Vehicle-mounted CCS integrated busbar fault diagnosis method and system based on multi-source data fusion

Abstract

The invention discloses a vehicle-mounted CCS integrated busbar fault diagnosis method and system based on multi-source data fusion. The method comprises the steps that multi-source real-time data and working condition labels are collected; preprocessing the data and extracting a multi-dimensional fault sensitive feature vector; inputting the feature vector, the working condition label and the historical fault label into a model fusing CNN and LSTM, and determining a fault state and confidence in combination with a working condition weight; and the confidence coefficient is adjusted according to the environment temperature and the working duration, and graded early warning is executed. According to the method, the fault diagnosis accuracy and the early warning reliability are improved, the safety risk is effectively avoided, the maintenance cost is reduced, and the engineering application requirement of a vehicle-mounted scene is completely met.

Description

Technical Field

[0001] This application belongs to the field of fault diagnosis technology, specifically relating to a fault diagnosis method and system for an on-board CCS integrated busbar based on multi-source data fusion. Background Technology

[0002] As new energy vehicles iterate towards longer range, faster charging, and longer lifespan, the safety redundancy design and dynamic monitoring capabilities of power battery systems have become core competitive advantages in the industry. The onboard CCS integrated busbar, as the nerve center connecting the cell array, transmitting power current, and collecting status signals, directly determines the vehicle's power output and safety baseline through its operational stability. It not only bears the large current transmission from the cells to the electronic control system but also achieves precise acquisition of individual cell voltage and temperature via FPC / PCB, making it a key carrier for BMS (Battery Management System) to achieve overcurrent protection and thermal runaway early warning.

[0003] Unlike traditional discrete busbars, integrated busbars highly integrate copper and aluminum conductors, insulation encapsulation, and signal interfaces, operating in an extremely complex environment: they must withstand current surges and temperature fluctuations during charge and discharge cycles, resist strong electromagnetic interference from onboard motors and controllers, and adapt to mechanical stresses caused by vehicle vibrations. Their potential faults (such as increased contact resistance due to poor contact, conductor melting caused by overcurrent, and insulation aging and damage) often exhibit a "hidden development, sudden outbreak" characteristic, directly threatening the safety of passengers. Therefore, implementing "real-time monitoring, early warning, and precise location" fault diagnosis for CCS busbars has become a rigid requirement for the upgrading of safety technology in new energy vehicles.

[0004] Currently, fault diagnosis of vehicle-mounted CCS busbars mainly relies on the basic monitoring functions built into the BMS and post-incident repair methods. Specific solutions include single-parameter threshold monitoring and generalized intelligent diagnostic models. Both of these solutions have shortcomings, as detailed below: The single-parameter threshold monitoring method is the most commonly used diagnostic method in BMS. It achieves fault judgment by setting fixed thresholds for current, voltage, and temperature (such as triggering an overcurrent alarm when the current is >600A). The core defects are: (1) It relies only on a single physical quantity threshold and cannot associate multi-parameter coupling characteristics. For example, "poor contact" will simultaneously cause local temperature rise, voltage drop increase, and current fluctuation. Monitoring only temperature is prone to misjudgment due to differences in heat dissipation conditions, and monitoring only voltage is prone to interference from cell consistency fluctuations; (2) Fixed thresholds cannot match different operating conditions; (3) For latent faults such as early poor contact and micro-damage to insulation layer, the change of a single parameter does not reach the threshold and cannot be effectively identified. Such faults are the source of subsequent safety accidents.

[0005] The generalized intelligent diagnostic model uses a single CNN and LSTM model to diagnose busbar data, but it is not optimized for vehicle operating conditions. In practical applications, there are obvious bottlenecks: (1) The model training does not distinguish between operating conditions such as starting, acceleration, and charging. The fault characteristics under different operating conditions are significantly different, resulting in a sharp drop in cross-operating condition diagnostic accuracy; (2) The original electrical signal is directly input into the model without special processing for vehicle electromagnetic interference. The high-frequency noise generated by the motor operation will mask the fault-sensitive signal, causing the model to misjudge the interference fluctuation as a fault feature; (3) Only simple time-domain features (such as the mean current) are extracted, without integrating frequency-domain features (such as current harmonic distortion) and time-series features (such as voltage change slope), which cannot construct a complete fault feature profile and makes it difficult to identify early weak faults.

[0006] Therefore, current technologies are insufficient to meet the comprehensive requirements of full-condition coverage, early fault identification, and real-time accurate diagnosis. Summary of the Invention

[0007] To address the aforementioned problems in the existing technology, this application provides a method and system for fault diagnosis of an integrated busbar in a vehicle-mounted CCS based on multi-source data fusion. The technical problem to be solved by this application is achieved through the following technical solution: Firstly, this application provides a fault diagnosis method for an in-vehicle CCS integrated busbar based on multi-source data fusion, including: The sensor array is used to collect multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar and simultaneously record various operating condition tags. Multi-source real-time data is preprocessed to obtain preprocessed real-time data, and multi-dimensional fault-sensitive feature vectors are extracted from the preprocessed real-time data. Multi-dimensional fault-sensitive feature vectors, multiple working condition labels, and historical fault labels are input into a fault diagnosis model that integrates CNN and LSTM to obtain multiple fault probability distributions of the CCS integrated busbar. Based on the weights corresponding to each working condition and multiple fault probability distributions, the final fault probability distribution is determined. Based on the final fault probability distribution, the fault state and corresponding confidence level of the CCS integrated busbar are determined. The confidence level is dynamically adjusted based on the ambient temperature and operating time of the CCS integrated busbar, and graded early warnings are implemented based on the fault status of the CCS integrated busbar.

[0008] Optionally, before utilizing a sensor array to collect multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar and simultaneously record various operating condition tags, the vehicle-mounted CCS integrated busbar fault diagnosis method based on multi-source data fusion also includes: At the copper-aluminum busbar electrical connection node in the CCS integrated busbar, a Hall current sensor, a first differential voltage sensor, and a first patch-type temperature sensor are deployed. A second differential voltage sensor and a second surface-mount temperature sensor are deployed at the FPC / PCB signal acquisition component interface in the CCS integrated busbar. A third patch-type temperature sensor is deployed at the critical stress point of the plastic structural component in the CCS integrated busbar. The Hall current sensor, the first differential voltage sensor, the first surface mount temperature sensor, the second differential voltage sensor, the second surface mount temperature sensor, and the third surface mount temperature sensor are calibrated and clock-synchronized. Then, the output terminals are connected to the high-speed data acquisition module, and communication is established with the vehicle battery management system through the bus interface of the high-speed data acquisition module.

