Real-time data-based intelligent real-time warning system for wiring harness connection state

Abstract

The present application belongs to the technical field of wire harness connection early warning, and in particular to a wire harness connection state intelligent real-time early warning system based on real-time data. In view of the problems of the existing sensing dimension being single, the static threshold adaptive ability being absent, the network topology cognition being absent, and the active maintenance being unable to be supported, the present application proposes the following scheme, which comprises a multi-source sensing fusion module connected with an adaptive preprocessing module, and the adaptive preprocessing module is connected with a health index construction module. The present application realizes online measurement of electrochemical impedance spectroscopy and synchronous sensing of multiple physical quantities through multi-source sensing fusion, realizes individualized collaborative evolution of early warning threshold through dynamic baseline learning, realizes early warning de-redundancy through topology sensing, realizes bidirectional coupling of prediction and simulation and residual life prediction through digital twin simulation, forms a complete technical closed loop, and comprehensively improves the fault detection accuracy, false alarm rate and active maintenance ability.

Description

Technical Field

[0001] This invention relates to the field of wire harness connection early warning technology, and in particular to an intelligent real-time early warning system for wire harness connection status based on real-time data. Background Technology

[0002] Wiring harness connectors are critical fundamental components of automotive electrical systems, typically numbering in the hundreds throughout a vehicle, and undertaking the dual core functions of power transmission and signal interconnection. With the rapid popularization of new energy vehicles and the continuous increase in the complexity of vehicle electrical architecture, the reliability of wiring harness connectors has become the primary source of vehicle electrical failures. Statistics show that over 30% of vehicle electrical failures can be attributed to poor contact, oxidation / corrosion, or mechanical loosening of wiring harness connectors. While existing technologies for monitoring the condition of wire harness connectors have accumulated some expertise, they still suffer from the following significant shortcomings: Firstly, the sensing dimensions are limited and the information is severely insufficient. Existing solutions typically only detect contact resistance or monitor current / voltage anomalies, making it difficult to distinguish the essential differences between electrical and mechanical degradation, resulting in persistently high false alarm and false negative rates. In particular, electrochemical impedance spectroscopy, as the most accurate means of characterizing the electrochemical state of connector interfaces, is completely unsuitable for online application in automotive scenarios due to the complexity of its measurement equipment and the need for offline operation. Secondly, static threshold warning lacks adaptive capability. Existing warning schemes generally use fixed thresholds, which cannot adapt to the individual characteristics of different connectors due to differences in material batches, installation processes and usage environments. They also cannot automatically adjust the judgment criteria in real time to follow changes in operating conditions, resulting in a large number of false alarms due to operating conditions and missed alarms for real faults. Third, there is a lack of overall understanding of the connector network topology. Existing solutions treat each connector as an independent entity and monitor them separately, ignoring the electrical coupling and mechanical vibration transmission relationships between nodes in the harness network. This makes it impossible to identify the risk of cascading failures or suppress the early warning storm caused by single-point misjudgment. Fourth, there is a lack of lifespan prediction and proactive maintenance capabilities. Existing solutions are mainly based on current status judgment and lack the ability to predict the health degradation trend of connectors. They cannot provide a confident estimate of the remaining lifespan and cannot support proactive maintenance decisions. The maintenance strategy remains at the stage of passive periodic replacement.

[0003] Existing technologies suffer from limitations such as limited perception dimensions, lack of static threshold adaptive capability, lack of network topology awareness, and inability to support proactive maintenance. To address these issues, we propose a wire harness-connected intelligent real-time early warning system based on real-time data. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies, such as limited perception dimensions, lack of static threshold adaptive capability, lack of network topology knowledge, and inability to support proactive maintenance, by proposing a wire harness-connected intelligent real-time early warning system based on real-time data.

[0005] The intelligent real-time early warning system based on real-time data in the wire harness connection structure provided in this application adopts the following technical solution: A wire harness-connected intelligent real-time early warning system based on real-time data includes a multi-source sensing fusion module, which is connected to an adaptive preprocessing module, a health index construction module, a dynamic baseline learning module, a decline trend prediction module, a digital twin simulation module, and a topology sensing module. An edge inference module is connected to a topology sensing module, an edge cloud decision module is connected to an edge cloud decision module, a dynamic early warning module is connected to a dynamic early warning module, and an early warning suppression module is connected to an early warning suppression module. The root cause diagnosis module is connected to the early warning and suppression module, the root cause diagnosis module is connected to the lifetime prediction module, the lifetime prediction module is connected to the sampling control module, and the sampling control module is connected to the environmental decoupling module. The knowledge transfer module is connected to the environment decoupling module, the knowledge transfer module is connected to the security execution module, the security execution module is connected to the visual interaction module, the visual interaction module is connected to the maintenance decision module, and the maintenance decision module is connected to the closed-loop evolution module.

