A sensitive environmental element extraction method and device

Abstract

This application discloses a method and apparatus for extracting sensitive environmental elements, relating to the fields of product reliability verification and environmental testing technology. It constructs a multi-dimensional service environment element space including physical stress, operating conditions, and damage response. Target environmental elements are obtained through Z-score standardization, principal component analysis for decorrelation, and Morris algorithm for rapid screening. Latin hypercube sampling is used to generate initial training samples to train a damage response prediction model. Based on the target environmental elements and the damage response prediction model, a first identification result is obtained using sensitive element identification based on Sobol variance decomposition. A second identification result is obtained by selectively using a spatial coupling coordination degree model to obtain the coupling coordination degree. This method can accurately distinguish between single sensitive elements and synergistically coupled sensitive damage-causing elements, providing a scientific basis for environmental test tailoring, reliability simulation iteration, and predictive maintenance, significantly improving the efficiency and accuracy of environmental sensitivity analysis and assessment for complex equipment.

Description

Technical Field

[0001] This application relates to the field of product reliability verification and environmental testing technology, specifically to a method and apparatus for extracting sensitive environmental elements. Background Technology

[0002] As industrial equipment becomes increasingly complex, precise, and long-lasting, products often operate in complex environments deeply intertwined with multiple physical stresses, dynamic loads, and human activities. Identifying and extracting sensitive damage-causing factors under these operating conditions is a prerequisite for achieving iterative reliability simulation, tailored environmental testing, and predictive maintenance. However, existing reliability engineering methods face significant challenges when processing massive amounts of real-world operating environment data.

[0003] First, there is a significant gap between laboratory single-environment stress tests and real-world service conditions. Complex equipment typically faces multiple stress coupling effects, such as the combined fatigue effects of temperature cycling and random vibration. Traditional "single-variable" analysis methods cannot capture the nonlinear interactions between different stresses, leading to identified sensitive damage-causing factors that often deviate from the actual failure mechanism. Second, data collected by conventional sensor systems covers hundreds of dimensions, including temperature, humidity, pressure, vibration acceleration, salt spray concentration, and operating load. Only a small portion of this massive amount of service data leads to product failure, and traditional local sensitivity analysis is computationally inefficient when dealing with high-dimensional parameter spaces, making it difficult to identify and extract deep-seated factors causing product damage. Third, high-fidelity reliability simulation models are computationally expensive; for example, a single run of a thermal runaway dynamics model for a power battery may take several hours. Global sensitivity analysis to identify key sensitive damage-causing factors typically requires tens of thousands of iterations, which is difficult to achieve in practical engineering.

[0004] To address the current need for extracting environmentally sensitive elements and identifying sensitive damage-causing factors in complex systems, existing static models and methods struggle to dynamically reflect individual differences during manufacturing, service, and maintenance, leading to overly conservative reliability assessments or potential safety risks. Therefore, there is an urgent need for a framework capable of efficiently processing multi-source, heterogeneous service environment data and systematically quantifying stress interactions for element extraction. Summary of the Invention

[0005] The purpose of this application is to provide a method and apparatus for extracting sensitive environmental elements, which solves the problem that the static models and methods in the prior art are difficult to dynamically reflect individual differences in the manufacturing, service and maintenance processes, resulting in overly conservative reliability assessment results or safety risks.

[0006] This application is achieved through the following technical solution:

[0007] The first aspect of this application provides a method for extracting sensitive environmental elements, including:

[0008] Construct a multi-dimensional service environment element space; the multi-dimensional service environment element space includes physical stress, operating conditions, and damage response;

[0009] The physical stress and working condition are Z-score standardized to eliminate dimensional differences, and principal component analysis algorithm is used to decorrelate the standardized data to obtain standardized element vectors.

[0010] The Morris algorithm is used to perform rapid step-by-step filtering on each environmental element in the standardized element vector to obtain the target environmental element.

[0011] The Latin hypercube sampling algorithm is used to sample the target environmental elements and the damage response, generate initial training samples in the multidimensional service environment element space, and use the initial training samples to train the damage response prediction model.

[0012] Based on the damage response prediction model, the identification of sensitive elements based on Sobol variance decomposition is used to determine the identification result of the first sensitive environmental element corresponding to the target environmental element.

[0013] Based on the first sensitive environmental element identification result, the spatial coupling coordination degree model is selectively used to obtain the coupling coordination degree corresponding to the target environmental element, and the second sensitive environmental element identification result is obtained according to the coupling coordination degree.

[0014] In one possible implementation, the physical stress includes temperature, vibration, and / or humidity; the operation control system includes rotational speed, load, and / or switching frequency; and the damage response includes crack length, impedance drift, and / or failure probability.

[0015] In one possible implementation, the Morris algorithm is used to perform rapid step-by-step filtering on each environmental element in the standardized element vector to obtain the target environmental elements, including:

[0016] The Morris algorithm was used to calculate the absolute mean and standard deviation of each environmental element in the standardized element vector across all trajectories.

[0017] Environmental elements whose absolute average value is less than a preset average value threshold and whose standard deviation is less than a preset standard deviation threshold are removed to obtain the target environmental elements.

[0018] In one possible implementation, the Latin hypercube sampling algorithm is used to sample the target environmental elements and the damage response, generating initial training samples in a multidimensional service environment element space, and the damage response prediction model is trained using the initial training samples, including:

[0019] The Latin hypercube sampling algorithm is used to sample the target environmental elements and the damage response to generate initial training samples in the multidimensional service environment element space.

[0020] Based on the initial training samples, a machine learning surrogate model based on artificial neural networks is trained. When the change in the full-order sensitivity index is lower than a preset change threshold in several consecutive iterations, the sampling process is stopped, the surrogate model is locked, and the damage response prediction model is obtained.

[0021] In one possible implementation, based on the damage response prediction model, sensitive element identification based on Sobol variance decomposition is used to determine the first sensitive environmental element identification result corresponding to the target environmental element, including:

[0022] Based on the damage response prediction model, calculate the total variance of the damage response prediction model output;

[0023] The first-order sensitivity index corresponding to the target environmental element is obtained based on the total variance, and the full-order sensitivity index corresponding to the target environmental element is obtained based on the total variance.

[0024] Based on the first-order sensitivity index and the full-order sensitivity index, the target environmental element is determined to be a single sensitive element or a coupled sensitive element, thus obtaining the first sensitive environmental element identification result.