[0009] Optionally, a sensor array can be used to collect multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar and simultaneously record various operating condition tags, including: The Hall current sensor, the first differential voltage sensor, the first patch temperature sensor, the second differential voltage sensor, the second patch temperature sensor, and the third patch temperature sensor are triggered to collect data at a predetermined sampling frequency. The results are obtained as follows: the busbar operating current collected by the Hall current sensor, the cell voltage collected by the first differential voltage sensor, the busbar surface temperature collected by the first patch temperature sensor, the signal line voltage collected by the second differential voltage sensor, the interface temperature collected by the second patch temperature sensor, and the structural component surface temperature collected by the third patch temperature sensor. Simultaneously with temperature collection, vehicle operating condition labels are collected from the on-board system; The busbar operating current, individual cell voltage, busbar surface temperature, signal line voltage, interface temperature, and structural component surface temperature are defined as multi-source real-time data.

[0010] Optionally, preprocessing of multi-source real-time data yields preprocessed real-time data, and multi-dimensional fault-sensitive feature vectors are extracted from the preprocessed real-time data, including: Preprocessed real-time data is obtained by performing noise filtering, missing value imputation, and outlier correction on multi-source real-time data. Time-domain features, frequency-domain features, and time-series features are extracted from preprocessed real-time data and concatenated into a multi-dimensional fault-sensitive feature vector. The time-domain features include current features, voltage features, and temperature features; the frequency-domain features include spectrum features, main frequency band energy ratio, and harmonic distortion rate; and the time-series features include current jump slope, voltage continuous fluctuation duration, and Pearson correlation coefficient of temperature change trend.

[0011] Optionally, the fault diagnosis model integrating CNN and LSTM includes fault diagnosis sub-models integrating CNN and LSTM corresponding to different working conditions and a decision layer fusion sub-module. Each fault diagnosis sub-model includes a feature layer fusion module and a CNN-LSTM hybrid diagnosis module connected in sequence. The CNN-LSTM hybrid diagnosis module includes a CNN sub-module, two LSTM sub-modules, and a fault classification head connected in sequence. The feature layer fusion module concatenates the input vectors in different dimensions to form initial high-dimensional features, and then reduces and fuses these initial high-dimensional features to obtain a fused feature matrix. The CNN-LSTM hybrid diagnosis module reshapes the fused feature matrix into a two-dimensional feature map through the CNN sub-module, and then captures the temporal features of the two-dimensional feature map through the two LSTM sub-modules. The fault classification head predicts the fault probability distribution of the CCS integrated busbar based on historical fault labels and the captured features. The decision layer fusion sub-module assigns weights to each fault diagnosis sub-model according to different working conditions, and calculates the fault state and initial confidence of the CCS integrated busbar based on the weights and the fault probability distribution predicted by each fault diagnosis sub-model.

[0012] Optionally, the decision-level fusion submodule is used to assign weights to each fault diagnosis sub-model according to different operating conditions, including: By training corresponding fault diagnosis sub-models that fuse CNN and LSTM using historical multi-source data under a single working condition, and assigning weights to each fault diagnosis sub-model based on real-time data under different working conditions, the weights corresponding to each fault diagnosis sub-model are obtained.

[0013] Optionally, based on the weights and the fault probability distribution of the CCS integrated busbar predicted by each fault diagnosis sub-model, the fault state and initial confidence level of the CCS integrated busbar are calculated, including: Each item in the fault probability distribution of the CCS integrated busbar predicted by each fault diagnosis sub-model is multiplied by its corresponding weight, and then summed to obtain the final fault probability distribution of the CCS integrated busbar. The fault state with the highest probability is selected as the fault state of the CCS integrated busbar and its corresponding fault probability is used as the initial confidence level.

[0014] Optionally, multi-dimensional fault-sensitive feature vectors, multiple operating condition labels, and historical fault labels are input into a fault diagnosis model that integrates CNN and LSTM to obtain multiple fault probability distributions of the CCS integrated busbar. Based on the weights corresponding to each operating condition and the multiple fault probability distributions, the final fault probability distribution is determined. Based on the final fault probability distribution, the fault state of the CCS integrated busbar and the corresponding confidence level are determined, including: Multi-dimensional fault-sensitive feature vectors, multiple working condition labels, and historical fault labels are input into a fault diagnosis model that integrates CNN and LSTM. The feature layer fusion module in the fault diagnosis sub-model corresponding to each working condition concatenates the input multi-dimensional fault-sensitive feature vectors in terms of dimensions to form initial high-dimensional features. Then, the initial high-dimensional features are reduced in dimension and fused to obtain a fused feature matrix. The CNN-LSTM hybrid diagnosis module in the fault diagnosis sub-model corresponding to each working condition is used to reshape the fused feature matrix into a two-dimensional feature map, and then the two-dimensional feature map is used to capture temporal features through two LSTM sub-modules. Using a fault classification head, it can predict the fault probability distribution of the CCS integrated busbar based on the captured features, historical fault labels, and operating condition labels. The decision layer fusion submodule is used to assign weights to each fault diagnosis sub-model according to different working conditions. Based on the weights and the fault probability distribution of the CCS integrated busbar predicted by each fault diagnosis sub-model, the fault state and initial confidence of the CCS integrated busbar are calculated. The fault diagnosis model is validated using a validation set to calculate the classification accuracy. The classification accuracy is then multiplied by the initial confidence level to obtain the confidence level corresponding to the fault state of the CCS integrated busbar.

[0015] Optionally, the confidence level can be dynamically adjusted based on the ambient temperature and operating duration of the CCS integrated busbar, and graded early warnings can be implemented based on the fault status of the CCS integrated busbar, including: Determine whether the ambient temperature and / or operating time of the CCS integrated busbar are greater than a predetermined threshold. If so, increase the confidence level of the fault state of the CCS integrated busbar by the corresponding step size to dynamically adjust the confidence level. A tiered early warning system is implemented based on the dynamically adjusted confidence level and the fault status of the CCS integrated busbar.

