding, specifically integrating the material constitutive prediction parameters and the metal product feature optimization data, performing feature coding for mechanical property quality prediction, and obtaining state variable feature data;

[0024] Step S32: Constructing a mechanical causal graph for metal products, specifically by using structural equation modeling and a standard causal structure learning algorithm to extract mechanical features of metal products based on the state variable feature data to obtain causal dependency mechanical feature data, and then constructing graph data based on the causal dependency mechanical feature data to obtain mechanical causal graph data for metal products;

[0025] The nodes of the metal product mechanical cause-effect diagram data are used to represent the metal product body;

[0026] The edges of the metal product causal graph data are used to represent the causal influence of process and structural characteristics on mechanical properties, and specifically the causal influence is represented by the causal dependent mechanical characteristic data;

[0027] The edge weight of the metal product causal graph data is used to represent the causal contribution value of each state variable feature;

[0028] Step S33: Constructing a reinforcement learning environment, specifically sequentially constructing state space parameters, action space parameters, state transfer function, and reward function to construct a basic reinforcement learning environment;

[0029] The state space parameters are specifically constructed based on the state variable characteristic data;

[0030] The action space parameters are specifically constructed by a set of adjustment amplitudes of process variables;

[0031] The state transfer function is specifically constructed by simulating the influence of fine-tuning process parameters;

[0032] The reward function is specifically constructed by predicting the error of the mechanical performance value;

[0033] Step S34: constructing a causal embedding policy network, specifically by introducing a causal attention embedding module as a policy network weight in the reinforcement learning basic environment, obtaining causal weighted state vector data, and calculating and outputting current action space parameters using a policy function to obtain causal embedding action parameters;

[0034] Step S35: constructing a mechanical causal embedding reward, specifically introducing a grade function to improve the causal embedding reward, and training the model parameters of the reinforcement learning basic environment through a policy gradient optimization algorithm to obtain a mechanical quality prediction model;

[0035] Step S36: Mechanical performance quality prediction, specifically using the mechanical quality prediction model to perform mechanical performance quality prediction to obtain mechanical performance prediction parameters;

[0036] The mechanical performance prediction parameters specifically include mechanical performance index prediction values and mechanical performance grade rating parameters.

[0037] Furthermore, in step S4, the mechanical property test of the metal product is used to combine the model output with the actual test data for comprehensive analysis and testing, specifically combining the mechanical property prediction parameters with the actual metal product test experimental samples to perform comparative analysis, error analysis and prediction deviation analysis to obtain comprehensive reference data for mechanical property testing.

[0038] The present invention provides an artificial intelligence-based metal product mechanical property detection system, which includes a data fusion analysis module, a material constitutive property prediction module, a mechanical property quality prediction module, and a metal product mechanical property detection module;

[0039] The data fusion analysis module is used for data fusion analysis, obtains metal product feature optimization data through data fusion analysis, and sends the metal product feature optimization data to the material constitutive performance prediction module;

[0040] The material constitutive performance prediction module is used to predict the material constitutive performance, obtain material constitutive prediction parameters through the material constitutive performance prediction, and send the material constitutive prediction parameters to the mechanical performance quality prediction module;

[0041] The mechanical performance quality prediction module is used for mechanical performance quality prediction, obtains mechanical performance prediction parameters through mechanical performance quality prediction, and sends the mechanical performance prediction parameters to the mechanical performance detection module;

[0042] The mechanical property detection module is used for detecting the mechanical properties of metal products, and obtains comprehensive reference data of the mechanical property detection of metal products through the mechanical property detection of metal products.

[0043] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0044] (1) In the existing methods for testing the mechanical properties of metal products, there is a technical problem that the process information is disconnected from the final performance, which makes it difficult to trace the material causes and process paths. For example, in actual production, when a batch of cold-rolled steel plates shows performance fluctuations in the tensile test, traditional testing methods can only feedback abnormal results, but cannot clearly determine whether the problem is caused by alloy ratio deviation, excessive annealing temperature, or insufficient release of surface stress. This makes quality control extremely dependent on experience and lacks replicability. This solution creatively proposes a dual-path analysis framework that combines material constitutive modeling and performance prediction. It can predict the performance range based on historical process data and material characteristics before testing, and combine model deduction to analyze the source of deviation after testing, realizing the transition from "result judgment" to "cause explanation" testing mode, greatly improving the efficiency of locating and responding to quality anomalies;

