Method for predicting performance degradation and estimating service life of tin-based metal material driven by recurrent neural network

Abstract

The invention discloses a tin-based metal material performance degradation prediction and life estimation method driven by a recurrent neural network, and the method comprises the steps: constructing a correlation model of microstructure and macroscopic performance mapping through the fusion of a microscopic feature parameter set and a dynamic behavior capture result, carrying out the deep learning of a nonlinear law of the material behavior change, and carrying out the prediction of the performance degradation of a tin-based metal material. A mapping relation model capable of comprehensively reflecting multiple environmental influences is obtained; according to the mapping relation model, analog simulation is carried out on material behavior changes under complex environmental factors, performance degradation trend prediction values under different environmental condition combinations are obtained, a simulation prediction data set is generated, the accuracy of prediction model construction is evaluated, and according to the dynamic behavior capture requirement of the tin-based metal material in practical application, the performance degradation trend prediction value is obtained. And constructing a real-time monitoring data input interface, obtaining comparison analysis of online environment factor data and a prediction result, and determining a final prediction model output result. The invention provides an effective method for reliability evaluation and life prediction of the material.

Description

Technical Field

[0001] This invention relates to the field of metal material prediction technology, and in particular to a method for predicting the performance degradation and lifespan of tin-based metal materials driven by recurrent neural networks. Background Technology

[0002] Tin-based metals are indispensable key materials in modern industry, widely used in electronics, energy, and machinery manufacturing. Their performance stability and service life directly affect the safe operation of equipment and the long-term reliability of systems. However, with the increasing complexity of application environments, material performance degradation and lifespan prediction have become major issues that urgently need to be addressed. Researching how to accurately predict material behavior changes under various complex conditions is of great significance for improving industrial production efficiency and reducing maintenance costs.

[0003] Currently, although some methods have been developed to analyze and predict the performance degradation of tin-based metal materials, these methods often neglect the dynamic response characteristics of materials under the coupling of multiple environmental factors, especially when faced with the interactive effects of variables such as temperature, humidity, and mechanical stress, lacking a comprehensive characterization of changes in material behavior. This limitation leads to significant deviations between the predicted results and actual operating conditions, making it difficult to meet the application requirements in complex environments.

[0004] Against this backdrop, the core challenges facing this field are becoming increasingly apparent. The performance degradation of tin-based metals under different environmental conditions exhibits highly nonlinear characteristics. This nonlinearity makes it difficult for traditional methods to accurately capture the dynamic behavior of materials as they change with time and external stimuli. Furthermore, this nonlinearity is closely related to the complex mapping relationship between the material's microstructure and macroscopic properties. Due to a lack of in-depth understanding of this mapping relationship, predictive models often fail to reflect the material's true response under real-world operating conditions, leading to insufficient accuracy in lifetime prediction.

[0005] Therefore, how to construct a predictive model that can comprehensively consider the influence of multiple environmental factors and accurately characterize the mapping relationship between the microstructure and macroscopic properties of materials has become a key issue in improving the accuracy of performance degradation prediction and lifetime estimation of tin-based metal materials. Summary of the Invention

[0006] To address the technical problems existing in the prior art, this invention proposes a method for predicting the performance degradation and lifetime of tin-based metal materials driven by a recurrent neural network, so as to accurately predict the performance degradation behavior of tin-based metal materials under complex environments.

[0007] To achieve the above objectives, this invention provides a method for predicting performance degradation and lifetime of tin-based metal materials driven by a recurrent neural network, comprising:

[0008] Experimental data of tin-based metal materials under different environmental conditions were collected to construct an initial dataset. Based on the initial dataset, a multi-environmental factor coupling model based on machine learning was constructed to obtain preliminary dynamic behavior capture results.

[0009] Based on the preliminary dynamic behavior capture results, the potential correlation between microstructure relationships and performance degradation characteristics is determined, and a set of micro-feature parameters is obtained. The set of micro-feature parameters is fused with the preliminary dynamic behavior capture results to construct a correlation model between microstructure and macro performance. Based on the correlation model between microstructure and macro performance, deep learning training is performed on the nonlinear laws of material behavior changes to obtain a mapping relationship model that can comprehensively reflect multiple environmental influences.

[0010] The material behavior changes under complex environmental factors are simulated using the mapping relationship model to obtain the predicted values ​​of performance degradation trends under different combinations of environmental conditions and generate a simulation prediction dataset.

[0011] The accuracy of the prediction model is optimized using the simulation prediction dataset to obtain an optimized performance degradation prediction framework. Based on the optimized performance degradation prediction framework, a real-time monitoring data input interface is constructed to meet the dynamic behavior capture requirements of tin-based metal materials in practical applications. The online environmental factor data and prediction results are compared and analyzed to determine the final prediction model output.