[0016] Secondly, this application provides an on-board CCS integrated busbar fault diagnosis system based on multi-source data fusion, comprising: The acquisition module is configured to use a sensor array to acquire multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar and simultaneously record various operating condition tags. The processing module is configured to preprocess multi-source real-time data to obtain preprocessed real-time data, and extract multi-dimensional fault-sensitive feature vectors from the preprocessed real-time data. The prediction module is configured to input multi-dimensional fault-sensitive feature vectors, multiple working condition labels, and historical fault labels into a fault diagnosis model that integrates CNN and LSTM to obtain multiple fault probability distributions of the CCS integrated busbar. Based on the weights corresponding to each working condition and the multiple fault probability distributions, the final fault probability distribution is determined, and based on the final fault probability distribution, the fault state of the CCS integrated busbar and the corresponding confidence level are determined. The grading module is configured to dynamically adjust the confidence level based on the ambient temperature and operating time of the CCS integrated busbar, and to execute grading early warnings based on the fault status of the CCS integrated busbar.

[0017] Beneficial effects: This application provides a fault diagnosis method and system for vehicle-mounted CCS integrated busbars based on multi-source data fusion. The method includes using a sensor array to collect multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar and simultaneously recording various operating condition labels, overcoming the limitation of traditional single-threshold monitoring which can only identify visible faults. The multi-source real-time data is preprocessed to obtain preprocessed real-time data, and multi-dimensional fault-sensitive feature vectors are extracted from the preprocessed real-time data to capture multi-parameter coupling features. The multi-dimensional fault-sensitive feature vectors, various operating condition labels, and historical fault labels are input into a fusion CNN and... In the LSTM fault diagnosis model, multiple fault probability distributions of the CCS integrated busbar are obtained. Based on the weights corresponding to each operating condition and the multiple fault probability distributions, the final fault probability distribution is determined. Based on the final fault probability distribution, the fault state of the CCS integrated busbar and the corresponding confidence level are determined. The confidence level is dynamically adjusted in combination with the ambient temperature and working duration of the CCS integrated busbar. The graded early warning is executed in combination with the fault state of the CCS integrated busbar, realizing the upgrade from fixed threshold judgment to operating condition adaptive decision-making. The accuracy of full-scenario diagnosis is greatly improved, truly adapting to the complex and ever-changing operating scenarios of vehicles.

[0018] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for diagnosing faults in an onboard CCS integrated busbar based on multi-source data fusion, as provided in this application. Figure 2 This is a schematic diagram of the fault diagnosis model provided in this application; Figure 3 This is a schematic diagram of the vehicle-mounted CCS integrated busbar fault diagnosis system based on multi-source data fusion provided in this application. Detailed Implementation

[0020] The present application will be described in further detail below with reference to specific embodiments, but the implementation of the present application is not limited thereto.

[0021] like Figure 1 As shown, this application provides a fault diagnosis method for an in-vehicle CCS integrated busbar based on multi-source data fusion, including: The S100 uses a sensor array to collect multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar and simultaneously record various working condition tags. S200 preprocesses multi-source real-time data to obtain preprocessed real-time data, and extracts multi-dimensional fault-sensitive feature vectors from the preprocessed real-time data. S300 inputs multi-dimensional fault-sensitive feature vectors, multiple working condition labels, and historical fault labels into a fault diagnosis model that integrates CNN and LSTM to obtain multiple fault probability distributions of the CCS integrated busbar. Based on the weights corresponding to each working condition and multiple fault probability distributions, the final fault probability distribution is determined. Based on the final fault probability distribution, the fault status and corresponding confidence level of the CCS integrated busbar are determined. The S400 dynamically adjusts the confidence level based on the ambient temperature and operating time of the CCS integrated busbar, and performs graded early warning based on the fault status of the CCS integrated busbar.

[0022] This application provides a fault diagnosis method for vehicle-mounted CCS integrated busbars based on multi-source data fusion. The method includes using a sensor array to collect multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar and simultaneously recording various operating condition labels, overcoming the limitation of traditional single-threshold monitoring which can only identify explicit faults. The multi-source real-time data is preprocessed to obtain preprocessed real-time data, and multi-dimensional fault-sensitive feature vectors are extracted from the preprocessed real-time data to capture multi-parameter coupling features. The multi-dimensional fault-sensitive feature vectors, various operating condition labels, and historical fault labels are input into a fusion CNN and L... In the STM fault diagnosis model, multiple fault probability distributions of the CCS integrated busbar are obtained. Based on the weights corresponding to each operating condition and the multiple fault probability distributions, the final fault probability distribution is determined. Based on the final fault probability distribution, the fault state of the CCS integrated busbar and the corresponding confidence level are determined. The confidence level is dynamically adjusted in combination with the ambient temperature and working duration of the CCS integrated busbar. The graded early warning is executed in combination with the fault state of the CCS integrated busbar, realizing the upgrade from fixed threshold judgment to operating condition adaptive decision-making. The accuracy of full-scenario diagnosis is greatly improved, truly adapting to the complex and ever-changing operating scenarios of vehicles.

[0023] In one specific embodiment of this application, prior to S100, the on-board CCS integrated busbar fault diagnosis method based on multi-source data fusion further includes: S001, at the copper-aluminum busbar electrical connection node in the CCS integrated busbar, a Hall current sensor, a first differential voltage sensor and a first patch temperature sensor are deployed. The copper-aluminum busbar connection node is a key monitoring point. Since this location is a critical node for current transmission, the core risks are poor contact (abnormal voltage drop), overcurrent (excessive current), and overheating (sudden temperature rise). All three parameters need to be monitored together.

[0024] S002, at the FPC / PCB signal acquisition component interface in the CCS integrated busbar, a second differential voltage sensor and a second surface-mount temperature sensor are deployed; The core risks of CCS integrated busbars are excessive interface contact resistance (abnormal voltage signal) and localized overheating. No current monitoring is required, as the FPC / PCB signal acquisition component interface is for signal transmission with minimal current.

[0025] S003, at the critical stress point of the plastic structural component in the CCS integrated busbar, a third patch-type temperature sensor is deployed; At the critical stress point of the plastic structural component, the structural deformation compresses the busbar, causing indirect heating. There is no direct electrical connection, so there is no need for current or voltage sensors; only a temperature sensor is needed to monitor the temperature.

[0026] S004 calibrates and clocks the Hall current sensor, the first differential voltage sensor, the first surface mount temperature sensor, the second differential voltage sensor, the second surface mount temperature sensor, and the third surface mount temperature sensor, and then connects the output to the high-speed data acquisition module to establish communication with the vehicle battery management system through the bus interface of the high-speed data acquisition module.

[0027] This application connects all sensor outputs to a high-speed data acquisition module, establishes communication with the vehicle battery management system via a bus interface, and realizes the exchange of operating condition information. Then, it performs range zero-point calibration on each sensor and records the calibration coefficients; it also synchronously configures a clock synchronization mechanism for the acquisition module to ensure that the timestamp error of current, voltage, and temperature data is less than a threshold, such as ≤1ms, and establishes a unified acquisition time reference.