[0006] Furthermore, the digital twin simulation module includes a three-coupled twin unit, a prediction-driven simulation unit, a simulation back-correction unit, and a UKF data assimilation unit. The three-coupled twin unit is connected to the prediction-driven simulation unit, and the prediction-driven simulation unit is connected to the UKF data assimilation unit.

[0007] Furthermore, the edge-cloud decision module includes a Bayesian fusion unit, a confidence assessment unit, a consistency verification unit, and a hardware signature linkage unit. The Bayesian fusion unit is connected to the confidence assessment unit, the confidence assessment unit is connected to the consistency verification unit, and the consistency verification unit is connected to the hardware signature linkage unit. The confidence assessment unit employs a dynamic confidence assessment algorithm; Marginal confidence : ; in, The current signal-to-noise ratio quality score is given (0~1, calculated from the signal amplitude and noise power spectrum). The uncertainty of the edge lightweight model is 0~1, calculated from the entropy of the softmax probability distribution after temperature scaling. A score is given for consistency of historical reasoning (0~1, with a higher score for smaller variance in the N most recent reasoning results). Recommendation weight ; Cloud-based side confidence : ; in, Current communication latency (unit: ms); This is the delay decay constant (typical value 500ms, confidence level drops significantly when delay exceeds this value); This is the time difference between the last update of the cloud model; The model freshness half-life (typical value 24 hours); The standard deviation of the Monte Carlo Dropout prediction in the cloud (normalized from 0 to 1). Recommendation weight .

[0008] Furthermore, the topology sensing module includes a topology graph construction unit, a GAT risk propagation unit, a risk node marking unit, and an early warning and suppression coordination unit. The topology graph construction unit is connected to the GAT risk propagation unit, the GAT risk propagation unit is connected to the risk node marking unit, and the risk node marking unit is connected to the early warning and suppression coordination unit.

[0009] Furthermore, the dynamic baseline learning module includes an OGD online update unit, a learning rate adaptive unit, a multi-condition baseline unit, and a dynamic threshold driving unit. The OGD online update unit is connected to the learning rate adaptive unit, the learning rate adaptive unit is connected to the multi-condition baseline unit, and the multi-condition baseline unit is connected to the dynamic threshold driving unit.

[0010] Furthermore, the dynamic baseline learning module receives the four-dimensional health vector output by the health index construction module, maintains an individualized dynamic baseline model for each monitored connector, and drives the dynamic early warning module to update the warning threshold synchronously in real time using the baseline parameters. The dynamic baseline model updates the baseline parameters in real time in each sampling period using an online incremental gradient descent algorithm, and maintains an independent set of baseline parameters for different working conditions. The warning thresholds at each level are dynamically calculated based on Mahalanobis distance. Whenever the baseline parameters are updated, the warning threshold is automatically updated synchronously, realizing the co-evolution of the warning threshold and the individual state of the connector.

[0011] Furthermore, the topology perception module is used to construct a wiring harness network topology map based on the vehicle wiring harness design data, and to perform risk propagation analysis on the dynamic health map using a graph attention network to identify high-risk nodes and paths. The topology analysis results are then transmitted in real time to the early warning propagation suppression module to achieve early warning redundancy removal. The graph attention network aggregates neighborhood health information through multi-layer graph convolution and outputs a comprehensive risk score for each node. Safety-critical functional domain nodes are assigned a risk weight coefficient no less than three times that of ordinary nodes. Multiple early warnings within the same physical cluster are merged into a single associated early warning, and only the root cause node with the highest risk score is reported.

