Electric connector fault prediction method

Abstract

The invention discloses an electric connector fault prediction method, and the method comprises the steps: synchronously obtaining multiple types of time series data of a contact area of a target electric connector in a preset time window based on a preset data collection device, and the time series data comprise temperature spatial and temporal distribution time series data, contact resistance transient time series data and a surface image sequence; performing feature analysis on the multi-source time sequence data by using a preset time sequence coupling sign extraction mechanism to obtain a multi-source phase change feature vector of the target electric connector in the current time window; inputting the multi-source phase change feature vector into a pre-trained electric connector health state prediction model, and outputting a prediction result of the health state of the target connector at the current moment, including a health state index; and performing differential early warning based on the predicted health state index of the target connector. Therefore, early and quantitative accurate prediction of the state of the electric connector is realized.

Description

Technical Field

[0001] This invention relates to the field of electrical signal processing technology, and in particular to a method for predicting electrical connector faults. Background Technology

[0002] Electrical connectors, as key components for electrical connections and signal transmission in electronic devices and systems, directly impact the safe and stable operation of the entire device and system. In fields with high reliability requirements, such as aerospace, high-speed rail, new energy vehicles, and precision industrial equipment, sudden failures of electrical connectors can lead to signal interruptions, control malfunctions, or even catastrophic accidents, causing significant economic losses and safety risks. Therefore, accurate fault prediction and health management of electrical connectors, shifting from "reactive maintenance" or "periodic replacement" to "predictive maintenance," has significant engineering value and is an urgent need.

[0003] Existing technologies for predicting or assessing the lifespan of electrical connectors mainly revolve around the following two categories: The first type is threshold monitoring methods based on macroscopic electrical parameters. These methods continuously monitor macroscopic electrical parameters such as contact resistance and insulation resistance, and determine performance degradation or impending failure when the measured value exceeds a preset threshold. However, this type of method suffers from significant early warning lag. Significant changes in macroscopic electrical parameters usually result from substantial damage to the internal materials (such as oxide layer thickening or microcrack propagation). At this point, the failure is often in its middle to late stages, the early warning window is extremely short, and predictive maintenance becomes meaningless.

[0004] The second category is a comprehensive evaluation method that combines multi-dimensional detection. For example, it integrates multiple means such as electrical performance monitoring, surface optical image recognition, material hardness testing, and shell sealing testing to weighted score the condition of electrical connectors and thus assess their remaining lifespan. However, its essence is still a post-processing analysis of multiple macroscopic, static, and superficial parameters. These methods fail to deeply reveal the inherent and dynamic correlation mechanisms between different parameters, and fail to deeply analyze the dynamic correlation mechanisms between different physical quantities in space and time. Such correlations are often key early characteristics of specific failure modes (such as electrothermal ablation and fretting wear), leading to inaccurate identification of potential faults and a high false alarm rate. Therefore, their early warning capabilities remain limited, and they struggle to accurately distinguish between harmless surface defects and potential precursors to internal faults. Summary of the Invention

[0005] This application provides a method for predicting electrical connector failures, which enables accurate prediction of the health status of a single electrical connector by fusing multi-source heterogeneous data.

[0006] This application provides a method for predicting electrical connector failures, including: S101, based on a preset data acquisition device, synchronously acquires multiple types of time-series data of the target electrical connector contact area within a preset time window, including temperature spatiotemporal distribution time-series data, contact resistance transient time-series data, and surface image sequence. S102, using a preset time-series coupling feature extraction mechanism, perform feature analysis on multi-source time-series data to obtain the multi-source phase transition feature vector of the target electrical connector within the current time window; S103, input the multi-source phase change feature vector into the pre-trained electrical connector health status prediction model, and output the prediction result of the current health status of the target connector, including the health status index. S104 provides differentiated early warning based on the health status index predicted by the target connector.

[0007] Preferably, Preferably, the thermal characteristics at each moment in the thermal characteristic sequence include average temperature rise, local maximum temperature rise, temperature dispersion index, and hot spot statistical characteristics; The electrical characteristics at each moment in the electrical characteristic sequence include the resistance change rate and the resistance noise characteristic index; The visual feature sequence includes surface texture complexity, edge sharpness change rate, and morphological anomaly score of the area corresponding to hot spots.

[0008] Preferably, the calculation of the thermal-morphological correlation degree based on the thermal feature sequence and the visual feature sequence includes: Extracting the local maximum temperature rise sequence from the thermal feature sequence With visual feature sequence, shape anomaly scoring sequence Calculate the time window for both. The Pearson correlation coefficient within the range yields the thermal-morphological correlation: , This represents the i-th moment within a preset time window, and k represents the number of elements extracted according to a preset time interval within the preset time window. Indicates time The local maximum temperature rise, Indicates time The morphological anomaly scoring sequence, This represents the average value of the local maximum temperature rise sequence. This represents the average value of the morphological anomaly scoring sequence.