[0012] Compared with the prior art, the present invention has the following advantages and technical effects:

[0013] This invention constructs an initial dataset by collecting experimental data under various environmental conditions; preprocesses and extracts features from the data to obtain a standardized feature dataset; builds a multi-environmental factor coupling model based on machine learning to capture dynamic behavior; establishes a micro-macro mapping relationship model by combining microstructure data; conducts simulations to predict performance degradation trends under different environments; optimizes the prediction model through cross-validation; and finally, constructs a real-time monitoring interface to achieve online prediction. This invention can comprehensively consider the coupled effects of multiple environmental factors such as temperature, humidity, and mechanical stress, accurately predicting the performance degradation behavior of tin-based metal materials under complex environments, providing an effective method for material reliability assessment and lifetime prediction. Attached Figure Description

[0014] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0015] Figure 1This is a flowchart of a method for predicting performance degradation and lifetime of tin-based metal materials driven by a recurrent neural network, according to an embodiment of the present invention. Detailed Implementation

[0016] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0017] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0018] This embodiment proposes a method for predicting performance degradation and lifetime of tin-based metal materials driven by a recurrent neural network, such as... Figure 1 ,include:

[0019] A method for predicting performance degradation and lifetime of tin-based metal materials driven by recurrent neural networks, characterized by comprising:

[0020] Experimental data of tin-based metal materials under different environmental conditions were collected to construct an initial dataset. Based on the initial dataset, a multi-environmental factor coupling model based on machine learning was constructed to obtain preliminary dynamic behavior capture results.

[0021] Based on the preliminary dynamic behavior capture results, the potential correlation between microstructure relationships and performance degradation characteristics is determined, and a set of micro-feature parameters is obtained. The set of micro-feature parameters is fused with the preliminary dynamic behavior capture results to construct a correlation model between microstructure and macro performance. Based on the correlation model between microstructure and macro performance, deep learning training is performed on the nonlinear laws of material behavior changes to obtain a mapping relationship model that can comprehensively reflect multiple environmental influences.

[0022] The material behavior changes under complex environmental factors are simulated using the mapping relationship model to obtain the predicted values ​​of performance degradation trends under different combinations of environmental conditions and generate a simulation prediction dataset.

[0023] The accuracy of the prediction model is optimized using the simulation prediction dataset to obtain an optimized performance degradation prediction framework. Based on the optimized performance degradation prediction framework, a real-time monitoring data input interface is constructed to meet the dynamic behavior capture requirements of tin-based metal materials in practical applications. The online environmental factor data and prediction results are compared and analyzed to determine the final prediction model output.

[0024] Specifically, this embodiment constructs an initial dataset by collecting experimental data under various environmental conditions; preprocesses and extracts features from the data to obtain a standardized feature dataset; constructs a multi-environmental factor coupling model based on machine learning to capture dynamic behavior; establishes a micro-macro mapping relationship model by combining microstructure data; performs simulation to predict performance degradation trends under different environments; optimizes the prediction model through cross-validation; and finally constructs a real-time monitoring interface to achieve online prediction.

[0025] Furthermore, preliminary dynamic behavior capture results are obtained, including:

[0026] By collecting experimental data on tin-based metal materials under different environmental conditions, test results including environmental factors such as temperature, humidity, and mechanical stress are obtained, and the initial dataset is constructed to obtain preliminary information on the distribution of environmental impacts.

[0027] Based on the preliminary environmental impact distribution information, data preprocessing techniques are used to clean and standardize the dataset in the initial dataset. Feature extraction is performed to target the effects of multiple coupled environmental factors, and the weight of the key environmental variable combinations on the performance degradation features is determined to obtain the standardized feature dataset.

[0028] Based on the standardized feature dataset, the machine learning-based multi-environmental factor coupling model is constructed. The model is trained and its parameters are optimized for highly nonlinear performance degradation features to obtain a prediction framework that can reflect the interactive effects of complex environmental factors, thus obtaining the preliminary dynamic behavior capture results.

[0029] Specifically, for experimental data acquisition of tin-based metals, a systematic experimental procedure can be designed to record multi-dimensional environmental factors such as temperature, humidity, and mechanical stress. Assuming the experiment is conducted in a laboratory, the temperature range is set to 20℃ to 80℃, the humidity range to 30% to 90%, and the mechanical stress is applied using a tensile testing machine, ranging from 50MPa to 500MPa. The data acquisition equipment includes a high-precision temperature sensor, a hygrometer, and a stress testing instrument. Data is recorded once per second, generating an initial dataset containing timestamps and environmental parameters. This approach ensures the comprehensiveness of the data, providing a reliable foundation for subsequent analysis.

[0030] If temperature data shows values ​​above 100℃ or below 0℃, which are significantly outside the experimental range, these can be considered outliers and removed directly. For noisy data, such as fluctuations caused by transient interference from humidity sensors, median filtering can be used for smoothing. The cleaned dataset needs to be standardized, such as converting temperature and humidity to a uniform unit, to ensure data quality meets analytical requirements.

[0031] Data cleaning can improve the accuracy of subsequent modeling and reduce interference from invalid data.

[0032] Specifically, when marking data points that exceed the threshold, the temperature threshold can be set to 25℃ to 75℃, and the humidity threshold to 40% to 80%. If a data point has a temperature of 80℃ or a humidity of 90%, it is marked as a "high-risk point." The marked dataset can intuitively reflect the deviation of environmental factors.

[0033] For example, the labeling results show that 10% of the data points are generated under high temperature and humidity conditions, indicating that environmental deviations may cause performance anomalies. This labeling method helps to quickly locate key issues and improve analysis efficiency. In the feature extraction stage, key features such as temperature change rate, humidity fluctuation amplitude, and stress peak can be extracted from the labeled dataset.

[0034] Under high temperature and humidity conditions, the tensile strength of tin-based metals decreases by 10%, indicating a correlation between environmental factors and performance degradation. When using support vector machines for modeling, temperature, humidity, and stress can be used as input features, with the performance degradation rate as the output, to train the model and identify key influence weights. Assuming the model shows a weight of 0.6 for temperature, 0.3 for humidity, and 0.1 for stress, it indicates that temperature is the dominant factor. This modeling approach can accurately quantify environmental impacts, providing a basis for optimized design. In simulation analysis, different environmental combinations can be simulated based on the predictive model. For example, at a temperature of 70℃ and humidity of 80%, the performance degradation rate is predicted to be 15%; while at a temperature of 50℃ and humidity of 50%, the degradation rate is only 5%. This simulation reveals the significant negative impact of high temperature and humidity on tin-based metals, providing guidance for optimizing material application scenarios.

