Multi-modal mechanical coupling multi-working-condition pressure-bearing equipment combined use evaluation method

Abstract

The invention discloses a multi-modal mechanical coupling multi-working condition pressure-bearing equipment combined use evaluation method, and relates to the technical field of multi-modal mechanical coupling, and the method comprises the following steps: obtaining structure response data and working condition state data of pressure-bearing equipment under multiple working conditions, and dividing the structure response data features to obtain a first response feature fragment set; performing consistency screening on the working condition state data to obtain a first state sequence, and performing association matching on the first state sequence and the first response feature fragment set to obtain a second response feature fragment set; clustering according to a preset coupling physical quantity type to obtain a coupling feature subset; extracting a change rate between subsets and dividing a dynamic response region; constructing a graph model by taking the dynamic response region as a node, and performing community division to obtain a community feature set; and inputting the result into the neural network model, and outputting the combined use evaluation result of the equipment under multiple working conditions, thereby solving the problem of low evaluation accuracy caused by complex structure response and strong coupling characteristic under multiple working conditions.

Description

Technical Field

[0001] This invention relates to the field of multimodal mechanical coupling technology, and more specifically, to a multimodal mechanical coupling method for evaluating the suitability of multi-condition pressure-bearing equipment. Background Technology

[0002] Multi-condition pressure vessels are pressure vessels or equipment capable of operating safely under various conditions, including different temperatures, pressures, media states, and operating conditions. Compared to single-condition equipment, multi-condition pressure vessels require greater design flexibility and safety margins to adapt to changing process demands and environmental variations. Their design typically follows relevant pressure vessel codes and standards, incorporating material selection, structural strength analysis, and fatigue life assessment to ensure structural integrity and sealing performance under all conditions. Multi-condition pressure vessels are widely used in industries such as petrochemicals, natural gas processing, pharmaceuticals, and energy, effectively improving production continuity and safety while reducing the risk of equipment failure due to frequent switching between operating conditions.

[0003] However, pressure equipment will experience a variety of different operating conditions during actual operation, such as pressure fluctuations, temperature changes and alternating loads. These conditions cause the equipment structure to produce complex responses, including stress distribution, deformation and vibration and other mechanical behaviors. At the same time, there are significant coupling relationships between these responses, which are manifested as mutual influence and dynamic coupling between different physical quantities and structural parts, making it difficult to directly evaluate the overall structural state through a single index.

[0004] To address the above problems, this invention proposes a solution. Summary of the Invention

[0005] To overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide a multimodal mechanical coupling method for evaluating the suitability of pressure-bearing equipment under multiple working conditions. This method addresses the problem of low accuracy in evaluating the suitability of pressure-bearing equipment when its structural response is complex and its coupling characteristics are strong under multiple working conditions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A multimodal mechanical coupling method for evaluating the suitability of pressure-bearing equipment under multiple operating conditions includes the following steps: acquiring structural response data and operating condition state data of the pressure-bearing equipment under multiple operating conditions, and performing feature segmentation on the structural response data to obtain a first set of response feature segments; performing consistency screening on the operating condition state data to obtain a first state sequence and associating it with the first set of response feature segments to obtain a second set of response feature segments; clustering the second set of response feature segments according to a preset type of coupled physical quantity to obtain several coupled feature subsets; extracting the rate of change between coupled feature subsets and dividing the coupled feature subsets into several dynamic response regions based on the rate of change; constructing a graph model using several dynamic response regions as nodes, and performing modular decomposition of the graph model using a community partitioning algorithm to obtain a community feature set; inputting the community feature set into a preset neural network model and outputting the suitability evaluation results of the pressure-bearing equipment under multiple operating conditions.

[0007] In a preferred embodiment, the step of feature segmentation of the structural response data to obtain a first set of response feature segments specifically involves: performing empirical wavelet decomposition on the structural response data to obtain several intrinsic mode components; performing singular spectrum analysis on the several intrinsic mode components to identify abrupt change points of the several intrinsic mode components; cutting the time series with the abrupt change points as boundaries to obtain several segmented segments; calculating the Hausdorff distance between each segmented segment to construct a feature similarity matrix; using a spectral clustering algorithm to divide the feature similarity matrix into several feature segment subsets; calculating and fusing the time-frequency domain statistics of the several feature segment subsets to obtain the first set of response feature segments.