[0025] In one possible implementation, based on the first-order sensitivity index and the full-order sensitivity index, the target environmental element is determined to be a single sensitive element or a coupled sensitive element, and a first sensitive environmental element identification result is obtained, including:

[0026] Obtain the coupling contribution ratio of the target environmental element: the coupling contribution ratio is the ratio of the difference between the full-order sensitivity index and the first-order sensitivity index to the full-order sensitivity index; when the coupling contribution ratio is less than a preset coupling ratio threshold, the target environmental element is determined to be a single sensitive element.

[0027] When the proportion of the coupling contribution is greater than or equal to the corresponding coupling proportion threshold, the target environmental element is determined to be a coupling-sensitive element.

[0028] The results of the full-level sensitivity index and the determination that the target environmental element is a single sensitive element or a coupled sensitive element are used together as the identification result of the first sensitive environmental element.

[0029] In one possible implementation, based on the first sensitive environmental element identification result, a spatial coupling coordination degree model is selectively used to obtain the coupling coordination degree corresponding to the target environmental element, and a second sensitive environmental element identification result is obtained based on the coupling coordination degree, including:

[0030] Based on the first sensitive environmental element identification results, the full-level sensitivity indices corresponding to all target environmental elements are collected.

[0031] Based on the full-order sensitivity indices corresponding to all target environmental elements, the Otsu algorithm is used to obtain the high and low thresholds of the sensitivity indices through iterative calculation.

[0032] For any target environmental element, if the full-order sensitivity index corresponding to the target environmental element is greater than the high threshold of the sensitivity index, then the target environmental element is determined to be a key sensitive damaging element; if the full-order sensitivity index corresponding to the target environmental element is less than the low threshold of the sensitivity index, then the target environmental element is determined to be background noise and is removed; if the full-order sensitivity index corresponding to the target environmental element is greater than or equal to the low threshold of the sensitivity index and less than or equal to the high threshold of the sensitivity index, then the selection is made according to the determination result of whether the target environmental element is a single sensitive element or a coupled sensitive element in the first sensitive environmental element identification result. When the target environmental element is a coupled sensitive element, the coupling coordination degree model is used to obtain the coupling coordination degree corresponding to the target environmental element; when the target environmental element is a single sensitive element, the target environmental element is determined to be an insensitive element and is removed, and the coupling coordination degree is no longer calculated; the second sensitive environmental element identification result is obtained according to the coupling coordination degree.

[0033] In one possible implementation, a spatial coupling coordination degree model is used to obtain the coupling coordination degree corresponding to the target environmental elements, including:

[0034] The range standardization method is used to standardize the target environmental element values ​​of different dimensions and obtain the standardized comprehensive score of the target environmental element.

[0035] Based on the standardized comprehensive scores of the identified key sensitive damage-causing factors and the standardized comprehensive scores of the undetermined target environmental factors, the coupling degree of the undetermined target environmental factors is obtained.

[0036] A comprehensive evaluation index is obtained based on the standardized comprehensive scores of the identified key sensitive damage-causing factors and the standardized comprehensive scores of the undetermined target environmental factors.

[0037] The coupling coordination degree corresponding to the target environmental element is processed using the comprehensive evaluation index to obtain the coupling coordination degree corresponding to the target environmental element.

[0038] In one possible implementation, obtaining the identification result of the second sensitive environmental element based on the coupling coordination degree includes:

[0039] If the coupling coordination degree is greater than or equal to a preset threshold, the target environmental element is determined to be a sensitive damaging element with coordinated coupling; otherwise, it is determined to be an insensitive element and eliminated. The sensitive damaging element with coordinated coupling and the critical most sensitive damaging element are jointly output as the second sensitive environmental element identification result, and the two together constitute the sensitive damaging element spectrum.

[0040] A second aspect of this application provides a sensitive environmental element extraction device, comprising:

[0041] The data space construction module is used to construct a multi-dimensional service environment element space; the multi-dimensional service environment element space includes physical stress, operating conditions, and damage response.

[0042] The standardization module is used to perform Z-score standardization on the physical stress and working condition to eliminate dimensional differences, and to use principal component analysis algorithm to decorrelate the standardized data to obtain a standardized element vector.

[0043] The element filtering module is used to perform rapid step-by-step filtering on each environmental element in the standardized element vector using the Morris algorithm to obtain the target environmental element.

[0044] The response prediction module is used to sample the target environmental elements and the damage response using the Latin hypercube sampling algorithm, generate initial training samples in the multidimensional service environment element space, and train the damage response prediction model using the initial training samples.

[0045] The first identification module is used to determine the first sensitive environmental element identification result corresponding to the target environmental element by using the sensitive element identification based on Sobol variance decomposition based on the damage response prediction model.

[0046] The second identification module is used to selectively use a spatial coupling coordination degree model to obtain the coupling coordination degree corresponding to the target environmental element based on the identification result of the first sensitive environmental element, and to obtain the identification result of the second sensitive environmental element according to the coupling coordination degree.

[0047] Compared with the prior art, this application has the following advantages and beneficial effects:

[0048] This application provides a method and apparatus for extracting sensitive environmental elements. It constructs a multi-dimensional service environment element space including physical stress, operating conditions, and damage response. Target environmental elements are obtained through Z-score standardization, principal component analysis for decorrelation, and Morris algorithm for rapid screening. Latin hypercube sampling is used to generate initial training samples to train a damage response prediction model. Based on the target environmental elements and the damage response prediction model, a first identification result is obtained using sensitive element identification based on Sobol variance decomposition. A second identification result is obtained by selectively using a spatial coupling coordination degree model to obtain the coupling coordination degree. This method can accurately distinguish between single sensitive elements and synergistically coupled sensitive damage-causing elements, providing a scientific basis for environmental test tailoring, reliability simulation iteration, and predictive maintenance, significantly improving the efficiency and accuracy of environmental sensitivity analysis and assessment for complex equipment. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the exemplary embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0050] Figure 1 A flowchart illustrating a method for extracting sensitive environmental elements provided in this application embodiment;

[0051] Figure 2 An overall technical roadmap for examples provided in the embodiments of this application;

[0052] Figure 3 Flowchart for calculating the first-order and full-order exponents of electronic stress in Sobol global sensitivity analysis provided for embodiments of this application;

[0053] Figure 4 A logic diagram for determining collaboratively sensitive damage-causing factors based on the spatial coupling coordination degree model (CCDM) provided in this application embodiment;

[0054] Figure 5 This is a schematic diagram of the structure of a sensitive environmental element extraction device provided in an embodiment of this application;

[0055] The attached diagram shows the markings and corresponding component names:

[0056] 501 - Data Space Construction Module, 502 - Standardization Module, 503 - Element Screening Module, 504 - Response Prediction Module, 505 - First Identification Module, 506 - Second Identification Module. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this application are only for explaining this application and are not intended to limit this application.