[0028] In one specific embodiment of this application, S100 includes: S110, triggers the Hall current sensor, the first differential voltage sensor, the first patch temperature sensor, the second differential voltage sensor, the second patch temperature sensor, and the third patch temperature sensor to collect data at a predetermined sampling frequency, and obtains the busbar operating current collected by the Hall current sensor, the cell voltage collected by the first differential voltage sensor, the busbar surface temperature collected by the first patch temperature sensor, the signal line voltage collected by the second differential voltage sensor, the interface temperature collected by the second patch temperature sensor, and the structural component surface temperature collected by the third patch temperature sensor; This application allows for a preset sampling frequency, such as 1kHz, to synchronously trigger sensors to collect data such as busbar operating current, cell voltage, and busbar surface temperature.

[0029] S120, synchronized with S110, collects vehicle condition tags from the on-board system; This application covers typical operating conditions of vehicles, such as starting, accelerating, constant speed, fast charging, and slow charging, and records the vehicle operating condition label at the time of collection.

[0030] S130 defines the busbar operating current, individual cell voltage, busbar surface temperature, signal line voltage, interface temperature, and structural component surface temperature as multi-source real-time data. This application can temporarily store raw multi-source real-time data and operating condition labels in the cache area of ​​the vehicle edge computing unit to ensure that no data is lost.

[0031] In one specific embodiment of this application, S200 includes: S210, performs noise filtering, missing value imputation and outlier correction on multi-source real-time data to obtain preprocessed real-time data; This application performs standardized preprocessing on the raw real-time multi-source data stored in the cache. The specific process is as follows: (1) Noise filtering: Wavelet transform is used to decompose the current and voltage data into three levels to filter out high-frequency noise from vehicle electromagnetic interference, resulting in denoised data: For example, the Haar wavelet basis function is used to perform three-level wavelet decomposition on the raw current and voltage data. After decomposition, one set of low-frequency approximation coefficients and three sets of high-frequency detail coefficients are obtained. The three sets of high-frequency detail coefficients are thresholded using a default threshold rule, and the thresholds of the remaining coefficients are set to 0, otherwise they remain unchanged. Finally, the unprocessed low-frequency approximation coefficients and the corrected high-frequency detail coefficients are reconstructed through inverse wavelet transform to obtain denoised data, which provides a high-quality data foundation for subsequent fault-sensitive feature extraction.

[0032] (2) Missing value imputation: Missing data values ​​are imputed based on the values ​​of five consecutive sampling points before and after using linear interpolation. (3) Outlier correction: A sliding window method is used to determine the absolute deviation of the median. The specific steps are as follows: A. Set the sliding window size to 50 sampling points; B. Calculate the median M and absolute deviation of the data within each window. Then, the median of the absolute deviations is calculated as MAD; C. Set an anomaly detection threshold, and obtain the absolute deviation threshold based on MAD and the preset multiplier coefficient coef. ,when hour, This is an outlier.

[0033] D. Outlier Replacement: For identified outliers, a linear weighted interpolation of the first three valid data points and the last three valid data points within the window is used for replacement, as shown in the following formula: Indicates outlier The corrected value after replacement , , Indicates outlier sampling points The previous three consecutive valid data (forward valid data within the window). , , These represent outlier sampling points. The next 3 consecutive valid data (valid data in the window backwards). , , , This represents the weighting coefficient for the corresponding valid data, set according to the principle of "the closer to the outlier, the greater the weight," specifically: .

[0034] (4) Data standardization: using the Z-score formula Multidimensional data is mapped to a uniform dimension interval with a mean of 0 and a standard deviation of 1. Indicates the global standard deviation. The mean, This is the original data.

[0035] This application integrates a full-process data optimization scheme that combines Haar wavelet denoising, sliding window-MAD outlier correction, and Z-score standardization. It effectively filters out vehicle electromagnetic interference and noise signals caused by vibration, corrects occasional sensor outliers, unifies multi-dimensional data units, and achieves optimization from coarse data input to fine preprocessing. The anti-interference capability and feature effectiveness are significantly enhanced.

[0036] S220 extracts time-domain features, frequency-domain features, and time-series features from preprocessed real-time data, and concatenates them into a multi-dimensional fault-sensitive feature vector. The time-domain features include current features, voltage features, and temperature features; the frequency-domain features include spectrum features, main frequency band energy ratio, and harmonic distortion rate; and the time-series features include current jump slope, voltage continuous fluctuation duration, and Pearson correlation coefficient of temperature change trend.

[0037] This application extracts multi-dimensional fault-sensitive features from preprocessed real-time data to form a vector. These multi-dimensional fault-sensitive features include: time-domain features, frequency-domain features, and time-series features. Time-domain features include current features, voltage features, and temperature features. The current features are the mean, peak value, valley value, variance, and kurtosis within the sampling period; the voltage features are the maximum and minimum values ​​of the cell voltage difference and the coefficient of variation; the temperature features are the maximum busbar temperature, temperature difference, and heating rate. The frequency-domain features are the peak values, main frequency band energy proportion, and harmonic distortion rate extracted from the current and voltage data using Fast Fourier Transform (FFT); the time-series features are the current jump slope, the duration of continuous voltage fluctuations, and the Pearson correlation coefficient of temperature change trends. This application concatenates these features into a 30-dimensional feature vector, which serves as the input basis for the fault diagnosis model. For example: time domain (12 dimensions) + frequency domain (10 dimensions) + time-series (8 dimensions) = 30 "1-dimensional features".

[0038] This application uses the obtained 30-dimensional fault-sensitive feature vector, the corresponding vehicle operating condition label, and the historical fault labels of the CCS busbar ("normal / poor contact / overcurrent damage / insulation failure") as network input. The vehicle operating condition labels are shown in the table below: This application employs a differentiated deployment of Hall current sensors, differential voltage sensors, and patch-type temperature sensors. Multi-source real-time data is collected through this differentiated sensor array. The multi-source real-time data is preprocessed to extract a multi-dimensional fault-sensitive feature vector that integrates time domain, frequency domain, and time sequence data. This multi-dimensional fault-sensitive feature vector can construct a complete fault feature profile, overcoming the limitation of traditional single-threshold monitoring which can only identify explicit faults. It can capture the multi-parameter coupling characteristics of early latent faults such as poor contact and micro-damage to the insulation layer, such as the combined effects of increased contact resistance leading to voltage drop, temperature rise, and current fluctuations, thus mitigating safety risks at the source.