[0035] Furthermore, the preliminary environmental impact distribution information is obtained, including:

[0036] Based on the initial dataset, outliers and noisy data are removed using data cleaning methods to obtain a processed standardized dataset, and the data quality is determined to meet the requirements of subsequent analysis.

[0037] If the temperature or humidity conditions in the standardized dataset exceed the preset threshold range, the relevant data points are marked, the marked dataset is obtained, and it is determined whether there is a significant environmental factor bias.

[0038] By extracting features from the labeled dataset, the correlation between performance degradation and material behavior is analyzed to obtain the mapping relationship between key environmental factors and performance degradation.

[0039] Based on the mapping relationship, the support vector machine algorithm is used to model the change law of performance degradation, obtain the prediction model, and determine the influence weight of environmental factors on material behavior;

[0040] By simulating and analyzing preliminary information on environmental distribution through predictive models, the performance degradation trend under different combinations of environmental factors is obtained, and the behavior of tin-based metals under different conditions is judged.

[0041] Based on the performance degradation trend obtained from the simulation analysis, data visualization processing is performed on the change pattern to obtain an intuitive environmental impact distribution map, thus obtaining the preliminary environmental impact distribution information.

[0042] Specifically, the initial dataset is preprocessed using cleaning techniques, employing denoising methods and missing value imputation to obtain a cleaned dataset, ensuring data integrity and consistency. If the distribution of environmental factors in the cleaned dataset deviates from a preset threshold, cluster analysis is used to group the data points, resulting in a grouped dataset to assess the heterogeneity of environmental factors. Based on the grouped dataset, principal component analysis is used to extract key variables, analyze the coupling effects of multiple environmental factors, and obtain a dimensionality-reduced feature dataset to identify the main influencing factors. Using the dimensionality-reduced feature dataset, correlation analysis is used to calculate the correlation strength between key variables and performance degradation, obtaining the quantitative results of variable influence weights and assessing weight distribution characteristics. If significant coupling effects exist in the quantitative results of variable influence weights, a decision tree algorithm is used to model the combination of key variables, obtaining a performance degradation prediction model and determining the impact path of environmental factors. Based on the performance degradation prediction model, data visualization techniques are used to display the influence trend of the combination of key variables, obtaining a visualized distribution of environmental factor effects and assessing the interaction effects of multiple factors. By visualizing the distribution, statistical testing methods were used to verify the interaction effects of environmental factors, obtain significance test results, and determine the contribution of the combination of key variables to performance degradation.

[0043] Furthermore, based on the standardized feature dataset, the machine learning-based multi-environmental factor coupling model is constructed. Training and parameter optimization are performed on the highly nonlinear performance degradation characteristics to obtain a predictive framework that reflects the interactive effects of complex environmental factors, resulting in the preliminary dynamic behavior capture results, including:

[0044] Based on the standardized feature dataset, a multi-environmental factor coupling model is constructed using the support vector machine algorithm;

[0045] During training, if the feature dimension is higher than the preset feature dimension threshold, principal component analysis is used to reduce the dimension and obtain an initial coupled model. The initial coupled model is then used to train the model to target the nonlinear performance degradation feature.

[0046] If the training error exceeds the preset error threshold, the hyperparameters are adjusted using a grid search method to obtain the optimized coupled model.

[0047] Based on the optimized coupling model, the interaction weights of environmental factors are obtained;

[0048] If the weight value is lower than the preset weight threshold, the corresponding feature is removed to obtain a simplified prediction framework;

[0049] Using the simplified prediction framework, the random forest algorithm is employed to extract dynamic behavioral features;

[0050] If the feature importance is lower than the preset importance, the model is retrained to obtain the dynamic behavior prediction results;

[0051] Based on the dynamic behavior prediction results, the performance degradation trend is obtained and smoothed using a sliding window method to obtain preliminary capture results;

[0052] Based on the preliminary capture results, the stability of the prediction framework is evaluated using a cross-validation method. If the validation error is lower than a preset error threshold, the final prediction framework is determined, and the preliminary dynamic behavior capture results are obtained.

[0053] Specifically, when constructing a multi-environmental factor coupling model using the support vector machine algorithm, a radial basis function kernel can be selected to handle nonlinear relationships. Assuming the dataset contains 1000 records involving features such as temperature, humidity, and wind speed, the model seeks the optimal hyperplane to separate data from those exhibiting performance degradation and those in normal conditions.

[0054] It should be noted that if the feature dimension is too high, such as exceeding 20 dimensions, principal component analysis is used to reduce the dimensionality, retaining 90% of the variance to generate an initial coupled model. This dimensionality reduction method reduces computational complexity while preserving key information.

[0055] In one embodiment, if the training error exceeds a preset threshold, such as an error greater than 0.1, a grid search can be used to adjust the regularization parameter C and kernel function parameter γ of the support vector machine. For example, the optimal combination of C between 0.1 and 10 and γ between 0.01 and 1 is searched to obtain an optimized coupled model. This method can improve the model's ability to fit complex environmental factors. By calculating the interaction weights of environmental factors through the model, if the weight of humidity on performance degradation is less than 0.05, this feature is removed, generating a simplified prediction framework and reducing the risk of overfitting.