[0058] like Figure 1 As shown in the figure, this application provides a method for extracting sensitive environmental elements, including:

[0059] S101. Construct a multi-dimensional service environment element space; the multi-dimensional service environment element space includes physical stress, operating conditions, and damage response.

[0060] In one possible implementation, the physical stress includes temperature, vibration, and / or humidity; the operation control system includes rotational speed, load, and / or switching frequency; and the damage response includes crack length, impedance drift, and / or failure probability.

[0061] Multi-source heterogeneous data fusion technology can be used to construct a multi-dimensional element space covering three dimensions: physical stress, operating conditions, and damage response. This integrates raw data collected from sensor systems (physical stress: temperature, vibration, humidity), operation control systems (operating conditions: speed, load, switching frequency), and health monitoring systems (damage response: crack length, impedance drift, failure probability).

[0062] S102. Perform Z-score standardization on the physical stress and working condition to eliminate dimensional differences, and use principal component analysis algorithm to decorrelate the standardized data to obtain standardized element vectors.

[0063] Optionally, a deep neural network (DNN) combined with a self-attention mechanism can be introduced. By learning the contribution of various physical stresses to the observed damage response under different operating conditions, the dynamic influence weights of different environmental factors can be automatically calculated, thereby identifying which stress combinations are potential damage sources and establishing a multidimensional correlation weight between physical stresses and operating conditions. For example, the correlation weights can serve as a priori reference for subsequent Morris rapid step-by-step screening: environmental factors with extremely low correlation weights are prioritized for elimination in the initial screening stage to further improve the screening efficiency of high-dimensional factor space; stress combinations with high correlation weights are considered potential damage sources and are given special attention in subsequent sensitivity identification.

[0064] First, Z-score standardization (x*=(x−μ) / σ, where μ and σ are the sample mean and sample standard deviation of the element, respectively) is performed on the physical stress and operating conditions to eliminate dimensional differences. Specifically, for each environmental element, the sample mean μ and sample standard deviation σ are calculated on all its observed sample sequences, and each original observed value x of the element is transformed into x* according to the above formula. Each environmental element is transformed independently without affecting others. After the transformation, each environmental element is transformed into a dimensionless numerical sequence with a mean of 0 and a standard deviation of 1, so that environmental elements with different dimensions and orders of magnitude, such as temperature, vibration acceleration, and humidity, fall into the same comparable scale, and the relative differences between samples within the same element remain unchanged. On this basis, principal component analysis (PCA) is used to decorrelate the standardized data, transforming the multidimensional and cross-dimensional service environment data into a unified standardized element vector. This provides a unified and standardized input technique for establishing a mathematical mapping relationship between the service environment element space and the damage response space Y (this mapping relationship is specifically implemented by the damage response prediction model trained in S104). This eliminates the impact of dimensional differences on the subsequent identification of sensitive damage-causing factors.

[0065] S103. The Morris algorithm is used to perform rapid step-by-step filtering on each environmental element in the standardized element vector to obtain the target environmental element.

[0066] In one possible implementation, the Morris algorithm is used to perform rapid step-by-step filtering on each environmental element in the standardized element vector to obtain the target environmental elements, including:

[0067] The Morris algorithm was used to calculate the absolute mean and standard deviation of each environmental element in the standardized element vector across all trajectories.

[0068] Environmental elements whose absolute average value is less than a preset average value threshold and whose standard deviation is less than a preset standard deviation threshold are removed to obtain the target environmental elements.

[0069] For example, a rapid step-by-step screening based on the Morris method: The Morris method, as a highly efficient screening tool, performs One-At-A-Time (OAT) sampling by constructing random trajectories in an n-dimensional parameter space. An improved Morris method is used to rapidly scan and randomly sample a space encompassing multi-dimensional parameters such as temperature, humidity, pressure, vibration acceleration, salt spray concentration, and operating load. The primary effects of each environmental element are then calculated. For each environmental element parameter By changing the tiny step size Observe the change in output Y:

[0070]

[0071] Calculate all trajectories absolute average and standard deviation .in This represents the overall intensity of the environmental factor's impact on product damage. This represents the strength of the coupling effect between this element and other elements. To avoid the accidental removal of environmental elements with low overall influence but strong interaction with other elements in the initial screening stage, the removal criterion is set as follows: an environmental element is removed only when its absolute average value is less than a preset average value threshold and its standard deviation is less than a preset standard deviation threshold; elements with larger standard deviations are retained and proceeded to subsequent coupling analysis.

[0072] Identify and remove Minimal environmental factors have minimal impact on product damage response and are therefore non-sensitive, thus significantly reducing the dimensions of the model input.

[0073] S104. The Latin hypercube sampling algorithm is used to sample the target environmental elements and the damage response, generate initial training samples in the multidimensional service environment element space, and use the initial training samples to train the damage response prediction model.

[0074] In one possible implementation, the Latin hypercube sampling algorithm is used to sample the target environmental elements and the damage response, generating initial training samples in a multidimensional service environment element space, and the damage response prediction model is trained using the initial training samples, including:

[0075] The Latin hypercube sampling algorithm is used to sample the target environmental elements and the damage response to generate initial training samples in the multidimensional service environment element space.

[0076] Based on the initial training samples, a machine learning surrogate model based on artificial neural networks is trained. When the change in the full-order sensitivity index is lower than a preset change threshold in several consecutive iterations, the sampling process is stopped, the surrogate model is locked, and the damage response prediction model is obtained.

[0077] For example, within the simplified multidimensional service environment element space, a machine learning agent model based on artificial neural networks (ANN) is constructed. (This surrogate model is the specific implementation of the mathematical mapping relationship between the service environment element space and the damage response space Y described in S102.) It employs the Latin hypercube sampling (LHS) algorithm, which has excellent space-filling properties, to generate initial training samples within the service element space, enabling the prediction of complex failure probabilities and damage effects under massive service environment data. A convergence criterion based on the change in the full-order sensitivity index is set (for example, this change threshold can be on the order of 0.01). When the change in the full-order sensitivity index is lower than the preset threshold in several consecutive iterations, the sampling process is stopped, and the surrogate model is locked.