[0012] Furthermore, the edge-cloud decision module is used to fuse the local inference results of the edge inference module with the cloud-based deep analysis results using a Bayesian credibility weighted approach. When the edge-cloud results are inconsistent, a consistency anomaly handling process is triggered, and the fused decision is output to the secure execution module for hardware signature before driving the dynamic early warning module. The weights of the Bayesian fusion are dynamically calculated based on signal quality, communication latency, data freshness, and model uncertainty. Before fusion, an edge-cloud consistency check is performed. When the divergence exceeds the threshold, root cause diagnosis is used to eliminate sensor faults and the sampling rate is increased to obtain verification data. After convergence, the fused decision is output. The fused decision is digitally signed by the hardware security element to generate a trusted decision package. The early warning can only be triggered after the downstream module's signature verification is passed.

[0013] Furthermore, the digital twin simulation module receives the future health vector prediction value output by the degradation trend prediction module, drives the connector-level digital twin to perform accelerated time-step simulation, and simultaneously inputs the simulation results back into the degradation trend prediction module to correct its prediction confidence. It also achieves assimilation and update with the measured data through unscented Kalman filtering. The digital twin is composed of a geometric contact mechanics sub-model, a material aging sub-model, and an electrical characteristic sub-model coupled together. The three sub-models are coupled with each other through state vectors containing contact stress, oxide film thickness, corrosion current density, and contact resistance, which together characterize the global health state of the connector.

[0014] Furthermore, the multi-source sensing fusion module is used to simultaneously collect four types of physical parameters for each monitored connector in the wiring harness: contact resistance, micro-vibration acceleration, local temperature rise, and electrochemical impedance spectroscopy. The collected data is then packaged and output to the adaptive preprocessing module according to a unified time reference. Among them, the electrochemical impedance spectroscopy is measured online in real time across the entire frequency band within a single sampling period by applying multi-frequency sinusoidal excitation through a direct digital frequency synthesis chip and combining it with fast Fourier transform. Characteristic parameters for characterizing the oxide film thickness and corrosion degree at the contact interface are extracted through equivalent circuit fitting.

[0015] In summary, this application includes at least one of the following beneficial technical effects: 1. This solution achieves simultaneous acquisition and fusion sensing of four types of physical parameters—contact resistance, micro-vibration acceleration, local temperature rise, and electrochemical impedance spectroscopy—through a multi-source sensing fusion module. It transplants electrochemical impedance spectroscopy technology from the laboratory to the vehicle-mounted online monitoring scenario, effectively distinguishing between electrochemical degradation and mechanical degradation, and significantly improving the accuracy of fault detection. 2. This solution establishes an individualized dynamic baseline model for each connector independently through a dynamic baseline learning module. Based on the online incremental gradient descent algorithm, the baseline parameters are updated in real time, and the warning threshold evolves synchronously with the baseline, completely replacing the static threshold solution and significantly reducing the false alarm rate due to operating conditions. 3. This solution achieves bidirectional coupling between prediction-driven simulation and simulation-verified prediction through a digital twin simulation module. Combined with the remaining lifetime confidence interval output by the lifetime prediction module, it supports data-driven proactive maintenance decisions and effectively reduces unplanned downtime.

[0016] This invention achieves online measurement of electrochemical impedance spectroscopy and synchronous sensing of multiple physical quantities through multi-source sensing fusion, realizes individualized and collaborative evolution of early warning thresholds through dynamic baseline learning, realizes redundancy removal of early warning through topological sensing, and realizes bidirectional coupling of prediction and simulation and remaining lifetime prediction through digital twin simulation, forming a complete technical closed loop, which comprehensively improves the accuracy of fault detection, false alarm rate and proactive maintenance capability. Attached Figure Description

