Power supply fault detection method and system for battery of electric vehicle

Abstract

The invention relates to the technical field of battery power supply fault detection, and particularly discloses a power supply fault detection method and system for an electric vehicle battery, and the method comprises the steps: collecting the multi-modal data of the electric vehicle battery, carrying out the preprocessing and feature extraction of the multi-modal data of the electric vehicle battery, and constructing an intelligent diagnosis model. And carrying out potential fault propagation path prediction, predicting the diffusion path and speed of the fault in a battery system, carrying out electric vehicle battery fault positioning and early warning, and outputting a three-dimensional power supply fault thermodynamic diagram and a hierarchical autonomous disposal instruction of the electric vehicle battery. According to the invention, the problems of poor detection effect, emphasis on single fault diagnosis, large diagnosis difficulty, influenced accuracy and the like caused by incapability of accurately adapting to a fault detection method based on a model in a complex and changeable actual working condition in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery power supply fault detection, and specifically to a method and system for detecting power supply faults of an electric vehicle battery. Background Art

[0002] As a clean and efficient means of transportation, electric vehicles have been widely promoted and applied globally. The rapid growth in the number of electric vehicles in use has made the reliability and safety of the battery power supply system crucial. Once a fault occurs in the battery power supply, it will not only affect the normal operation of the vehicle but may also lead to safety accidents. Therefore, effective fault detection methods are needed to ensure the reliable operation of electric vehicles.

[0003] The battery system of an electric vehicle consists of multiple battery modules, a battery management system, and various sensors, etc., and its structure and working principle are relatively complex. During the charging and discharging process of the battery, it is affected by various factors such as temperature, humidity, and charging and discharging rate, which may cause a decline in battery performance, faults, or even safety problems.

[0004] Currently, there are still some deficiencies in the detection of power supply faults of electric vehicle batteries. Specifically, some existing model-based fault detection methods may not be able to accurately adapt to complex and changing actual working conditions, resulting in poor detection effects. Multiple faults may exist simultaneously in the electric vehicle battery system. For example, battery aging and local overheating may occur simultaneously, or battery cell faults and connection line faults may interact with each other. Existing fault detection methods often focus on the diagnosis of single faults, and for the case of multi-fault coupling, the diagnosis is difficult and the accuracy will also be affected. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a method and system for detecting power supply faults of an electric vehicle battery, which can effectively solve the problems involved in the above background art.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: In the first aspect of the present invention, a method for detecting power supply faults of an electric vehicle battery is provided, including the following steps: collecting multimodal data of the electric vehicle battery through sensors and outputting the multimodal data of the electric vehicle battery, where the multimodal data includes voltage data, current data, temperature data, vibration data, acoustic data, and gas data of the electric vehicle battery; preprocessing and feature extraction of the multimodal data of the electric vehicle battery to obtain a multimodal feature vector of the electric vehicle battery; based on the multimodal feature vector of the electric vehicle battery, predicting the potential fault propagation path of the electric vehicle battery through an intelligent diagnosis model, predicting the diffusion path and speed of the fault in the battery system of the electric vehicle battery, and obtaining the prediction result of the potential fault propagation path; based on the prediction result of the potential fault propagation path, performing fault location and early warning on the electric vehicle battery, and outputting a three-dimensional power supply fault heat map and a hierarchical autonomous disposal instruction of the electric vehicle battery.

[0007] As a further method, collecting multimodal data of the electric vehicle battery through sensors and outputting the multimodal data of the electric vehicle battery includes: using high-precision voltage sensors and current sensors to collect voltage data and current data of each battery cell in the electric vehicle battery; arranging multiple high-precision temperature sensors inside and on the surface of the battery pack of the electric vehicle battery to monitor the temperature distribution of the electric vehicle battery and obtain temperature data of each battery cell in the electric vehicle battery; using an in-vehicle high-sensitivity acceleration sensor to collect vibration data of each battery cell in the electric vehicle battery; installing a high-sensitivity microphone near the battery pack of the electric vehicle battery to collect acoustic data of each battery cell in the electric vehicle battery; installing a gas sensor near the battery pack of the electric vehicle battery, and detecting hydrogen and carbon monoxide released by the electric vehicle battery during the battery thermal runaway process through the gas sensor to obtain gas data of each battery cell in the electric vehicle battery; determining the voltage data, current data, vibration data, acoustic data, and gas data of each battery cell in the electric vehicle battery as the multimodal data of each battery cell in the electric vehicle battery.

[0008] As a further method, it further includes: dynamically adjusting the multimodal data collection frequency of the electric vehicle battery based on the change rate of the multimodal data; deploying redundant sensors for the electric vehicle battery.