[0009] Preferably, the pre-trained electrical connector health status prediction model is obtained in the following way: B1. Collect a large number of historical operation data packets for similar electrical connectors, including complete time-series datasets from commissioning to functional failure and normal decommissioning, and extract every time point from each data packet. The corresponding multi-source phase transition feature vector is obtained, and the health status within the future time window at that point in time is acquired, generating a health status index. , as a label; B2. Using all the multi-source phase transition feature vectors after labeling as the training sample set, train the pre-selected neural network structure, continuously optimize the model parameters, and obtain the final health status prediction model.

[0010] Preferably, the health status index is determined as follows: Based on each historical multi-source phase transition feature vector, at the corresponding time point If the electrical connector malfunctions within the subsequent preset time window, record the malfunction type and the actual time of occurrence. The health status index is: in, The preset decay time constant, This represents the actual time when the fault occurred.

[0011] Preferably, S104 includes: Based on the health status index of the target connector, a first-level warning is issued: when the health status index is not 1, a warning is issued according to the preset status threshold.

[0012] Preferably, the target electrical connector is a plurality of node electrical connectors in the target electrical network, and S104 may further include: S201, for each node i in the target electrical network The key state data of each target electrical connector at time t are obtained, including the rate of change of resistance, the thermal-morphological time-series correlation, and the predicted value of the health state index. The key state data of each target electrical connector are then used to form the global state matrix of the target electrical network at time t. ; S202, construct a topology model of the target electrical network, and calculate the dynamic inter-node fault propagation impact coefficients based on the global state matrix. To form a propagation influence coefficient matrix ; S203, input the real-time acquired propagation influence coefficient matrix into the pre-trained cascade propagation path prediction model, and output the predicted node failure sequence; S204, Based on the predicted node failure sequences of all target electrical connectors in the target electrical network, calculate the frequency of each node appearing in the node failure sequence. As a vulnerability indicator for the node, the failure sequence of the node with the highest vulnerability index is taken as the critical failure propagation path, and the cascading failure risk value of the critical failure propagation path is calculated. .

[0013] Preferably, the training method for the cascading propagation path prediction model includes: C1. Collect a large amount of historical operating data in the target electrical network, including the propagation influence coefficient matrix of each electrical connector at different historical time t, filter all historical operating data of faults occurring within a future time window from time t, and label each propagation influence coefficient matrix. C2. Using all the labeled propagation influence coefficient matrices as the training dataset, train the pre-selected neural network structure, continuously optimize the model parameters, and obtain the cascade propagation path prediction model.

[0014] Preferably, the labeling content of the propagation influence coefficient matrix is ​​set as follows: the actual network-wide status information at the time when the corresponding electrical connector fails within a future time window, that is, the failure sequence of all network nodes within a future time window starting from the time of the failure.

[0015] One or more technical solutions provided in this application have at least the following technical effects or advantages: By integrating thermal (temperature field), electrical (contact resistance), and visual (surface image) sensors, a high-dimensional time-series data stream reflecting the electro-thermal-morphological coupling behavior is synchronously acquired under a unified clock. Deep features are extracted from the raw data, including statistical features of each dimension (such as temperature rise, resistance noise, and texture complexity), as well as cross-dimensional "thermal-morphological correlation," which quantifies the synchronicity between heat generation and surface degradation. A prediction model is trained using historical data, which learns the mapping relationship from "multi-source phase transition feature vectors" to "health status index" over a future period, thereby achieving a quantitative assessment of the current health status.

[0016] By treating all critical electrical connectors in the network as nodes, their health status indices and key characteristics are synchronously acquired to form a "global state matrix," achieving a real-time snapshot of the entire network's health status. Combining the network's static topology (electrical connections, physical distances) and the dynamic health status of nodes, a real-time "fault propagation impact coefficient matrix" is calculated. This coefficient quantifies the ability of a degraded node to exert "risk pressure" on neighboring nodes through electrical and physical coupling. Using historical cascading fault data to train a model, and inputting the current propagation impact coefficient matrix, the most likely future node failure sequences and their probabilities are predicted. Based on the predicted failure sequences, the "appearance frequency" of nodes is statistically analyzed as a vulnerability indicator to identify critical fault propagation paths and calculate their overall risk value, implementing network-level early warning. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the electrical connector fault prediction method according to an embodiment of the present invention. Detailed Implementation

[0018] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0019] It should be noted that the terms "vertical," "horizontal," "up," "down," "left," "right," and similar expressions used in this article are for illustrative purposes only and do not represent the only possible implementation.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] Example 1: Figure 1 This is a flowchart illustrating the electrical connector fault prediction method according to an embodiment of the present invention.

[0022] like Figure 1 As shown, a method for predicting electrical connector faults includes the following steps: S101, based on a preset data acquisition device, synchronously acquires multiple types of time-series data of the target electrical connector contact area within a preset time window, including temperature spatiotemporal distribution time-series data. Contact resistance transient timing data Surface image sequence .