[0045] (2) In the existing analysis of the structural properties of materials, there is a technical problem that the continuous influence of the process on the evolution of the material microstructure is ignored, resulting in poor generalization of the constitutive model. For example, in the manufacture of hot-formed high-strength steel, different cooling paths will significantly affect the final ratio of martensite to bainite, thereby affecting the yield strength and plasticity. Traditional constitutive models often only fit the material behavior based on a single temperature or strain rate curve, ignoring the intermediary variable of organizational evolution, and are difficult to meet the modeling requirements under multi-batch and asynchronous process conditions. This solution introduces a "structural evolution improvement factor" to explicitly incorporate dynamic processes such as phase transformation behavior and organizational morphology changes into the model parameters. Combined with a multi-factor training strategy, it achieves accurate prediction of the stress-strain behavior of materials under different thermal paths. It is particularly suitable for the performance analysis of metal materials such as steel and aluminum alloys that are manufactured in parallel with multiple specifications and multiple paths;

[0046] (3) In the existing mechanical performance analysis process, there are technical problems such as the lack of interpretability of performance prediction results and the lack of modeling support for the causal chain between process and performance. For example, in complex curved die forgings or welded structures, mechanical properties are not only determined by material parameters, but also affected by spatial characteristics such as the distribution of stress concentration zones and the width of the heat-affected zone of the weld. Traditional methods often use black box prediction models. Although the error is controllable, it cannot answer the question "why the strength is lower than expected" and it is difficult to guide the direction of process optimization. This solution constructs a "metal product mechanical causal diagram" based on process path and material state, and embeds a reinforcement learning strategy network to dynamically adjust the prediction strategy from the causal chain, realizing the "contribution factor weight" analysis behind the performance results, thereby possessing the triple capabilities of "prediction, explanation and optimization", significantly improving the performance assurance and design feedback efficiency of complex structural parts before service.

[0047] (4) In order to solve the technical problems of data fragmentation, isolated links and inability of prediction results to directly serve quality control decision-making in the existing metal product performance analysis process, this solution realizes a closed-loop analysis of the entire chain from raw data to performance evaluation by constructing an integrated linkage system of four modules: "data fusion - material modeling - performance prediction - detection and evaluation". In the system structure designed by the present invention, each module can operate independently and achieve efficient collaboration through a unified data interface and scheduling logic. It supports the rapid injection of real-time collected process parameters (such as temperature field and stress state) into the model for online performance estimation. At the same time, it combines actual detection data for deviation correction and feedback learning, thereby constructing an intelligent detection system with the self-circulation capability of "prediction - detection - correction". BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic flow chart of a method for detecting the mechanical properties of metal products based on artificial intelligence provided by the present invention;

[0049] Figure 2 A schematic diagram of a metal product mechanical property testing system based on artificial intelligence provided by the present invention;

[0050] Figure 3 Schematic diagram of the process of predicting the constitutive properties of materials in step S2;

[0051] Figure 4 Schematic diagram of the process of mechanical property quality prediction in step S3.

[0052] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0054] In the description of the present invention, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0055] Example 1, see Figure 1 The present invention provides a method for detecting the mechanical properties of metal products based on artificial intelligence, which includes the following steps:

[0056] Step S1: data fusion analysis;

[0057] Step S2: prediction of material constitutive properties;

[0058] Step S3: mechanical property quality prediction;

[0059] Step S4: Testing the mechanical properties of metal products.

[0060] By performing the above operations, the technical problem of the disconnection between process information and final performance in existing methods for testing the mechanical properties of metal products, which makes it difficult to trace the test results to the material causes and process paths, is addressed. For example, in actual production, when a batch of cold-rolled steel plates exhibits performance fluctuations in a tensile test, traditional testing methods can only feedback abnormal results, but cannot clearly determine whether the problem is caused by alloy ratio deviation, excessive annealing temperature, or insufficient release of surface stress. This makes quality control extremely dependent on experience and lacks replicability. This solution creatively proposes a dual-path analysis framework that combines material constitutive modeling and performance prediction. It can predict the performance range based on historical process data and material characteristics before testing, and analyze the source of deviation based on model deduction after testing, realizing the transition from a "result judgment" to a "cause explanation" testing mode, greatly improving the efficiency of locating and responding to quality anomalies.

