A method and system for equipment state monitoring and analysis and evaluation based on big data

Abstract

The application relates to the field of equipment state monitoring in an industrial Internet of Things, and discloses a method and system for equipment state monitoring and analysis and evaluation based on big data, the method comprising the following steps: multi-source heterogeneous data protocol adaptation and dynamic mask cleaning to generate a standardized data stream; based on a physical topology, a dynamic hypergraph embedding constraint equation is constructed; an incremental tensor decomposition is combined with manifold constraint to update a core tensor; based on a singular value distribution and a hyperedge propagation algorithm, an abnormal association is located; through topology optimal transmission and knowledge distillation, cross-equipment model migration is realized, and a target equipment evaluation model is generated; the system comprises a data preprocessing module, a hypergraph modeling module, a tensor analysis module, a state evaluation module, a migration learning module and a dynamic optimization module. Through multi-source data dynamic cleaning, physical constraint hypergraph modeling, incremental tensor decomposition and manifold constraint, combined with abnormal positioning closed loop and cross-equipment topology migration, the application realizes equipment state monitoring and rapid model adaptation.

Description

Technical Field

[0001] This invention relates to the field of equipment condition monitoring in the Industrial Internet of Things (IIoT), specifically to a method and system for equipment condition monitoring, analysis, and evaluation based on big data. Background Technology

[0002] In the field of industrial equipment condition monitoring, traditional methods often rely on single-type sensor data or simple fusion techniques for anomaly detection, making it difficult to effectively handle significant differences in protocol formats, sampling frequencies, and noise levels among multi-source heterogeneous sensor data. Raw data, without deep cleaning and standardization, is prone to noise interference and outliers, leading to insufficient reliability in subsequent modeling and analysis. Furthermore, existing models often employ purely data-driven graph neural networks or statistical methods, lacking explicit embedding of the equipment's physical laws, resulting in a disconnect between state representation and the actual physical mechanisms, weak interpretability, and limited adaptability to operating conditions.

[0003] Furthermore, most methods employ static modeling frameworks, failing to dynamically respond to data distribution drift caused by changes in operating conditions and component degradation during equipment operation, resulting in significant performance degradation of the model over long-term monitoring. In cross-device knowledge transfer scenarios, existing technologies struggle to quantify the topological similarity between devices and neglect the consistency alignment of physical constraints, leading to insufficient generalization ability of the transfer model on the target device. More critically, existing systems generally lack closed-loop optimization mechanisms, making it impossible to adjust model parameters in reverse based on real-time monitoring results, hindering adaptive dynamic evolution and continuous performance improvement.

[0004] Therefore, this invention proposes a method and system for equipment status monitoring, analysis and evaluation based on big data to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for equipment status monitoring, analysis, and evaluation based on big data. This solves the problems of difficult multi-source data cleaning, disconnect between physical laws and data models, insufficient adaptability to dynamic operating conditions, and low efficiency of cross-equipment knowledge transfer in industrial equipment monitoring.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for equipment status monitoring, analysis, and evaluation based on big data, the method comprising the following steps:

[0007] S1. Perform protocol adaptation and dynamic mask cleaning on multi-source heterogeneous sensor data to generate a standardized data stream;

[0008] S2. Based on the standardized data stream and the physical topology relationship of the device, construct a dynamic hypergraph topology structure and embed the physical constraint equations of device operation into the hypergraph convolution operation;

[0009] S3. Perform incremental tensor decomposition on the dynamic hypergraph topology, update the core tensor based on the sliding window, and maintain the consistency of physical laws through manifold space constraints;

[0010] S4. Calculate the device anomaly score based on the singular value distribution of the core tensor, and locate the associated superedge that caused the anomaly based on the superedge influence propagation algorithm model.

[0011] S5. Based on the topological attributes of the associated hyperedges, the similarity between devices is measured by the optimal transmission measure of the hypergraph topology, and the cross-device analysis model is transferred by combining the knowledge distillation loss function to generate the state assessment model of the target device.

[0012] Preferably, in step S1, the step of performing protocol adaptation and dynamic masking cleaning on multi-source heterogeneous sensor data to generate a standardized data stream includes:

[0013] Mask generation: Generating a dynamic binary mask matrix M based on a sliding median filter and the device's historical health status. mask ∈{0,1} N×D And masking of abnormal periods or sensor channels, where N represents the number of sensors and D represents the dimension of the original data;

[0014] Noise suppression: for the original input data Channel-by-channel sliding median filtering is applied to eliminate impulse noise and generate element-by-element multiplication fusion filtering results;

[0015] Data cleaning: Based on element-wise multiplication and filtering results, a data cleaning model, and a dynamic binary mask matrix, multi-source heterogeneous sensor data is cleaned to generate cleaned data. The data cleaning model is as follows:

[0016] X clean =MedianFilter(X raw )⊙M mask ;

[0017] in, The original input data; ⊙ represents element-wise multiplication, M mask It is a dynamic binary mask matrix;

[0018] Using the joint characteristic matrix F edge The cleaned data is subjected to joint time-frequency domain features to generate the standardized data stream.

[0019] The joint feature matrix F edge Based on the short-time Fourier transform feature F STF1 With continuous wavelet transform feature F CWT The concatenation is obtained by concatenating along the feature dimension, and the concatenation satisfies the relational expression: F edge=F STFT ⊕F CWT ;

[0020] Where ⊕ represents the feature concatenation operation; F STFT and F CWT The results were obtained by extracting signal segments using a sliding window.