[0008] In a preferred embodiment, the process of consistency screening of the operating condition data to obtain a first state sequence and association matching with a first response feature fragment set to obtain a second response feature fragment set specifically involves: performing t-SNE dimensionality reduction on the operating condition data to obtain dimensionality-reduced operating condition data; aligning the time axis of the dimensionality-reduced operating condition data with that of the first response feature fragment set using a dynamic time warping algorithm to obtain aligned operating condition data; performing state-based discretization processing on the aligned operating condition data to obtain an initial state sequence; calculating the transition probabilities between states in the initial state sequence based on a preset Markov chain model, and screening the initial state sequence based on the transition probabilities to obtain a first state sequence; performing spatiotemporal association matching between the first state sequence and the first response feature fragment set based on a preset maximum mutual information entropy criterion; and adding operating condition state labels to the successfully matched feature fragments to obtain the second response feature fragment set.

[0009] In a preferred embodiment, the step of clustering the second response feature fragment set according to a preset type of coupled physical quantity to obtain several coupled feature subsets specifically involves: decomposing the second response feature fragments into several time gradient tensor fields using a high-order singular value decomposition algorithm; calculating the local curvature covariance matrix of the tensor fields of several time gradients to obtain a feature similarity measure; calculating the Wasserstein distance between the second response feature fragments using an entropy-regularized Sinkhorn algorithm based on the feature similarity measure; optimizing the spectral clustering algorithm based on the Wasserstein distance; and clustering the second response feature fragments based on the optimized spectral clustering algorithm to obtain coupled feature subsets.

[0010] In a preferred embodiment, the step of optimizing the spectral clustering algorithm based on Wasserstein distance and clustering the second response feature segments based on the optimized spectral clustering algorithm to obtain a coupled feature subset specifically involves: constructing a weighted adjacency matrix based on Wasserstein distance; constructing a normalized graph Laplacian for the weighted adjacency matrix and extracting the eigenvectors corresponding to several minimum eigenvalues; using the eigenvectors to map each second response feature segment to the spectral embedding space, and performing initial clustering in the spectral embedding space to obtain several initial clusters; solving for the Wasserstein centroid within each initial cluster and repeatedly performing cluster partitioning with the centroid as the cluster center until a preset convergence condition is met to obtain the coupled feature subset.

[0011] In a preferred embodiment, the step of extracting the rate of change among coupled feature subsets and dividing the coupled feature subsets into several dynamic response regions based on the rate of change specifically involves: extracting the time-series feature signals of the coupled feature subsets and performing time-frequency domain analysis on the time-series feature signals using Fourier transform to obtain the rate of change among the coupled feature subsets; constructing a rate field based on the rate of change among the coupled feature subsets and calculating the eigenvalues ​​of the Jacobian matrix at the boundaries of the coupled feature subsets; identifying the rate abrupt change hypersurface of the rate field according to the eigenvalue sign distribution and marking the continuous region enclosed by the hypersurface as the first dynamic response region; calculating the sum of the partial derivatives of the rate of change at each spatial location of the rate field and the corresponding spatial coordinates to obtain the rate field divergence; and merging the first dynamic response regions based on the positive or negative sign of the rate field divergence to obtain several dynamic response regions.

[0012] In a preferred embodiment, the modular decomposition of the graph model using a community partitioning algorithm to obtain a community feature set specifically involves: initializing each node in the graph model as an independent community and calculating the modularity of the community under the initial partition; traversing each node in the graph model and calculating the modularity increment after the node moves into an adjacent community; merging the community pairs with the largest modularity increments in each iteration of community partitioning until a preset constraint condition is met to obtain a first community; extracting the intrinsic vibration eigenvalues ​​of the nodes in the first community to construct a community feature vector; and performing tensor condensation on the community feature vector to obtain the community feature set.

[0013] In a preferred embodiment, the step of inputting the community feature set into a preset neural network model and outputting the suitability evaluation result of the pressure-bearing equipment under multiple working conditions specifically involves: performing a persistent homology transformation on the community feature set to obtain a first feature value; inputting the first feature value into the preset neural network model for neighborhood feature aggregation to obtain a hidden state vector; and using a radial basis function kernel to map the hidden state vector into a failure probability distribution as the suitability evaluation result.