[0061] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S1, the data fusion analysis is used to integrate multi-source structured data of metal products during the manufacturing, processing and inspection process and perform data fusion and optimization. Specifically, the original data set of the metal product manufacturing and inspection process is obtained through multi-source data collection, and the metal product feature optimization data is obtained by performing cleaning, standardization and association modeling operations.

[0062] The original data set of the metal product manufacturing and inspection process, specifically including material composition data, process parameter data, processing data, historical actual inspection data and environmental condition data;

[0063] The metal product feature optimization data specifically includes structured material composition feature vectors, structured process parameter feature vectors, structured machining process feature vectors, structured environmental condition feature vectors and process material coding feature data.

[0064] Example 3, see Figure 1 、 Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the material constitutive property prediction is used to intelligently predict the stress-strain relationship parameters of the metal material under different process conditions. Specifically, based on the metal product feature optimization data, a multi-factor metal material constitutive modeling method combined with structural evolution improvement is used to predict the material constitutive property and obtain the material constitutive prediction parameters, including the following steps:

[0065] Step S21: input feature processing, specifically extracting the structured material composition feature vector and the structured process parameter feature vector from the metal product feature optimization data, and performing normalization processing to obtain normalized material constitutive feature data;

[0066] Step S22: defining constitutive targets, specifically defining target parameters for predicting the structural properties of the material by defining a constitutive relationship target prediction vector, and obtaining a constitutive target function; the constitutive target function specifically includes an elastic modulus target, a yield strength target, a strength coefficient target, and a strain hardening exponent target;

[0067] The calculation formula of the constitutive objective function is:

[0068] ;

[0069] Where Y is the constitutive objective function, E is the elastic modulus target, is the yield strength target, K is the strength coefficient target, and n is the strain hardening exponent target;

[0070] Step S23: constructing a basic prediction model, specifically constructing a standard XGBoost model as the basic model of the regression prediction model, and optimizing the model prediction by minimizing the square error loss function;

[0071] The calculation formula for minimizing the square error loss function is:

[0072] ;

[0073] Where, is the minimization square error loss function, N is the total number of targets, i is the first sample index, j is the target index, is the predicted value corresponding to the i-th input sample and the j-th target variable, is the true value corresponding to the i-th input sample and the j-th target variable;

[0074] Step S24: constructing a structural evolution improvement factor, specifically by defining a process trend function and constructing a trend deviation index function to represent the trend deviation of the material structure body affected by the process parameters, obtaining a trend deviation loss term, and using the trend deviation loss term to construct a structural evolution improvement factor, and performing a weighted loss fusion by weighting the trend deviation loss term and the minimization of square error loss to obtain a comprehensive loss function;

[0075] The process trend function specifically assumes that the influence of the process parameters on the constitutive objective function is a single-peak trend, and defines the trend characteristics. The calculation formula is:

[0076] ;

[0077] Where, The overall yield strength Heat treatment time t h The first derivative of is the process trend function, t h is the heat treatment time, t * It is the critical time point, which is used to indicate the time when the material performance reaches its peak;

[0078] The trend deviation loss term is specifically defined by taking the difference in the trend change direction, and the calculation formula is:

[0079] ;

[0080] Where, is the trend deviation loss term, which is used to indicate whether the predicted value violates the actual law in the trend direction. t is the number of valid trend sample pairs, i is the sample index, It is the process trend function model that predicts the trend of constitutive value change. is the constitutive target prediction value of the i+1th sample, is the constitutive target prediction value of the i-th sample, is the trend of heat treatment time change direction, t h(i+1) is the heat treatment time of the i+1th sample, t h(i) is the heat treatment time of the i-th sample, is the theoretical trend direction parameter of the i-th sample;

[0081] Step S25: Material constitutive prediction, specifically, training a material constitutive prediction model based on the comprehensive loss function to obtain a material constitutive prediction model, and using the material constitutive prediction model to perform material constitutive prediction to obtain material constitutive prediction parameters;

[0082] The material constitutive prediction parameters specifically include material constitutive stress-strain relationship characteristics and a combination of predicted constitutive parameters;

[0083] Preferably, Table 1 is an output example table of the predicted constitutive parameter combination. As shown in the table, the predicted constitutive parameter combination includes elastic modulus, yield strength, strength coefficient and strain hardening exponent. The output range of the elastic modulus is [50000, 220000]; the output range of the yield strength is [100, 1200]; the output range of the strength coefficient is [300, 2500]; and the output range of the strain hardening exponent is [0.05, 0.6].