[0021] Preferably, in step S2, the step of constructing a dynamic hypergraph topology based on the standardized data stream and the device physical topology relationship, and embedding the device operation physical constraint equations into the hypergraph convolution operation includes:

[0022] The dynamic hypergraph topology is defined as a hypergraph tensor. in, For the set of sensor nodes; ε t It is a set of hyperedges, and each hyperedge connects at least three sensor nodes; For the multimodal interaction weight tensor, K represents the number of sensor types and M represents the number of operating conditions;

[0023] The physical constraint equations for the device operation are generated by discretizing the partial differential equations of the device's physical laws and embedded into the hypergraph convolution operation. The formula for the convolution operation is:

[0024]

[0025] Among them, H (l+1) Let (l+1) be the feature matrix of the vertices in the (l+1)th layer. Let d represent the feature matrix of the vertices in the l-th layer, N represent the number of sensors, and d l This represents the original data dimension in the feature matrix of the vertices at layer l;

[0026] The learnable parameter matrix; σ represents the physical constraint term; σ is the nonlinear activation function; ×1 and ×2 represent the 1-modulus product and 2-modulus product of the tensor, respectively.

[0027] Preferably, the physical constraint term Generation through conservation or kinetic equations of operation of discretized devices, including heat conduction equations. Or the finite difference form of the Navier-Stokes equations;

[0028] In the hypergraph convolution operation, 1 modulo product × 1 represents the weight tensor. With vertex feature matrix H (l) The line-by-line interaction, where 2 modulo product × 2 represents the weight tensor. With parameter matrix Θ (l)The column-by-column interaction results are mapped to a high-dimensional space through a non-linear activation function σ.

[0029] Preferably, step S3 includes:

[0030] Extracting temporal feature tensors using a sliding window An optimization objective function is constructed, where N is the number of sensors, D′ is the feature dimension, t is the sliding window, and Δ is the window length. The optimization objective function is:

[0031]

[0032] The constraints are and in, For the core tensor; U (n) It is a factor matrix; R is the discretization matrix of the physical equations; max The preset rank upper limit;

[0033] The core tensor is updated based on a sliding window, and the parameters are updated through a manifold space constraint model and Stiefel manifold projection. The manifold space constraint model is as follows:

[0034]

[0035] Where η is the learning rate; For the objective function pair gradient, Let n be the factor matrix of the nth mode at the kth iteration; the QR decomposition operation projects the gradient-updated matrix onto the Stiefel manifold projection, which is...

[0036] Preferably, step S4 includes:

[0037] Through the core tensor The distribution calculation of singular values ​​and anomaly scoring formulas is used to obtain the equipment anomaly score. The anomaly scoring formula is as follows:

[0038]

[0039] Among them, S anomaly The value for the equipment malfunction score. This represents a slice of the core tensor in the r-th mode; σ is the physical constraint factor matrix; r It is the r-th singular value; The mean of the singular values;

[0040] Based on the hyperedge influence propagation algorithm model, the gradient contribution of hyperedge weights to anomaly scoring is calculated, and the associated hyperedges that trigger anomalies are determined by combining the importance metric of sensor nodes. The hyperedge influence propagation algorithm model is as follows:

[0041]

[0042] Among them, I(e j For associated superedges that trigger anomalies, Indicates the superedge e j Weight parameters; PageRank(v i ) is the sensor node v i The global topological importance score is selected based on I(e) j )>θ I The hyperedges form a set of related hyperedges, v i For the superedge e j The i-th sensor node included belongs to the sensor node set. θ I The abnormal threshold is dynamically adjusted. It is a partial derivative.

[0043] Preferably, in step S5:

[0044] The hypergraph topology optimal transmission is achieved by solving the source device hypergraph. SuperMap of Target Device Optimal transfer matrix between To achieve this, the transmission distance is defined as:

[0045]

[0046] Where S represents the hypergraph topology of the source device; T represents the target device; Π(μ S ,μ T To satisfy the marginal distribution μ S ,μ T The set of transmission matrices; and All are hypergraph topology embedding functions;

[0047] The knowledge distillation loss function is achieved by constraining the consistency of the core tensor distributions of the source and target devices, and is defined as follows:

[0048]

[0049] in, The optimal transmission matrix; D represents the i-th and j-th core tensor slices of the source and target devices, respectively; KL The divergence is Kullback-Leibler.

[0050] Preferably, the total loss function for cross-device analysis model transfer is a weighted combination of knowledge distillation loss and optimal transfer loss, wherein the knowledge distillation loss function is:

[0051]

[0052] in, Let be the total loss function for transfer learning; For the optimal transmission loss term, and These are the node embedding vectors for the source and target devices, respectively; α∈(0,1) is the balancing weight coefficient.

[0053] This invention also provides a big data-based equipment status monitoring and analysis evaluation system, the system comprising the following modules:

[0054] The data preprocessing module performs protocol conversion and dynamic mask cleaning of multi-source heterogeneous sensor data, and outputs a standardized data stream containing joint time and frequency domain features;

[0055] The hypergraph modeling module constructs a dynamic hypergraph based on the physical topology of the devices and standardized data flow, and embeds the physical constraint equations of device operation into the hypergraph convolution;

[0056] The tensor analysis module performs incremental tensor decomposition on dynamic hypergraphs and updates the core tensors through manifold space constraints.

[0057] The status assessment module calculates device anomaly scores based on the singular value distribution of the core tensor and locates associated hyperedges using the hyperedge influence propagation algorithm.

[0058] The transfer learning module, based on the topological attributes of the associated hyperedges and the optimal transmission distance, enables cross-device knowledge transfer and generates a target device evaluation model.