[0039] refer to Figure 2As shown, the fault diagnosis model integrating CNN and LSTM in this application includes fault diagnosis sub-models integrating CNN and LSTM corresponding to different working conditions and a decision layer fusion sub-module. Each fault diagnosis sub-model includes a feature layer fusion module and a CNN-LSTM hybrid diagnosis module connected in sequence. The CNN-LSTM hybrid diagnosis module includes a CNN sub-module, two LSTM sub-modules and a fault classification head connected in sequence. The feature layer fusion module is used to concatenate the input vectors in terms of dimensions to form initial high-dimensional features, and then reduce the dimensions of the initial high-dimensional features to obtain a fused feature matrix. The CNN-LSTM hybrid diagnosis module is used to reshape the fused feature matrix into a two-dimensional feature map through the CNN sub-module, and then capture the temporal features of the two-dimensional feature map through the two LSTM sub-modules. The fault classification head predicts the fault probability distribution of the CCS integrated bus based on historical fault labels and captured features. The decision layer fusion sub-module is used to assign weights to each fault diagnosis sub-model according to different working conditions, and calculate the fault state and initial confidence of the CCS integrated bus based on the weights and the fault probability distribution of the CCS integrated bus predicted by each fault diagnosis sub-model.

[0040] Among them, the historical fault labels are derived from the fault status records in the historical operation data of the busbar, which have been calibrated by experts or experiments.

[0041] This application addresses the differences in busbar load characteristics under various operating conditions such as start-up, acceleration, charging, and stationary conditions. It constructs a fault diagnosis sub-model specific to each operating condition and dynamically weights and fuses them, completely solving the pain points of poor adaptability of fixed thresholds and high misjudgment rate across operating conditions in traditional methods. It achieves an upgrade from fixed threshold judgment to operating condition adaptive decision-making, significantly improving the accuracy of full-scenario diagnosis and truly adapting to the complex and ever-changing operating scenarios of vehicles.

[0042] The main functions of the feature layer fusion module in this application include feature concatenation and dimensionality reduction fusion. Feature concatenation involves concatenating time-domain, frequency-domain, and time-series feature vectors in terms of dimensions to form initial high-dimensional features. Dimensionality reduction fusion uses principal component analysis (PCA) to reduce the dimensionality to a lower dimension (e.g., 256 dimensions, resulting in a fusion feature matrix of shape T×256, where T is the length of the time series, i.e., T sampling times).

[0043] The CNN-LSTM hybrid diagnostic module of this application mainly performs spatial feature extraction (CNN submodule), temporal feature capture (LSTM submodule), and fault classification. Spatial feature extraction is implemented by the CNN submodule, whose input is a T×256×1 two-dimensional feature map reconstructed from the fused feature matrix. Its structure is as follows: The first layer uses a Conv2D convolutional layer (configured with 64 3×3 kernels, using "same" padding to ensure the output feature map size matches the input). Then, it uses a ReLU activation function to introduce non-linear features. Finally, it uses a BatchNorm (batch normalization) layer to standardize the features, improving training stability. The second layer follows immediately after the first, again using a Conv2D convolutional layer (with identical parameters: 64 3×3 kernels, "same" padding). It then sequentially uses a ReLU activation function, a BatchNorm layer, and finally a MaxPool2D max pooling layer (using 2×2 kernels) to downsample the feature map, preserving key features and reducing computation. The third layer, after the second pooling layer, uses a Conv2D convolutional layer (configured with 128 3×3 kernels, "same" padding to increase the number of feature channels and capture more complex features). It then uses a ReLU activation function and a BatchNorm layer for standardization. The fourth layer follows the third layer, continuing to use the Conv2D convolutional layer (with the same parameters as the third layer: 128 3×3 convolutional kernels, "same" padding). Then, it passes through the ReLU activation function, the BatchNorm batch normalization layer, and finally the MaxPool2D max pooling layer (2×2 pooling kernel) to complete the second downsampling, outputting the final spatial feature map.

[0044] The temporal feature capture process is implemented by the LSTM submodule, which flattens the CNN output into a T×128 temporal vector and inputs it into LSTM(64, return_sequences=True)→ReLU and LSTM(32, return_sequences=False)→ReLU.

[0045] Fault classification is implemented by the fault classification header, and its implementation process is Dense(16)→ReLU→Dense(4)→Softmax. The four working conditions correspond to four states.

[0046] The decision-level fusion submodule trains corresponding fault diagnosis sub-models that fuse CNN and LSTM using historical multi-source data under a single working condition. It assigns weights to each fault diagnosis sub-model based on real-time data under different working conditions, thus obtaining the weights corresponding to each fault diagnosis sub-model. Each item in the fault probability distribution of the CCS integrated busbar predicted by each fault diagnosis sub-model is multiplied by its corresponding weight, and then summed to obtain the final fault probability distribution of the CCS integrated busbar. The fault state with the highest probability is selected as the fault state of the CCS integrated busbar and its corresponding fault probability as the initial confidence level.

[0047] This application can train four CNN-LSTM sub-models based on startup / acceleration / charging / stationary operating conditions; for example: sub-model M1: adapts to startup conditions (cold / normal / hot startup), focusing on identifying poor contact during startup; sub-model M2: adapts to acceleration conditions (high current surge), focusing on identifying overcurrent damage during acceleration; sub-model M3: adapts to charging conditions (slow charging / fast charging), focusing on identifying overcurrent damage and insulation failure during charging; sub-model M4: adapts to stationary conditions (no load), focusing on identifying insulation failure during stationary operation.

[0048] The decision-making fusion submodule of this application assigns weights based on the matching degree of the current working condition and obtains the final fault probability distribution by weighting. Specifically, it includes: determining the working condition and assigning weights, calculating the final fault probability distribution by weighting, and fusing all fault probability distributions.

[0049] In determining the operating condition and allocating weights, this application uses real-time data (charging power = 50kW, vehicle speed = 0km / h) to determine that the current operating condition is charging. Following the rule that "the current operating condition sub-model has the highest weight, and other sub-models are weighted according to their correlation with the operating condition," a simple example allocation (total weight = 1) is performed: Charging sub-model M3 (current operating condition): weight 0.6 (core dominant); Acceleration sub-model M2 (high current correlation): weight 0.2 (auxiliary supplement); Start-up sub-model M1 (low correlation): weight 0.1 (basic reference); Static sub-model M4 (no correlation): weight 0.1 (interference suppression).