[0056] When using the Random Forest algorithm to extract dynamic behavioral features, the importance of features can be evaluated using 100 decision trees. Assuming temperature and air pressure have importance values ​​of 0.4 and 0.3 respectively, while wind speed is only 0.05, the wind speed feature is removed, and the model is retrained. This approach focuses on key environmental factors, improving prediction accuracy.

[0057] It should be noted that if the trend fluctuation exceeds the preset threshold, such as the variance of the performance degradation prediction value being greater than 0.2, the data can be smoothed by taking the 7-day average value using the sliding window method to obtain a stable capture result.

[0058] In one possible implementation, cross-validation is used to evaluate the stability of the predictive framework. For example, using 5-fold cross-validation, the dataset is divided into 5 parts, with 4 parts used for training and 1 part for validation each time. If the average error is below 0.05, the model's stability is confirmed. This method ensures the model's generalization ability across different subsets of data. The final dynamic behavior capture results can be visualized, such as plotting trends in temperature and performance degradation, intuitively reflecting the impact of environmental factors. This approach facilitates analysts' understanding of the interactive effects of multiple factors.

[0059] Furthermore, based on the preliminary dynamic behavior capture results, the potential correlation between microstructure relationships and performance degradation characteristics is determined, and a set of micro-feature parameters is obtained, including:

[0060] The microstructure of tin-based metal materials is scanned using image acquisition technology to obtain preliminary microstructure image data. Based on the preliminary microstructure image data, image processing technology is used to segment and extract morphological features to determine the boundaries and distribution information of morphological features.

[0061] Based on the distribution information of morphological features, feature parameters are calculated through quantitative analysis methods to obtain a set of geometric and spatial distribution parameters of the microstructure. If some parameters in the set of geometric and spatial distribution parameters exceed the preset range, local image enhancement processing is performed on the relevant areas to obtain clearer local feature image data.

[0062] Based on local feature image data and dynamic behavior datasets, we analyze the potential relationship between structural relationships and performance degradation, and identify the key influencing factors of performance degradation.

[0063] By comparing and analyzing the key influencing factors and feature parameter sets, the support vector machine algorithm is used to predict the performance degradation trend, and the quantitative result of the degradation trend is obtained.

[0064] Based on the quantitative results of the aforementioned decline trend, and combined with the characteristics of material information, the direction and strategy for microstructure adjustment are determined, and the set of microscopic feature parameters is output.

[0065] Specifically, an initial dynamic behavior dataset is obtained by capturing the correlation between data and dynamic behavior. When studying the performance degradation of tin-based metal materials, high-precision sensors can be used to collect dynamic response data of the material under different temperature, humidity, and stress conditions, such as resistivity changes or mechanical strength decay, generating a dynamic behavior dataset containing time series data. This data reflects the behavioral characteristics of the material in complex environments, providing a foundation for subsequent analysis. Image acquisition technology is used to scan the microstructure of tin-based metal materials, obtaining preliminary microstructure image data.

[0066] For example, scanning electron microscopy can be used to perform high-resolution imaging of tin-based metal surfaces, acquiring microscopic morphological information such as grain size and grain boundary distribution. It is important to note that image acquisition must ensure sufficiently high resolution, such as a magnification of 10,000x, to capture micron-level defects. Based on the preliminary microstructure image data, image processing techniques are used to segment and extract morphological features, determining their boundaries and distribution information. Grain boundaries and pores can be identified using edge detection algorithms, combined with threshold segmentation techniques to extract grain shape and distribution features, ensuring clear boundaries and accurate distribution information. Noise filtering techniques can be introduced during segmentation to improve image quality. For the distribution information of morphological features, feature parameters are calculated using quantitative analysis methods to obtain the set of geometric and spatial distribution parameters of the microstructure.

[0067] The average diameter, area ratio, and porosity of the grains are calculated to generate a quantized dataset containing geometric features. If some parameters exceed preset thresholds, such as porosity exceeding 5%, further analysis of their impact on performance is required. If some parameters in the geometric and spatial distribution parameter set exceed preset threshold ranges, local image enhancement processing is performed on the relevant areas to obtain clearer local feature image data.

[0068] If the image contrast of the grain boundary region is insufficient, histogram equalization can be used to enhance local features, facilitating subsequent analysis of key defect areas. Based on the local feature image data, combined with dynamic behavior datasets, the potential relationship between structural relationships and performance degradation is analyzed to identify key influencing factors of performance degradation.

[0069] By combining data on grain size and resistivity variations, it was found that smaller grain sizes may lead to increased resistivity, indicating that grain boundary density is a key factor in performance degradation. Through comparative analysis of key influencing factors and characteristic parameter sets, a support vector machine algorithm was used to predict the performance degradation trend, yielding a quantitative result of the degradation trend.

[0070] For example, based on features such as grain boundary density and porosity, a support vector machine model is trained to predict the performance degradation of a material after 1000 hours of use, such as a 10% reduction in strength. Based on the quantified results of the degradation trend and combined with the characteristics of the material information, the direction and strategy for microstructure adjustment are determined, and an optimized set of feature parameters is output. Adjusting the grain size to the 5-10 micrometer range reduces grain boundary density, and optimizing porosity to below 2% improves the long-term stability of the material.

[0071] It should be noted that the above method, through multi-dimensional data collection and analysis, combined with image processing and machine learning techniques, can effectively identify key factors of performance degradation and propose optimization strategies, providing a reliable technical path for improving the performance of tin-based metal materials.