[0059] The dynamic optimization module adjusts the hypergraph weight parameters in real time based on anomaly scores and transfer learning results to drive the dynamic evolution of the system.

[0060] Preferably, the dynamic optimization module is specifically used to perform the following operations:

[0061] Based on the anomaly score and associated hyperedge identifier, update the hyperedge weight tensor using a gradient-driven mechanism;

[0062] By fusing transfer learning loss and core tensor features, a topological evolution gradient signal is generated;

[0063] Adaptively adjust the physical constraint discretization parameters of the hypergraph modeling module;

[0064] By synchronously updating the hypergraph weights and physical constraints through cross-module interfaces, a closed-loop optimization is formed.

[0065] This invention provides a method and system for equipment status monitoring, analysis, and evaluation based on big data. It has the following beneficial effects:

[0066] 1. This invention effectively solves the problems of heterogeneous data formats and high noise interference from multi-source sensors by using protocol conversion and dynamic masking and cleaning techniques, generating a standardized data stream containing joint time-frequency domain features. The dynamic masking mechanism, combined with the device's health status, adaptively masks abnormal data points, ensuring that the data input to subsequent modules has both integrity and reliability, providing a high-quality data foundation for device status analysis.

[0067] 2. This invention constructs a hypergraph structure based on the physical topology of equipment, embedding discretized physical equations of equipment operation into hypergraph convolution operations to achieve dual constraints of data-driven approaches and physical laws. This design breaks through the limitations of traditional graph models that rely solely on data associations, making state representations more closely match the actual operating characteristics of equipment and significantly improving the interpretability and generalization ability of the model.

[0068] 3. This invention employs a sliding window mechanism to achieve incremental tensor decomposition of dynamic hypergraphs, combined with Stiefel manifold projection to maintain parameter orthogonality, effectively avoiding mode aliasing. Physical constraint terms force the decomposition results to conform to conservation equations, ensuring the physical consistency of core tensors during temporal evolution and enhancing adaptability to dynamic changes in equipment operating conditions.

[0069] 4. This invention quantifies anomaly scoring through the singular value distribution of the core tensor and combines gradient contribution and topological importance using the hyperedge influence propagation algorithm to achieve multi-dimensional localization of anomaly sources. The dynamic optimization module adjusts the hypergraph weights in real time based on anomaly feedback, forming an "evaluation-localization-optimization" closed loop, improving the system's response speed and processing accuracy for complex faults.

[0070] 5. This invention, based on the optimal transport and knowledge distillation loss function of hypergraph topology, achieves alignment of core tensor distributions and node feature mapping between source and target devices. This mechanism overcomes the limitation of data silos for single devices, rapidly generating state evaluation models adapted to target devices using topological similarity, significantly reducing the need for labeled data and training costs for deploying new devices. Attached Figure Description

[0071] Figure 1 This is a flowchart of the method of the present invention;

[0072] Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0073] The technical solutions in 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] Please see Figure 1 This invention provides a method for equipment status monitoring and analysis evaluation based on big data, the method comprising the following steps:

[0075] S1. Perform protocol adaptation and dynamic mask cleaning on multi-source heterogeneous sensor data to generate a standardized data stream;

[0076] In this embodiment, the protocol adaptation and dynamic masking cleaning process for multi-source heterogeneous sensor data is achieved by generating a standardized data stream through the following technical solution:

[0077] For sensor devices using different communication protocols (such as Modbus, CAN, MQTT, etc.), a multi-protocol parsing rule base is established to convert raw data streams into a unified structured format. Through a pre-defined field mapping table, function codes, register addresses, and packet identifiers in the protocols are converted into standard field names, and time-series data, device IDs, and timestamp information are extracted. For unstructured data (such as JSON or binary streams), a recursive parsing algorithm is used to extract key parameters, forming an initial data matrix with consistent dimensions. Where N represents the number of sensors, and D represents the raw data dimension of a single sensor.

[0078] To eliminate sensor noise and outlier interference, a sliding median filter is used to filter the original data matrix X. raw Channel-by-channel filtering is performed. Specifically, for the time-series data of each sensor... The data segment is truncated using a window length W, the median of the data within the window is calculated, and the filtered data matrix X is generated. filtered .

[0079] Based on this, an autoencoder model is trained using historical health data of the device to calculate the current data reconstruction error ∈ in real time. When ∈ exceeds a preset threshold, a dynamic binary mask matrix M is generated. mask ∈{0,1} N×D Abnormal data points are masked. The cleaning process is implemented through element-wise multiplication:

[0080] X clean =MedianFilter(X raw )⊙Mmask ;

[0081] Where ⊙ represents the element-wise multiplication operation of the matrix, the final output is the cleaned data matrix X after removing noise and outliers. clean .

[0082] To fully characterize the equipment's operating status, the cleaned data matrix X... clean Joint time-frequency domain feature extraction is performed. First, short-time Fourier transform (STFT) is used to perform time-frequency analysis on the time-series data segments of each sensor. The signal segments are truncated using the Hanning window function, and their spectral energy distribution is calculated to generate a time-frequency feature matrix. Here, F1 represents the frequency domain feature dimension. Simultaneously, Continuous Wavelet Transform (CWT) is used to perform multi-scale decomposition on the same signal segment, and Morlet wavelet basis functions are used to generate a scale coefficient matrix with adjustable time-frequency resolution. Where F2 is the wavelet scale number. The two types of features are concatenated along the feature dimension to construct a joint feature matrix:

[0083] F edge =F STFT ⊕F CWT ;

[0084] Here, ⊕ represents the column concatenation operation of the matrix, ultimately resulting in a matrix with dimension . Standardized data flow.