[0017] Figure 1 This is a structural block diagram of the wire harness-connected intelligent real-time early warning system based on real-time data proposed in this invention; Figure 2 This is a structural block diagram of the dynamic baseline learning module of the intelligent real-time early warning system based on real-time data in the wire harness connection structure proposed in this invention. Figure 3 This is a structural block diagram of the digital twin simulation module of the wire harness-connected intelligent real-time early warning system based on real-time data proposed in this invention. Figure 4 This is a structural block diagram of the topology sensing module of the wire harness-connected intelligent real-time early warning system based on real-time data proposed in this invention. Figure 5 This is a structural block diagram of the edge-cloud decision-making module of the wire harness-connected intelligent real-time early warning system based on real-time data proposed in this invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] Example 1 Reference Figures 1-5The intelligent real-time early warning system based on real-time data includes a multi-source sensing fusion module, which is connected to an adaptive preprocessing module, which is connected to a health index construction module, which is connected to a dynamic baseline learning module, which is connected to a decline trend prediction module, which is connected to a digital twin simulation module, and which is connected to a topology sensing module. The edge inference module is connected to the topology perception module. The edge inference module is connected to the edge cloud decision module. The edge cloud decision module is connected to the dynamic early warning module. The dynamic early warning module is connected to the early warning suppression module. The root cause diagnosis module is connected to the early warning and suppression module. The root cause diagnosis module is connected to the lifetime prediction module. The lifetime prediction module is connected to the sampling control module. The sampling control module is connected to the environmental decoupling module. The knowledge transfer module is connected to the environment decoupling module, which in turn connects to the safety execution module. The safety execution module is connected to the visual interaction module, which in turn connects to the maintenance decision module. The maintenance decision module is connected to the closed-loop evolution module. The multi-source sensing fusion module is used to synchronously collect four types of physical parameters—contact resistance, micro-vibration acceleration, local temperature rise, and electrochemical impedance spectroscopy—from each monitored connector in the wiring harness. The collected data is packaged and output to the adaptive preprocessing module according to a unified time base. Specifically, the electrochemical impedance spectroscopy is performed by applying multi-frequency sinusoidal excitation through a direct digital frequency synthesis chip and combining it with fast Fourier transform to complete online real-time measurement across the entire frequency band within a single sampling period. Characteristic parameters for characterizing the oxide film thickness and corrosion degree at the contact interface are extracted through equivalent circuit fitting.

[0020] In this embodiment, the digital twin simulation module includes a three-coupled twin unit, a prediction-driven simulation unit, a simulation back-correction unit, and a UKF data assimilation unit. The three-coupled twin unit is connected to the prediction-driven simulation unit, and the prediction-driven simulation unit is connected to the UKF data assimilation unit. The digital twin simulation module receives the future health vector prediction value output by the degradation trend prediction module, drives the connector-level digital twin to perform accelerated time-step simulation, and simultaneously inputs the simulation results back into the degradation trend prediction module to correct its prediction confidence. It also achieves assimilation and update with measured data through unscented Kalman filtering. The digital twin is composed of a geometric contact mechanics sub-model, a material aging sub-model, and an electrical characteristic sub-model coupled together. These three sub-models are coupled through state vectors containing contact stress, oxide film thickness, corrosion current density, and contact resistance, jointly characterizing the global health state of the connector. The topology perception module includes a topology graph construction unit, a GAT risk propagation unit, a risk node marking unit, and an early warning and suppression coordination unit. The topology graph construction unit is connected to the GAT risk propagation unit, the GAT risk propagation unit is connected to the risk node marking unit, and the risk node marking unit is connected to the early warning and suppression coordination unit. The topology perception module is used to construct a wiring harness network topology graph based on the vehicle wiring harness design data. It uses a graph attention network to perform risk propagation analysis on the dynamic health graph, identify high-risk nodes and paths, and transmit the topology analysis results to the early warning propagation and suppression module in real time to achieve early warning redundancy removal. The graph attention network aggregates neighborhood health information through multi-layer graph convolution and outputs a comprehensive risk score for each node. Safety critical functional domain nodes are assigned a risk weight coefficient of no less than three times that of ordinary nodes. Multiple early warnings within the same physical cluster are merged into a single associated early warning, and only the root cause node with the highest risk score is reported.