[0078] S105. Based on the damage response prediction model, the identification of sensitive elements based on Sobol variance decomposition is adopted to determine the identification result of the first sensitive environmental element corresponding to the target environmental element.

[0079] In one possible implementation, based on the damage response prediction model, sensitive element identification based on Sobol variance decomposition is used to determine the first sensitive environmental element identification result corresponding to the target environmental element, including:

[0080] Based on the damage response prediction model, calculate the total variance of the damage response prediction model output;

[0081] The first-order sensitivity index corresponding to the target environmental element is obtained based on the total variance, and the full-order sensitivity index corresponding to the target environmental element is obtained based on the total variance.

[0082] Based on the first-order sensitivity index and the full-order sensitivity index, the target environmental element is determined to be a single sensitive element or a coupled sensitive element, thus obtaining the first sensitive environmental element identification result.

[0083] In one possible implementation, based on the first-order sensitivity index and the full-order sensitivity index, the target environmental element is determined to be a single sensitive element or a coupled sensitive element, and a first sensitive environmental element identification result is obtained, including:

[0084] Obtain the coupling contribution ratio of the target environmental element: the coupling contribution ratio is the ratio of the difference between the full-order sensitivity index and the first-order sensitivity index to the full-order sensitivity index; when the coupling contribution ratio is less than a preset coupling ratio threshold, the target environmental element is determined to be a single sensitive element.

[0085] When the proportion of the coupling contribution is greater than or equal to the corresponding coupling proportion threshold, the target environmental element is determined to be a coupling-sensitive element.

[0086] The results of the full-level sensitivity index and the determination that the target environmental element is a single sensitive element or a coupled sensitive element are used together as the identification result of the first sensitive environmental element.

[0087] Based on the locked surrogate model, the Sobol global sensitivity analysis method based on variance decomposition can be applied to quantitatively analyze the sensitivity of each factor. The method is as follows:

[0088] 1. Calculate the total variance V(Y) of the model output Y:

[0089] ;

[0090] in, Let represent the variance of the individual contribution of the i-th environmental element. The variance of individual contributions is represented by the interaction contribution among environmental elements, where n represents the total number of environmental elements. The variance of individual contributions is obtained by fixing the i-th environmental element and calculating the variance of the expected value of the model output condition, i.e., V. i =Var(E[Y|X i The variance of the second-order interaction contribution among environmental factors is V. ij =Var(E[Y|X i ,X j ])−V i -V j Higher-order interaction terms are similarly recursively derived; the aforementioned conditional variance is estimated in engineering by performing Monte Carlo sampling (such as the Saltelli sampling scheme) on the locked surrogate model. Here, Y represents the output of the damage response prediction model, i.e., the predicted damage response value; X... i X represents the i-th environmental element; j This represents the j-th environmental element.

[0091] Calculate the first-order sensitivity index and full-order sensitivity index :

[0092] First-order sensitivity index reflecting the direct destructive power of a certain environmental factor The calculation formula is as follows:

[0093] ;

[0094] A full-order sensitivity index that reflects the total destructive power of the environmental element and all its coupling effects. The calculation formula is as follows:

[0095]

[0096] in, This represents a vector consisting of all environmental elements except the i-th environmental element. E[·] represents the variance of the expected value of the model output condition under the condition that all other factors except the i-th environmental factor are fixed, where E[·] and Var(·) are the expected value and variance operations, respectively. This represents the total variance of the damage response prediction model output.

[0097] Determine whether an environmental element is a single sensitive element or a coupled sensitive element: Calculate - The larger the difference, the more likely the factor is causing product failure through multi-stress coupling rather than a single effect. Since the overall sensitivity index is never less than the first-order sensitivity index, the coupling contribution ratio η is defined as the ratio of the difference between the two to the overall sensitivity index. When η is less than a preset coupling ratio threshold ε (for example, ε can be 0.2, which can be adjusted according to engineering needs), i.e., the first-order sensitivity index... Approaching the full-order sensitivity index If η is a single, dominant sensitive damaging factor (such as failure caused by extreme high temperature), then the environmental factor is the single dominant sensitive damaging factor (such as failure caused by extreme high temperature); when η is greater than or equal to ε, that is, the full-order sensitivity index. Significantly greater than the first-order sensitivity index If the environmental factor is a sensitive damage factor dominated by synergistic coupling (such as brittle fracture caused by the coupling of "high frequency vibration-low temperature environment").

[0098] S106. Based on the first sensitive environmental element identification result, selectively use the spatial coupling coordination degree model to obtain the coupling coordination degree corresponding to the target environmental element, and obtain the second sensitive environmental element identification result according to the coupling coordination degree.

[0099] In one possible implementation, based on the first sensitive environmental element identification result, a spatial coupling coordination degree model is selectively used to obtain the coupling coordination degree corresponding to the target environmental element, and a second sensitive environmental element identification result is obtained based on the coupling coordination degree, including:

[0100] Based on the first sensitive environmental element identification results, the full-level sensitivity indices corresponding to all target environmental elements are collected.

[0101] Based on the full-order sensitivity indices corresponding to all target environmental elements, the Otsu algorithm is used to obtain the high and low thresholds of the sensitivity indices through iterative calculation.

[0102] For any target environmental element, if the full-order sensitivity index corresponding to the target environmental element is greater than the high threshold of the sensitivity index, then the target environmental element is determined to be a key sensitive damaging element; if the full-order sensitivity index corresponding to the target environmental element is less than the low threshold of the sensitivity index, then the target environmental element is determined to be background noise and is removed; if the full-order sensitivity index corresponding to the target environmental element is greater than or equal to the low threshold of the sensitivity index and less than or equal to the high threshold of the sensitivity index, then the selection is made according to the determination result of whether the target environmental element is a single sensitive element or a coupled sensitive element in the first sensitive environmental element identification result. When the target environmental element is a coupled sensitive element, the coupling coordination degree model is used to obtain the coupling coordination degree corresponding to the target environmental element; when the target environmental element is a single sensitive element, the target environmental element is determined to be an insensitive element and is removed, and the coupling coordination degree is no longer calculated; the second sensitive environmental element identification result is obtained according to the coupling coordination degree.

[0103] In one possible implementation, a spatial coupling coordination degree model is used to obtain the coupling coordination degree corresponding to the target environmental elements, including:

[0104] The range standardization method is used to standardize the target environmental element values ​​of different dimensions and obtain the standardized comprehensive score of the target environmental element.