[0023] For example, the pre-installed data acquisition device includes a thermal acquisition unit, an electrical acquisition unit, and a visual acquisition unit. Specific examples can be found in existing technologies, such as miniature thermocouple arrays or infrared thermal imaging units, four-terminal sampling circuit units, and industrial camera units. This invention does not elaborate on or limit these features. For example: Thermal acquisition unit: Employs a miniature thermocouple array or a low-cost infrared thermal imaging module to monitor the temperature of the contact area. The thermocouple array is embedded in a grid pattern near the contact area, with a spatial resolution of no less than 200 micrometers. The infrared thermal imaging module is fixedly installed in the observation direction of the contact. Pixel-temperature relationships are established through calibration, and the acquired thermal radiation images are converted into temperature distribution data. Specifically, the thermocouple matrix is ​​embedded in a grid pattern (e.g., 10 rows × 10 columns) on the insulating substrate of the target connector, adjacent to the metal contacts, to capture real-time two-dimensional temperature field distribution time-series data of the contact surface and near-surface. It is understandable that the grid of the thermocouple matrix corresponds to the pixel area of ​​the thermal radiation image.

[0024] Electrical acquisition unit: Employs a four-terminal sampling circuit to synchronously measure the transient timing data of the contact resistance flowing through the contacts at a sampling frequency of not less than 1 MHz. To eliminate lead resistance errors and capture instantaneous fluctuations in resistance, as well as operating current. ; Visual acquisition unit: A fixed-focal-length visible light industrial camera that acquires visible light images of the contact surface at a preset interval (e.g., 1 frame per minute). The industrial camera is mounted in a fixed position so that its field of view can cover the contact area.

[0025] It should be noted that the visual acquisition unit and the thermal acquisition unit are installed in fixed positions. The fields of view of the thermal acquisition unit and the visual acquisition unit are mapped to spatial coordinates through offline calibration. This mapping relationship is used to associate the position in the thermal information with the physical position in the visible light image. Thus, the coordinate mapping relationship between the fields of view of the two units is pre-established through offline calibration, without the need for strict optical axis alignment. This invention will not elaborate on this, but can refer to the spatial coordinate mapping of relevant existing technologies, such as using a calibration board or feature point matching. For example, the offline calibration can be performed by placing a checkerboard calibration board with known physical dimensions in the contact area, simultaneously acquiring thermal imaging images and visible light images, identifying corner points in the images, and establishing an affine transformation or perspective transformation matrix for the pixels between the two images, thereby completing the establishment of the coordinate mapping relationship.

[0026] Therefore, all acquisition units are controlled and driven by a unified central clock source, ensuring the time alignment of multi-source data, that is, each sampling point of temperature, resistance, and image data has a strictly consistent timestamp. This constitutes a synchronous multi-source time-series data stream. The acquisition of this data stream provides original high-dimensional time-series data reflecting the electro-thermal-visual behavior of the micro-regions of the contact, offering accurate and synchronous input for subsequent feature extraction.

[0027] S102, using a preset time-series coupling feature extraction mechanism, performs feature analysis on multi-source time-series data to obtain the multi-source phase transition feature vector of the target electrical connector within the current time window, which is used to comprehensively quantify the degradation state of the target electrical connector.

[0028] In some embodiments, the pre-defined temporal coupling feature extraction mechanism specifically includes: extracting temporal features from temporal data of temperature spatiotemporal distribution, transient temporal data of contact resistance, and temporal data of surface images to obtain thermal feature sequences, electrical feature sequences, and visual feature sequences; calculating the thermal-morphological correlation degree based on the thermal feature sequences and visual feature sequences; and fusing the statistical features and thermal-morphological correlation degree of the thermal feature sequences, electrical feature sequences, and visual feature sequences to obtain a multi-source phase transition feature vector. A1. Extract the thermal features at corresponding time t from the time series data of temperature spatiotemporal distribution according to the preset time interval, including average temperature rise, local maximum temperature rise, temperature dispersion index, and hot spot statistical features, to form a thermal feature sequence.

[0029] Among them, average temperature rise , For ambient temperature, The average temperature of the contact area at time t; the local maximum temperature rise. Used to identify localized overheating; calculates the temperature dispersion index. The dispersion index is used to quantify the uniformity of heat distribution; an increase in the dispersion index often indicates non-uniform degradation of the contact surface. The effective temperature measurement points are defined; the hot spot statistical characteristics are set as the number and proportion of hot spots, specifically: hot spots exceeding a preset temperature threshold are identified in continuous areas (the threshold is set to...). The pixels are clustered into hot spots, and the number of hot spots is counted. And calculate the ratio of the total area of ​​all hot spots to the total area of ​​the observed area. .

[0030] A2. Extract electrical characteristics at corresponding times from the transient time-series data of contact resistance according to preset time intervals, including resistance change rate and resistance noise characteristic index, to form an electrical characteristic sequence.