[0021] In this embodiment, the edge-cloud decision module includes a Bayesian fusion unit, a confidence evaluation unit, a consistency verification unit, and a hardware signature linkage unit. The Bayesian fusion unit is connected to the confidence evaluation unit, the confidence evaluation unit is connected to the consistency verification unit, and the consistency verification unit is connected to the hardware signature linkage unit. The confidence assessment unit employs a dynamic confidence assessment algorithm; Marginal confidence : ; in, The current signal-to-noise ratio quality score is given (0~1, calculated from the signal amplitude and noise power spectrum). The uncertainty of the edge lightweight model is 0~1, calculated from the entropy of the softmax probability distribution after temperature scaling. A score is given for consistency of historical reasoning (0~1, with a higher score for smaller variance in the N most recent reasoning results). Recommendation weight ; Cloud-based side confidence : ; in, Current communication latency (unit: ms); This is the delay decay constant (typical value 500ms, confidence level drops significantly when delay exceeds this value); This is the time difference between the last update of the cloud model; The model freshness half-life (typical value 24 hours); The standard deviation of the Monte Carlo Dropout prediction in the cloud (normalized from 0 to 1). Recommendation weight ; The edge-cloud decision module is used to fuse the local inference results of the edge inference module with the cloud-based deep analysis results using a Bayesian credibility weighted approach. When the edge-cloud results are inconsistent, a consistency anomaly handling process is triggered, and the fused decision is output to the secure execution module for hardware signing before driving the dynamic early warning module. The weights of the Bayesian fusion are dynamically calculated based on signal quality, communication latency, data freshness, and model uncertainty. Before fusion, an edge-cloud consistency check is performed. When the divergence exceeds the threshold, root cause diagnosis is used to eliminate sensor faults and the sampling rate is increased to obtain verification data. After convergence, the fused decision is output. The fused decision is digitally signed by the hardware security element to generate a trusted decision package. The early warning can only be triggered after the downstream module's signature verification is passed. The dynamic baseline learning module includes an OGD online update unit, a learning rate adaptive unit, a multi-condition baseline unit, and a dynamic threshold driving unit. The OGD online update unit is connected to the learning rate adaptive unit, which is connected to the multi-condition baseline unit, and the multi-condition baseline unit is connected to the dynamic threshold driving unit. The dynamic baseline learning module receives the four-dimensional health vector output by the health index construction module, maintains an individualized dynamic baseline model for each monitored connector independently, and drives the dynamic early warning module to update the warning threshold synchronously in real time using the baseline parameters. The dynamic baseline model updates the baseline parameters in real time in each sampling period using an online incremental gradient descent algorithm, and maintains independent baseline parameter sets for different conditions. The warning thresholds at each level are dynamically calculated based on Mahalanobis distance. Whenever the baseline parameters are updated, the warning threshold is automatically updated synchronously, realizing the co-evolution of the warning threshold and the individual state of the connector.

[0022] The implementation principle in this embodiment is as follows: During use, the multi-source sensing fusion module synchronously collects four types of physical parameters for each monitored connector: contact resistance, micro-vibration acceleration, local temperature rise, and electrochemical impedance spectroscopy. The electrochemical impedance spectroscopy is measured online in real-time across the entire frequency band within a single sampling period by applying multi-frequency sinusoidal excitation through a direct digital frequency synthesis chip and combining it with fast Fourier transform. The collected data is packaged and output to the adaptive preprocessing module according to a unified time reference. The adaptive preprocessing module performs wavelet threshold denoising purification and dynamic filtering based on operating condition self-identification on the original signal, and then outputs the purified signal to the health index construction module. The health index construction module extracts features from the purified signal and maps them into a four-dimensional health vector. The environment decoupling module obtains the vehicle's current... Environmental parameters are calculated by subtracting the environmental disturbance bias from the health vector to obtain the net health vector, which is then passed to the dynamic baseline learning module. This module maintains individualized dynamic baseline parameters for each connector using an online incremental gradient descent algorithm, and maintains independent baseline parameter sets for different operating conditions. Warning thresholds at each level are dynamically calculated based on Mahalanobis distance, and are automatically updated synchronously whenever baseline parameters are updated. Simultaneously, the health index construction module outputs the enhanced health feature matrix to the decline trend prediction module. This module combines the dynamic baseline output to predict the future health vector evolution trajectory. The prediction results are injected into the digital twin simulation module to drive accelerated time-step simulation of the connector-level digital twin. The simulation results are then used as the reverse input for the decline trend prediction. The prediction module corrects its prediction confidence, forming a two-way coupling between prediction and simulation. The node health status output by the digital twin is transmitted to the topology perception module to update the node features of the harness topology graph. The topology analysis results of the topology perception module are sent to the edge inference module. The edge inference module runs a lightweight inference model and outputs local inference results, which are input into the edge-cloud decision module together with the cloud-based deep analysis results. The edge-cloud decision module fuses the data using a Bayesian confidence-weighted method. The fusion weights are dynamically calculated based on signal quality, communication latency, data freshness, and model uncertainty. Before fusion, an edge-cloud consistency check is performed. When the divergence exceeds the threshold, an anomaly handling process is triggered. After the fusion decision is hardware-signed by the security execution module, it drives the dynamic early warning module to generate a four-level dynamic early warning signal. The system, along with the topology graph, inputs an early warning suppression module to suppress early warning propagation. Multiple early warnings within the same physical cluster are merged into a single associated early warning, and only the root cause node with the highest risk score is reported. The root cause diagnosis module combines electrochemical impedance spectroscopy characteristics and twin simulation to identify the fault type. The lifetime prediction module integrates fatigue, corrosion, and thermal aging mechanisms to output the remaining lifetime confidence interval. The maintenance decision module outputs the optimal maintenance window suggestion based on the remaining lifetime. The closed-loop evolution module collects structured fault data after a confirmed fault occurs and triggers three parallel updates: incremental fine-tuning of the prediction model, baseline learning rate strategy correction, and twin parameter re-identification, completing the system's closed-loop evolution. The sampling control module dynamically adjusts the sampling frequency of the multi-source sensing fusion module according to the early warning level, achieving on-demand resource allocation.The knowledge transfer module uses federated learning to share early warning model knowledge across different vehicle models, accelerating model convergence in small-sample scenarios for new models. The visual interaction module provides 3D wiring harness topology visualization and AR-assisted maintenance guidance.