[0105] Based on the standardized comprehensive scores of the identified key sensitive damage-causing factors and the standardized comprehensive scores of the undetermined target environmental factors, the coupling degree of the undetermined target environmental factors is obtained.

[0106] A comprehensive evaluation index is obtained based on the standardized comprehensive scores of the identified key sensitive damage-causing factors and the standardized comprehensive scores of the undetermined target environmental factors.

[0107] The coupling coordination degree corresponding to the target environmental element is processed using the comprehensive evaluation index to obtain the coupling coordination degree corresponding to the target environmental element.

[0108] In one possible implementation, obtaining the identification result of the second sensitive environmental element based on the coupling coordination degree includes:

[0109] If the coupling coordination degree is greater than or equal to a preset threshold, the target environmental element is determined to be a sensitive damaging element with coordinated coupling; otherwise, it is determined to be an insensitive element and eliminated. The sensitive damaging element with coordinated coupling and the critical most sensitive damaging element are jointly output as the second sensitive environmental element identification result, and the two together constitute the sensitive damaging element spectrum.

[0110] For example, this invention combines statistical distribution and coupling coordination theory to establish an adaptive extraction mechanism based on the sensitivity index distribution, summarizing the full-order sensitivity indices of all environmental elements. A histogram of its numerical distribution is constructed. An improved Otsu's algorithm, based on the maximum inter-class variance method, is applied to iteratively calculate the high threshold of the sensitivity index. and low threshold .satisfy Environmental factors are key sensitive damage factors, and vice versa. Environmental factors are removed as background noise; due to and For environmental elements in the intermediate zone, a spatial coupling coordination degree model (CCDM) is introduced for secondary verification. Whether to introduce the CCDM into intermediate zone environmental elements is determined by the judgment result obtained in S105, which determines whether the environmental element is a single sensitive element or a coupled sensitive element: secondary verification is only performed when the environmental element is determined to be a coupled sensitive element; intermediate zone environmental elements determined to be single sensitive elements, whose average level of full-order sensitivity index mainly originates from their own independent action rather than coupled amplification, are neither sufficient to be judged as key sensitive damage-causing elements nor have extraction value for synergistic damage, and are directly eliminated as insensitive elements. Therefore, the single / coupled judgment result obtained in S105 constitutes the criterion for selectively introducing the CCDM in this step, avoiding redundant calculations caused by indiscriminately performing coupling coordination degree calculations on all intermediate zone environmental elements, and ensuring that the decomposition results of the first-order sensitivity index and the full-order sensitivity index are actually used in subsequent extraction stages.

[0111] Based on the spatial coupling coordination degree model (CCDM), the coupling degree C between the unknown environmental element and the identified key sensitive damage-causing element is calculated to measure the intensity of the coupling influence between the two.

[0112] First, the range standardization method is used to standardize the element values ​​of different dimensions. The formula for calculating the standardized composite score of a certain environmental element is:

[0113] ;

[0114] in, This represents the standardized comprehensive evaluation value of the i-th observation sample within the j-th environmental element. Its value ranges from 0 to 1, and it is a dimensionless relative intensity index used in the subsequent calculation of the coupling degree C. This represents the original observed value of the j-th environmental element in the i-th sample. In a reliability verification scenario, this could be the original physical stress value collected by the sensor, specific operating condition parameters, or a preliminarily calculated sensitivity index. . This represents the minimum value of the j-th environmental element among all the sample sequences involved in the calculation. This represents the maximum value of the j-th environmental element among all the sample sequences involved in the calculation. The standardized composite score of the j-th environmental element is obtained by aggregating the composite evaluation values ​​of all m observed samples of this element. For example, an arithmetic mean is used, which is the sum of the composite evaluation values ​​of each sample and then divided by the number of samples m.

[0115] Secondly, calculate the coupling degree C between the undetermined element and the identified key sensitive damage-causing element, using the following formula:

[0116] ;

[0117] in, The standardized composite score represents the identified key sensitive damage-causing factors. The standardized score represents the environmental element to be determined. When there are multiple key sensitive damage-causing elements, the coupling degree C is calculated by pairing the environmental element to be determined with each key sensitive damage-causing element one by one, and the maximum value is taken as the coupling degree of the environmental element to be determined; that is, as long as the environmental element to be determined has a strong coupling relationship with any key sensitive damage-causing element, it is considered to have the extraction value of synergistic damage.

[0118] The value of C ranges from 0 to 1. A larger C value indicates a stronger dynamic correlation between the environmental element to be determined and the key sensitive damage-causing element, demonstrating a clear synergistic coupling in the product damage process. However, the coupling degree C alone cannot directly determine whether the environmental element has extraction value, because C only reflects the strength of the correlation, not the level of coordination in the system's progression towards severe damage. Therefore, the coupling coordination degree D needs to be further calculated.

[0119]

[0120]

[0121] Where T is the comprehensive evaluation index, and... α and β are weighting coefficients (in damage verification, the weight of key elements α is usually taken as 0.6-0.8; α and β are normalized weighting coefficients, satisfying α+β=1, so β=1−α, usually taken as 0.2–0.4. They respectively characterize the relative importance of key sensitive damage-causing elements and undetermined environmental elements in synergistic damage, and can be determined by expert experience, analytic hierarchy process or entropy weight method, and are exemplary adjustable parameters).

[0122] If D ≥ 0.5 (0.5 is an example preset threshold, which can be adjusted according to product reliability level requirements), it indicates that although the independent sensitivity of the undetermined element is not high, it can significantly enhance the total damage of the system in the coupling mechanism. Therefore, it is extracted as a synergistically coupled sensitive damage-causing element. If D < 0.5, it is determined to be an insensitive element and is removed. Thus, the final extracted spectrum of sensitive damage-causing elements consists of two parts: first, key sensitive damage-causing elements whose overall sensitivity index is higher than the high threshold and are directly extracted by Otsu's determination; second, synergistically coupled sensitive damage-causing elements whose overall sensitivity index is between the high and low thresholds and are supplemented after secondary verification by the spatial coupling coordination degree model. The latter supplements rather than replaces the former, and both serve as the output of the second sensitive environmental element identification result. In addition, the single / coupled determination results obtained from S105 are also output along with the sensitive damage element spectrum, serving as an identifier for the damage mechanism type of each sensitive damage element: for key sensitive damage elements determined to be single sensitive elements, their damage is mainly determined by the value of the element itself, and single-stress accelerated testing can be used in subsequent environmental test tailoring, with a focus on controlling the value of the element; for key sensitive damage elements determined to be coupled sensitive elements and all synergistically coupled sensitive damage elements, their damage originates from the synergistic effect of multiple stresses, and subsequent multi-stress comprehensive testing must be used while maintaining the application sequence and combination relationship of the corresponding stresses, and they cannot be separated into single-stress tests.