[0009] As a further method, the multi-modal data acquisition frequency of the electric vehicle battery is dynamically adjusted based on the change rate of the multi-modal data, and redundant sensors are deployed for the electric vehicle battery, including: obtaining the change rate of the multi-modal data of the electric vehicle battery; when the change rate exceeds the first change threshold stored in the database, adjusting the multi-modal data acquisition frequency of the electric vehicle battery to twice the initial acquisition frequency; when the change rate exceeds the second change threshold stored in the database, adjusting the multi-modal data acquisition frequency of the electric vehicle battery to three times the initial acquisition frequency; obtaining the service life and the number of single cells of the electric vehicle battery, and storing the service life and the number of single cells of the electric vehicle battery as specified tags; determining a redundant sensor deployment scheme matching the specified tags based on the specified tag-redundant sensor deployment scheme mapping set stored in the database; and deploying redundant sensors for the electric vehicle battery according to the redundant sensor deployment scheme.

[0010] As a further method, preprocessing and feature extraction are performed on the multi-modal data of the electric vehicle battery to obtain the multi-modal feature vector of the electric vehicle battery, including: preprocessing the multi-modal data of the electric vehicle battery to obtain preprocessed voltage data, current data, temperature data, vibration data, acoustic data, and gas data; calculating the ratio of the standard deviation to the average value of the preprocessed voltage data, and calculating the voltage stability of the electric vehicle battery based on the ratio to obtain the voltage volatility of the electric vehicle battery; obtaining the derivative of the preprocessed current data by capturing the mutation of the current data to obtain the current change rate of the electric vehicle battery; analyzing the uniformity of the internal heat distribution of the electric vehicle battery based on the preprocessed temperature data, and calculating the difference between the preprocessed temperature data of adjacent batteries to obtain the temperature gradient of the electric vehicle battery; analyzing the frequency components of the preprocessed vibration data by Fourier transform based on the preprocessed vibration data to extract the main vibration frequency and intensity of the electric vehicle battery; identifying abnormal acoustic patterns inside the electric vehicle battery based on the preprocessed acoustic data, and extracting the features of the acoustic signal of the electric vehicle battery using Mel-frequency cepstral coefficients; extracting the change rates of the hydrogen concentration and carbon monoxide concentration of the electric vehicle battery based on the preprocessed gas data; and determining the voltage volatility, current change rate, temperature gradient, main vibration frequency and intensity, features of the acoustic signal, and change rates of the hydrogen concentration and carbon monoxide concentration of the electric vehicle battery as the multi-modal feature vector of the electric vehicle battery.

[0011] As a further method, based on the multi-modal feature vectors of the electric vehicle battery, the potential fault propagation path of the electric vehicle battery is predicted through an intelligent diagnosis model, and the diffusion path and speed of the fault in the battery system of the electric vehicle battery are predicted. The obtained potential fault propagation path prediction results include: through the dynamic spatio-temporal graph convolutional network of the intelligent diagnosis model, the battery system of the electric vehicle battery is modeled as a dynamic graph, the battery cells carrying multi-modal feature vectors are determined as the nodes of the dynamic graph, and based on the multi-modal feature vectors of each node and the multi-modal feature vector-edge weight mapping set stored in the database, the edge weights of the dynamic graph are determined; based on the dynamic graph, the time convolutional network of the intelligent diagnosis model is used to capture the state evolution trend of the battery cells, and the features of neighboring battery cells are aggregated through the graph attention mechanism layer to identify the fault propagation path of the electric vehicle battery; through the uncertainty quantification module of the intelligent diagnosis model, based on the Bayesian deep learning layer, the fault category of the electric vehicle battery and the confidence value of the corresponding fault probability are determined from the multi-modal feature vectors of the electric vehicle battery; the diffusion path and speed of the fault in the battery system of the electric vehicle battery are predicted based on the cellular automaton model; the fault propagation path of the electric vehicle battery, the fault category and the confidence value of the corresponding fault probability, and the diffusion path and speed of the fault in the battery system of the electric vehicle battery are determined as the potential fault propagation path prediction results.

[0012] As a further method, based on the improved cellular automaton model, the diffusion path and speed of the electric vehicle battery fault in the battery system are predicted. The specific analysis process is as follows: the battery system of the electric vehicle battery is discretized into grid cells, and the grid cells are denoted as cells. The cell state of each cell is determined by its own and the previous iteration states of its neighbors. Among them, the cell state = [temperature, SOC, internal resistance, fault flag]; in each iteration process, the cell state is updated. If the temperature or current of the cell exceeds the threshold, or a neighboring cell of the cell fails, the fault flag of the cell will be marked as the fault state; the iteration process is repeated until the battery system reaches the preset stable state in the database or reaches the preset maximum number of iterations; record the positions of the fault cells in each time step, and draw the fault diffusion path diagram according to the position changes of the fault cells. Among them, the fault cell refers to the cell whose fault flag is marked as the fault state; obtain the time and distance from the initial position change of the fault cell to the position of the corresponding fault cell after at least one time step, and calculate the fault diffusion speed according to the time and distance.