[0014] The technical effects and advantages of the multimodal mechanical coupling multi-condition pressure-bearing equipment of the present invention in using the evaluation method are as follows: This invention integrates structural response and operational status data of pressure equipment under different operating conditions at multiple levels and dimensions. It employs advanced algorithms such as empirical wavelet decomposition, singular spectrum analysis, dynamic time warping, and the maximum mutual information entropy criterion to achieve accurate feature extraction and state correlation of complex structural responses, effectively capturing the dynamic changes in mechanical coupling effects under multiple operating conditions. Based on high-order singular value decomposition and the entropy-regularized Sinkhorn algorithm, deep clustering of response features is performed. Combined with Fourier transform and rate field analysis, the dynamic response region is accurately delineated. Community-level key features are extracted through community partitioning and persistent homology transform. Finally, a neural network model is used for efficient prediction of failure probability. This method significantly improves the accuracy and robustness of assessing the operational status of pressure equipment under complex structural responses and strong coupling characteristics in multi-condition environments. It overcomes the bottleneck of traditional evaluation methods in handling multimodal information fusion and dynamic coupling responses, providing scientific and reliable technical support for equipment safety management and risk control. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the evaluation method for multi-modal mechanical coupling and multi-condition pressure-bearing equipment according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1, Figure 1 This invention presents a method for evaluating the suitability of multi-modal mechanically coupled, multi-condition pressure-bearing equipment for use, comprising the following steps: S1, acquire structural response data and operating status data of pressure equipment under multiple operating conditions, and perform feature segmentation on the structural response data to obtain the first response feature fragment set; In this example, the structural response data is segmented into features to obtain the first set of response feature fragments, specifically: Empirical wavelet decomposition is performed on the structural response data to obtain several intrinsic mode components; Singular spectrum analysis is performed on several intrinsic mode components to identify abrupt change points in several intrinsic mode components; The time series was divided into several segments by using the mutation point as the boundary; Calculate the Hausdorff distance between each segment and construct a feature similarity matrix; The feature similarity matrix is ​​divided into several feature segment subsets using a spectral clustering algorithm; Calculate and fuse the time-frequency domain statistics of several feature fragment subsets to obtain the first response feature fragment set.

[0018] It should be noted that the structural response data of pressure equipment under multiple operating conditions mainly includes the mechanical response signals generated by the equipment under different operating conditions, such as vibration, stress, strain, displacement, acceleration, etc. These data reflect the dynamic behavior and physical response characteristics of the equipment structure under various operating conditions.

[0019] Furthermore, for each intrinsic mode component, its corresponding time-series data matrix is ​​first constructed. Then, the main energy features of the signal are extracted using singular value decomposition (SVD) to obtain the singular value spectrum. Abrupt or significant changes in the singular values ​​are observed within the singular value spectrum. These change points are used as candidate locations for abrupt changes, reflecting significant transformations in the structure or dynamic characteristics of the mode component, thus enabling effective identification of abrupt changes in the mode component. The identified abrupt changes are used as the cutting boundaries of the time series. Starting from the first abrupt change point, the entire time-series data of the mode component is cut into multiple continuous and non-overlapping segments. Within each segment, the signal changes are relatively stable or maintain similar characteristics, ensuring that the cut segments better reflect the structural response characteristics of the mode component in different time intervals.

[0020] Furthermore, the Hausdorff distance is an index that measures the maximum and shortest distance between two sets of points. Specifically, it is defined as the maximum value of the nearest distance from each point in one set to another set of points and the maximum value of the reverse calculation of the two. It is used to measure the similarity between two cut segments in terms of shape or trajectory. Based on this, the Hausdorff distance between all pairs of segments is calculated, and the distance values ​​are converted into similarity indices to construct a feature similarity matrix, where the matrix elements represent the strength of similarity between different segments.

[0021] Finally, for each feature segment subset, time-domain statistics (such as mean, variance, kurtosis, etc.) and frequency-domain statistics (such as spectral energy distribution, dominant frequency, etc.) are calculated separately. Combining the time and frequency characteristics of the signal, multidimensional statistical indicators that can comprehensively reflect the internal characteristics of the segment are extracted. Then, these statistics are integrated through fusion algorithms (such as weighted average or feature dimensionality reduction techniques) to form a comprehensive and stable feature representation, and finally obtain the first response feature segment set for subsequent analysis.