[0072] Furthermore, based on the correlation model between the microstructure and macroscopic performance mapping, deep learning training is performed to target the nonlinear laws governing changes in material behavior, thereby obtaining a mapping relationship model that comprehensively reflects multiple environmental influences, including:

[0073] Data fusion technology is used to integrate the set of micro-feature parameters and the preliminary dynamic behavior capture results to obtain a preliminary feature behavior dataset;

[0074] Based on the preliminary feature behavior dataset, feature extraction is performed on the nonlinear features of behavior changes to obtain a feature subset that can reflect the change pattern. Through the correspondence information between the feature subset and the macro performance, the initial structure of the mapping framework is constructed, and the association weights of each key node in the framework are determined.

[0075] The initial structure of the mapping framework is optimized and trained using deep learning methods, and the parameters are adjusted to account for the influence of environmental factors, resulting in a mapping relationship model that can comprehensively reflect multiple environmental influences.

[0076] Specifically, when integrating data on the correlation between microscopic features and dynamic behavior, data fusion techniques can be used to achieve multi-source data registration and feature alignment. First, grain size distribution data of tin-based metal materials are collected at the microscopic level, combined with strain response time series in dynamic behavior.

[0077] In one embodiment, principal component analysis (PCA) is used to reduce the dimensionality of multidimensional data on grain size and strain response, generating a fused dataset containing key features. For example, under conditions of an average grain size of 8 micrometers and a strain response time of 0.2 seconds, a comprehensive feature vector reflecting the material's deformation sensitivity can be extracted from the fused dataset. This method effectively integrates multidimensional information, retains key characteristics, and provides a reliable foundation for subsequent analysis. Specifically, for the extraction of nonlinear features related to behavioral changes, wavelet transform technology can be used to decompose the nonlinear signals in the dynamic behavioral data. For example, by performing wavelet transform on the resistivity change data of tin-based metal materials at 50°C, high-frequency noise and low-frequency trend signals can be separated, and a key feature subset reflecting resistivity abrupt changes, such as a signal component with an abrupt change frequency of 0.5 Hz, can be extracted. This feature subset can clearly characterize the nonlinear response law of the material under specific conditions, facilitating subsequent mapping analysis.

[0078] In one embodiment, when constructing the initial structure of the mapping framework, the weights of key nodes can be determined through correlation analysis. For example, analyzing the relationship between grain size and material fatigue life reveals that when the grain size is in the range of 5-10 micrometers, the fatigue life is extended to over 10,000 cycles. Based on this, the weight of the grain size node can be set to 0.6, and the weight of the porosity node to 0.3. This weight distribution reflects the degree of influence of each feature on macroscopic performance, providing a quantitative basis for framework construction. When optimizing the mapping framework using deep learning methods, a convolutional neural network can be used to train the fused dataset.

[0079] In one possible implementation, grain size, porosity, and stress response data are input into the network, with a learning rate of 0.001, and trained for 100 epochs to optimize the model's ability to predict material strength degradation. After training, the model can adapt to performance changes under different stress conditions, improving prediction accuracy.

[0080] Furthermore, by using a mapping relationship model to simulate the changes in material behavior under complex environmental factors, predicted values ​​of performance degradation trends under different combinations of environmental conditions are obtained, generating a simulation prediction dataset, including:

[0081] The combined effects of environmental conditions are analyzed using simulation technology to obtain the performance degradation trend of materials under different conditions. A prediction model is constructed based on the performance degradation trend, and the trend prediction is trained using the support vector machine algorithm to obtain preliminary prediction results.

[0082] If the deviation between the preliminary prediction results and the actual performance degradation trend exceeds the preset deviation value, information enhancement processing is performed on the environmental condition-related data to obtain an optimized dataset.

[0083] By adjusting the parameter distribution of the prediction model using the optimized dataset, the model's adaptability under different working conditions is judged. Based on the analysis results of the adaptability, the mapping relationship between material behavior changes and environmental condition combinations is dynamically updated to obtain the updated simulation dataset, thereby generating the simulation prediction dataset.

[0084] Specifically, during data acquisition in complex environments, changes in material behavior can be monitored in real time using a multi-sensor system. High-precision sensor arrays are deployed to acquire multi-dimensional information, addressing various environmental factors such as temperature, humidity, and stress. Assuming an experimental scenario with a temperature range of -20 to 80 degrees Celsius and humidity controlled between 30% and 90%, data is collected every minute by the sensors to form an initial dataset. This approach comprehensively captures the impact of environmental conditions on material behavior, laying the foundation for subsequent analysis.

[0085] When building a predictive model and training it using the Support Vector Machine (SVM) algorithm, the dataset can be divided into a training set and a validation set in an 8:2 ratio. During training, the focus is on the nonlinear relationship between environmental variables and performance degradation. Initial predictions might show a 15% bias in the degradation rate prediction under extreme temperatures. This method helps identify weaknesses in the model and provides direction for subsequent optimization. For example, if the initial prediction bias exceeds a preset threshold, such as 10%, information augmentation can be performed on the environmental data. Data quality can be optimized by increasing the sampling frequency or introducing noise filtering techniques. Assuming that after augmentation, the number of data points increases from 5000 to 8000, and the prediction bias decreases to 8%, this demonstrates that data augmentation has a significant effect on improving model accuracy.