[0123] Compared with the prior art, the present invention has the following significant advantages:

[0124] (1) It can accurately extract the most critical and sensitive damage-causing factors from hundreds of dimensions of raw service data, which not only reduces the pressure of data analysis, storage and transmission, but also provides a scientific basis for the "load spectrum trimming" of environmental tests;

[0125] (2) By replacing expensive finite element simulation or physical fatigue test with machine learning proxy model, the analysis cycle of environmental sensitivity factors of complex equipment is greatly shortened, and the product design iteration speed is significantly improved.

[0126] (3) It can not only identify the influence of single environmental stress, but also quantitatively characterize the nonlinear effect of multi-stress coupling through Sobol global sensitivity analysis and CCDM model, and obtain the synergistic coupling damage elements triggered under specific working conditions, thus solving the problem of large deviation in damage mechanism identification under real complex environment.

[0127] (4) By utilizing digital twin technology, this method can automatically adapt to the sensitivity changes of different products and different service stages, thereby improving the health monitoring accuracy of complex systems under variable operating conditions. The extracted sensitive damage factors can be used as the core driving input of the digital twin model to achieve near real-time accurate prediction of the physical entity's health status and remaining useful life (RUL).

[0128] Based on the above-described method and technical solution, this application also provides an example based on electronic components, as detailed below.

[0129] Figure 2 This invention provides an overall technical roadmap for a method for extracting sensitive environmental elements in complex environments based on multidimensional coupling and sensitivity analysis, which is divided into an input end, a process end, and an output end.

[0130] First, at the input end, raw data collected from electronic component sensor systems (physical stress: temperature, vibration, humidity), operation control systems (operating conditions: speed, load, switching frequency), and health monitoring systems (damage response: crack length, impedance drift, failure probability) are integrated.

[0131] Secondly, at the process end, a multi-dimensional element space covering physical stress, operating condition, and damage response is constructed using multi-source heterogeneous data fusion technology. A deep neural network (DNN) combined with a self-attention mechanism is introduced to learn the contribution of various physical stresses to the observed damage response under different operating conditions and identify dynamic change patterns, establishing a correlation weight between multi-dimensional physical stress and operating condition. This correlation weight serves as a priori reference for subsequent Morris rapid step-by-step screening: environmental elements with extremely low weights are prioritized for elimination in the initial screening stage, while stress combinations with higher weights are considered potential damage sources and given priority in subsequent sensitivity identification. First, Z-score standardization is applied to physical stress and operating condition elements to eliminate dimensional differences. Then, principal component analysis (PCA) is used to decorrelate the standardized data, transforming the multi-dimensional and cross-dimensional service environment data into a unified standardized element vector. Establish a mathematical mapping relationship between the service environment element space and the damage response space Y. This eliminates the impact of dimensional differences on the subsequent identification of sensitive damage-causing factors.

[0132] Finally, in the first part of the output, to address the problem of the extremely time-consuming nature of complex finite element thermo-mechanical coupling simulation calculations, a sensitive damage-causing element acceleration identification scheme based on a surrogate model is adopted, as follows:

[0133] (1) Initial screening stage - rapid step-by-step screening based on Morris method

[0134] Morris sampling trajectories were executed across 80 potential environmental factors (high temperature, vibrational components in specific frequency ranges, etc.) involving electronic components. The primary effects of each environmental factor were calculated. And the absolute average value representing the overall impact of a certain environmental factor on product damage. It can quickly eliminate parameters that have minimal impact on the internal circuitry, such as "external air pressure of the casing", reducing the parameter dimension to less than 15 dimensions, thereby achieving a significant reduction in the model input dimension.

[0135] (2) Modeling stage – Damage response prediction based on ANN machine learning

[0136] Within the simplified multidimensional service environment element space, a machine learning agent model based on artificial neural networks (ANN) is constructed. Latin hypercube sampling (LHS), which has excellent space-filling properties, was used to generate 150 initial training samples in a 15-dimensional space. The damage index was then obtained by calling a simulation model. Convergence criteria were set, and the sensitivity index across all orders was monitored. The fluctuation situation, when in 5 consecutive iterations of sampling The change was lower than If the model is deemed to have stably characterized the complex damaging effects, sampling is stopped and the surrogate model is locked.

[0137] (3) Analysis stage – Identification of sensitive elements based on Sobol variance decomposition

[0138] Run 50,000 Sobol global sensitivity analyses on the locked agent model. Calculate the first-order sensitivity index. and full-order sensitivity index The results show that temperature stress ( =0.52) is the main cause of loss, while the full-order exponent of the "temperature-vibration" coupling is... The value reached 0.82. This indicates that although vibration is not the primary factor, it significantly enhances the crack initiation rate of the solder layer by coupling with high temperature, revealing that vibration accelerates the accumulation of strain energy in the solder layer at high temperature.

[0139] In the second part of the output, an improved Otsu algorithm is applied to iteratively calculate the high threshold of the sensitivity index. and low threshold .satisfy Environmental factors are key sensitive damage factors, and vice versa. Environmental factors are removed as background noise; due to and The environmental factors between them are then assessed using the Spatial Coupling Coordination Degree Model (CCDM) for secondary verification. Only intermediate zone environmental factors identified as coupling-sensitive factors undergo secondary verification; those identified as single-sensitive factors are directly removed as insensitive factors. The coupling coordination degree D between the undetermined factor and the identified key sensitive damage-causing factors is calculated. If D ≥ 0.5, it indicates that although the undetermined factor has low independent sensitivity, it can significantly enhance the total system damage through the coupling mechanism, and is thus extracted as a synergistically coupled sensitive damage-causing factor. If D < 0.5, it is identified as an insensitive factor and removed.

[0140] The method for extracting sensitive environmental elements in complex environments based on multidimensional coupling and sensitivity analysis proposed in this invention is characterized by not only focusing on the destructive power of a single environmental element, but also on the coupling and synergistic effect between environmental elements, thereby forming a complete spectrum of sensitive damage-causing elements for products and avoiding the omission of secondary sensitive damage-causing factors under complex stress environments.