[0022] S2, perform consistency screening on the operating condition data to obtain the first state sequence and associate and match it with the first response feature fragment set to obtain the second response feature fragment set; In this example, consistency filtering is performed on the operating condition data to obtain a first state sequence, which is then associated and matched with a first response feature fragment set to obtain a second response feature fragment set, specifically: The operating condition data is reduced to dimensionality using t-SNE to obtain the reduced operating condition data. The time axis of the dimension-reduced operating condition data and the first response feature fragment set are aligned using a dynamic time warping algorithm to obtain aligned operating condition data. The aligned operating condition data is then subjected to state-based discretization to obtain the initial state sequence. The transition probabilities between states in the initial state sequence are calculated based on the preset Markov chain model, and the initial state sequence is filtered based on the transition probabilities to obtain the first state sequence.

[0023] Based on the preset maximum mutual information entropy criterion, the first state sequence and the first response feature fragment set are spatiotemporally correlated and matched. Add working condition status labels to the successfully matched feature fragments to obtain the second response feature fragment set.

[0024] It should be noted that the operating condition data covers environmental and process parameter information during equipment operation, such as temperature, pressure, flow rate, speed, load status, medium properties, and other external or internal conditions that affect the operating status of the equipment. These data describe the specific operating environment and status changes of the equipment. The combination of these two data is used to comprehensively evaluate the operating status and safety of pressure equipment under multiple operating conditions.

[0025] Furthermore, the aligned operating condition data is usually continuous multidimensional data. By setting predefined state division rules or thresholds, the continuous data is mapped to a finite discrete state space. Specific methods may include clustering or interval partitioning to map the state data at each time point to the corresponding discrete state label, thereby generating a sequence of discrete states. This sequence is the initial state sequence, which reflects the discrete state trajectory of the operating condition changing over time.

[0026] Furthermore, by using the initial state sequence, the number of transitions between each discrete state is counted, and then the state transition probability matrix is ​​calculated to form the transition probability matrix of the Markov chain model. Subsequently, based on this model, by setting a transition probability threshold or utilizing probability distribution characteristics, state transitions with low or abnormal transition probabilities in the initial state sequence are screened or eliminated to remove noise and inconsistent states, and finally the optimized and screened first state sequence is obtained to more accurately reflect the effective changes in the working condition.

[0027] Finally, the time alignment relationship between the first state sequence and each feature segment in the first response feature segment set is calculated to ensure that they are synchronized on the time axis. Then, the degree of information sharing between the state sequence and the response feature segments is evaluated by calculating their mutual information. Based on the maximum mutual information entropy criterion, the pairing relationship between the state sequence and the feature segment that maximizes mutual information is selected. The specific steps include: The state sequence and feature segments are mapped to each other, and the mutual information value corresponding to each segment is calculated. The maximum mutual information matching scheme is determined using a greedy algorithm. After the matching is completed, the successfully matched feature fragments are assigned corresponding status labels to form a set of closely related second response feature fragments.

[0028] S3, cluster the second response feature fragment set according to the preset coupling physical quantity type to obtain several coupling feature subsets; In this example, the set of second response feature fragments is clustered according to a preset type of coupled physical quantity to obtain several subsets of coupled features, specifically: The second response feature segment is decomposed into several time gradient tensors using a high-order singular value decomposition algorithm. Calculate the local curvature covariance matrix of the tensor field with several time gradients to obtain the feature similarity measure; Based on feature similarity measurement, the Wasserstein distance between the second response feature fragments is calculated using the entropy-regularized Sinkhorn algorithm. The second response feature fragments are clustered based on the Wasserstein distance-optimized spectral clustering algorithm, and the coupled feature subsets are obtained based on the optimized spectral clustering algorithm.

[0029] It should be noted that Higher-order Singular Value Decomposition (HOSVD) is a method for decomposing multidimensional data tensors. It constructs a multidimensional tensor representation for the second response feature fragment, covering multiple dimensions such as time, space, and physical quantities. HOSVD decomposes this higher-order tensor into a core tensor and several orthogonal factor matrices, thereby extracting gradient information in the time dimension to form a temporal gradient tensor field describing the changing characteristics at different time scales. These tensor fields reflect the multi-scale coupling structure of the response features in the spatiotemporal domain, providing a foundation for subsequent feature similarity measurement.

[0030] For each temporal gradient tensor field, the second-order statistical feature of the gradient change, i.e., the local curvature, is calculated in its local neighborhood. By obtaining the spatial gradient and second derivative of the tensor field, the covariance matrix is ​​calculated to quantify its local morphology and change characteristics. This covariance matrix reflects the consistency and complexity of the tensor field's changes in the local region. Based on these local curvature covariance matrices, a similarity metric between features can be defined to measure the degree of similarity between different response feature segments in spatiotemporal change patterns.