[0050] At the same time, the diagnostic outputs of the four fault diagnosis sub-models on the busbar status (probability distribution: [normal, poor contact, overcurrent damage, insulation failure]) are simplified as follows: M1 (start-up) output: [0.1, 0.7, 0.1, 0.1] (misjudged as poor contact); M2 (acceleration) output: [0.1, 0.1, 0.7, 0.1] (judged as overcurrent damage); M3 (charging) output: [0.05, 0.05, 0.8, 0.1] (judged as overcurrent damage); M4 (stationary) output: [0.8, 0.05, 0.05, 0.1] (misjudged as normal).

[0051] This application sums the results item by item according to the formula "output of each sub-model × corresponding weight" to obtain the final result: (1) Normal probability: (0.1×0.1) + (0.1×0.2) + (0.05×0.6) + (0.8×0.1) = 0.01 + 0.02 + 0.03 + 0.08 = 0.14; (2) Probability of poor contact: (0.7×0.1) + (0.1×0.2) + (0.05×0.6) + (0.05×0.1) = 0.07 + 0.02 + 0.03 + 0.005 = 0.125; (3) Probability of overcurrent damage: (0.1×0.1) + (0.7×0.2) + (0.8×0.6) + (0.05×0.1) =0.01 + 0.14 + 0.48 + 0.005 = 0.635; (4) Insulation failure probability: (0.1×0.1) + (0.1×0.2) + (0.1×0.6) + (0.1×0.1) =0.01 + 0.02 + 0.06 + 0.01 = 0.1.

[0052] The final failure probability distribution of this application is [0.14, 0.125, 0.635, 0.1]. The maximum probability (0.635) corresponds to overcurrent damage, which means that after fusion, it is accurately determined that the CCS integrated busbar is in this state.

[0053] In one specific embodiment of this application, S300 includes: S310 inputs multi-dimensional fault-sensitive feature vectors, multiple working condition labels, and historical fault labels into a fault diagnosis model that integrates CNN and LSTM. The feature layer fusion module in the fault diagnosis sub-model corresponding to each working condition concatenates the input multi-dimensional fault-sensitive feature vectors in terms of dimensions to form initial high-dimensional features. Then, the initial high-dimensional features are reduced in dimension and fused to obtain a fused feature matrix. S320 uses the CNN-LSTM hybrid diagnosis module in the fault diagnosis sub-model corresponding to each working condition to reshape the fused feature matrix into a two-dimensional feature map. Then, the two-dimensional feature map is used to capture temporal features through two LSTM sub-modules. Finally, the fault probability distribution of the CCS integrated busbar is predicted through the fault classification head and the working condition label. The S330 uses a fault classification head to predict the fault probability distribution of the CCS integrated busbar based on captured features, historical fault labels, and operating condition labels. S340 utilizes the decision-level fusion submodule to assign weights to each fault diagnosis sub-model according to different working conditions, and calculates the fault state and initial confidence level of the CCS integrated busbar based on the weights and the fault probability distribution predicted by each fault diagnosis sub-model. S350 uses the validation set to validate the fault diagnosis model to calculate the classification accuracy, and multiplies the classification accuracy by the initial confidence level to obtain the confidence level corresponding to the fault state of the CCS integrated busbar.

[0054] This application outputs the probability distribution of four states, for example, [0.02, 0.88, 0.05, 0.05], corresponding to poor contact, and also outputs the initial confidence level corrected by the validation set accuracy. The correspondence between the fault probability distribution array index and the CCS busbar state is: [Index 0: Normal, Index 1: Poor contact, Index 2: Overcurrent damage, Index 3: Insulation failure].

[0055] Example 1: Normal state output fault probability distribution: [0.96, 0.02, 0.01, 0.01]. Interpretation: The model determines that the busbar is normal with a probability of 96%, poor contact with a probability of 2%, overcurrent damage with a probability of 1%, and insulation failure with a probability of 1%. The highest probability corresponds to index 0 (normal state). Validation set accuracy correction logic: Assume that the model's classification accuracy for the normal state on the validation set is 95% (i.e., the proportion of historical data in which this state is correctly classified). Initial confidence: 0.96 (maximum probability) × 95% (corresponding state classification accuracy) = 0.912 (91.2%). The final output result is the fault probability distribution [0.96, 0.02, 0.01, 0.01] + initial confidence 91.2% (classified as normal state, high reliability).

[0056] Example 2: Fault probability distribution for poor contact status: [0.03, 0.89, 0.05, 0.03]. Interpretation: The model determines the probability of poor contact as 89%, with all other statuses having probabilities below 10%. The highest probability corresponds to index 1 (poor contact). Validation accuracy correction logic: The model's classification accuracy for poor contact is 93%. Initial confidence level: 0.89 × 93% = 0.8277 (82.77%). The final output is the probability distribution [0.03, 0.89, 0.05, 0.03] + initial confidence level 82.77% (classified as poor contact, high reliability).

[0057] This application not only outputs fault types and probability distributions, but also calculates the confidence level corresponding to the fault state of the CCS integrated busbar by multiplying the maximum probability by the corresponding operating condition classification accuracy. This provides a quantitative basis for tiered early warning, avoiding user panic or neglect caused by a one-size-fits-all approach to early warning. Simultaneously, it provides maintenance personnel with a fault reliability reference, significantly reducing unnecessary maintenance costs and achieving closed-loop management from accurate early warning to efficient maintenance, fully meeting the engineering application requirements of automotive scenarios. This application achieves a complete transformation from single result output to confidence level-supported decision-making, significantly improving early warning reliability and engineering practicality.

[0058] In one specific embodiment of this application, S400 includes: S410, determine whether the ambient temperature and / or working time of the CCS integrated busbar are greater than a predetermined threshold. If so, increase the confidence level of the fault state of the CCS integrated busbar by the corresponding step size on the original basis to dynamically adjust the confidence level. This application dynamically adjusts the confidence level by incorporating ambient temperature and busbar operating time (e.g., when the temperature is >40℃, the confidence level for thermal failure is increased by 0.1).

[0059] S420 performs graded early warning based on the dynamically adjusted confidence level and the fault status of the CCS integrated busbar.