[0085] Dynamic binary mask matrix M mask The generation of this data depends on real-time assessment of the device's health status. Preferably, an autoencoder model is used for unsupervised training on historical normal data to reconstruct the error ∈=∥X. raw -Decoder(Encoder(X raw ))∥2 serves as a health status indicator. When the real-time data ∈ exceeds the dynamic threshold τ (adaptively adjusted according to the 3σ principle of historical error distribution), the corresponding data point is determined to be abnormal, and M is then entered into an anomaly. mask The corresponding position is set to zero, thereby preserving valid data while shielding against abnormal interference.

[0086] The final generated standardized data stream is represented by the feature matrix F edge The feature matrix serves as the carrier, with row vectors corresponding to each sensor node and column vectors containing joint time-frequency domain features. Preferably, to adapt to subsequent hypergraph modeling requirements, the feature matrix also retains the physical attributes of the sensor nodes (such as location coordinates and type labels) as metadata for constructing the physical topology of the devices.

[0087] S2. Based on the standardized data stream and the physical topology relationship of the device, construct a dynamic hypergraph topology structure and embed the physical constraint equations of device operation into the hypergraph convolution operation;

[0088] In this embodiment, a technical solution is implemented by constructing a dynamic hypergraph topology based on the relationship between standardized data flow and device physical topology, and by embedding device operation physical constraint equations to achieve hypergraph convolution operations. The specific implementation process is as follows:

[0089] Hypergraph topology is defined in tensor form. It comprises a set of sensor nodes, a set of hyperedges, and a multimodal interaction weight tensor. The vertex set... Generated by mapping the physical topology of the devices, each vertex corresponds to a sensor node and carries position coordinates (such as 3D spatial coordinates) and type attributes (such as sensor type, such as temperature, vibration, current, etc.). Hyperedge set ε t The construction is based on the coupling relationship between device operation logic and physical state. Each hyperedge connects at least three sensor nodes to characterize complex interactions between multiple nodes (e.g., the vibration transmission path of a bearing-gear-shaft in a gearbox). Multimodal interaction weight tensor The generation depends on the joint feature matrix F of the standardized data stream. edge Where K represents the number of sensor types and M represents the number of operating conditions (e.g., no load, full load, overload, etc.). Preferably, the weight values ​​are calculated using the mutual information of the feature matrix to reflect the correlation strength of different sensor types under specific operating conditions.

[0090] To incorporate the physical laws governing device operation into hypergraph convolution operations, the conservation equations or dynamic equations need to be discretized. For example, the heat conduction equation... Discretize it into matrix form using the finite difference method. Where Q is the number of spatial discretization points, and R3 is the modal dimension of the constraint equations. This discretization matrix serves as the physical constraint term. Through learnable projection matrix Embedded into the hypergraph convolutional layer, ensuring that the model output conforms to the physical laws of the device.

[0091] Hypergraph convolution operations achieve multimodal interaction and feature propagation through tensor modulus multiplication. The convolution formula for the l-th layer is defined as:

[0092]

[0093] in, This represents the vertex feature matrix of the l-th layer, where N is the total number of sensor nodes, and d l For feature dimensions; σ is a learnable parameter matrix used for feature dimension transformation; σ is a non-linear activation function (such as ReLU). A product of 1 modulo 1 represents the weight tensor. H by row and vertex feature matrix (l)Interact to capture correlation patterns between sensor types; 2 modulo product × 2 represents the column-wise interaction between the weight tensor and the parameter matrix Θ. (l) Interact to achieve adaptive mapping of feature dimensions. Physical constraints. By embedding discretized physical equations into the convolutional output through linear projection, the model is ensured to maintain a balance between data-driven approaches and physical laws.

[0094] Weight Tensor Dynamically adjusted according to changes in equipment operating conditions. Preferably, based on the time-frequency feature matrix F of the standardized data stream. edge The real-time correlation strength between sensor types is calculated using mutual information, and the parameters of the third dimension (operating condition mode dimension) of the weight tensor are updated. For a specific operating condition mode m, its corresponding weight slice... Calculated using the following formula:

[0095]

[0096] Where I(·) represents conditional mutual information, used to quantify the correlation strength between sensor types k1 and k2 under operating condition m.

[0097] Physical constraints The generation of the constraint matrix Φ requires selecting the appropriate physical equation based on the type of equipment. Taking rotating machinery as an example, the constraint matrix Φ is generated by discretizing the Navier-Stokes equations using the finite difference form. PDE Its discretization process includes the following steps:

[0098] The velocity and pressure field variables in the fluid dynamics equations are meshed in the equipment space domain;

[0099] A linear system of equations is constructed by applying the central difference approximation to the derivative terms in the partial differential equations.

[0100] Convert the system of equations into matrix form Φ PDE , serving as the basic template for physical constraints.

[0101] Through learnable parameter matrix Physical constraint template Φ PDE It dynamically adapts to the feature space of different convolutional layers, achieving a deep fusion of physical laws and data features.

[0102] Hypergraph convolution operations overcome the limitation of traditional graph neural networks, where edges connect only two nodes, by connecting multiple nodes through hyperedges. For example, when describing bearing-gear-shaft coupled vibration, a single hyperedge can simultaneously connect vibration sensors, temperature sensors, and speed sensors, through weight tensors. The three-modal interaction synchronously models the coupling effects of mechanical vibration, heat conduction, and rotational speed changes. The embedding of physical constraint terms further constrains the rationality of the convolution output, such as ensuring that the temperature field prediction results conform to the attenuation characteristics of the heat conduction equation.