[0086] When adjusting the parameter distribution of the prediction model, the model weights can be recalibrated based on the optimized dataset to test the model's adaptability under various operating conditions. For example, assuming the model's predicted performance degradation rate deviates from the actual value by only 5% in a low-temperature, low-humidity environment, and by 7% in a high-temperature, high-humidity environment, this indicates good model adaptability. This adjustment ensures the model's reliability in different scenarios.

[0087] Furthermore, using the simulation prediction dataset, the accuracy of the mapping relationship model is optimized to obtain an optimized performance degradation prediction framework, including:

[0088] By simulating the prediction dataset, cross-validation technology is used to divide it into training and validation sets to determine the preliminary accuracy of the prediction model.

[0089] If the initial accuracy is lower than the preset accuracy threshold, then for the environmental factor coupling scenario, high-bias samples are extracted from the training set to obtain the combination of environmental conditions with large deviations.

[0090] Based on high-biased samples, the gradient descent algorithm is used to adjust the parameters of the prediction model, resulting in an updated model parameter distribution. The updated model parameter distribution is then used to predict the performance degradation trend of the validation set, and the optimized prediction results are obtained.

[0091] If the deviation between the optimized prediction results and the actual performance degradation trend still exceeds the preset accuracy threshold, feature enhancement processing is performed on the combination of environmental conditions to obtain the enhanced feature dataset.

[0092] By using the enhanced feature dataset, the prediction model is retrained using the random forest algorithm to obtain a further optimized performance degradation prediction framework. Based on the optimized performance degradation prediction framework, the simulation prediction dataset is continuously validated to determine the accuracy of the final prediction model.

[0093] Specifically, assuming that under high temperature and high humidity conditions (e.g., temperature 80 degrees Celsius, humidity 90%), the model's predicted material performance degradation rate deviates from the actual value by 12%. Analysis of these samples reveals that the coupling effect of temperature and humidity is the main source of this deviation. For example, these high-deviation samples can be separately categorized to form a subset dataset containing approximately 1000 samples for subsequent parameter adjustment. When optimizing model parameters using the gradient descent algorithm, the weights are adjusted primarily for these high-deviation samples.

[0094] For example, the impact of temperature-humidity coupling effects on the model can be reduced through iterative updates. Assuming that after 10 iterations, the model parameters converge, the optimized parameter distribution reduces the prediction bias on the validation set from 12% to 8%. This optimization method can improve the model's adaptability to complex environmental conditions.

[0095] Furthermore, based on the optimized performance degradation prediction framework, and considering the need to capture the dynamic behavior of tin-based metal materials in practical applications, a real-time monitoring data input interface is constructed. This interface is used to obtain online environmental factor data and compare it with the prediction results to determine the final prediction model output, including:

[0096] Dynamic behavior data related to tin-based metals are acquired from the online environment, and environmental factor information is obtained using a pre-set sensor network to obtain a preliminary environmental data set;

[0097] Based on the initial environmental dataset, data cleaning tools were used to remove noise and outliers. The cleaned data was then standardized to determine a structured environmental feature dataset.

[0098] If certain key indicators in the structured environmental feature dataset exceed the preset threshold, the anomaly detection mechanism is triggered. The abnormal data is labeled according to the preset logical rules to obtain the labeled feature dataset. For the labeled feature dataset, the random forest algorithm is used to perform preliminary predictive analysis to obtain prediction result data related to dynamic behavior and determine the distribution characteristics of the prediction results.

[0099] Based on the distribution characteristics of the prediction results, a comparative analysis is performed with the actual data obtained from real-time monitoring. If the difference exceeds the preset difference range, the parameters of the prediction model are adjusted to obtain optimized prediction parameters.

[0100] By optimizing the prediction parameters and updating the configuration of the prediction model, continuous predictions are made on the dynamic behavior of tin-based metals in practical applications to determine the final output results.

[0101] Specifically, in the implementation of the anomaly detection mechanism, if the temperature index in the structured dataset exceeds the preset threshold of 45 degrees Celsius for three consecutive days, anomaly detection is triggered. These data are then labeled as potential risk points using preset logical rules, forming a labeled feature dataset. This labeling method facilitates rapid location of problematic data during subsequent analysis, enhancing the targeted nature of data processing.

[0102] When performing preliminary predictive analysis on the labeled feature dataset, the random forest algorithm can effectively capture the nonlinear relationship between the dynamic behavior of tin-based metals and environmental factors. Assuming the prediction results show that the metal surface corrosion rate is faster under high temperature and humidity conditions, and the distribution characteristics show that the predicted values ​​are concentrated in the higher-risk range, this analytical approach helps to reveal the influence of the environment on metal behavior, providing a basis for subsequent optimization.

[0103] During the comparative analysis phase, if the predicted results differ significantly from the real-time monitoring data—for example, if the predicted corrosion rate is 20% higher than the actual value—then the prediction model parameters need to be adjusted. This can be achieved by increasing the weight of high-temperature and high-humidity scenarios to optimize the prediction parameters and make the model more closely match the actual application environment. This adjustment method can improve the accuracy of predictions and reduce bias.

[0104] After updating the prediction model configuration, continuous predictions are made regarding the dynamic behavior of tin-based metals in practical applications, with prediction results output at regular intervals. For example, if a prediction indicates a potential significant performance degradation in a certain region within the next week, timely output facilitates the implementation of protective measures. This continuous prediction mechanism provides real-time references for practical applications, enhancing response capabilities.