[0023] Example 2 The difference between this embodiment and Embodiment 1 is that the multi-source sensing fusion module is connected to a sensor self-diagnosis and verification module. The sensor self-diagnosis and verification module is used to perform real-time self-diagnosis and verification of the operating status of each sensor. By injecting self-test excitation signals and cross-consistency verification, it identifies faults such as sensor offset, saturation or failure, and transmits the sensor health status identifier to the preprocessing module for subsequent confidence correction.

[0024] Example 3 The difference between this embodiment and Embodiment 1 is that the visual interaction module is connected to a driving behavior perception module. The driving behavior perception module is used to acquire driving behavior characteristics such as accelerator pedal opening, brake pedal opening, and vehicle speed change rate, classify driving behavior into aggressive, standard, and mild types, and transmit them to the dynamic warning module. The dynamic warning module dynamically adjusts the warning threshold according to the driving behavior category.

[0025] Example 4 The difference between this embodiment and Embodiment 1 is that the digital twin simulation module is connected to an electrical stress monitoring module. The electrical stress monitoring module is used to collect the working current and voltage drop flowing through the connector, monitor current impact events, and transmit the impact amplitude and duration to the digital twin simulation module. The digital twin simulation module updates the electromigration damage and thermal fatigue accumulation calculations based on the impact accumulation effect.

[0026] Example 5 The difference between this embodiment and Embodiment 1 is that the sampling control module is connected to the energy management module. The energy management module is used to monitor the overall energy consumption of the system in real time. When the vehicle is parked, it sends an energy consumption optimization scheduling command to the sampling control module to extend the electrochemical impedance spectroscopy measurement period of the low-risk connector, reduce the sampling frequency, and push a low battery reminder when the battery power is lower than the threshold.

[0027] Experimental Example I. Experimental Objective To verify the effectiveness of this system compared to existing technologies, comparative experiments were conducted on this system and a traditional wire harness connector condition monitoring solution under the same test conditions to evaluate the differences in performance of the two in key indicators such as fault detection accuracy, false alarm rate, early warning response delay, and remaining life prediction error.

[0028] II. Experiment Setup Sixty identical wiring harness connectors from the same vehicle model were selected as test subjects and randomly divided into three groups (A, B, and C), with 20 connectors in each group. The following three deterioration conditions were simulated: ① Oxidative deterioration group – accelerated oxidation under high temperature and humidity; ② Mechanical loosening group – gradual loosening of the connectors through cyclic vibration applied on a vibration table; ③ Electrochemical corrosion group – accelerated corrosion through a trace amount of salt spray applied to the contact interface. Within each group, three simultaneous monitoring schemes were used: Scheme 1: traditional single physical quantity (contact resistance) + static threshold early warning scheme; Scheme 2: multi-physical quantity (contact resistance + vibration + temperature rise) + static threshold early warning scheme; Scheme 3: the system scheme of this invention. The experiment ran continuously for 240 hours, and the early warning performance of each scheme under the three deterioration conditions was recorded. The evaluation metrics are defined as follows: Fault detection accuracy = (Number of correct warnings / Actual number of faults) × 100%; False alarm rate = (Number of false alarms / Total number of alarms) × 100%; Warning response delay is the time difference between the actual occurrence of the fault (based on the degradation start point confirmed by physical inspection) and the issuance of the first warning by the system; Remaining lifetime prediction error is the ratio of the absolute value of the deviation between the predicted remaining lifetime and the actual remaining lifetime to the actual remaining lifetime. III. Data Comparison