[0141] Figure 3 First-order sensitivity index in Sobol global sensitivity analysis With full-order sensitivity index Calculation flowchart. The mapping relationship between the service environment element space X and the damage response space Y is known. Based on a machine learning surrogate model using artificial neural networks (ANN), this paper applies the Sobol global sensitivity analysis method based on variance decomposition to quantitatively analyze the sensitivity of each factor. First, the total variance V(Y) of the model output Y is calculated, and then decomposed into the sum of the variances of individual parameters and the variances of parameter interactions. Next, the first-order sensitivity index reflecting the direct destructive power of a certain environmental factor is calculated. And the full-order sensitivity index that reflects the total destructive power of the environmental element and all its coupling effects. Finally, by comparing the first-order sensitivity indices... and full-order sensitivity index The magnitude of the value is used to determine whether an environmental element is a single sensitive element or a coupled sensitive element. Since the full-order sensitivity index is never less than the first-order sensitivity index, the coupling contribution ratio η is defined as the ratio of the difference between the two to the full-order sensitivity index. When η is less than a preset coupling ratio threshold ε (for example, ε can be 0.2, which can be adjusted according to project needs), i.e., the first-order sensitivity index is less than the threshold value, the coupling contribution ratio is less than the threshold value. Approaching the full-order sensitivity index If η is a single, dominant, sensitive, and damaging factor, then that environmental factor is the primary factor causing the damage; when η is greater than or equal to ε, i.e., the full-order sensitivity index... Significantly greater than the first-order sensitivity index If so, then the environmental element is a sensitive damaging element dominated by synergistic coupling.

[0142] Figure 4This is a logic diagram for determining collaboratively sensitive damaging factors based on the Spatial Coupling and Coordination Degree Model (CCDM). First, the full-order sensitivity indices of all environmental factors are summarized. A histogram of its numerical distribution is constructed. The improved Otsu's algorithm, based on the maximum inter-class variance method, is applied to iteratively calculate the high threshold of the sensitivity index. and low threshold The sensitivity index of the entire order obtained in the previous period. Make a preliminary classification to meet the requirements. Environmental factors (such as peak temperature of the heat sink substrate and root mean square value of random vibration in the Z direction) are key sensitive damage factors, and vice versa. Environmental factors are removed as background noise; due to and Environmental factors (such as ambient humidity) are introduced into the Spatial Coupling Coordination Degree Model (CCDM). If an environmental factor (such as ambient humidity) is identified as a coupling-sensitive factor in S105, it is then introduced into the CCDM for secondary verification. Secondly, the range standardization method is used to standardize the factor values ​​of different dimensions to eliminate the influence of dimensional differences on subsequent calculations, obtaining the standardized comprehensive score of the identified key sensitive damage-causing factors. Standardized comprehensive score of undetermined environmental factors The coupling degree C between the undetermined environmental element and the identified key sensitive damage-causing element (such as temperature stress) is calculated to measure the strength of the coupling effect between them. The coupling coordination degree D is further calculated to determine the synergistic damage level of the undetermined environmental element and the key sensitive damage-causing element on the system. If D ≥ 0.5, it indicates that although the undetermined environmental element has low independent sensitivity, it can significantly enhance the total damage of the system through the coupling mechanism, and is therefore extracted as a synergistically coupled sensitive damage-causing element; if D < 0.5, it is judged as an insensitive element and is removed. Taking electronic components as an example, since humidity has a significant corrosion-promoting catalytic effect in high-temperature environments, the coupling coordination degree D = 0.68 > 0.5 is calculated, therefore humidity is included as a synergistic sensitive element in the spectrum of electronic sensitive damage-causing elements.

[0143] like Figure 5 As shown in the figure, this application embodiment also provides a sensitive environmental element extraction device, including:

[0144] The data space construction module 501 is used to construct a multi-dimensional service environment element space; the multi-dimensional service environment element space includes physical stress, operating conditions, and damage response.

[0145] The standardization module 502 is used to perform Z-score standardization on the physical stress and working condition to eliminate dimensional differences, and to use principal component analysis algorithm to decorrelate the standardized data to obtain a standardized element vector.

[0146] The element filtering module 503 is used to perform rapid step-by-step filtering on each environmental element in the standardized element vector using the Morris algorithm to obtain the target environmental element.

[0147] The response prediction module 504 is used to sample the target environmental elements and the damage response using the Latin hypercube sampling algorithm, generate initial training samples in the multidimensional service environment element space, and train the damage response prediction model using the initial training samples.

[0148] The first identification module 505 is used to determine the first sensitive environmental element identification result corresponding to the target environmental element by using the sensitive element identification based on Sobol variance decomposition based on the damage response prediction model.

[0149] The second identification module 506 is used to selectively use a spatial coupling coordination degree model to obtain the coupling coordination degree corresponding to the target environmental element based on the first sensitive environmental element identification result, and to obtain the second sensitive environmental element identification result based on the coupling coordination degree.

[0150] The sensitive environmental element extraction device provided in this application embodiment can execute the method and technical solution in any of the embodiments. Its principle and beneficial effects are similar, and will not be described again here.

[0151] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only a specific embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for extracting sensitive environmental elements, characterized in that, include: Constructing a multi-dimensional service environment element space; The multidimensional service environment element space includes physical stress, operating conditions, and damage response; The physical stress and working condition are Z-score standardized to eliminate dimensional differences, and principal component analysis algorithm is used to decorrelate the standardized data to obtain standardized element vectors. The Morris algorithm is used to perform rapid step-by-step filtering on each environmental element in the standardized element vector to obtain the target environmental element. The Latin hypercube sampling algorithm is used to sample the target environmental elements and the damage response, generate initial training samples in the multidimensional service environment element space, and use the initial training samples to train the damage response prediction model. Based on the damage response prediction model, the identification of sensitive elements based on Sobol variance decomposition is used to determine the identification result of the first sensitive environmental element corresponding to the target environmental element. Based on the first sensitive environmental element identification result, the spatial coupling coordination degree model is selectively used to obtain the coupling coordination degree corresponding to the target environmental element, and the second sensitive environmental element identification result is obtained according to the coupling coordination degree.

2. The method for extracting sensitive environmental elements according to claim 1, characterized in that, The physical stresses include temperature, vibration, and / or humidity; the operation control system includes rotational speed, load, and / or switching frequency; the damage response includes crack length, impedance drift, and / or failure probability.

3. The method for extracting sensitive environmental elements according to claim 1, characterized in that, The Morris algorithm is used to perform rapid step-by-step filtering on each environmental element in the standardized element vector to obtain the target environmental elements, including: The Morris algorithm was used to calculate the absolute mean and standard deviation of each environmental element in the standardized element vector across all trajectories. Environmental elements whose absolute average value is less than a preset average value threshold and whose standard deviation is less than a preset standard deviation threshold are removed to obtain the target environmental elements.