[0105] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting performance degradation and lifetime of tin-based metal materials driven by recurrent neural networks, characterized in that, include: Experimental data of tin-based metal materials under different environmental conditions were collected to construct an initial dataset. Based on the initial dataset, a multi-environmental factor coupling model based on machine learning was constructed to obtain preliminary dynamic behavior capture results. Based on the preliminary dynamic behavior capture results, the potential correlation between microstructure relationships and performance degradation characteristics is determined, and a set of micro-feature parameters is obtained. The set of micro-feature parameters is fused with the preliminary dynamic behavior capture results to construct a correlation model between microstructure and macro performance. Based on the correlation model between microstructure and macro performance, deep learning training is performed on the nonlinear laws of material behavior changes to obtain a mapping relationship model that can comprehensively reflect multiple environmental influences. The material behavior changes under complex environmental factors are simulated using the mapping relationship model to obtain the predicted values ​​of performance degradation trends under different combinations of environmental conditions and generate a simulation prediction dataset. The accuracy of the prediction model is optimized using the simulation prediction dataset to obtain an optimized performance degradation prediction framework. Based on the optimized performance degradation prediction framework, a real-time monitoring data input interface is constructed to meet the dynamic behavior capture requirements of tin-based metal materials in practical applications. The online environmental factor data and prediction results are compared and analyzed to determine the final prediction model output.

2. The method for predicting performance degradation and lifetime of tin-based metal materials driven by recurrent neural networks according to claim 1, characterized in that, Obtaining the preliminary dynamic behavior capture results includes: By collecting experimental data on tin-based metal materials under different environmental conditions, test results including environmental factors such as temperature, humidity, and mechanical stress are obtained, and the initial dataset is constructed to obtain preliminary information on the distribution of environmental impacts. Based on the preliminary environmental impact distribution information, data preprocessing techniques are used to clean and standardize the dataset in the initial dataset. Feature extraction is performed to target the effects of multiple coupled environmental factors, and the weight of the key environmental variable combinations on the performance degradation features is determined to obtain the standardized feature dataset. Based on the standardized feature dataset, the machine learning-based multi-environmental factor coupling model is constructed. The model is trained and its parameters are optimized for highly nonlinear performance degradation features to obtain a prediction framework that can reflect the interactive effects of complex environmental factors, thus obtaining the preliminary dynamic behavior capture results.

3. The method for predicting performance degradation and lifetime of tin-based metal materials driven by recurrent neural networks according to claim 2, characterized in that, The preliminary environmental impact distribution information is obtained, including: Based on the initial dataset, outliers and noisy data are removed using data cleaning methods to obtain a processed standardized dataset, and the data quality is determined to meet the requirements of subsequent analysis. If the temperature or humidity conditions in the standardized dataset exceed the preset threshold range, the relevant data points are marked, the marked dataset is obtained, and it is determined whether there is a significant environmental factor bias. By extracting features from the labeled dataset, the correlation between performance degradation and material behavior is analyzed to obtain the mapping relationship between key environmental factors and performance degradation. Based on the mapping relationship, the support vector machine algorithm is used to model the change law of performance degradation, obtain the prediction model, and determine the influence weight of environmental factors on material behavior; By simulating and analyzing preliminary information on environmental distribution through predictive models, the performance degradation trend under different combinations of environmental factors is obtained, and the behavior of tin-based metals under different conditions is judged. Based on the performance degradation trend obtained from the simulation analysis, data visualization processing is performed on the change pattern to obtain an intuitive environmental impact distribution map, thus obtaining the preliminary environmental impact distribution information.

4. The method for predicting performance degradation and lifetime of tin-based metal materials driven by recurrent neural networks according to claim 2, characterized in that, Based on the standardized feature dataset, a machine learning-based multi-environmental factor coupling model is constructed. Training and parameter optimization are performed on the highly nonlinear performance degradation characteristics to obtain a predictive framework that reflects the interactive effects of complex environmental factors, resulting in the preliminary dynamic behavior capture results, including: Based on the standardized feature dataset, a multi-environmental factor coupling model is constructed using the support vector machine algorithm; During training, if the feature dimension is higher than the preset feature dimension threshold, principal component analysis is used to reduce the dimension and obtain an initial coupled model. The initial coupled model is then used to train the model to target the nonlinear performance degradation feature. If the training error exceeds the preset error threshold, the hyperparameters are adjusted using a grid search method to obtain the optimized coupled model. Based on the optimized coupling model, the interaction weights of environmental factors are obtained; If the weight value is lower than the preset weight threshold, the corresponding feature is removed to obtain a simplified prediction framework; Using the simplified prediction framework, the random forest algorithm is employed to extract dynamic behavioral features; If the feature importance is lower than the preset importance, the model is retrained to obtain the dynamic behavior prediction results; Based on the dynamic behavior prediction results, the performance degradation trend is obtained and smoothed using a sliding window method to obtain preliminary capture results; Based on the preliminary capture results, the stability of the prediction framework is evaluated using a cross-validation method. If the validation error is lower than a preset error threshold, the final prediction framework is determined, and the preliminary dynamic behavior capture results are obtained.