[0029] IV. Data Analysis The experimental data above show that, in terms of fault detection accuracy, the system of this invention achieves over 95% accuracy under all three adverse chemical conditions, an improvement of approximately 27-47 percentage points compared to Scheme 1 and approximately 15-30 percentage points compared to Scheme 2. Its advantage is most significant under electrochemical corrosion conditions, thanks to the introduction of online electrochemical impedance spectroscopy measurement technology, which can directly capture early changes in the electrochemical state of the contact interface. Regarding the false alarm rate, the system of this invention controls it below 8%, a reduction of approximately 22-34 percentage points compared to Scheme 1 and approximately 14-19 percentage points compared to Scheme 2. This verifies that the dynamic baseline learning module, through individualized baselines and dynamic threshold mechanisms... The system effectively suppresses false alarms due to operating conditions. Regarding early warning response delay, the system can detect potential faults in as little as 1.8 hours under mechanical loosening conditions and only 4.1 hours under electrochemical corrosion conditions, approximately 14-22 hours earlier than Solution 1. This is attributed to the predictive capability of the degradation trend prediction module and the bidirectional coupling inference mechanism of the digital twin simulation module. In terms of remaining life prediction, the system has an average prediction error of 8.7%, providing a reliable quantitative basis for proactive maintenance decisions. These experimental data fully demonstrate that the system of this invention is significantly superior to existing technologies in terms of detection accuracy, false alarm suppression, early warning timeliness, and life prediction capability.

[0030] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A wire harness-connected intelligent real-time early warning system based on real-time data, characterized in that: include: A multi-source sensing fusion module is connected to an adaptive preprocessing module, which in turn is connected to a health index construction module. The health index construction module is connected to a dynamic baseline learning module, which is connected to a decline trend prediction module. The decline trend prediction module is connected to a digital twin simulation module, which is connected to a topology sensing module. An edge inference module is connected to a topology sensing module, an edge cloud decision module is connected to an edge cloud decision module, a dynamic early warning module is connected to a dynamic early warning module, and an early warning suppression module is connected to an early warning suppression module. The root cause diagnosis module is connected to the early warning and suppression module, the root cause diagnosis module is connected to the lifetime prediction module, the lifetime prediction module is connected to the sampling control module, and the sampling control module is connected to the environmental decoupling module. The knowledge transfer module is connected to the environment decoupling module, the knowledge transfer module is connected to the security execution module, the security execution module is connected to the visual interaction module, the visual interaction module is connected to the maintenance decision module, and the maintenance decision module is connected to the closed-loop evolution module.

2. The intelligent real-time early warning system based on real-time data in the form of a wire harness as described in claim 1, characterized in that: The dynamic baseline learning module includes an OGD online update unit, a learning rate adaptive unit, a multi-condition baseline unit, and a dynamic threshold driving unit. The OGD online update unit is connected to the learning rate adaptive unit, the learning rate adaptive unit is connected to the multi-condition baseline unit, and the multi-condition baseline unit is connected to the dynamic threshold driving unit.

3. The intelligent real-time early warning system based on real-time data in the form of a wire harness as described in claim 2, characterized in that: The digital twin simulation module includes a three-coupled twin unit, a prediction-driven simulation unit, a simulation back-correction unit, and a UKF data assimilation unit. The three-coupled twin unit is connected to the prediction-driven simulation unit, and the prediction-driven simulation unit is connected to the UKF data assimilation unit.

4. The intelligent real-time early warning system based on real-time data in the form of a wire harness as described in claim 3, characterized in that: The topology sensing module includes a topology graph construction unit, a GAT risk propagation unit, a risk node marking unit, and an early warning and suppression coordination unit. The topology graph construction unit is connected to the GAT risk propagation unit, the GAT risk propagation unit is connected to the risk node marking unit, and the risk node marking unit is connected to the early warning and suppression coordination unit.