[0013] As a further method, based on the prediction results of potential fault propagation paths, fault location and early warning are carried out on the electric vehicle battery, and a three-dimensional power supply fault heat map and a hierarchical autonomous disposal instruction of the electric vehicle battery are output, including: obtaining the coordinates of suspected fault points based on the prediction results of potential fault propagation paths; inputting the coordinates of suspected fault points into the digital twin of the battery system of the electric vehicle to simulate in real time the impact of the abnormality at this position on the global state and verify the physical rationality of the positioning result. If the physical rationality of the positioning result passes, the coordinates of the suspected fault points are retained; if the physical rationality of the positioning result does not pass, the coordinates of the suspected fault points are eliminated to obtain a set of credible suspected fault point positioning coordinates; based on the fault category of the electric vehicle battery and the confidence value of the corresponding fault probability and in combination with the set of credible suspected fault point positioning coordinates, a three-dimensional power supply fault heat map of the electric vehicle battery is output; based on the fault category of the electric vehicle battery and the confidence value of the corresponding fault probability, a power supply fault early warning signal of the electric vehicle battery is generated and a hierarchical autonomous disposal instruction is output.

[0014] As a further method, based on the fault category of the electric vehicle battery and the confidence value of the corresponding fault probability, generating a power supply fault early warning signal of the electric vehicle battery and outputting a hierarchical autonomous disposal instruction includes: storing the fault category of the electric vehicle battery and the confidence value of the corresponding fault probability as a power supply fault label of the electric vehicle battery; determining the power supply fault early warning level matching the power supply fault label according to the power supply fault label - power supply fault early warning level mapping set stored in the database; when the power supply fault early warning level is a minor fault, sending a yellow early warning signal to prompt the state of the electric vehicle battery and sending a maintenance inspection instruction for the electric vehicle battery; when the power supply fault early warning level is a serious fault, sending a red early warning signal and sending an instruction to disconnect the battery from the vehicle system to make the vehicle enter the safe mode.

[0015] The second aspect of the present invention provides a power supply fault detection system for an electric vehicle battery, including: a multimodal data acquisition module, which is used to collect multimodal data of the electric vehicle battery through sensors and output the multimodal data of the electric vehicle battery. The multimodal data includes voltage data, current data, temperature data, vibration data, acoustic data, and gas data of the electric vehicle battery; a multimodal feature vector acquisition module, which is used to preprocess and extract features from the multimodal data of the electric vehicle battery to obtain the multimodal feature vector of the electric vehicle battery; a fault propagation path prediction module, which is used to predict the potential fault propagation path of the electric vehicle battery through an intelligent diagnosis model based on the multimodal feature vector of the electric vehicle battery, predict the diffusion path and speed of the fault in the battery system of the electric vehicle battery, and obtain the prediction result of the potential fault propagation path; a hierarchical autonomous disposal instruction output module, which is used to locate and warn of faults in the electric vehicle battery based on the prediction result of the potential fault propagation path, and output a three-dimensional power supply fault heat map and hierarchical autonomous disposal instructions for the electric vehicle battery.

[0016] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) By providing a power supply fault detection method and system for an electric vehicle battery, the present invention can comprehensively reflect the working state of the battery from multiple dimensions by collecting multimodal data such as voltage, current, temperature, vibration, acoustics, and gas of the electric vehicle battery. Compared with a single data source, multimodal data can provide richer and more accurate information, which helps to detect battery faults more precisely. Preprocessing and feature extraction of the multimodal data can remove noise and redundant information in the data and extract feature vectors that can better reflect the essence of battery faults. These feature vectors can more accurately describe the state of the battery and provide a more reliable basis for subsequent fault diagnosis, thereby improving the accuracy and reliability of fault detection.

[0017] (2) By constructing an intelligent diagnosis model and predicting the potential fault propagation path based on the multimodal feature vector, the present invention can understand in advance the diffusion path and speed of the fault in the battery system. Fault location and warning based on the prediction result of the potential fault propagation path can accurately indicate the location where the fault occurs and send out a warning signal in a timely manner. The output three-dimensional power supply fault heat map of the electric vehicle battery can intuitively display the distribution and severity of the fault. At the same time, the hierarchical autonomous disposal instructions can automatically trigger corresponding disposal measures according to the severity of the fault, improve the efficiency and automation of fault handling, and ensure the safe operation of the electric vehicle. Description of the Drawings

[0018] The present invention will be further described with reference to the accompanying drawings. However, the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the following drawings without creative efforts.

[0019] Figure 1 It is a schematic flow chart of the method steps of the present invention; Figure 2 It is a flow chart of steps for performing electric vehicle battery fault location and warning based on the prediction result of the potential fault propagation path, and outputting a three-dimensional power supply fault heat map and a hierarchical autonomous disposal instruction for the electric vehicle battery; Figure 3 It is a schematic diagram of the connection of the system modules of the present invention. Detailed implementation manners

[0020] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0021] Referring to Figure 1 As shown, the first aspect of the present invention provides a method for detecting power supply faults of an electric vehicle battery, including: steps S101-104.

[0022] S101, performing multi-modal data acquisition on the electric vehicle battery through sensors, and outputting multi-modal data of the electric vehicle battery. The multi-modal data includes voltage data, current data, temperature data, vibration data, acoustic data, and gas data of the electric vehicle battery.