[0084] Table 1 Example output of predicted constitutive parameter combinations

[0085]

[0086] By performing the above operations, we can address the technical problem that in the existing material bulk structure and performance analysis process, the continuous influence of the process on the evolution of the material's microstructure is ignored, resulting in poor generalization of the constitutive model. For example, in the manufacture of hot-formed high-strength steel, different cooling paths will significantly affect the final martensite-bainite ratio, thereby affecting the yield strength and plasticity. Traditional constitutive models often only fit the material behavior based on a single temperature or strain rate curve, ignoring the intermediary variable of tissue evolution, and are unable to meet the modeling requirements under multi-batch and asynchronous process conditions. This solution introduces a "structural evolution improvement factor" to explicitly incorporate dynamic processes such as phase transformation behavior and tissue morphology changes into the model parameters. Combined with a multi-factor training strategy, it achieves accurate prediction of the stress-strain behavior of materials under different thermal paths. It is particularly suitable for performance analysis scenarios of metal materials such as steel and aluminum alloys that are manufactured in parallel with multiple specifications and multiple paths.

[0087] Example 4, see Figure 1 、 Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the mechanical property quality prediction is used to predict the final mechanical properties of the product under actual use or testing conditions. Specifically, based on the material constitutive prediction parameters and the metal product feature optimization data, a reinforcement learning performance prediction method combined with causal graph improved embedding is used to perform mechanical property quality prediction to obtain mechanical property prediction parameters, including the following steps:

[0088] Step S31: inputting feature coding, specifically integrating the material constitutive prediction parameters and the metal product feature optimization data, performing feature coding for mechanical property quality prediction, and obtaining state variable feature data;

[0089] Step S32: Constructing a mechanical causal graph for metal products, specifically by using structural equation modeling and a standard causal structure learning algorithm to extract mechanical features of metal products based on the state variable feature data to obtain causal dependency mechanical feature data, and then constructing graph data based on the causal dependency mechanical feature data to obtain mechanical causal graph data for metal products;

[0090] The nodes of the metal product mechanical cause-effect diagram data are used to represent the metal product body;

[0091] The edges of the metal product causal graph data are used to represent the causal influence of process and structural characteristics on mechanical properties, and specifically the causal influence is represented by the causal dependent mechanical characteristic data;

[0092] The edge weight of the metal product causal graph data is used to represent the causal contribution value of each state variable feature;

[0093] Step S33: constructing a reinforcement learning environment, specifically constructing state space parameters, action space parameters, state transfer function and reward function in sequence to build a basic reinforcement learning environment;

[0094] The state space parameters are specifically constructed based on the state variable characteristic data;

[0095] The action space parameters are specifically constructed by a set of adjustment amplitudes of process variables;

[0096] The state transfer function is specifically constructed by simulating the influence of fine-tuning process parameters;

[0097] The reward function is specifically constructed by predicting the error of the mechanical performance value;

[0098] Step S34: constructing a causal embedding policy network, specifically by introducing a causal attention embedding module as a policy network weight in the reinforcement learning basic environment, obtaining causal weighted state vector data, and calculating and outputting current action space parameters using a policy function to obtain causal embedding action parameters;

[0099] Step S35: constructing a mechanical causal embedding reward, specifically introducing a grade function to improve the causal embedding reward, and training the model parameters of the reinforcement learning basic environment through a policy gradient optimization algorithm to obtain a mechanical quality prediction model;

[0100] The calculation formula for improving the causal embedding reward by introducing the level function is:

[0101] ;

[0102] Where, is the reward function value improved by causal embedding reward, is the target performance prediction value of the reinforcement learning model, It is the reference standard value of the target performance. When it is +1, it is used to indicate that the model prediction result is better than 110% of the standard, which is the excellent class. When it is +0.2, it is used to indicate that the prediction value is within the range of the standard line to 110%, which is the qualified class. When it is -1, it is used to indicate that the prediction value is lower than the standard, which is the unqualified class.