[0031] The distribution representations corresponding to two response feature fragments are constructed using the aforementioned feature similarity metric. The Wasserstein distance is used to measure the optimal transmission cost between these two distributions. The entropy-regularized Sinkhorn algorithm makes the computation process more stable and efficient by adding an entropy term to the traditional optimal transmission problem. Specifically, it first calculates the cost matrix of the two sets of feature distributions, then iteratively updates the transport matrix to minimize the objective function with the entropy regularization term, and finally obtains an approximate Wasserstein distance as a distance metric between the second response feature fragments, providing a basis for subsequent clustering analysis.

[0032] In this example, the Wasserstein distance-optimized spectral clustering algorithm is used, and the optimized spectral clustering algorithm is used to cluster the second response feature fragments to obtain a coupled feature subset, specifically: Construct a weighted adjacency matrix based on Wasserstein distance; Construct a normalized graph Laplacian for the weighted adjacency matrix and extract the eigenvectors corresponding to several minimum eigenvalues. Each second response feature fragment is mapped to a spectral embedding space using eigenvectors, and initial clustering is performed within the spectral embedding space to obtain several initial clusters; Solve for the Wasserstein centroid within each initial cluster and repeat the cluster partitioning with the centroid as the cluster center until the preset convergence condition is met, thus obtaining a coupled feature subset.

[0033] It should be noted that, based on the Wasserstein distance, the distance value between each pair of feature segments is converted into a similarity weight. Typically, a Gaussian kernel function or other decay function is used to map the distance to a weight value. The smaller the distance, the greater the weight, thus constructing a weighted adjacency matrix. The elements of this matrix represent the connection strength and similarity between feature segments, which reflects the relationship structure between feature segments and provides a foundation for subsequent spectral analysis.

[0034] Furthermore, extracting the eigenvectors corresponding to several minimum eigenvalues ​​of the normalized graph Laplacian operator of the weighted adjacency matrix is ​​a key step in spectral clustering. These eigenvectors can capture the main clustering information and the low-dimensional embedding space of community partitioning in the graph structure, reflecting the intrinsic similarity and clustering characteristics of data points, thus facilitating the mapping of high-dimensional complex data to a low-dimensional space for effective clustering.

[0035] By using the graph Laplacian eigenvector corresponding to each second response feature fragment as the new feature coordinates, a low-dimensional spectral embedding space is formed, where each fragment is mapped to a vector representation of a point. Then, traditional clustering algorithms (such as k-means) are used to cluster these points in this low-dimensional space to divide them into several initial clusters, which reflect the natural grouping of feature fragments on the graph structure.

[0036] Finally, the Wasserstein centroid within each initial cluster refers to the optimal representative distribution of the distributions corresponding to all feature segments in that cluster. The centroid is calculated by minimizing the sum of the Wasserstein distances from all segments to the centroid, and represents the collective characteristics of the cluster. The preset convergence condition is usually that the change in the centroid position in continuous iterations is less than a set threshold, or the cluster division no longer changes significantly, to ensure that the clustering process is stable and the final result has good representativeness.

[0037] S4, extract the rate of change between coupled feature subsets, and divide the coupled feature subsets into several dynamic response regions based on the rate of change; In this example, the rate of change among coupled feature subsets is extracted, and the coupled feature subsets are divided into several dynamic response regions based on the rate of change, specifically: The temporal feature signals of the coupled feature subsets are extracted, and the Fourier transform is used to perform time-frequency domain analysis on the temporal feature signals to obtain the rate of change between the coupled feature subsets. A rate field is constructed based on the rate of change between coupled feature subsets, and the eigenvalues ​​of the Jacobian matrix at the boundary of the coupled feature subsets are calculated. Identify the rate abrupt change hypersurface of the rate field based on the eigenvalue sign distribution, and mark the continuous region enclosed by the hypersurface as the first dynamic response region; At each spatial location in the velocity field, the sum of the partial derivatives of the rate of change of the spatial location and the corresponding spatial coordinates is calculated to obtain the velocity field divergence. The first dynamic response region is merged based on the positive and negative values ​​of the rate field divergence to obtain several dynamic response regions.