[0023] In some embodiments of the present disclosure, multimodal data of an electric vehicle battery is collected through sensors, and the output multimodal data of the electric vehicle battery includes: using high-precision voltage sensors and current sensors to collect voltage data and current data of each battery cell in the electric vehicle battery; arranging a plurality of high-precision temperature sensors inside and on the surface of the battery pack of the electric vehicle battery to monitor the temperature distribution of the electric vehicle battery and obtain the temperature data of each battery cell in the electric vehicle battery; using an in-vehicle high-sensitivity acceleration sensor to collect vibration data of each battery cell in the electric vehicle battery; installing a high-sensitivity microphone near the battery pack of the electric vehicle battery to collect acoustic data of each battery cell in the electric vehicle battery; installing a gas sensor near the battery pack of the electric vehicle battery, and detecting hydrogen and carbon monoxide released by the electric vehicle battery during battery thermal runaway through the gas sensor to obtain gas data of each battery cell in the electric vehicle battery; determining the voltage data, current data, vibration data, acoustic data, and gas data of each battery cell in the electric vehicle battery as the multimodal data of each battery cell in the electric vehicle battery.

[0024] In some embodiments, the method further includes: dynamically adjusting the multimodal data collection frequency of the electric vehicle battery based on the change rate of the multimodal data; deploying redundant sensors for the electric vehicle battery.

[0025] In one embodiment, the change rate of the multimodal data of the electric vehicle battery is obtained; when the change rate exceeds the first change threshold stored in the database, the multimodal data collection frequency of the electric vehicle battery is adjusted to twice the initial collection frequency; when the change rate exceeds the second change threshold stored in the database, the multimodal data collection frequency of the electric vehicle battery is adjusted to three times the initial collection frequency; the service life and the number of single cells of the electric vehicle battery are obtained, and the service life and the number of single cells of the electric vehicle battery are stored as specified tags; based on the specified tag-redundant sensor deployment scheme mapping set stored in the database, a redundant sensor deployment scheme matching the specified tag is determined; according to the redundant sensor deployment scheme, redundant sensors are deployed for the electric vehicle battery.

[0026] Specifically, for deploying redundant sensors, the same parameter is measured simultaneously by multiple sensors. For example, normally only one sensor is placed at this position to detect current, but to prevent a single-point sensor from failing, redundant sensors may be deployed at this position. By deploying two or more sensors, it is ensured that the subsequent analysis will not be affected due to the failure of a single-point sensor. The redundant sensors and the normal sensors monitor the battery together to prevent excessive data errors or data loss. Outliers are excluded through data fusion to reduce the risk of false alarms.

[0027] In one implementation, multi-modal data such as voltage, current, temperature, vibration, acoustics, and gas are collected to comprehensively capture battery state information. For example, abnormal noises caused by subtle internal structure changes of the battery can be detected through acoustic data, and characteristic gases released at the initial stage of thermal runaway can be detected through gas data, compensating for the limitations of single-data detection and improving the timeliness and accuracy of fault discovery.

[0028] Specifically, the acquisition frequency is dynamically adjusted according to the data change rate. When the change rate exceeds the threshold, the frequency is increased, enabling dense data acquisition when the battery state changes drastically, not missing key information, and accurately capturing weak signals at the initial stage of faults. At the same time, the redundant sensor deployment scheme is determined based on the battery usage years and the number of battery cells, which can ensure the reliability of data acquisition. Even when some sensors fail, data can still be effectively collected. By comprehensively considering various factors, the trend of fault spread can be known in advance, enabling maintenance personnel to plan countermeasures in advance and reducing the harm of faults.

[0029] S102. Preprocess and extract features from the multi-modal data of the electric vehicle battery to obtain the multi-modal feature vector of the electric vehicle battery.

[0030] In some embodiments, S102 includes: preprocessing the multi-modal data of the electric vehicle battery to obtain preprocessed voltage data, current data, temperature data, vibration data, acoustic data, and gas data; calculating the ratio of the standard deviation to the average value of the preprocessed voltage data, and calculating the voltage stability of the electric vehicle battery based on the ratio to obtain the voltage volatility of the electric vehicle battery; by capturing the mutation of the current data, calculating the derivative of the preprocessed current data to obtain the current change rate of the electric vehicle battery; based on the preprocessed temperature data, analyzing the uniformity of the internal heat distribution of the electric vehicle battery, and calculating the difference between the preprocessed temperature data of adjacent batteries to obtain the temperature gradient of the electric vehicle battery; based on the preprocessed vibration data, analyzing the frequency components of the preprocessed vibration data through Fourier transform to extract the main vibration frequency and intensity of the electric vehicle battery; based on the preprocessed acoustic data, identifying abnormal acoustic patterns inside the electric vehicle battery, and using Mel frequency cepstral coefficients to extract the characteristics of the acoustic signal of the electric vehicle battery; based on the preprocessed gas data, extracting the change rates of hydrogen concentration and carbon monoxide concentration of the electric vehicle battery; determining the voltage volatility, current change rate, temperature gradient, main vibration frequency and intensity, characteristics of the acoustic signal, and change rates of hydrogen concentration and carbon monoxide concentration of the electric vehicle battery as the multi-modal feature vector of the electric vehicle battery.