[0103] The causal weighted embedding reward function is obtained by further weighting the causal contribution weight based on the result of the causal embedding reward improvement of the introduced level function. The calculation formula is:

[0104] ;

[0105] Where, is the causal weighted embedding reward function, I is the set of input variable indices adjusted by the current policy action, is the input variable index adjusted by the current policy action, It corresponds to The causal contribution weights of the input variable indexes;

[0106] By constructing the optimization objective function of the policy gradient optimization algorithm based on the causal weighted embedding reward function, the calculation formula is:

[0107] ;

[0108] Where, is the optimization objective function of the policy network, is the parameter set of the policy network, is the expected value under the current strategy, is the current policy function, which is defined based on the current state space parameters and the current action space parameters. T is the total training time, and t is the time index. is the reward discount factor, the default value is 0.95, is the causally weighted embedding reward function;

[0109] Step S36: Mechanical performance quality prediction, specifically using the mechanical quality prediction model to perform mechanical performance quality prediction to obtain mechanical performance prediction parameters;

[0110] The mechanical performance prediction parameters specifically include mechanical performance index prediction values and mechanical performance grade rating parameters;

[0111] Preferably, Table 2 is an output example table of the predicted values of the mechanical property indicators. As shown in the table, the mechanical property prediction parameters include heat treatment temperature, holding time, yield strength prediction value and prediction grade. The output range of the heat treatment temperature is [600, 1100]; the output range of the holding time is [30, 120]; the output range of the yield strength prediction value is [350, 1200]; the output range of the prediction grade is [excellent, qualified, unqualified].

[0112] Table 2 Output example of mechanical performance index prediction values

[0113]

[0114] By performing the above operations, we can address the technical problems in the existing mechanical performance analysis process, such as the lack of interpretability of performance prediction results and the lack of modeling support for the causal chain between process and performance. For example, in complex curved die forgings or welded structures, the mechanical properties are not only determined by material parameters, but also affected by spatial characteristics such as the distribution of stress concentration areas and the width of the heat-affected zone of the weld. Traditional methods often use black box prediction models. Although the error is controllable, it cannot answer the question "why the strength is lower than expected" and it is difficult to guide the direction of process optimization. This solution constructs a "metal product mechanical causal diagram" based on process path and material state, and embeds a reinforcement learning strategy network to dynamically adjust the prediction strategy from the causal chain, realizing the "contribution factor weight" analysis behind the performance results, thereby possessing the triple capabilities of "prediction, explanation and optimization", significantly improving the performance assurance and design feedback efficiency of complex structural parts before service.

[0115] Example 5, see Figure 1 、 Figure 2 This embodiment is based on the above embodiment. In step S4, the mechanical property test of the metal product is used to combine the model output with the actual test data for comprehensive analysis and testing. Specifically, the mechanical property prediction parameters and the actual metal product test experimental samples are combined to perform comparative analysis, error analysis and prediction deviation analysis to obtain comprehensive reference data for mechanical property testing.

[0116] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The present invention provides an artificial intelligence-based metal product mechanical property detection system, including a data fusion analysis module, a material constitutive property prediction module, a mechanical property quality prediction module and a metal product mechanical property detection module;

[0117] The data fusion analysis module is used for data fusion analysis, obtains metal product feature optimization data through data fusion analysis, and sends the metal product feature optimization data to the material constitutive performance prediction module;

[0118] The material constitutive performance prediction module is used to predict the material constitutive performance, obtain material constitutive prediction parameters through the material constitutive performance prediction, and send the material constitutive prediction parameters to the mechanical performance quality prediction module;

[0119] The mechanical performance quality prediction module is used for mechanical performance quality prediction, obtains mechanical performance prediction parameters through mechanical performance quality prediction, and sends the mechanical performance prediction parameters to the mechanical performance detection module;

[0120] The mechanical property detection module is used for detecting the mechanical properties of metal products, and obtains comprehensive reference data of the mechanical property detection of metal products through the mechanical property detection of metal products.