[0031] The rate of change of resistance is defined as the transient value of the current contact resistance. Relative to the initial resistance The relative change in resistance characterizes the macroscopic drift of resistance. , The average resistance value of the target electrical connector under stable conditions during the initial stage of use; the resistance noise characteristic index is defined as: for Power spectral density (PSD) estimation is performed within a short time window, and its power spectral density (PSD) is calculated. Linear fitting was performed on PSDs in the low-frequency range (e.g., 0.1 Hz to 10 Hz). The amplitude exponent of the 1 / f noise (flicker noise), i.e., the characteristic exponent, is obtained. This index reflects the stability of the contact surface state and is related to the formation kinetics of amorphous layers such as oxide films on the contact surface. In metal contacts, An increase in the value is strongly correlated with an increase in the density of surface oxide films, adsorption layers, or defect states, making it a sensitive indicator of microscopic phase transitions.

[0032] A3. For the surface image sequence, after image registration and extraction of the region of interest (ROI, which can be referred to in the relevant technical description, but will not be elaborated here in this invention), the visual features at the corresponding time are extracted according to the preset time interval, including the surface texture complexity, the rate of change of edge sharpness, and the morphological anomaly score of the area corresponding to the hot spot, to form a visual feature sequence.

[0033] The surface texture complexity is calculated as follows: the gray-level co-occurrence matrix (GLCM) is computed within the ROI, and its entropy is calculated. , For normalized values, an increase in entropy indicates that the surface has become rougher due to wear and corrosion; the edge sharpness change rate is obtained by calculating the difference between the average floating-point values ​​of the edge gradients of adjacent frames, reflecting the generation or expansion rate of defects such as scratches. The Sobel operator is applied to the ROI to calculate the gradient magnitude map. Calculate its mean Then the rate of change of edge sharpness , The average value is the value at the time corresponding to the previous time interval; the anomaly score of the hot spot corresponding to the region is set as follows: based on the coordinate mapping (the coordinate mapping relationship established in S101), locate each hot spot on the surface image. The corresponding visible light region is used to calculate the difference between it and a pre-set standard template region (the region corresponding to a pre-prepared standard surface image, evaluated by experts and practical data experience): texture contrast. Color differences Brightness difference The morphological anomaly score is obtained by weighted summation. If there is more than one hot spot, the average value can be taken. This score takes into account the difference between the region and the normal reference region.

[0034] A4. Extracting the local maximum temperature rise sequence from the thermal feature sequence. With visual feature sequence, shape anomaly scoring sequence Calculate the time window for both. The Pearson correlation coefficient within the range yields the thermal-morphological correlation: , This represents the i-th moment within a preset time window, and k represents the number of elements extracted according to a preset time interval within the preset time window. Indicates time The local maximum temperature rise, Indicates time The morphological anomaly scoring sequence, This represents the average value of the local maximum temperature rise sequence. The average value of the morphological anomaly score sequence is denoted as . The thermal-morphological correlation quantifies the degree of synchronization between heating and surface morphological degradation over time. The closer the absolute value is to 1, the stronger the coupling between the two.

[0035] A5. Calculate the statistical characteristics of the thermal characteristic sequence, electrical characteristic sequence, and visual characteristic sequence, including the mean, variance, and first-order linear regression slope value. Combine the statistical characteristics, thermal-morphological correlation degree, and current operating condition characteristics to form a multi-source phase change characteristic vector.

[0036] The current operating condition characteristics are set as the average current and average ambient temperature within a preset time window.

[0037] S103, input the multi-source phase transition feature vector into the pre-trained electrical connector health status prediction model, and output the prediction result of the target connector's current health status, including the health status index.

[0038] In some embodiments, the pre-trained electrical connector health status prediction model is obtained in the following way: B1. Collect a large number of historical operational data packets for similar electrical connectors in this scenario, including complete time-series datasets from commissioning to the occurrence of functional failures (such as excessive contact resistance or connection failure) or normal decommissioning. Extract each time point from each data packet. The corresponding multi-source phase transition feature vector is obtained, and the health status within the future time window at that point in time is acquired, generating a health status index. And the fault type, as a label.

[0039] For example, the health status index is determined as follows: Based on each historical multi-source phase transition feature vector, at the corresponding time point If the electrical connector malfunctions within the subsequent preset time window, record the malfunction type and the actual time of occurrence. The health status index is: in, The preset decay time constant (e.g., set to) / 3). This refers to the actual time when the fault occurred; therefore, the closer the time of fault occurrence is to the current time, the better. ,but The lower the value, the worse the health status, indirectly reflecting the "remaining effective time" before the failure occurs; if at the corresponding time point If no fault occurs within the preset time window, its health status index is defined as 1, indicating that there is no risk of fault in the short term at the current moment, and the fault type can be set to null.

[0040] B2. Using all labeled multi-source phase transition feature vectors as the training sample set, train the pre-selected neural network structure, continuously optimizing the model parameters. The training objective is to minimize the loss function between the predicted and true values, such as mean squared error (MSE). , The number of training samples. For predicting health status index, The actual health status index is used to obtain the final health status prediction model. The model learns early "feature fingerprints" that can predict failure trends.