[0031] Specifically, parameters such as the above voltage fluctuations and current change rates are measured for each cell, and then integrated into a multi-modal feature vector. This multi-modal feature vector should include the multi-modal feature vector of the first cell, which is the voltage volatility rate, current change rate, temperature gradient, main vibration frequency and intensity, characteristics of the acoustic signal, and the change rates of hydrogen concentration and carbon monoxide concentration of the first cell; the multi-modal feature vector of the second cell, which is the voltage volatility rate, current change rate, temperature gradient, main vibration frequency and intensity, characteristics of the acoustic signal, and the change rates of hydrogen concentration and carbon monoxide concentration of the second cell; the multi-modal feature vector of the nth cell, which is the voltage volatility rate, current change rate, temperature gradient, main vibration frequency and intensity, characteristics of the acoustic signal, and the change rates of hydrogen concentration and carbon monoxide concentration of the nth cell.

[0032] In one implementation, preprocessing is performed on the multi-modal data of the electric vehicle battery. The specific analysis process is as follows: Wavelet transform is used to denoise the electric vehicle battery voltage data and electric vehicle battery current data. The wavelet basis function is selected, and wavelet decomposition is performed on the electric vehicle battery voltage data and electric vehicle battery current data to extract low-frequency and high-frequency components. Threshold processing is performed on the high-frequency components to remove noise, and the denoised electric vehicle battery voltage data and electric vehicle battery current data are obtained through wavelet reconstruction; Gaussian filtering is used to denoise the electric vehicle battery temperature data to obtain the denoised electric vehicle battery temperature data; a band-pass filter is used to denoise the electric vehicle battery vibration data to remove high-frequency and low-frequency noise; spectral subtraction is used to denoise the electric vehicle battery acoustic data, and the electric vehicle battery noise spectrum is subtracted from the signal spectrum; noise reduction processing is performed on the electric vehicle battery gas data, and high-frequency noise in the data is removed through discrete wavelet transform while retaining the main characteristics of gas concentration changes.

[0033] The moving average method is used to smooth the multi-modal data of the electric vehicle battery. The average value of the data points within each window and several data points before and after them is taken to obtain the smoothed multi-modal data of the electric vehicle battery; the Z-Score method is used to detect and process outliers in the multi-modal data of the electric vehicle battery, and the average value of adjacent normal data points is used for filling.

[0034] By applying various filtering methods such as wavelet transform, Gaussian filtering, band-pass filter, and spectral subtraction to denoise according to the characteristics of different types of data (voltage, current, temperature, vibration, acoustic, gas), various noise interferences such as high-frequency noise and low-frequency noise can be effectively removed, enabling the data to more accurately reflect the true state of the battery and laying a good foundation for subsequent analysis. Discrete wavelet transform is used for gas data noise reduction, which can remove high-frequency noise while retaining key features.

[0035] The moving average method is used to smooth the multi-modal data, which can reduce the random fluctuations of the data, make the data change trend clearer, facilitate the analysis of data patterns, and reduce the risk of misjudgment caused by data fluctuations. The Z-Score method is adopted to detect and process outliers, and the average value of adjacent normal data points is used to fill them, which can avoid the adverse effects of outliers on subsequent analysis and ensure the reliability and stability of the data.

[0036] The preprocessed voltage data, current data, temperature data, vibration data, acoustic data, and gas data are obtained.

[0037] In one implementation, by calculating characteristic quantities such as voltage volatility, current change rate, and temperature gradient, the operating state of the battery is quantified from different perspectives. Voltage volatility reflects voltage stability, and the current change rate can capture sudden changes in current to help determine whether the battery charging and discharging are abnormal. Fourier transform analysis of the vibration data for frequency components and extraction of features from the acoustic data using Mel-frequency cepstral coefficients can uncover potential information such as changes in the internal structure of the battery and component looseness, which are not easily detected in conventional inspections and contribute to the early detection of battery early faults. Extracting the change rates of hydrogen and carbon monoxide concentrations in the gas data can monitor in real time the fault-related indicators such as battery thermal runaway and provide key evidence for fault warning.

[0038] S103. Based on the multi-modal feature vectors of the electric vehicle battery, predict the potential fault propagation path of the electric vehicle battery through an intelligent diagnosis model, predict the diffusion path and speed of the fault in the battery system of the electric vehicle battery, and obtain the prediction result of the potential fault propagation path.