5. The method for predicting performance degradation and lifetime of tin-based metal materials driven by a recurrent neural network according to claim 1, characterized in that, Based on the preliminary dynamic behavior capture results, the potential correlation between microstructure relationships and performance degradation characteristics is determined, and a set of micro-feature parameters is obtained, including: The microstructure of tin-based metal materials is scanned using image acquisition technology to obtain preliminary microstructure image data. Based on the preliminary microstructure image data, image processing technology is used to segment and extract morphological features to determine the boundaries and distribution information of morphological features. Based on the distribution information of morphological features, feature parameters are calculated through quantitative analysis methods to obtain a set of geometric and spatial distribution parameters of the microstructure. If some parameters in the set of geometric and spatial distribution parameters exceed the preset range, local image enhancement processing is performed on the relevant areas to obtain clearer local feature image data. Based on local feature image data and dynamic behavior datasets, we analyze the potential relationship between structural relationships and performance degradation, and identify the key influencing factors of performance degradation. By comparing and analyzing the key influencing factors and feature parameter sets, the support vector machine algorithm is used to predict the performance degradation trend, and the quantitative result of the degradation trend is obtained. Based on the quantitative results of the aforementioned decline trend, and combined with the characteristics of material information, the direction and strategy for microstructure adjustment are determined, and the set of microscopic feature parameters is output.

6. The method for predicting performance degradation and lifetime of tin-based metal materials driven by recurrent neural networks according to claim 1, characterized in that, Based on the correlation model between the microstructure and macroscopic performance, deep learning training is performed to target the nonlinear laws governing changes in material behavior, resulting in a mapping relationship model that comprehensively reflects multiple environmental influences, including: Data fusion technology is used to integrate the set of micro-feature parameters and the preliminary dynamic behavior capture results to obtain a preliminary feature behavior dataset; Based on the preliminary feature behavior dataset, feature extraction is performed on the nonlinear features of behavior changes to obtain a feature subset that can reflect the change pattern. Through the correspondence information between the feature subset and the macro performance, the initial structure of the mapping framework is constructed, and the association weights of each key node in the framework are determined. The initial structure of the mapping framework is optimized and trained using deep learning methods, and the parameters are adjusted to account for the influence of environmental factors, resulting in a mapping relationship model that can comprehensively reflect multiple environmental influences.

7. The method for predicting performance degradation and lifetime of tin-based metal materials driven by a recurrent neural network according to claim 1, characterized in that, The mapping relationship model is used to simulate the changes in material behavior under complex environmental factors, obtain predicted values ​​of performance degradation trends under different combinations of environmental conditions, and generate a simulation prediction dataset, including: The combined effects of environmental conditions are analyzed using simulation technology to obtain the performance degradation trend of materials under different conditions. A prediction model is constructed based on the performance degradation trend, and the trend prediction is trained using the support vector machine algorithm to obtain preliminary prediction results. If the deviation between the preliminary prediction results and the actual performance degradation trend exceeds the preset deviation value, information enhancement processing is performed on the environmental condition-related data to obtain an optimized dataset. By adjusting the parameter distribution of the prediction model using the optimized dataset, the model's adaptability under different working conditions is judged. Based on the analysis results of the adaptability, the mapping relationship between material behavior changes and environmental condition combinations is dynamically updated to obtain the updated simulation dataset, thereby generating the simulation prediction dataset.

8. The method for predicting performance degradation and lifetime of tin-based metal materials driven by a recurrent neural network according to claim 7, characterized in that, Using the simulation prediction dataset, the accuracy of the prediction model is optimized to obtain an optimized performance degradation prediction framework, including: By simulating the prediction dataset, cross-validation technology is used to divide it into training and validation sets to determine the preliminary accuracy of the prediction model. If the initial accuracy is lower than the preset accuracy threshold, then for the environmental factor coupling scenario, high-bias samples are extracted from the training set to obtain the combination of environmental conditions with large deviations. Based on high-biased samples, the gradient descent algorithm is used to adjust the parameters of the prediction model, resulting in an updated model parameter distribution. The updated model parameter distribution is then used to predict the performance degradation trend of the validation set, and the optimized prediction results are obtained. If the deviation between the optimized prediction results and the actual performance degradation trend still exceeds the preset accuracy threshold, feature enhancement processing is performed on the combination of environmental conditions to obtain the enhanced feature dataset. By using the enhanced feature dataset, the prediction model is retrained using the random forest algorithm to obtain a further optimized performance degradation prediction framework. Based on the optimized performance degradation prediction framework, the simulation prediction dataset is continuously validated to determine the accuracy of the final prediction model.

9. The method for predicting performance degradation and lifetime of tin-based metal materials driven by a recurrent neural network according to claim 8, characterized in that, Based on the optimized performance degradation prediction framework, and addressing the need to capture the dynamic behavior of tin-based metal materials in practical applications, a real-time monitoring data input interface is constructed. This interface is used to obtain online environmental factor data and compare it with the prediction results to determine the final prediction model output, including: Dynamic behavior data related to tin-based metals are acquired from the online environment, and environmental factor information is obtained using a pre-set sensor network to obtain a preliminary environmental data set; Based on the initial environmental dataset, data cleaning tools were used to remove noise and outliers. The cleaned data was then standardized to determine a structured environmental feature dataset. If certain key indicators in the structured environmental feature dataset exceed the preset threshold, the anomaly detection mechanism is triggered. The abnormal data is labeled according to the preset logical rules to obtain the labeled feature dataset. For the labeled feature dataset, the random forest algorithm is used to perform preliminary predictive analysis to obtain prediction result data related to dynamic behavior and determine the distribution characteristics of the prediction results. Based on the distribution characteristics of the prediction results, a comparative analysis is performed with the actual data obtained from real-time monitoring. If the difference exceeds the preset difference range, the parameters of the prediction model are adjusted to obtain optimized prediction parameters. By optimizing the prediction parameters and updating the configuration of the prediction model, continuous predictions are made on the dynamic behavior of tin-based metals in practical applications to determine the final output results.

Images