[0041] For example, the pre-selected neural network structure can be set as an LSTM model architecture or a Transformer encoder architecture. This invention will not elaborate on these. For example, the pre-selected neural network structure is a temporal regression model based on a Transformer encoder. First, a fully connected layer projects the multi-source phase transition feature vector sequence to a high-dimensional space. Then, sine-cosine position encoding is added to inject temporal sequence information. Next, N (e.g., 2) Transformer encoder layers are stacked to form the core processing module. Each layer contains a multi-head self-attention sub-layer and a feedforward neural network sub-layer to capture the long-term dependencies and dynamic patterns within the feature sequence. Finally, the sequence representation output by the encoder is aggregated in the time dimension (e.g., averaged) and outputs a predicted value of the health status index for a future preset time window through a regression head composed of a fully connected layer and an activation function.

[0042] S104 provides differentiated early warning based on the health status index predicted by the target connector.

[0043] Specifically, based on the health status index of the target connector, a first-level early warning (node ​​level) is issued: when the health status index is not 1, an early warning is issued according to a preset status threshold (determined based on expert experience and historical data). When the health status index is less than the preset status threshold (e.g., set to 0.5), a first-level early warning message (including the health status index of the target electrical connector and the multi-source phase change feature vector) is generated to notify the management personnel and implement relevant interventions.

[0044] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: By integrating thermal (temperature field), electrical (contact resistance), and visual (surface image) sensors, a high-dimensional time-series data stream reflecting the electro-thermal-morphological coupling behavior is synchronously acquired under a unified clock. Deep features are extracted from the raw data, including not only the statistical features of each dimension (such as temperature rise, resistance noise, and texture complexity) but also the cross-dimensional "thermal-morphological correlation," which quantifies the synchronicity between heating and surface degradation. A predictive model is trained using historical data, which learns the mapping relationship from "multi-source phase transition feature vectors" to "health status index" over a future period, thereby achieving a quantitative assessment of the current health status.

[0045] It can detect microscopic signs of degradation long before functional failure (such as open circuits), such as oxide film growth on contact surfaces, texture changes caused by fretting wear, and localized micro-area overheating. By quantifying these early signals, it provides a valuable time window for preventative maintenance. It outputs a continuous "health status index" (0-1), replacing the traditional binary judgment of "normal / fault," enabling maintenance personnel to accurately grasp the "sub-health" level of connectors and achieve refined condition assessment. It integrates multi-dimensional evidence for decision-making. For example, a single resistance fluctuation may be noise, but if it is accompanied by hot spots and morphological abnormalities in the area (high "thermal-morphological correlation"), it is very likely a strong signal of real degradation, thus significantly improving the reliability of diagnosis. Early warnings are no longer based on fixed threshold alarms, but on predictions of future health trends. Managers can scientifically plan maintenance timing and resources based on the changing trends and current values ​​of the health index over multiple consecutive time windows, shifting from "reactive maintenance" or "periodic inspection" to "predictive maintenance."

[0046] By simultaneously collecting three types of heterogeneous time-series data—temperature, resistance, and visual images—of the contact area, and extracting dynamic features that comprehensively reflect material degradation, contact deterioration, and surface morphology damage, a prediction model trained based on historical outcomes is used to achieve quantitative assessment of the health status of electrical connectors and early risk warning. This approach overcomes the limitations of traditional methods that only monitor resistance thresholds, integrating multi-dimensional information from electrical, thermal, and visual dimensions. In particular, by defining the thermal-morphological time-series correlation, it can capture the coupled degradation process of localized overheating caused by poor contact, which accelerates surface wear in that area. Warnings can be issued even before single-dimensional symptoms are obvious, significantly advancing the warning window. The extracted features all have clear physical or statistical meanings (e.g., resistance noise index α correlates with surface states, temperature dispersion index correlates with contact uniformity, and texture entropy correlates with surface roughness), making the prediction results highly interpretable and providing maintenance personnel with direct clues about problems (e.g., "hot spots and scratch areas highly overlap"), rather than just an abstract risk score.

[0047] By leveraging microscopic and multidimensional features, early detection of degradation processes is achieved, transforming "post-event alarms" into "pre-event predictions." A single temperature rise may be caused by the environment, and a single scratch may not affect conductivity. Through multi-source fusion and correlation analysis (such as thermal-morphological correlation), benign anomalies and malignant degradation can be effectively distinguished, improving the accuracy of early warnings. At the same time, it avoids relying on complex physical equations that are difficult to calibrate or destructive detection that cannot be performed online. It adopts a purely data-driven approach with easily obtainable tags, making high-precision prediction technology economical and feasible in industrial settings.

[0048] Example 2: In practical industrial applications, electrical connectors typically exist in a networked form within complex electrical systems (such as power distribution cabinets, airborne avionics systems, and battery management systems). The health degradation of a single connector not only affects local functionality but can also, through electrical connections and physical proximity, exert stress on other connectors in the network, triggering a cascading failure. Traditional single-point prediction methods cannot assess this systemic risk. This example aims to quantify the fault coupling and dynamic propagation mechanisms between nodes, simulating the cascading diffusion of risk in the network, and identifying critical paths and bottleneck nodes that could lead to system collapse in advance, building upon the single-node health prediction provided in Example 1.