[0103] S3. Perform incremental tensor decomposition on the dynamic hypergraph topology, update the core tensor based on the sliding window, and maintain the consistency of physical laws through manifold space constraints;

[0104] In this embodiment, the technical solution of incremental tensor decomposition of the dynamic hypergraph topology and maintaining the consistency of physical laws through manifold space constraints is implemented as follows:

[0105] Temporal feature tensors generated based on dynamic hypergraph topology (N is the number of sensors, D) ′ (where T is the time series length and T is the feature dimension) A sliding window mechanism is used to extract local time series segments. Where Δ is the window length. The core tensor is realized by constructing an optimization objective function that includes reconstruction error and incremental smoothing terms. With factor matrix U (n) The joint update. The objective function is defined as:

[0106]

[0107] in, For the core tensor; U (n) It is a factor matrix; R is the discretization matrix of the physical equations; max The preset rank upper limit;

[0108] The first term is the reconstruction error term, ensuring the accuracy of the decomposition result's fit to the window data; the second term is the incremental smoothing term, which controls the smoothness of adjacent window factor matrices through a regularization factor λ, avoiding abrupt parameter changes. Constraints include:

[0109] Rank constraint: The modal rank of the core tensor satisfies To control model complexity;

[0110] Physical consistency constraint: The product of the core tensor and the factor matrix must satisfy the discretized physical equations. in, is the discretization matrix of the partial differential equation.

[0111] The core tensor is updated iteratively using the Alternating Direction Multiplier Method (ADMM). With factor matrix U (n) Specifically:

[0112] Core tensor update: fixed factor matrix U(n) Solve using the least squares method Satisfying rank constraints

[0113] Factor matrix update: fixed For each factor matrix U (n) Gradient descent optimization is performed, and the parameter update direction is adjusted by combining an incremental smoothing term.

[0114] Factor matrix U (n) The update must satisfy the Stiefel manifold constraint U. (n)T U (n) =I, to ensure the orthogonality of the decomposition results is consistent with the physical laws. Specifically, this is achieved through the projective gradient descent method:

[0115]

[0116] Where η is the learning rate; For the objective function pair gradient, The factor matrix of the nth mode (tensor decomposition dimension) at the kth iteration; the QR decomposition operation projects the gradient-updated matrix onto the Stiefel manifold. Preferably, the projection process is implemented through Gram-Schmidt orthogonalization or Householder transformation to ensure the orthogonality of the matrix column vectors.

[0117] Discretization matrix Φ of physical equations PDE The generation is based on the conservation equations (such as mass conservation and energy conservation) or dynamic equations (such as the Navier-Stokes equations) of the equipment operation. Taking the fluid dynamics equations as an example, their discretization process includes:

[0118] The continuous equation is meshed in the device space domain to generate a spatial discrete point set;

[0119] A system of linear equations is constructed by approximating the partial derivative terms using finite difference.

[0120] Convert the system of equations into matrix form Φ PDE , serving as a template for physical constraints.

[0121] Through constraints The core tensor is forced to remain consistent with the discretization equation in the third mode (physical constraint mode), thereby ensuring that the decomposition results conform to the physical laws of the equipment.

[0122] To adapt to changes in equipment operating conditions, the window length Δ and regularization factor λ are dynamically adjusted according to the data distribution. Preferably, adaptive parameters are calculated based on the feature similarity ρ of the data within the sliding window.

[0123] λ=λ base ·exp(-γρ);

[0124] Where, ρ th ρ is the similarity threshold; γ is the decay coefficient. When the data distribution changes drastically (ρ < ρ0), th Shorten the window length to respond quickly to sudden changes; conversely, lengthen the window to enhance stability.

[0125] S4. Calculate the device anomaly score based on the singular value distribution of the core tensor, and locate the associated superedge that caused the anomaly based on the superedge influence propagation algorithm model.

[0126] In this embodiment, the technical solution of calculating device anomaly scores based on the singular value distribution of the core tensor and locating the associated hyperedges that caused the anomalies using the hyperedge influence propagation algorithm is implemented as follows:

[0127] Singular value analysis of the core tensor:

[0128] Core tensors obtained from incremental tensor decomposition Perform modal slicing and singular value decomposition. Slicing is performed on the third mode (physically constrained mode). Calculate its relationship with the physical constraint factor matrix. 3-modulus product:

[0129]

[0130] in, U represents a slice of the core tensor in the k-th physical constraint mode; (3) (k,:) is the k-th row vector of the factor matrix. Singular Value Decomposition (SVD) is performed on the product result to obtain the set of singular values. Characterize the energy distribution characteristics of each mode.

[0131] Equipment Anomaly Score anomaly The formula is defined by using the statistical properties of singular value distributions and the energy weighting of physical constraint modes:

[0132]

[0133] in, is the mean of the singular values; ∥·∥2 is the spectral norm of the matrix. This formula achieves anomaly detection through the following mechanism:

[0134] Energy weighting term It reflects the energy intensity under physical constraints. In abnormal states, the energy distribution deviates from the normal pattern because the physical laws are broken.

[0135] Exponential decay term exp(-σ) r / σavg ): Suppress the contribution of high singular values, amplify the anomaly sensitivity of low singular values, and enhance the ability to capture subtle anomaly features.