[0038] It should be noted that, for the time-series characteristic signal of each coupled feature subset, Fourier transform is used to convert it from the time domain to the frequency domain. By analyzing the dominant frequency component and amplitude changes in the spectrum, the energy distribution and variation law of the signal at different frequencies are quantified, and then the rate of change of the signal is calculated. This rate reflects the dynamic change intensity and frequency characteristics of the feature subset response, providing basic data for the subsequent construction of a rate field describing its spatial changes.

[0039] Furthermore, the rate field is a scalar or vector field describing the distribution of the rate of change of coupled feature subsets in the spatial domain, representing the speed and direction of change at each spatial location. By mapping the rate of change of each coupled feature subset to its corresponding location in space, an overall rate distribution map is constructed. The purpose of constructing the rate field is to reveal the spatial heterogeneity and abrupt change characteristics of the rate of change in space. At the boundary of the rate field, the Jacobian matrix is ​​constructed by calculating the partial derivatives of the local rate of change, and its eigenvalues ​​are solved to quantitatively describe the change properties and stability at the boundary, providing a mathematical basis for identifying abrupt change regions.

[0040] By analyzing the sign distribution of the Jacobian matrix eigenvalues ​​at the boundary of the velocity field, the locations where the eigenvalue signs change are identified. These locations correspond to critical points or faults in the velocity change, forming a hypersurface structure in space, namely the velocity abrupt change hypersurface. This hypersurface divides different regions with significant changes in the velocity field, marking the boundary of the dynamic response of the coupled feature subset, and is the key basis for the division of the dynamic response region.

[0041] Furthermore, at each spatial location of the velocity field, the divergence of the velocity field, i.e. the degree of divergence of the rate of change vector field, is calculated. The positive or negative nature of the divergence is used to determine the convergence or divergence behavior of the local flow. Based on the continuous regions with positive or negative divergence, the previously identified first dynamic response regions are spatially merged. Regions with similar and adjacent divergence properties are merged into larger dynamic response regions, thereby obtaining several dynamic response regions with inherent consistency and spatial continuity, achieving effective division of the dynamic behavior of the coupled feature subset.

[0042] S5, construct a graph model by using several dynamic response regions as nodes, and use a community partitioning algorithm to modularly decompose the graph model to obtain a community feature set; In this example, a community partitioning algorithm is used to modularly decompose the graph model to obtain the community feature set, specifically: Initialize each node in the graph model as an independent community and calculate the community modularity under the initial partition; Traverse each node in the graph model and calculate the module degree increment after the node moves into an adjacent community; In each iteration of community partitioning, the community pairs with the largest modularity increment are merged until the preset constraints are met, thus obtaining the first community; Extract the intrinsic vibration eigenvalues ​​of nodes within the first community and construct the community feature vector; Tensor condensation is performed on the community feature vectors to obtain the community feature set.

[0043] It should be noted that during the community division process, the modularity increment resulting from merging communities under the current division state is first calculated. The two adjacent communities with the largest modularity increment are selected for merging. By iteratively executing this process, the overall modularity index is continuously improved until the modularity no longer increases significantly or reaches the preset constraints (such as maximum community size, number of communities, or modularity threshold), thus ultimately forming the first community. This merging strategy based on modularity optimization ensures the quality and stability of community division.

[0044] The intrinsic vibration eigenvalues ​​of nodes within the first community refer to the natural frequencies or eigenvalues ​​calculated by constructing corresponding vibration system models (such as Laplace operators or dynamic matrices) based on the connection relationships and dynamic response characteristics of nodes within the community. These eigenvalues ​​reflect the vibration modes and dynamic characteristics of the structure within the community. These intrinsic vibration eigenvalues ​​are combined into a vector form in a certain order, that is, the community feature vector is constructed to describe the overall dynamic behavior characteristics of the community.

[0045] Finally, tensor condensation is performed on multiple community feature vectors, which means compressing and fusing these vectors according to a multidimensional tensor structure. The dominant features and latent patterns are extracted by tensor dimensionality reduction methods (such as higher-order singular value decomposition, principal component analysis, etc.), thereby integrating and simplifying the dynamic features of different communities and ultimately forming a representative and distinctive set of community features, providing efficient and accurate input features for subsequent usability evaluation.

[0046] S6 inputs the community feature set into the preset neural network model and outputs the evaluation results of the suitability of the pressure equipment under multiple working conditions.