[0049] In some embodiments, the target electrical connector is a plurality of node electrical connectors in the target electrical network, and S104 may further include: S201, for each node i in the target electrical network The key state data of each target electrical connector at time t is obtained, including at least the resistance change rate, thermal-morphological time series correlation, and predicted health state index. The key state data of each target electrical connector are used to form a global state matrix of the target electrical network at time t, which is used to reflect the microscopic health status of the entire network in real time. Specifically, in the target electrical network, all critical node electrical connectors are identified. The data acquisition process for all nodes is uniformly triggered by a high-precision synchronous clock signal located in the central control unit, ensuring that the data acquired by any two nodes in the entire network at the same physical moment has the same timestamp. Strict consistency is maintained; at each node, steps S102 to S103 are executed in parallel and independently to obtain the predicted health status index value and generate key status data; the key status data of all nodes are transmitted in real time to the central processing server through an industrial network (such as EtherCAT, TSN, or 5G private network); the central server receives and parses the key status data of all nodes in each processing cycle (e.g., per second), organizing it into a two-dimensional global status matrix according to the node number. The row index i corresponds to the node, and the column vector is the key state data of the node at time t, thereby realizing a synchronous snapshot of the health status of all nodes in the network.

[0050] S202, construct a topology model of the target electrical network, and based on the global state matrix. Calculate the dynamic inter-node fault propagation impact coefficient. To form a propagation influence coefficient matrix It reflects the direction and intensity of the impact of faults in the network in real time.

[0051] Specifically, constructing a topology model of the target electrical network includes: Based on the electrical topology (circuit schematic or wiring diagram) of the target electrical network, construct a directed weighted graph model with electrical connectors as nodes and electrical connections as edges. Abstract each electrical connector as a vertex. Vertex set Corresponding electrical connector nodes; if there is a direct electrical connection between two electrical connectors (such as through wires, buses, or PCB traces), then a directed edge is established. The direction is usually defined as the normal flow direction of current or the direction of power transfer, side set Represents electrical connections; weight matrix , This represents the electrical coupling strength from node i to node j, and assigns a weight to each edge. Its value can be the admittance value of the corresponding branch, the rated current ratio, or simplified to 1 (indicating a strong connection) and 0 (indicating no direct connection).

[0052] Specifically, based on the global state matrix Calculate the dynamic inter-node fault propagation impact coefficient: in, Let represent the electrical coupling strength from node i to node j in the weight matrix. The physical installation distance between nodes i and j is measured or obtained from CAD drawings and stored in a distance matrix. , , These are the resistance change rate, thermal-morphological temporal correlation degree, and predicted health status index of node i, extracted from the global state matrix. , These are preset normalization factors corresponding to the rate of change of resistance and the time-series correlation of thermal-morphology, namely the maximum rate of change of resistance and the maximum time-series correlation of thermal-morphology, which are set based on historical data and expert experience. , and The preset weighting coefficients can be determined from historical data or based on actual scenario requirements. The sum of these coefficients is 1, representing the importance of the resistance change rate, thermal-morphological time-series correlation, and predicted health status index in the fault propagation influence coefficient. A weighted sum is used to calculate the real-time degradation driving force of node i. Finally, the static coupling term, spatial attenuation term, and dynamic driving force are multiplied to obtain the final result. This quantifies a node's real-time ability to output "risk pressure" to neighboring nodes through electrical and physical channels due to its current deterioration state.

[0053] S203, the real-time acquired propagation influence coefficient matrix is ​​input into the pre-trained cascade propagation path prediction model, and the predicted node failure sequence is output. And the probability distribution of each node appearing at each position in the sequence, Indicates the predicted first A failed node.

[0054] In some embodiments, the training method for the cascading propagation path prediction model includes: C1. Collect a large amount of historical operational data from the target electrical network, including the propagation influence coefficient matrix of each electrical connector at different historical times t. Filter all historical operational data where faults occur within a future time window from time t, and label each propagation influence coefficient matrix. The label content is set to the actual network-wide state information at the time the corresponding electrical connector fails within the future time window, i.e., the failure sequence of all network nodes within the future time window starting from the fault time, generated in chronological order. It should be noted that if no network nodes fail within the future time window (which can be understood as a fault occurring), the failure sequence of all network nodes can be recorded as null.

[0055] It should be noted that the historical operating data can be obtained through actual operating records, or by constructing a digital twin model of the target electrical network in a simulation platform (such as MATLAB / Simulink, RT-LAB), injecting various faults and simulating their propagation process to generate simulated cascading fault data for model training.

[0056] C2. Using all the labeled propagation influence coefficient matrices as the training dataset, train the pre-selected neural network structure, continuously optimize the model parameters, and obtain the cascade propagation path prediction model.

[0057] For example, the pre-selected neural network structure can adopt a graph attention neural network architecture, which will not be elaborated upon in this invention. For example, the cascade propagation path prediction model adopts a graph neural network architecture, specifically an encoder-sequence decoder model based on a graph attention network. The encoder part: with the electrical connector as the node and the real-time fault propagation influence coefficient as the edge weight, a dynamic directed weighted graph is constructed. Message passing and feature aggregation are performed through a multi-layer graph attention network, so that each node can update its own representation according to the state and connection strength of its neighboring nodes, thereby simulating the diffusion process of fault pressure in the network. The decoder part: a recurrent neural network or a Transformer decoder is used. Starting from the graph global context vector generated by the encoder, it autoregressively predicts the next most likely failed node and gradually generates the node failure sequence P.