[0136] To locate the associated hyperedges that trigger anomalies, the gradient contribution of the hyperedge weights to the anomaly score is calculated based on the backpropagation mechanism, and then integrated with the node topological importance metric:

[0137] Gradient contribution calculation: Hyperedge weights are calculated using automatic differentiation techniques. For S anomaly partial derivatives It reflects the sensitivity of changes in hyperedge weights to outlier scores;

[0138] Node importance measurement: The PageRank algorithm is used to calculate node v. i Global topological importance PageRank (v i This measures the connectivity centrality and information propagation capability of nodes in a hypergraph.

[0139] Hyperedge Influence Integration: For each hyperedge e j The formula for aggregating the influence of its connected nodes is:

[0140]

[0141] in, Indicates the superedge e j Weight parameters; PageRank(v i ) is node v i Global topological importance score, v i For the superedge e j The i-th sensor node included belongs to the sensor node set. θ I The abnormal threshold is dynamically adjusted. Partial derivatives are used to describe the rate of change of a variable in a multivariate function.

[0142] This formula combines local gradient contributions with global topological importance to identify hyperedges that are both sensitive to anomalies and located in critical topological positions.

[0143] The threshold θ is dynamically adjusted based on the distribution characteristics of historical anomaly scores. I Preferably, the adaptive quantile method is used:

[0144] θ I =μ I +k·σ I ;

[0145] Where, μ I σ represents the historical average influence.I is the standard deviation; k is the sensitivity coefficient. Screening is performed to meet I(e... j )>θ I The hyperedges form the associated hyperedge set ε fault It outputs the set of sensor nodes it connects to and the physical coupling path, providing root cause information for subsequent maintenance decisions.

[0146] S5. Based on the topological attributes of the associated hyperedges, the similarity between devices is measured by the optimal transmission measure of the hypergraph topology, and the cross-device analysis model is transferred by combining the knowledge distillation loss function to generate the state assessment model of the target device.

[0147] In this embodiment, a technical solution for cross-device analysis model transfer based on the topological attributes of associated hyperedges is proposed. This solution generates a state assessment model for the target device through joint optimization of hypergraph topology optimal transfer and knowledge distillation loss function. The specific implementation process is as follows:

[0148] Define source device hypergraph SuperMap of Target Device Its topological units are sets of sub-hypergraphs. and Hypergraph Embedded Network Map the sub-hypergraph to low-dimensional feature vectors and calculate the topologically optimal transmission distance:

[0149]

[0150] Where S represents the hypergraph topology of the source device, including its sensor node set, hyperedge connections, and physical constraints; T represents the hypergraph topology of the target device, which must have similar physical topological characteristics to the source device; Π(μ S ,μ T To satisfy the marginal distribution μ S (Source device sub-hypergraph distribution) and μ T The set of transfer matrices for (target device sub-hypergraph distribution). Optimal transfer matrix Γ. * The Sinkhorn algorithm is used for iterative solution to balance transmission cost and entropy regularization term, ensuring that overfitting is avoided while preserving topological similarity.

[0151] To transfer knowledge from the source device to the target device, global consistency of the core tensor distributions of the two devices is constrained. The knowledge distillation loss function is defined as:

[0152]

[0153] in, The optimal transmission matrix; D represents the i-th and j-th core tensor slices of the source and target devices, respectively; KL The Kullback-Leibler divergence is used to measure the similarity of the distributions of the two core tensors. This is achieved by minimizing... Force the target device to approximate the physical constraint modal characteristics of the source device in the feature space.

[0154] To enhance cross-device node-level feature consistency, an optimal transmission loss term is defined. Local topological similarity is constrained by weighted node embedding distance:

[0155]

[0156] in, and These are the node embedding vectors for the source and target devices, respectively, obtained from the output of the last layer of the hypergraph convolutional network. The total loss function is a weighted combination of the knowledge distillation loss and the optimal transmission loss:

[0157]

[0158] in, α is the total loss function for transfer learning, used for cross-device knowledge transfer tasks, which jointly optimizes the objectives of knowledge distillation and topology alignment; α∈(0,1) is an adjustable balancing weight coefficient used to control the optimization ratio of global distribution alignment and local feature alignment.

[0159] Please see Figure 2 The present invention also provides a big data-based equipment status monitoring and analysis evaluation system, the system comprising the following modules:

[0160] The data preprocessing module is used to perform protocol conversion and dynamic mask cleaning on multi-source heterogeneous sensor data to generate a standardized data stream containing joint time-frequency domain features.

[0161] This module is responsible for protocol adaptation and cleaning of multi-source heterogeneous sensor data, generating a standardized data stream. Specifically, for sensor devices using different communication protocols (such as Modbus, CAN bus, MQTT, etc.), the raw data is converted into a unified structured format using a predefined protocol parsing rule base. The data cleaning process employs a sliding median filter to suppress impulse noise and combines it with a dynamic binary mask matrix generated based on the device's health status to mask outlier data points. The cleaned data undergoes joint time-frequency domain feature extraction, fusing frequency-domain energy features from short-time Fourier transform with time-frequency multi-scale features from continuous wavelet transform to generate a high-dimensional feature matrix, which serves as input for subsequent hypergraph modeling.

[0162] The hypergraph modeling module is used to construct a dynamic hypergraph structure based on the physical topology of the device and the standardized data flow, and to embed the physical constraints of device operation in the hypergraph convolution operation;

[0163] This module constructs a dynamic hypergraph topology based on the physical topology of the devices and standardized data flow. Hypergraph vertices represent sensor nodes, and hyperedges connect multiple sensors that are physically coupled or logically related. The hypergraph weight tensor is dynamically generated through a multimodal interaction mechanism, reflecting the correlation strength of different sensor types under specific operating conditions. In hypergraph convolution operations, discretized device physical equations are embedded, incorporating physical laws as constraint terms into the feature propagation process to ensure that the model output conforms to actual physical laws.