[0039] In some embodiments, S103 includes: modeling the battery system of the electric vehicle battery as a dynamic graph through the dynamic spatio-temporal graph convolutional network of the intelligent diagnosis model, determining the battery cells carrying multi-modal feature vectors as the nodes of the dynamic graph, and determining the edge weights of the dynamic graph based on the multi-modal feature vectors of each node and the multi-modal feature vector-edge weight mapping set stored in the database, specifically the temperature gradient-edge weight mapping set, and the edge weights of the dynamic graph are updated as the temperature gradient changes; based on the dynamic graph, capturing the state evolution trend of the battery cells through the time convolutional network of the intelligent diagnosis model, aggregating the features of neighboring battery cells through the graph attention mechanism layer, and identifying the fault propagation path of the electric vehicle battery; through the uncertainty quantification module of the intelligent diagnosis model, based on the Bayesian deep learning layer, determining the fault category of the electric vehicle battery and the confidence value of the corresponding fault probability from the multi-modal feature vectors of the electric vehicle battery; predicting the diffusion path and speed of the fault in the battery system of the electric vehicle battery based on the cellular automaton model; and determining the fault propagation path of the electric vehicle battery, the fault category and the confidence value of the corresponding fault probability, and the diffusion path and speed of the fault in the battery system of the electric vehicle battery as the prediction result of the potential fault propagation path.

[0040] In some embodiments, the battery system of an electric vehicle battery is discretized into grid cells, which are denoted as cells. The cell state of each cell is determined by the previous iteration states of itself and its neighbors. Among them, the cell state = [temperature, SOC, internal resistance, fault flag]; during each iteration process, the cell state is updated. Adjacent cells interact through Fourier's law of heat conduction and current distribution based on Kirchhoff's law. If the temperature or current of a cell exceeds the threshold, or a fault occurs in the adjacent cell of the cell, the fault flag of the cell will be marked as the fault state; the iteration process is repeated until the battery system reaches the preset stable state in the database or reaches the preset maximum number of iterations; record the positions of the faulty cells in each time step, and draw a fault propagation path diagram according to the position changes of the faulty cells, where the faulty cell refers to the cell whose fault flag is marked as the fault state; obtain the time and distance from the initial position of the faulty cell to the position of the corresponding faulty cell after at least one time step, and calculate the fault propagation speed according to the time and distance.

[0041] Specifically, thermistors are arranged on the surface of the battery cells to collect the surface temperature at a frequency of 50 Hz. If the temperature is abnormal, it is determined that a fault exists, and the fault flag is stored. A closed-loop Hall current sensor is used to detect the millisecond-level current fluctuation. If the current is abnormal, it is determined that a fault exists, and the fault flag is stored. A three-axis MEMS accelerometer is installed to capture abnormal vibrations. If the vibration is abnormal, it is determined that a fault exists, and the fault flag is stored. A catalytic combustion type hydrogen sensor and an electrochemical CO sensor are configured to detect the ppm-level gas concentration change. If the gas concentration is abnormal, it is determined that a fault exists, and the fault flag is stored.

[0042] In one implementation, the dynamic spatio-temporal graph convolutional network models the battery system as a dynamic graph. The nodes carry multi-modal feature vectors, and the edge weights are updated in real time according to the temperature gradient, etc., which can accurately reflect the dynamic characteristics of the battery system and the correlations between battery cells, providing an accurate basis for fault propagation analysis.

[0043] Specifically, the time convolutional network is combined to capture the evolution of the battery cell state, and the graph attention mechanism is used to analyze the cross-cell fault propagation, which can effectively aggregate the features of neighboring battery cells, accurately identify the fault propagation path, give full play to the advantages of different mechanisms, and improve the fault analysis ability.

[0044] In one implementation, a Bayesian deep learning layer is introduced to construct an uncertainty quantification module, which outputs the fault category and probability confidence value, can measure the reliability of the fault judgment, provides more comprehensive information for decision-making, and avoids misjudgment caused by uncertainty.

[0045] Specifically, the improved cellular automaton model discretizes the battery system. Based on physical laws such as Fourier's law of heat conduction and Kirchhoff's law, considering the heat and current interactions between battery cells, the fault diffusion simulation is made more in line with the actual physical process, accurately predicting the diffusion path and speed. By continuously updating the cell state, recording the positions of faulty cells, drawing the path diagram and calculating the speed, the development process of the fault in the battery system can be fully presented, facilitating the formulation of targeted maintenance strategies in advance and reducing the harm of faults.

[0046] In one implementation, the intelligent diagnosis model consists of three parts: a dynamic spatio-temporal graph convolutional network, an uncertainty quantification module, and a cellular automaton prediction module. The input of the temporal convolutional network is the dynamic graph node features: the cell nodes of the multi-modal feature vector (including voltage volatility, current change rate, temperature gradient, vibration frequency, acoustic features, gas concentration change rate), and the output is the fault propagation path.

[0047] The input of the uncertainty quantification module is the multi-modal feature vector, and the output is the confidence value of the electric vehicle battery fault category and the corresponding fault probability.

[0048] The input of the cellular automaton model is the initial cell state parameters: including temperature, SOC, internal resistance, fault flag, etc. The physical interaction rules are the heat diffusion rules based on Fourier's law of heat conduction and the current distribution rules based on Kirchhoff's law, and the output is the diffusion path and speed of the electric vehicle battery fault in the battery system.

[0049] The specific content of the fault propagation path prediction result includes the fault propagation path and speed, the confidence value of the electric vehicle battery fault category and the corresponding fault probability.