[0121] By performing the above operations, in order to address the technical problems in the existing metal product performance analysis process, such as data fragmentation, isolated links, and prediction results that cannot directly serve quality control decisions, this solution constructs an integrated linkage system of four modules: "data fusion - material modeling - performance prediction - detection and evaluation", thereby realizing a closed-loop analysis of the entire chain from raw data to performance evaluation. In the system structure designed by the present invention, each module can operate independently and achieve efficient collaboration through a unified data interface and scheduling logic. It supports the rapid injection of real-time collected process parameters (such as temperature field and stress state) into the model for online performance estimation, and at the same time combines actual detection data for deviation correction and feedback learning, thereby constructing an intelligent detection system with "prediction-detection-correction" self-circulation capability.

[0122] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0123] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0124] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A method for detecting the mechanical properties of metal products based on artificial intelligence, characterized by: The method comprises the following steps: Step S1: Data fusion analysis to obtain metal product feature optimization data; Step S2: Prediction of material constitutive properties, using a multi-factor metal material constitutive modeling method combined with structural evolution improvement to predict material constitutive properties and obtain material constitutive prediction parameters, including the following steps: Step S21: Input feature processing; Step S22: Constitutive target definition; Step S23: Basic prediction model construction; Step S24: Structural evolution improvement factor construction, definition of process trend function and construction of trend deviation index function; Step S25: Material constitutive prediction; Step S3: Mechanical performance quality prediction, using a reinforcement learning performance prediction method combined with causal graph improved embedding to perform mechanical performance quality prediction and obtain mechanical performance prediction parameters, including the following steps: Step S31: Input feature encoding; Step S32: Construct a mechanical causal graph for metal products; Step S33: Construct a reinforcement learning environment; Step S34: Construct a causal embedding strategy network; Step S35: Construct a mechanical causal embedding reward, and introduce a grade function to improve the causal embedding reward; Step S36: Mechanical performance quality prediction; Step S4: Testing the mechanical properties of metal products to obtain comprehensive reference data for mechanical properties testing.

2. The method for detecting mechanical properties of metal products based on artificial intelligence according to claim 1, characterized in that: In step S1 , the metal product feature optimization data specifically includes structured material composition feature vectors, structured process parameter feature vectors, structured machining process feature vectors, structured environmental condition feature vectors and process material coding feature data.

3. The method for detecting mechanical properties of metal products based on artificial intelligence according to claim 2, characterized in that: In step S2, the material constitutive property prediction is used to intelligently predict the stress-strain relationship parameters of the metal material under different process conditions. Specifically, based on the metal product feature optimization data, a multi-factor metal material constitutive modeling method combined with structural evolution improvement is used to predict the material constitutive property and obtain the material constitutive prediction parameters, including the following steps: Step S21: input feature processing, specifically extracting the structured material composition feature vector and the structured process parameter feature vector from the metal product feature optimization data, and performing normalization processing to obtain normalized material constitutive feature data; Step S22: defining constitutive targets, specifically defining target parameters for predicting the structural properties of the material by defining a constitutive relationship target prediction vector, and obtaining a constitutive target function; the constitutive target function specifically includes an elastic modulus target, a yield strength target, a strength coefficient target, and a strain hardening exponent target; Step S23: constructing a basic prediction model, specifically constructing a standard XGBoost model as the basic model of the regression prediction model, and optimizing the model prediction by minimizing the square error loss function; Step S24: constructing a structural evolution improvement factor, specifically by defining a process trend function and constructing a trend deviation index function to represent the trend deviation of the material structure body affected by the process parameters, obtaining a trend deviation loss term, and using the trend deviation loss term to construct a structural evolution improvement factor, and performing a weighted loss fusion by weighting the trend deviation loss term and the minimization of square error loss to obtain a comprehensive loss function; The process trend function specifically represents the assumption that the influence of the process parameters on the constitutive objective function is a unimodal monotonic trend, and defines the trend characteristic representation; The trend deviation loss term is specifically defined by taking the difference of the trend change direction; Step S25: Material constitutive prediction, specifically, training a material constitutive prediction model based on the comprehensive loss function to obtain a material constitutive prediction model, and using the material constitutive prediction model to perform material constitutive prediction to obtain material constitutive prediction parameters.

4. The method for detecting mechanical properties of metal products based on artificial intelligence according to claim 3, characterized in that: In step S25, the material constitutive prediction parameters specifically include material constitutive stress-strain relationship characteristics and a combination of predicted constitutive parameters.