[0060] The tiered early warning decision rules of this application are as follows: (1) Severe warning (safety mode), the fault status is insulation failure, confidence level > 0.85 and temperature difference > 5℃; the judgment result triggers high voltage power outage, the instrument panel color is red warning, and the voice broadcast "Please stop immediately"; (2) Moderate warning (maintenance reminder): The fault condition is overcurrent damage, with a confidence level of >0.75 and a peak current value >1.2 times the rated value; The judgment result is a yellow warning displayed on the instrument panel, and a maintenance reminder is pushed to the vehicle owner's terminal; (3) Mild warning (continuous monitoring): The fault status is poor contact, confidence level > 0.7 and voltage difference > 0.2V; the judgment result is indicated by a blue warning on the instrument panel, and the system collects data at high frequency; (4) Normal state: The fault state is normal, with a confidence level of >0.9; the judgment result has no warning, and the operation data is recorded.

[0061] Secondly, such as Figure 3 As shown, this application provides an in-vehicle CCS integrated busbar fault diagnosis system based on multi-source data fusion, comprising: The acquisition module 301 is configured to use a sensor array to acquire multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar and simultaneously record various working condition tags. Processing module 302 is configured to preprocess multi-source real-time data to obtain preprocessed real-time data, and extract multi-dimensional fault-sensitive feature vectors from the preprocessed real-time data; The prediction module 303 is configured to input multi-dimensional fault-sensitive feature vectors into a fault diagnosis model that integrates CNN and LSTM to obtain multiple fault probability distributions of the CCS integrated busbar, and determine the final fault probability distribution based on the weights corresponding to each working condition and the multiple fault probability distributions, and determine the fault state and corresponding confidence level of the CCS integrated busbar based on the final fault probability distribution. The grading module 304 is configured to dynamically adjust the confidence level based on the ambient temperature and operating time of the CCS integrated busbar, and to perform graded early warning based on the fault status of the CCS integrated busbar.

[0062] This application provides a vehicle-mounted CCS integrated busbar fault diagnosis system based on multi-source data fusion, including a data acquisition module configured to acquire multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar using a sensor array and simultaneously record various operating condition labels, breaking through the limitation of traditional single-threshold monitoring that can only identify explicit faults; a processing module configured to preprocess the multi-source real-time data to obtain preprocessed real-time data, and extract multi-dimensional fault-sensitive feature vectors from the preprocessed real-time data, which can capture multi-parameter coupling features; a prediction module configured to input the multi-dimensional fault-sensitive feature vectors, various operating condition labels, and historical fault labels into the system. In the fault diagnosis model that integrates CNN and LSTM, multiple fault probability distributions of the CCS integrated bus are obtained. Based on the weights corresponding to each operating condition and the multiple fault probability distributions, the final fault probability distribution is determined. Based on the final fault probability distribution, the fault state of the CCS integrated bus and the corresponding confidence level are determined. The grading module is configured to dynamically adjust the confidence level in combination with the ambient temperature and working time of the CCS integrated bus, and to perform graded early warning in combination with the fault state of the CCS integrated bus. This realizes the upgrade from fixed threshold judgment to operating condition adaptive decision-making, which greatly improves the accuracy of full-scenario diagnosis and truly adapts to the complex and ever-changing operating scenarios of vehicles.

[0063] It is worth noting that the terms "first" and "second" in this application are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0064] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of this application and should not be construed as limiting the specific implementation of this application to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of this application, and all such modifications or substitutions should be considered within the scope of protection of this application.

Claims

1. A fault diagnosis method for an on-board CCS integrated busbar based on multi-source data fusion, characterized in that, include: The sensor array is used to collect multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar and simultaneously record various operating condition tags. The multi-source real-time data is preprocessed to obtain preprocessed real-time data, and a multi-dimensional fault-sensitive feature vector is extracted from the preprocessed real-time data. The multi-dimensional fault-sensitive feature vector, the various working condition labels, and the historical fault labels are input into the fault diagnosis model that integrates CNN and LSTM to obtain multiple fault probability distributions of the CCS integrated busbar. Based on the weights corresponding to each working condition and the multiple fault probability distributions, the final fault probability distribution is determined. Based on the final fault probability distribution, the fault state and corresponding confidence level of the CCS integrated busbar are determined. The confidence level is dynamically adjusted based on the ambient temperature and operating time of the CCS integrated busbar, and graded early warning is executed based on the fault status of the CCS integrated busbar.

2. The method for fault diagnosis of vehicle-mounted CCS integrated busbar based on multi-source data fusion according to claim 1, characterized in that, Before utilizing a sensor array to collect multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar and simultaneously record various operating condition tags, the vehicle-mounted CCS integrated busbar fault diagnosis method based on multi-source data fusion also includes: At the copper-aluminum busbar electrical connection node in the CCS integrated busbar, a Hall current sensor, a first differential voltage sensor, and a first patch-type temperature sensor are deployed. A second differential voltage sensor and a second surface-mount temperature sensor are deployed at the FPC / PCB signal acquisition component interface in the CCS integrated busbar. A third patch-type temperature sensor is deployed at the critical stress point of the plastic structural component in the CCS integrated busbar; The Hall current sensor, the first differential voltage sensor, the first patch temperature sensor, the second differential voltage sensor, the second patch temperature sensor, and the third patch temperature sensor are calibrated and clock-synchronized. Then, the output terminals are connected to the high-speed data acquisition module, and communication is established with the vehicle battery management system through the bus interface of the high-speed data acquisition module.

3. The method for fault diagnosis of vehicle-mounted CCS integrated busbar based on multi-source data fusion according to claim 2, characterized in that, The method of using a sensor array to collect multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar and simultaneously record various operating condition tags includes: The Hall current sensor, the first differential voltage sensor, the first patch temperature sensor, the second differential voltage sensor, the second patch temperature sensor, and the third patch temperature sensor are triggered at a predetermined sampling frequency to collect data, thereby obtaining the busbar operating current collected by the Hall current sensor, the cell voltage collected by the first differential voltage sensor, the busbar surface temperature collected by the first patch temperature sensor, the signal line voltage collected by the second differential voltage sensor, the interface temperature collected by the second patch temperature sensor, and the structural component surface temperature collected by the third patch temperature sensor. Simultaneously with temperature collection, vehicle operating condition labels are collected from the on-board system; The busbar operating current, individual cell voltage, busbar surface temperature, signal line voltage, interface temperature, and structural component surface temperature are determined as multi-source real-time data.