5. The intelligent real-time early warning system based on real-time data in the form of a wire harness as described in claim 4, characterized in that: The edge-cloud decision module includes a Bayesian fusion unit, a confidence evaluation unit, a consistency verification unit, and a hardware signature linkage unit. The Bayesian fusion unit is connected to the confidence evaluation unit, the confidence evaluation unit is connected to the consistency verification unit, and the consistency verification unit is connected to the hardware signature linkage unit. The confidence assessment unit employs a dynamic confidence assessment algorithm; Marginal confidence : ； in, Score the current signal-to-noise ratio quality; Uncertainty in the edge lightweight model; Scoring for consistency of historical reasoning; Recommendation weight ; Cloud-based side confidence : ； in, This represents the current communication delay. The time delay decay constant; This is the time difference between the last update of the cloud model; The model's freshness half-life; The standard deviation of the Monte Carlo Dropout forecast in the cloud; Recommendation weight , .

6. The intelligent real-time early warning system based on real-time data in the form of a wire harness as described in claim 5, characterized in that: The multi-source sensing fusion module is used to synchronously collect four types of physical parameters for each monitored connector in the wiring harness: contact resistance, micro-vibration acceleration, local temperature rise, and electrochemical impedance spectroscopy. The collected data is packaged and output to the adaptive preprocessing module according to a unified time reference. Among them, the electrochemical impedance spectroscopy is performed by applying multi-frequency sinusoidal excitation through a direct digital frequency synthesis chip and combining it with fast Fourier transform to complete the online real-time measurement of the entire frequency band within a single sampling period. The feature parameters used to characterize the oxide film thickness and corrosion degree of the contact interface are extracted through equivalent circuit fitting.

7. The intelligent real-time early warning system based on real-time data in the form of a wire harness as described in claim 6, characterized in that: The dynamic baseline learning module receives the four-dimensional health vector output by the health index construction module, maintains an individualized dynamic baseline model for each monitored connector, and drives the dynamic early warning module to update the warning threshold synchronously in real time using the baseline parameters. The dynamic baseline model updates the baseline parameters in real time in each sampling period using an online incremental gradient descent algorithm, and maintains an independent set of baseline parameters for different operating conditions. The warning thresholds at each level are dynamically calculated based on Mahalanobis distance. Whenever the baseline parameters are updated, the warning threshold is automatically updated synchronously, realizing the co-evolution of the warning threshold and the individual state of the connector.

8. The intelligent real-time early warning system based on real-time data in the form of a wire harness as described in claim 7, characterized in that: The digital twin simulation module receives the future health vector prediction value output by the degradation trend prediction module, drives the connector-level digital twin to perform accelerated time-step simulation, and simultaneously inputs the simulation results back into the degradation trend prediction module to correct its prediction confidence. It also achieves assimilation and update with the measured data through unscented Kalman filtering. The digital twin is composed of a geometric contact mechanics sub-model, a material aging sub-model, and an electrical characteristic sub-model coupled together. The three sub-models are coupled with each other through state vectors containing contact stress, oxide film thickness, corrosion current density, and contact resistance, which together characterize the global health state of the connector.

9. The intelligent real-time early warning system based on real-time data in the form of a wire harness as described in claim 8, characterized in that: The topology perception module is used to construct a wiring harness network topology map based on the vehicle wiring harness design data. It uses a graph attention network to perform risk propagation analysis on the dynamic health map, identify high-risk nodes and paths, and transmit the topology analysis results to the early warning propagation suppression module in real time to achieve early warning redundancy removal. The graph attention network aggregates neighborhood health information through multi-layer graph convolution and outputs a comprehensive risk score for each node. Safety critical functional domain nodes are assigned a risk weight coefficient of no less than three times that of ordinary nodes. Multiple early warnings within the same physical cluster are merged into a single associated early warning, and only the root cause node with the highest risk score is reported.

10. The intelligent real-time early warning system based on real-time data in the form of a wire harness as described in claim 9, characterized in that: The edge-cloud decision module is used to fuse the local inference results of the edge inference module and the deep analysis results of the cloud in a Bayesian credibility weighted manner. When the edge-cloud results are inconsistent, a consistency anomaly handling process is triggered, and the fused decision is output to the secure execution module for hardware signature before driving the dynamic early warning module. The weight of the Bayesian fusion is dynamically calculated based on signal quality, communication latency, data freshness, and model uncertainty. Before fusion, an edge-cloud consistency check is performed. When the divergence exceeds the threshold, root cause diagnosis is used to eliminate sensor faults and the sampling rate is increased to obtain verification data. After convergence, the fused decision is output. The fused decision is digitally signed by the hardware security element to generate a trusted decision package. The early warning can only be triggered after the downstream module's signature verification is passed.

Images