5. The method for detecting mechanical properties of metal products based on artificial intelligence according to claim 4, characterized in that: In step S3, the mechanical property quality prediction is used to predict the final mechanical properties of the product under actual use or testing conditions. Specifically, based on the material constitutive prediction parameters and the metal product feature optimization data, a reinforcement learning performance prediction method combined with causal graph improved embedding is used to perform mechanical property quality prediction to obtain mechanical property prediction parameters, including the following steps: Step S31: inputting feature coding, specifically integrating the material constitutive prediction parameters and the metal product feature optimization data, performing feature coding for mechanical property quality prediction, and obtaining state variable feature data; Step S32: Constructing a mechanical causal graph for metal products, specifically by using structural equation modeling and a standard causal structure learning algorithm to extract mechanical features of metal products based on the state variable feature data to obtain causal dependency mechanical feature data, and then constructing graph data based on the causal dependency mechanical feature data to obtain mechanical causal graph data for metal products; The nodes of the metal product mechanical cause-effect diagram data are used to represent the metal product body; The edges of the metal product causal graph data are used to represent the causal influence of process and structural characteristics on mechanical properties, and specifically the causal influence is represented by the causal dependent mechanical characteristic data; The edge weight of the metal product causal graph data is used to represent the causal contribution value of each state variable feature; Step S33: constructing a reinforcement learning environment, specifically constructing state space parameters, action space parameters, state transfer function and reward function in sequence to build a basic reinforcement learning environment; The state space parameters are specifically constructed based on the state variable characteristic data; The action space parameters are specifically constructed by a set of adjustment amplitudes of process variables; The state transfer function is specifically constructed by simulating the influence of fine-tuning process parameters; The reward function is specifically constructed by predicting the error of the mechanical performance value; Step S34: constructing a causal embedding policy network, specifically by introducing a causal attention embedding module as a policy network weight in the reinforcement learning basic environment, obtaining causal weighted state vector data, and calculating and outputting current action space parameters using a policy function to obtain causal embedding action parameters; Step S35: constructing a mechanical causal embedding reward, specifically introducing a grade function to improve the causal embedding reward, and training the model parameters of the reinforcement learning basic environment through a policy gradient optimization algorithm to obtain a mechanical quality prediction model; Step S36: Mechanical performance quality prediction, specifically using the mechanical quality prediction model to perform mechanical performance quality prediction to obtain mechanical performance prediction parameters.

6. The method for detecting mechanical properties of metal products based on artificial intelligence according to claim 5, characterized in that: In step S36, the mechanical property prediction parameters specifically include mechanical property index prediction values and mechanical property grade rating parameters.

7. The method for detecting mechanical properties of metal products based on artificial intelligence according to claim 6, characterized in that: In step S4, the mechanical property test of the metal product is used to combine the model output with the actual test data for comprehensive analysis and testing. Specifically, the mechanical property prediction parameters and the actual metal product test experimental samples are combined to perform comparative analysis, error analysis and prediction deviation analysis to obtain comprehensive reference data for mechanical property testing.

8. An artificial intelligence-based metal product mechanical property testing system, used to implement the artificial intelligence-based metal product mechanical property testing method according to any one of claims 1 to 7, characterized in that: It includes data fusion analysis module, material constitutive performance prediction module, mechanical performance quality prediction module and metal product mechanical performance detection module.

9. The artificial intelligence-based metal product mechanical property detection system according to claim 8, characterized in that: The data fusion analysis module is used for data fusion analysis, obtains metal product feature optimization data through data fusion analysis, and sends the metal product feature optimization data to the material constitutive performance prediction module; The material constitutive performance prediction module is used to predict the material constitutive performance, obtain material constitutive prediction parameters through the material constitutive performance prediction, and send the material constitutive prediction parameters to the mechanical performance quality prediction module; The mechanical performance quality prediction module is used for mechanical performance quality prediction, obtains mechanical performance prediction parameters through mechanical performance quality prediction, and sends the mechanical performance prediction parameters to the mechanical performance detection module; The mechanical property detection module is used for detecting the mechanical properties of metal products, and obtains comprehensive reference data of the mechanical property detection of metal products through the mechanical property detection of metal products.

Images