4. The method for fault diagnosis of vehicle-mounted CCS integrated busbar based on multi-source data fusion according to claim 1, characterized in that, The step of preprocessing the multi-source real-time data to obtain preprocessed real-time data, and extracting multi-dimensional fault-sensitive feature vectors from the preprocessed real-time data, includes: The multi-source real-time data is subjected to noise filtering, missing value imputation, and outlier correction to obtain preprocessed real-time data. Time-domain features, frequency-domain features, and time-series features are extracted from the preprocessed real-time data, and the three are concatenated into a multi-dimensional fault-sensitive feature vector. The time-domain features include current features, voltage features, and temperature features; the frequency-domain features include spectrum features, main frequency band energy ratio, and harmonic distortion rate; and the time-series features include current jump slope, voltage continuous fluctuation duration, and Pearson correlation coefficient of temperature change trend.

5. The method for fault diagnosis of vehicle-mounted CCS integrated busbar based on multi-source data fusion according to claim 1, characterized in that, The fault diagnosis model integrating CNN and LSTM includes fault diagnosis sub-models integrating CNN and LSTM for different operating conditions and a decision layer fusion sub-module. Each fault diagnosis sub-model includes a feature layer fusion module and a CNN-LSTM hybrid diagnosis module connected in sequence. The CNN-LSTM hybrid diagnosis module includes a CNN sub-module, two LSTM sub-modules, and a fault classification head connected in sequence. The feature layer fusion module is used to concatenate the input vectors in terms of dimensions to form initial high-dimensional features, and then reduce the dimensions of the initial high-dimensional features to obtain a fused feature matrix. The CNN-LSTM hybrid diagnosis module is used to reshape the fused feature matrix into a two-dimensional feature map through the CNN sub-module, and then capture the temporal features of the two-dimensional feature map through the two LSTM sub-modules. The fault classification head predicts the fault probability distribution of the CCS integrated busbar based on the historical fault labels and the captured features. The decision layer fusion sub-module is used to assign weights to each fault diagnosis sub-model according to different operating conditions, and calculate the fault state and initial confidence of the CCS integrated busbar based on the weights and the fault probability distribution of the CCS integrated busbar predicted by each fault diagnosis sub-model.

6. The method for fault diagnosis of vehicle-mounted CCS integrated busbar based on multi-source data fusion according to claim 5, characterized in that, The decision-level fusion submodule is used to assign weights to each fault diagnosis sub-model according to different operating conditions, including: By training corresponding fault diagnosis sub-models that fuse CNN and LSTM using historical multi-source data under a single working condition, and assigning weights to each fault diagnosis sub-model based on real-time data under different working conditions, the weights corresponding to each fault diagnosis sub-model are obtained.

7. The method for fault diagnosis of vehicle-mounted CCS integrated busbar based on multi-source data fusion according to claim 5, characterized in that, The calculation of the fault state and initial confidence level of the CCS integrated busbar based on the fault probability distribution predicted by the weights and each fault diagnosis sub-model includes: Each item in the fault probability distribution of the CCS integrated busbar predicted by each fault diagnosis sub-model is multiplied by its corresponding weight, and then summed to obtain the final fault probability distribution of the CCS integrated busbar. The fault state with the highest probability is selected as the fault state of the CCS integrated busbar and its corresponding fault probability is used as the initial confidence level.

8. The method for fault diagnosis of vehicle-mounted CCS integrated busbar based on multi-source data fusion according to claim 5, characterized in that, The process of inputting the multi-dimensional fault-sensitive feature vector, the multiple operating condition labels, and historical fault labels into a fault diagnosis model that fuses CNN and LSTM to obtain multiple fault probability distributions of the CCS integrated busbar, and determining the final fault probability distribution based on the weights corresponding to each operating condition and the multiple fault probability distributions, and determining the fault state and corresponding confidence level of the CCS integrated busbar based on the final fault probability distribution includes: The multi-dimensional fault-sensitive feature vector, the various working condition labels, and the historical fault labels are input into the fault diagnosis model that integrates CNN and LSTM. The feature layer fusion module in the fault diagnosis sub-model corresponding to each working condition concatenates the input multi-dimensional fault-sensitive feature vector in terms of dimensions to form an initial high-dimensional feature. The initial high-dimensional feature is then dimensionality-reduced and fused to obtain a fused feature matrix. The CNN-LSTM hybrid diagnostic module in the fault diagnosis sub-model corresponding to each working condition is used to reshape the fused feature matrix into a two-dimensional feature map, and then the two-dimensional feature map is used to capture temporal features through two LSTM sub-modules. Using a fault classification head, it predicts the fault probability distribution of the CCS integrated busbar based on the captured features, the historical fault labels, and the operating condition labels. Using the decision layer fusion submodule, weights are assigned to each fault diagnosis sub-model according to different working conditions. Based on the weights and the fault probability distribution of the CCS integrated busbar predicted by each fault diagnosis sub-model, the fault state and initial confidence level of the CCS integrated busbar are calculated. The fault diagnosis model is validated using a validation set to calculate the classification accuracy, and the classification accuracy is multiplied by the initial confidence level to obtain the confidence level corresponding to the fault state of the CCS integrated busbar.

9. The method for fault diagnosis of vehicle-mounted CCS integrated busbar based on multi-source data fusion according to claim 1, characterized in that, The step of dynamically adjusting the confidence level based on the ambient temperature and operating time of the CCS integrated busbar, and performing graded early warning based on the fault status of the CCS integrated busbar, includes: Determine whether the ambient temperature and / or operating time of the CCS integrated busbar are greater than a predetermined threshold. If so, increase the confidence level corresponding to the fault state of the CCS integrated busbar by the corresponding step size to dynamically adjust the confidence level. A tiered early warning system is implemented based on the dynamically adjusted confidence level and the fault status of the CCS integrated busbar.

10. A vehicle-mounted CCS integrated busbar fault diagnosis system based on multi-source data fusion, characterized in that, include: The acquisition module is configured to use a sensor array to acquire multi-source real-time data from the core area of ​​the vehicle-mounted CCS integrated busbar and simultaneously record various operating condition tags. The processing module is configured to preprocess the multi-source real-time data to obtain preprocessed real-time data, and extract multi-dimensional fault-sensitive feature vectors from the preprocessed real-time data. The prediction module is configured to input the multi-dimensional fault-sensitive feature vector, the multiple working condition labels, and the historical fault labels into a fault diagnosis model that integrates CNN and LSTM to obtain multiple fault probability distributions of the CCS integrated busbar. Based on the weights corresponding to each working condition and the multiple fault probability distributions, the module determines the final fault probability distribution and, based on the final fault probability distribution, determines the fault state and corresponding confidence level of the CCS integrated busbar. The grading module is configured to dynamically adjust the confidence level based on the ambient temperature and operating time of the CCS integrated busbar, and to perform graded early warning based on the fault status of the CCS integrated busbar.

Images