[0058] S204, Based on the predicted node failure sequences of all target electrical connectors in the target electrical network, count the frequency of each node appearing in the node failure sequence. (Only when the sequence is not empty) As a vulnerability indicator for that node, the failure sequence of the node with the highest vulnerability indicator is used as the critical failure propagation path. (at least one), and calculate the cascading failure risk value of the critical failure propagation path. .

[0059] Specifically, the cascading failure risk value is set as follows: in, Indicates nodes in the critical fault propagation path. This represents the health status index of the node. This serves as a vulnerability indicator for the node. It should be noted that the cascading failure risk value can be normalized; however, the specific normalization process will not be elaborated upon in this invention.

[0060] S205, Cascaded Failure Risk Value Based on Critical Fault Propagation Path Secondary warning (path level): When the risk value of a critical propagation path exceeds the preset propagation threshold (set based on historical data and expert experience, for example, 0.8), secondary warning information is generated (including the risk value of the critical fault propagation path, the health status indicators of the electrical connectors involved, and the multi-source phase change characteristic vector), and managers are notified to implement relevant interventions.

[0061] Therefore, by treating all critical electrical connectors in the network as nodes, their health status indices and key characteristics are synchronously acquired to form a "global state matrix," achieving a real-time snapshot of the entire network's health status. Combining the network's static topology (electrical connections, physical distances) and the dynamic health status of nodes, a real-time "fault propagation impact coefficient matrix" is calculated. This coefficient quantifies the ability of a degraded node to exert "risk pressure" on neighboring nodes through electrical and physical coupling. Using historical cascading fault data to train the model, and inputting the current propagation impact coefficient matrix, the most likely future node failure sequences and their probabilities are predicted. Based on the predicted failure sequences, the "appearance frequency" of nodes is statistically analyzed as a vulnerability indicator to identify critical fault propagation paths and calculate their overall risk value, implementing network-level early warning.

[0062] By elevating the predictive perspective from individual components to the entire electrical system, this study reveals the complete chain of how localized faults dynamically propagate through network topology, potentially leading to system-wide paralysis. It not only identifies "unhealthy nodes" but also recognizes "high-risk nodes" with generally poor health but located on critical propagation paths. These nodes represent the weakest links in system resilience and are the "lever points" requiring priority intervention. The study outputs multiple possible fault propagation paths and their probabilities, essentially performing real-time risk simulations and stress tests on the system. This helps operations and maintenance personnel conduct targeted monitoring, guides maintenance resources in strategic and preventative deployment, and blocks the most likely paths to system failure at minimal cost, thereby improving the reliability and resilience of the entire network.

[0063] By introducing network topology and dynamic propagation models, a "systematic risk assessment" is achieved, focusing not on isolated failures but on the chain reaction and amplification effect of failures in the network. A "dynamic vulnerability index" is defined, where the vulnerability of a node depends not only on its static location but also on its own health status and its real-time potential impact on its neighbors. A slightly degraded node in a hub position may be more risky than a severely degraded node in a peripheral position. By identifying "critical failure propagation paths" and calculating "cascading failure risk values," a data-driven basis for global optimization is provided for maintenance decisions.

[0064] This approach elevates the predictive perspective from a single, isolated electrical connector to an electrical network system comprised of multiple connectors. It provides accurate health status predictions for each node in the network. Based on network topology and real-time status, it constructs an intelligent model capable of learning historical fault propagation patterns. This model can predict which nodes are most likely to be affected if a node in the network fails, thereby identifying the "critical vulnerable chain" that leads to system collapse. This is equivalent to providing the entire electrical network with the ability to predict the chain reaction of fault propagation, and to anticipate how a local fault can trigger a systemic disaster.

[0065] Traditional forecasting methods view each connector in isolation, failing to assess the cascading effects of a single failure. In complex electrical networks, human experience struggles to determine the potential propagation paths of a fault. Data-driven modeling quantifies and visualizes elusive systemic risks as specific critical paths and risk values, providing a basis for decision-making. Preventative maintenance strategies based on critical paths guide maintenance resources to be deployed most precisely to the links with the greatest impact on system security, achieving the optimization goal of maximizing reliability at the lowest cost. The predicted critical vulnerability paths can be fed back to the design phase to improve electrical topology, optimize connector layout, or add redundancy, thereby enhancing the inherent reliability of the system from the source.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting electrical connector faults, characterized in that, include: S101, based on a preset data acquisition device, synchronously acquires multiple types of time-series data of the target electrical connector contact area within a preset time window, including temperature spatiotemporal distribution time-series data, contact resistance transient time-series data, and surface image sequence. S102, using a preset time-series coupling feature extraction mechanism, perform feature analysis on multi-source time-series data to obtain the multi-source phase transition feature vector of the target electrical connector within the current time window; S103, input the multi-source phase change feature vector into the pre-trained electrical connector health status prediction model, and output the prediction result of the current health status of the target connector, including the health status index. S104 provides differentiated early warning based on the health status index predicted by the target connector.