[0164] The tensor analysis module is used to perform incremental tensor decomposition on dynamic hypergraphs and update the core tensors through manifold space constraints.

[0165] This module performs incremental tensor decomposition on the dynamic hypergraph topology, extracting time-series data segments through a sliding window and updating the core tensor with manifold space constraints. The core tensor decomposition process enforces the constraints of the discretized matrix of the physical equations and maintains the orthogonality of the factor matrices through Stiefel manifold projection, avoiding mode aliasing. The decomposition results update the core tensor and factor matrices in real time, providing a low-dimensional, high-fidelity state representation for subsequent anomaly detection.

[0166] The status assessment module is used to calculate the device anomaly score based on the singular value distribution of the core tensor and locate the associated hyperedges that caused the anomaly.

[0167] This module calculates device anomaly scores based on the singular value distribution of the core tensor. By analyzing the energy decay characteristics of singular values ​​and their deviation from physical constraint modes, it quantifies the probability of device anomalies. Combining the hyperedge influence propagation algorithm, it calculates the gradient contribution of the anomaly score to the hyperedge weights and integrates the global topological importance of nodes (such as PageRank scores) to locate the set of associated hyperedges that triggered the anomalies, outputting the sensor node and coupling path of the root cause of the anomaly.

[0168] The transfer learning module is used to achieve cross-device knowledge transfer based on the topological attributes of associated hyperedges and generate a state evaluation model for the target device.

[0169] This module generates an optimal transmission matrix based on the topological attributes of associated hyperedges and the similarity between sub-hypergraphs of devices using the hypergraph topological optimal transmission metric. It combines a knowledge distillation loss function to constrain the consistency of the core tensor distributions of the source and target devices and weighted aligns node-level embedded features. By jointly optimizing the knowledge distillation loss and the optimal transmission loss, it drives the target device model to inherit the knowledge of the source device while adapting to its unique operational characteristics, generating a customized state assessment model.

[0170] The dynamic optimization module is used to adjust the hypergraph weight parameters in real time based on anomaly scores and transfer learning results, driving the dynamic evolution of the system.

[0171] This module receives anomaly scores and associated hyperedge identifiers from the state assessment module in real time. It models the dynamic evolution of the hypergraph topology using neural differential equations, updating the hyperedge weight tensors to respond to changes in device state. Simultaneously, it integrates the model transfer loss from the transfer learning module and the core tensors from the tensor analysis module to generate topology evolution gradient signals, guiding iterative optimization of the hypergraph weight parameters. Based on changes in device operating conditions, it adaptively adjusts the discretized parameters of the physical constraint terms in the hypergraph modeling module, ensuring the real-time and accurate embedding of physical laws. Through a cross-module parameter synchronization interface, the updated hypergraph weights and physical constraint matrices are synchronized to both the hypergraph modeling and tensor analysis modules, forming a closed-loop feedback loop of "state assessment - parameter optimization - model update," driving the continuous dynamic evolution of the system.

[0172] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for equipment status monitoring, analysis, and evaluation based on big data, characterized in that, The method includes the following steps: S1. Perform protocol adaptation and dynamic mask cleaning on multi-source heterogeneous sensor data to generate a standardized data stream; S2. Based on the standardized data flow and device physical topology relationship, a dynamic hypergraph topology structure is constructed, and the device operation physical constraint equations are embedded in the hypergraph convolution operation, specifically including: The dynamic hypergraph topology is defined as a hypergraph tensor. ,in, A set of sensor nodes; It is a set of hyperedges, and each hyperedge connects at least three sensor nodes; For multimodal interaction weight tensors, Indicates the number of sensor types. Indicates the number of operating modes; The physical constraint equations for the device operation are generated by discretizing the partial differential equations of the device's physical laws and embedded into the hypergraph convolution operation. The formula for the convolution operation is: ； in, For the first Layer vertex feature matrix, Indicates the first Layer vertex feature matrix, Indicates the number of sensors. Indicates the first The original data dimension in the layer vertex feature matrix; The learnable parameter matrix; For physical constraints; It is a non-linear activation function; and These represent the product of tensors modulo 1 and modulo 2, respectively. S3. Perform incremental tensor decomposition on the dynamic hypergraph topology, update the core tensor based on a sliding window, and maintain the consistency of physical laws through manifold space constraints, including: Extracting temporal feature tensors using a sliding window And construct the optimization objective function, where, For the number of sensors, Let be the feature dimension, and t be the sliding window. Given the window length, the optimization objective function is: ； The constraints are and ,in, For the core tensor; It is a factor matrix; This is the discretization matrix of the physical equations; The preset rank upper limit; The core tensor is updated based on a sliding window, and the parameters are updated through a manifold space constraint model and Stiefel manifold projection. The manifold space constraint model is as follows: ； in, The learning rate; For the objective function pair gradient, For the first The mode in the th ... The factor matrix at the next iteration; The decomposition operation projects the gradient-updated matrix onto a Stiefel manifold projection, which is... ; S4. Calculate the device anomaly score based on the singular value distribution of the core tensor, and locate the associated superedge that caused the anomaly based on the superedge influence propagation algorithm model. S5. Based on the topological attributes of the associated hyperedges, the similarity between devices is measured by the optimal transmission measure of the hypergraph topology, and the cross-device analysis model is transferred by combining the knowledge distillation loss function to generate the state assessment model of the target device.