[0050] The difference between the fault propagation path and the diffusion path of the fault in the battery system is that the analysis dimension of the fault propagation path is the logical association between battery cells, and the analysis dimension of the diffusion path of the fault in the battery system is the spreading process in physical space.

[0051] S104. Based on the potential fault propagation path prediction result, conduct fault location and early warning for the electric vehicle battery, and output the three-dimensional power supply fault thermal map and hierarchical autonomous disposal instructions of the electric vehicle battery.

[0052] In some embodiments, S104 includes: such as Figure 2As shown, based on the prediction results of potential fault propagation paths, the coordinates of suspected fault points are obtained; the coordinates of suspected fault points are input into the digital twin of the battery system of the electric vehicle battery to simulate in real time the impact of the anomaly at this location on the global state, verify the physical rationality of the positioning results. If the physical rationality of the positioning results passes, the coordinates of the suspected fault points are retained. If the physical rationality of the positioning results fails, the coordinates of the suspected fault points are excluded to obtain a set of credible suspected fault point positioning coordinates; based on the fault categories of the electric vehicle battery and the confidence values of the corresponding fault probabilities and combined with the set of credible suspected fault point positioning coordinates, a three-dimensional power supply fault heat map of the electric vehicle battery is output; based on the fault categories of the electric vehicle battery and the confidence values of the corresponding fault probabilities, a power supply fault warning signal for the electric vehicle battery is generated and a hierarchical autonomous handling instruction is output.

[0053] In some embodiments, the fault categories of the electric vehicle battery and the confidence values of the corresponding fault probabilities are stored as power supply fault labels of the electric vehicle battery; according to the power supply fault label - power supply fault warning level mapping set stored in the database, the power supply fault warning level matching the power supply fault label is determined; when the power supply fault warning level is a minor fault, a yellow warning signal is issued to indicate the state of the electric vehicle battery, and a maintenance inspection instruction for the electric vehicle battery is issued; when the power supply fault warning level is a serious fault, a red warning signal is issued, and an instruction to disconnect the battery from the vehicle system is issued to make the vehicle enter the safe mode.

[0054] Specifically, the digital twin of the battery system is a virtual mirror system based on multi-physics field modeling and real-time data driving, and its architecture includes: The three-dimensional geometric model includes: cell level (single cell size, tab position), module level (cooling pipeline layout), system level (high and low voltage harness routing); Real-time data interface: Synchronization frequency: 100Hz / 10Hz; Multi-physics field simulation engine: Coupled solver support: Electrochemical model (P2D model), thermodynamics model (Fourier equation), structural mechanics model (modal analysis).

[0055] In some embodiments, generating a power supply fault warning signal for the electric vehicle battery and outputting a hierarchical autonomous handling instruction based on the fault categories of the electric vehicle battery and the confidence values of the corresponding fault probabilities includes: storing the fault categories of the electric vehicle battery and the confidence values of the corresponding fault probabilities as power supply fault labels of the electric vehicle battery; according to the power supply fault label - power supply fault warning level mapping set stored in the database, determining the power supply fault warning level matching the power supply fault label; based on the power supply fault warning level, issuing a power supply fault warning signal for the electric vehicle battery and outputting a hierarchical autonomous handling instruction.

[0056] In some embodiments, based on the power supply fault warning level, sending a power supply fault warning signal for the electric vehicle battery and outputting a hierarchical autonomous handling instruction includes: when the power supply fault warning level is a minor fault, sending a yellow warning signal to indicate the state of the electric vehicle battery, and sending a maintenance inspection instruction for the electric vehicle battery; when the power supply fault warning level is a serious fault, sending a red warning signal and sending an instruction to cut off the connection between the battery and the vehicle system, and the vehicle enters the safe mode.

[0057] Specifically, based on the prediction result of the potential fault propagation path, the coordinates of the suspected fault point are determined, and through the real-time simulation verification of the digital twin of the battery system, unreasonable positioning can be eliminated, and a set of credible fault point coordinates can be obtained, ensuring the accuracy of fault location, avoiding misjudgment, enabling maintenance personnel to accurately find the fault location, and improving the maintenance efficiency.

[0058] In one implementation, by comprehensively considering the electric vehicle battery fault category, the fault probability confidence value, and the set of credible fault point positioning coordinates, a three-dimensional power supply fault heat map is output, which can present the fault information in an intuitive and visual way, facilitating the operator to quickly understand the distribution range, severity, etc. of the battery faults and assisting in decision-making.

[0059] In one implementation, the power supply fault warning level is determined according to the battery fault category and the fault probability confidence value, and corresponding warning signals are sent and hierarchical autonomous handling instructions are output for different levels (minor fault, serious fault) to achieve differential processing. When it is a minor fault, maintenance inspection is prompted, and when it is a serious fault, the connection between the battery and the vehicle is cut off to make the vehicle enter the safe mode, which can not only timely remind to pay attention to the battery state, but also ensure the safety of the vehicle and personnel in serious situations.