2. The electrical connector fault prediction method as described in claim 1, characterized in that, The preset temporal coupling feature extraction mechanism specifically includes: By extracting time-series features from temperature spatiotemporal distribution time-series data, contact resistance transient time-series data, and surface image time-series data, thermal feature sequences, electrical feature sequences, and visual feature sequences are obtained. Calculate the thermal-morphological correlation degree based on thermal feature sequences and visual feature sequences; By integrating the statistical features and thermal-morphological correlation of thermal feature sequences, electrical feature sequences, and visual feature sequences, a multi-source phase transition feature vector is obtained.

3. The electrical connector fault prediction method as described in claim 2, characterized in that, The thermal characteristics at each time point in the thermal characteristic sequence include average temperature rise, local maximum temperature rise, temperature dispersion index, and hot spot statistical characteristics. The electrical characteristics at each moment in the electrical characteristic sequence include the resistance change rate and the resistance noise characteristic index; The visual feature sequence includes surface texture complexity, edge sharpness change rate, and morphological anomaly score of the area corresponding to hot spots.

4. The electrical connector fault prediction method as described in claim 3, characterized in that, The calculation of thermal-morphological correlation based on thermal feature sequences and visual feature sequences includes: Extracting the local maximum temperature rise sequence from the thermal feature sequence With visual feature sequence, shape anomaly scoring sequence Calculate the time window for both. The Pearson correlation coefficient within the range yields the thermal-morphological correlation: , This represents the i-th moment within a preset time window, and k represents the number of elements extracted according to a preset time interval within the preset time window. Indicates time The local maximum temperature rise, Indicates time The morphological anomaly scoring sequence, This represents the average value of the local maximum temperature rise sequence. This represents the average value of the morphological anomaly scoring sequence.

5. The electrical connector fault prediction method as described in claim 4, characterized in that, The pre-trained electrical connector health status prediction model is obtained as follows: B1. Collect a large number of historical operation data packets for similar electrical connectors, including complete time-series datasets from commissioning to functional failure and normal decommissioning, and extract every time point from each data packet. The corresponding multi-source phase transition feature vector is obtained, and the health status within the future time window at that point in time is acquired, generating a health status index. , as a label; B2. Using all the multi-source phase transition feature vectors after labeling as the training sample set, train the pre-selected neural network structure, continuously optimize the model parameters, and obtain the final health status prediction model.

6. The electrical connector fault prediction method as described in claim 4, characterized in that, The health status index is determined as follows: Based on each historical multi-source phase transition feature vector, at the corresponding time point If the electrical connector malfunctions within the subsequent preset time window, record the malfunction type and the actual time of occurrence. The health status index is: in, The preset decay time constant, This represents the actual time the fault occurred.

7. The electrical connector fault prediction method as described in claim 5, characterized in that, S104 includes: Based on the health status index of the target connector, a first-level warning is issued: when the health status index is not 1, a warning is issued according to the preset status threshold.

8. The electrical connector fault prediction method as described in claim 7, characterized in that, The target electrical connector is a plurality of node electrical connectors in the target electrical network, and S104 may further include: S201, for each node i in the target electrical network The key state data of each target electrical connector at time t are obtained, including the rate of change of resistance, the thermal-morphological temporal correlation, and the predicted value of the health state index. The key state data of each target electrical connector are then used to form the global state matrix of the target electrical network at time t. ; S202, construct a topology model of the target electrical network, and calculate the dynamic inter-node fault propagation impact coefficients based on the global state matrix. To form a propagation influence coefficient matrix ; S203, input the real-time acquired propagation influence coefficient matrix into the pre-trained cascade propagation path prediction model, and output the predicted node failure sequence; S204, Based on the predicted node failure sequences of all target electrical connectors in the target electrical network, calculate the frequency of each node appearing in the node failure sequence. As a vulnerability indicator for the node, the failure sequence of the node with the highest vulnerability index is taken as the critical failure propagation path, and the cascading failure risk value of the critical failure propagation path is calculated. .

9. The electrical connector fault prediction method as described in claim 8, characterized in that, The training method for the cascading propagation path prediction model includes: C1. Collect a large amount of historical operating data in the target electrical network, including the propagation influence coefficient matrix of each electrical connector at different historical time t, filter all historical operating data of faults occurring within a future time window from time t, and label each propagation influence coefficient matrix. C2. Using all the labeled propagation influence coefficient matrices as the training dataset, train the pre-selected neural network structure, continuously optimize the model parameters, and obtain the cascade propagation path prediction model.

10. The electrical connector fault prediction method as described in claim 8, characterized in that, The labeling content of the propagation influence coefficient matrix is ​​set as follows: the actual network-wide status information at the time when the corresponding electrical connector fails within a future time window, that is, the failure sequence of all network nodes within a future time window starting from the time of failure.

Images