[0047] In this example, the community feature set is input into a pre-defined neural network model, and the output is the evaluation result of the suitability of the pressure equipment under multiple operating conditions, specifically: The first eigenvalue is obtained by performing a persistent homology transformation on the community feature set. The first feature value is input into a preset neural network model to perform neighborhood feature aggregation, thereby obtaining the hidden state vector; The hidden state vector is mapped to a failure probability distribution using a radial basis function kernel, which serves as the evaluation result for suitability.

[0048] It should be noted that the persistent homology transformation of the community feature set is performed by using the persistent homology method in topological data analysis to extract topological invariants at different scales in the community feature set, capture the stability and change characteristics of the shape and structure of the community features, generate persistent barcodes or persistent maps, and then calculate the first feature value describing the overall topological structure of the community from this topological information. This feature value reflects the core topological attributes of the community feature set and serves as the key feature for subsequent neural network input.

[0049] The first feature value obtained from persistent cohomology is input into a preset neural network model. This model uses a designed neighborhood feature aggregation mechanism (such as the message passing mechanism in the neural network shown in the figure) to perform multi-level feature fusion and updates by combining the feature information of surrounding nodes. It gradually integrates the multi-dimensional features within the community and its neighborhood, and finally outputs a low-dimensional hidden state vector. The hidden state vector comprehensively reflects the local and global features of the community, providing a high-quality representation for subsequent failure probability prediction.

[0050] By employing a radial basis function (RBF) kernel, the hidden state vector output by the neural network is mapped to a high-dimensional feature space. By calculating the similarity between the hidden state vector and multiple basis function centers, a set of non-negative and normalized weights is generated. These weights can be interpreted as the probabilities of different failure modes, and then combined to form a complete failure probability distribution, reflecting the probability of various failure states that may occur in pressure equipment under multiple operating conditions.

[0051] The failure probability distribution is directly used as the output of the serviceability assessment to quantitatively describe the reliability and safety status of pressure equipment under multiple operating conditions. The assessment results reflect the failure risk level of the equipment in the current and future operating environment, providing a scientific basis for equipment maintenance decisions, risk management and safety assurance, ensuring that the equipment can operate safely and stably, and preventing potential accidents.

[0052] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0053] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0054] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0055] The above description is merely a specific embodiment 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.

[0056] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-modal mechanical coupling multi-condition pressure-bearing equipment suitable for use evaluation method, characterized in that, Includes the following steps: Obtain structural response data and operating condition data of pressure equipment under multiple operating conditions, and perform feature segmentation on the structural response data to obtain the first response feature fragment set; Consistency screening is performed on the operating condition data to obtain the first state sequence, which is then associated and matched with the first response feature fragment set to obtain the second response feature fragment set. The second response feature fragment set is clustered according to the preset coupling physical quantity type to obtain several coupling feature subsets; Extract the rate of change among the coupled feature subsets, and divide the coupled feature subsets into several dynamic response regions based on the rate of change; A graph model is constructed by using several dynamic response regions as nodes, and a community partitioning algorithm is used to modularly decompose the graph model to obtain a community feature set. The community feature set is input into a preset neural network model, and the output is the evaluation results of the suitability of the pressure equipment under multiple working conditions.

2. The multi-mode mechanical coupling multi-condition pressure-bearing equipment suitable for use evaluation method according to claim 1, characterized in that, The step of segmenting the structural response data to obtain a first set of response feature fragments is as follows: Empirical wavelet decomposition is performed on the structural response data to obtain several intrinsic mode components; Singular spectrum analysis is performed on several intrinsic mode components to identify abrupt change points in several intrinsic mode components; The time series was divided into several segments by using the mutation point as the boundary; Calculate the Hausdorff distance between each segment and construct a feature similarity matrix; The feature similarity matrix is ​​divided into several feature segment subsets using a spectral clustering algorithm; Calculate and fuse the time-frequency domain statistics of several feature fragment subsets to obtain the first response feature fragment set.

3. The multi-modal mechanical coupling multi-condition pressure-bearing equipment suitable for use evaluation method according to claim 2, characterized in that, The process of performing consistency screening on the operating condition data to obtain a first state sequence and then associating and matching it with a first set of response feature fragments to obtain a second set of response feature fragments is as follows: The operating condition data is reduced to dimensionality using t-SNE to obtain the reduced operating condition data. The time axis of the dimension-reduced operating condition data and the first response feature fragment set are aligned using a dynamic time warping algorithm to obtain aligned operating condition data. The aligned operating condition data is then subjected to state-based discretization to obtain the initial state sequence. The transition probabilities between states in the initial state sequence are calculated based on the preset Markov chain model, and the initial state sequence is filtered based on the transition probabilities to obtain the first state sequence. Based on the preset maximum mutual information entropy criterion, the first state sequence and the first response feature fragment set are spatiotemporally correlated and matched. Add working condition status labels to the successfully matched feature fragments to obtain the second response feature fragment set.