2. The equipment status monitoring and analysis evaluation method based on big data according to claim 1, characterized in that, In step S1, the step of performing protocol adaptation and dynamic masking cleaning on multi-source heterogeneous sensor data to generate a standardized data stream includes: Mask generation: Generating a dynamic binary mask matrix based on a sliding median filter and the device's historical health status. And shielding during abnormal periods or sensor channels, among which, Indicates the number of sensors. Indicates the original data dimension; Noise suppression: for the original input data Channel-by-channel sliding median filtering is applied to eliminate impulse noise and generate element-by-element multiplication fusion filtering results; Data cleaning: Based on element-wise multiplication and filtering results, a data cleaning model, and a dynamic binary mask matrix, multi-source heterogeneous sensor data is cleaned to generate cleaned data. The data cleaning model is as follows: ； in, This is the original input data; This represents element-wise multiplication. It is a dynamic binary mask matrix; Using the joint feature matrix The cleaned data is subjected to joint time-frequency domain features to generate the standardized data stream. The joint feature matrix Features of Short Time Fourier Transform Features of continuous wavelet transform The concatenation is obtained by concatenating along the feature dimension, and the concatenation satisfies the relational expression: ; in, Indicates feature concatenation operation; and The results were obtained by extracting signal segments using a sliding window.

3. The equipment status monitoring and analysis evaluation method based on big data according to claim 1, characterized in that, The physical constraints Generation through conservation or kinetic equations of operation of discretized devices, including heat conduction equations. Or the finite difference form of the Navier-Stokes equations; In the hypergraph convolution operation, the modulo-1 product... Represents the weight tensor With vertex feature matrix Line-by-line interaction, 2-modulo product Represents the weight tensor With parameter matrix The interaction is performed column by column, and the result is activated by a non-linear activation function. Mapped to a higher-dimensional space.

4. The equipment status monitoring and analysis evaluation method based on big data according to claim 1, characterized in that, The steps in S4 include: Through the core tensor The distribution calculation of singular values ​​and anomaly scoring formulas is used to obtain the equipment anomaly score. The anomaly scoring formula is as follows: ； in, The value for the equipment malfunction score. The core tensor represents the first... A slice of each modality; This is the physical constraint factor matrix; For the first One singular value; The mean of the singular values; Based on the hyperedge influence propagation algorithm model, the gradient contribution of hyperedge weights to anomaly scoring is calculated, and the associated hyperedges that trigger anomalies are determined by combining the importance metric of sensor nodes. The hyperedge influence propagation algorithm model is as follows: ； in, To trigger abnormal associated hyperedges, Indicates the superedge Weight parameters; For sensor nodes Global topological importance score, select those that meet the requirements The hyperedges form a set of related hyperedges. For super-edge The first included There are 1 sensor node, belonging to the sensor node set. ; The abnormal threshold is dynamically adjusted. It is a partial derivative.

5. The equipment status monitoring and analysis evaluation method based on big data according to claim 1, characterized in that, In step S5: The hypergraph topology optimal transmission is achieved by solving the source device hypergraph. SuperMap of Target Device Optimal transfer matrix between To achieve this, the transmission distance is defined as: ； in, Represents the hypergraph topology of the source device; Indicates the target device; To satisfy marginal distribution The set of transmission matrices; and All are hypergraph topology embedding functions; The knowledge distillation loss function is achieved by constraining the consistency of the core tensor distributions of the source and target devices, and is defined as follows: ； in, The optimal transmission matrix; The source device and the target device are respectively represented by their first and second parts. and A core tensor slice; The divergence is Kullback-Leibler.

6. The equipment status monitoring and analysis evaluation method based on big data according to claim 1, characterized in that, The total loss function for cross-device analysis model transfer is a weighted combination of knowledge distillation loss and optimal transfer loss, wherein the knowledge distillation loss function is: ； in, Let be the total loss function for transfer learning; For the optimal transmission loss term, and These are the node embedding vectors for the source device and the target device, respectively; These are the balancing weighting coefficients.

7. A big data-based equipment status monitoring and analysis evaluation system, applied to the method described in any one of claims 1-6, characterized in that, The system includes the following modules: The data preprocessing module performs protocol conversion and dynamic mask cleaning of multi-source heterogeneous sensor data, and outputs a standardized data stream containing joint time and frequency domain features; The hypergraph modeling module constructs a dynamic hypergraph based on the physical topology of the devices and standardized data flow, and embeds the physical constraint equations of device operation into the hypergraph convolution; The tensor analysis module performs incremental tensor decomposition on dynamic hypergraphs and updates the core tensors through manifold space constraints. The status assessment module calculates device anomaly scores based on the singular value distribution of the core tensor and locates associated hyperedges using the hyperedge influence propagation algorithm. The transfer learning module, based on the topological attributes of the associated hyperedges and the optimal transmission distance, enables cross-device knowledge transfer and generates a target device evaluation model. The dynamic optimization module adjusts the hypergraph weight parameters in real time based on anomaly scores and transfer learning results to drive the dynamic evolution of the system.

8. The equipment status monitoring and analysis evaluation system based on big data according to claim 7, characterized in that, The dynamic optimization module is specifically used to perform the following operations: Based on the anomaly score and associated hyperedge identifier, update the hyperedge weight tensor using a gradient-driven mechanism; By fusing transfer learning loss and core tensor features, a topological evolution gradient signal is generated; Adaptively adjust the physical constraint discretization parameters of the hypergraph modeling module; By synchronously updating the hypergraph weights and physical constraints through cross-module interfaces, a closed-loop optimization is formed.

Images