4. The method for extracting sensitive environmental elements according to claim 1, characterized in that, The Latin hypercube sampling algorithm is used to sample the target environmental elements and the damage response, generating initial training samples in a multidimensional service environment element space. The damage response prediction model is then trained using these initial training samples, including: The Latin hypercube sampling algorithm is used to sample the target environmental elements and the damage response to generate initial training samples in the multidimensional service environment element space. Based on the initial training samples, a machine learning surrogate model based on artificial neural networks is trained. When the change in the full-order sensitivity index is lower than a preset change threshold in several consecutive iterations, the sampling process is stopped, the surrogate model is locked, and the damage response prediction model is obtained.

5. The method for extracting sensitive environmental elements according to claim 1, characterized in that, Based on the damage response prediction model, the identification of sensitive elements based on Sobol variance decomposition is used to determine the first sensitive environmental element identification result corresponding to the target environmental element, including: Based on the damage response prediction model, calculate the total variance of the damage response prediction model output; The first-order sensitivity index corresponding to the target environmental element is obtained based on the total variance, and the full-order sensitivity index corresponding to the target environmental element is obtained based on the total variance. Based on the first-order sensitivity index and the full-order sensitivity index, the target environmental element is determined to be a single sensitive element or a coupled sensitive element, thus obtaining the first sensitive environmental element identification result.

6. The method for extracting sensitive environmental elements according to claim 5, characterized in that, Based on the first-order sensitivity index and the full-order sensitivity index, the target environmental element is determined to be a single sensitive element or a coupled sensitive element, and the first sensitive environmental element identification result is obtained, including: Obtain the coupling contribution ratio of the target environmental element: the coupling contribution ratio is the ratio of the difference between the full-order sensitivity index and the first-order sensitivity index to the full-order sensitivity index. When the proportion of the coupling contribution is less than the preset coupling proportion threshold, the target environmental element is determined to be a single sensitive element. When the proportion of the coupling contribution is greater than or equal to the corresponding coupling proportion threshold, the target environmental element is determined to be a coupling-sensitive element. The results of the full-level sensitivity index and the determination that the target environmental element is a single sensitive element or a coupled sensitive element are used together as the identification result of the first sensitive environmental element.

7. The method for extracting sensitive environmental elements according to claim 6, characterized in that, Based on the identification result of the first sensitive environmental element, a spatial coupling coordination degree model is selectively used to obtain the coupling coordination degree corresponding to the target environmental element, and the identification result of the second sensitive environmental element is obtained according to the coupling coordination degree, including: Based on the first sensitive environmental element identification results, the full-level sensitivity indices corresponding to all target environmental elements are collected. Based on the full-order sensitivity indices corresponding to all target environmental elements, the Otsu algorithm is used to obtain the high and low thresholds of the sensitivity indices through iterative calculation. For any target environmental element, if the full-order sensitivity index corresponding to the target environmental element is greater than the high threshold of the sensitivity index, then the target environmental element is determined to be a key sensitive damaging element; if the full-order sensitivity index corresponding to the target environmental element is less than the low threshold of the sensitivity index, then the target environmental element is determined to be background noise and is removed; if the full-order sensitivity index corresponding to the target environmental element is greater than or equal to the low threshold of the sensitivity index and less than or equal to the high threshold of the sensitivity index, then the selection is made according to the determination result of whether the target environmental element is a single sensitive element or a coupled sensitive element in the first sensitive environmental element identification result. When the target environmental element is a coupled sensitive element, the coupling coordination degree model is used to obtain the coupling coordination degree corresponding to the target environmental element; when the target environmental element is a single sensitive element, the target environmental element is determined to be an insensitive element and is removed, and the coupling coordination degree is no longer calculated; the second sensitive environmental element identification result is obtained according to the coupling coordination degree.

8. The method for extracting sensitive environmental elements according to claim 7, characterized in that, The spatial coupling coordination degree model is used to obtain the coupling coordination degree corresponding to the target environmental elements, including: The range standardization method is used to standardize the target environmental element values ​​of different dimensions and obtain the standardized comprehensive score of the target environmental element. Based on the standardized comprehensive scores of the identified key sensitive damage-causing factors and the standardized comprehensive scores of the undetermined target environmental factors, the coupling degree of the undetermined target environmental factors is obtained. A comprehensive evaluation index is obtained based on the standardized comprehensive scores of the identified key sensitive damage-causing factors and the standardized comprehensive scores of the undetermined target environmental factors. The coupling coordination degree corresponding to the target environmental element is processed using the comprehensive evaluation index to obtain the coupling coordination degree corresponding to the target environmental element.

9. The method for extracting sensitive environmental elements according to claim 8, characterized in that, The identification result of the second sensitive environmental element is obtained based on the coupling coordination degree, including: If the coupling coordination degree is greater than or equal to a preset threshold, the target environmental element is determined to be a sensitive damaging element with coordinated coupling; otherwise, it is determined to be an insensitive element and eliminated. The sensitive damaging element with coordinated coupling and the critical most sensitive damaging element are jointly output as the second sensitive environmental element identification result, and the two together constitute the sensitive damaging element spectrum.

10. A device for extracting sensitive environmental elements, characterized in that, include: The data space construction module is used to construct a multi-dimensional service environment element space; The multidimensional service environment element space includes physical stress, operating conditions, and damage response; The standardization module is used to perform Z-score standardization on the physical stress and working condition to eliminate dimensional differences, and to use principal component analysis algorithm to decorrelate the standardized data to obtain a standardized element vector. The element filtering module is used to perform rapid step-by-step filtering on each environmental element in the standardized element vector using the Morris algorithm to obtain the target environmental element. The response prediction module is used to sample the target environmental elements and the damage response using the Latin hypercube sampling algorithm, generate initial training samples in the multidimensional service environment element space, and train the damage response prediction model using the initial training samples. The first identification module is used to determine the first sensitive environmental element identification result corresponding to the target environmental element by using the sensitive element identification based on Sobol variance decomposition based on the damage response prediction model. The second identification module is used to selectively use a spatial coupling coordination degree model to obtain the coupling coordination degree corresponding to the target environmental element based on the identification result of the first sensitive environmental element, and to obtain the identification result of the second sensitive environmental element according to the coupling coordination degree.

Images