4. The multi-mode mechanical coupling multi-condition pressure-bearing equipment suitable for use evaluation method according to claim 3, characterized in that, The step of clustering the second response feature fragment set according to a preset coupling physical quantity type to obtain several coupling feature subsets is as follows: The second response feature segment is decomposed into several time gradient tensors using a high-order singular value decomposition algorithm. Calculate the local curvature covariance matrix of the tensor field with several time gradients to obtain the feature similarity measure; Based on feature similarity measurement, the Wasserstein distance between the second response feature fragments is calculated using the entropy-regularized Sinkhorn algorithm. The second response feature fragments are clustered based on the Wasserstein distance-optimized spectral clustering algorithm, and the coupled feature subsets are obtained based on the optimized spectral clustering algorithm.

5. The multi-mode mechanical coupling multi-condition pressure-bearing equipment suitable for use evaluation method according to claim 4, characterized in that, The method involves using a Wasserstein distance-optimized spectral clustering algorithm, and then clustering the second response feature fragments using the optimized algorithm to obtain a coupled feature subset. Specifically: Construct a weighted adjacency matrix based on Wasserstein distance; Construct a normalized graph Laplacian for the weighted adjacency matrix and extract the eigenvectors corresponding to several minimum eigenvalues; Each second response feature fragment is mapped to a spectral embedding space using eigenvectors, and initial clustering is performed within the spectral embedding space to obtain several initial clusters; Solve for the Wasserstein centroid within each initial cluster and repeat the cluster partitioning with the centroid as the cluster center until the preset convergence condition is met, thus obtaining a coupled feature subset.

6. The multi-mode mechanical coupling multi-condition pressure-bearing equipment suitable for use evaluation method according to claim 5, characterized in that, The extraction of the rate of change among the coupled feature subsets, and the division of the coupled feature subsets into several dynamic response regions based on the rate of change, specifically involves: The temporal feature signals of the coupled feature subsets are extracted, and the Fourier transform is used to perform time-frequency domain analysis on the temporal feature signals to obtain the rate of change between the coupled feature subsets. A rate field is constructed based on the rate of change between coupled feature subsets, and the eigenvalues ​​of the Jacobian matrix at the boundary of the coupled feature subsets are calculated. Identify the rate abrupt change hypersurface of the rate field based on the eigenvalue sign distribution, and mark the continuous region enclosed by the hypersurface as the first dynamic response region; At each spatial location in the velocity field, the sum of the partial derivatives of the rate of change of the spatial location and the corresponding spatial coordinates is calculated to obtain the velocity field divergence. The first dynamic response region is merged based on the positive and negative values ​​of the rate field divergence to obtain several dynamic response regions.

7. The multi-mode mechanical coupling multi-condition pressure-bearing equipment suitable for use evaluation method according to claim 6, characterized in that, The community partitioning algorithm is used to modularly decompose the graph model to obtain the community feature set, specifically: Initialize each node in the graph model as an independent community and calculate the community modularity under the initial partition; Traverse each node in the graph model and calculate the module degree increment after the node moves into an adjacent community; In each iteration of community partitioning, the community pairs with the largest modularity increment are merged until the preset constraints are met, thus obtaining the first community; Extract the intrinsic vibration eigenvalues ​​of nodes within the first community and construct the community feature vector; Tensor condensation is performed on the community feature vectors to obtain the community feature set.

8. The multi-mode mechanical coupling multi-condition pressure-bearing equipment suitable for use evaluation method according to claim 7, characterized in that, The process of inputting the community feature set into a preset neural network model and outputting the evaluation results of the suitability of the pressure equipment under multiple operating conditions is as follows: The first eigenvalue is obtained by performing a persistent homology transformation on the community feature set. The first feature value is input into a preset neural network model to perform neighborhood feature aggregation, thereby obtaining the hidden state vector; The hidden state vector is mapped to a failure probability distribution using a radial basis function kernel, which serves as the evaluation result for suitability.

Images