[0060] Specifically, the entire fault location, warning, and handling process is automatically completed based on the previous data processing and model prediction, without excessive manual intervention, improving the timeliness of fault response and the handling efficiency, reflecting the intelligent level of the electric vehicle battery management system, and enhancing the reliability and stability of the system operation.

[0061] Refer to Figure 3 As shown in the figure, the second aspect of the present invention provides a power supply fault detection system for an electric vehicle battery, including: a multi-modal data acquisition module, which is used to perform multi-modal data acquisition on the electric vehicle battery through sensors and output the multi-modal data of the electric vehicle battery, and the multi-modal data includes voltage data, current data, temperature data, vibration data, acoustic data, and gas data of the electric vehicle battery.

[0062] A multi-modal feature vector acquisition module, which is used to preprocess and extract features from the multi-modal data of the electric vehicle battery to obtain the multi-modal feature vector of the electric vehicle battery.

[0063] A fault propagation path prediction module, which is used to predict the potential fault propagation path of an electric vehicle battery based on the multi-modal feature vector of the electric vehicle battery through an intelligent diagnosis model, predict the diffusion path and speed of the fault in the battery system of the electric vehicle battery, and obtain the prediction result of the potential fault propagation path.

[0064] A hierarchical autonomous disposal instruction output module, which is used to locate and warn of faults in the electric vehicle battery based on the prediction result of the potential fault propagation path, and output a three-dimensional power supply fault heat map and hierarchical autonomous disposal instructions for the electric vehicle battery.

[0065] The above content is only an example and illustration of the structure of the present invention. Those skilled in the art of this technology can make various modifications or supplements to the described specific embodiments or use similar methods for substitution, as long as they do not deviate from the structure of the invention or exceed the scope defined by this claims, they should all fall within the protection scope of the present invention.

Claims

1. A method for detecting power supply failure of an electric vehicle battery, characterized in that: The following steps are involved: Performing multimodal data collection on an electric vehicle battery through a sensor, and outputting the multimodal data of the electric vehicle battery, wherein the multimodal data includes voltage data, current data, temperature data, vibration data, acoustic data, and gas data of the electric vehicle battery; Preprocessing and feature extraction are performed on the multimodal data of the electric vehicle battery to obtain a multimodal feature vector of the electric vehicle battery; Based on the multimodal feature vector of the electric vehicle battery, a potential fault propagation path of the electric vehicle battery is predicted through an intelligent diagnosis model, the diffusion path and speed of the fault in the battery system of the electric vehicle battery are predicted, and a potential fault propagation path prediction result is obtained; Based on the prediction result of the potential fault propagation path, the fault location and early warning of the electric vehicle battery are performed, and a three-dimensional power supply fault heat map and hierarchical autonomous handling instructions of the electric vehicle battery are output.

2. A method for detecting power supply failure of an electric vehicle battery according to claim 1, characterized in that: The method of collecting multimodal data of the electric vehicle battery by using a sensor and outputting the multimodal data of the electric vehicle battery includes: Using high-precision voltage sensors and current sensors to collect voltage data and current data of each battery cell in the electric vehicle battery; Arrange a plurality of high-precision temperature sensors inside and on the surface of the battery pack of the electric vehicle battery to monitor the temperature distribution of the electric vehicle battery and obtain the temperature data of each battery cell in the electric vehicle battery; Using a high-sensitivity on-board acceleration sensor to collect vibration data of each battery cell in the electric vehicle battery; Installing a high-sensitivity microphone near the battery pack of the electric vehicle battery to collect acoustic data of each battery cell in the electric vehicle battery; Installing a gas sensor near the battery pack of the electric vehicle battery, detecting hydrogen and carbon monoxide released by the electric vehicle battery during battery thermal runaway through the gas sensor, and obtaining gas data of each battery cell in the electric vehicle battery; The voltage data, current data, vibration data, acoustic data and gas data of each battery cell in the electric vehicle battery are determined as multimodal data of each battery cell in the electric vehicle battery.

3. The method for detecting power supply failure of an electric vehicle battery according to claim 1, characterized in that: The method further comprises: Dynamically adjusting the frequency of multimodal data collection of the electric vehicle battery based on the rate of change of the multimodal data; Redundant sensors are deployed on the electric vehicle battery.

4. A method for detecting power supply failure of an electric vehicle battery according to claim 3, characterized in that: The dynamically adjusting the multimodal data acquisition frequency of the electric vehicle battery based on the rate of change of the multimodal data and deploying redundant sensors on the electric vehicle battery includes: Obtaining a rate of change of multimodal data of the electric vehicle battery; When the change rate exceeds a first change threshold stored in the database, adjusting the multimodal data acquisition frequency of the electric vehicle battery to twice the initial acquisition frequency; When the change rate exceeds a second change threshold stored in the database, adjusting the multimodal data acquisition frequency of the electric vehicle battery to three times the initial acquisition frequency; Obtaining the service life and the number of cells of the electric vehicle battery, and storing the service life and the number of cells of the electric vehicle battery as a designated tag; Determine a redundant sensor deployment scheme matching the specified tag based on a specified tag-redundant sensor deployment scheme mapping set stored in a database; According to the redundant sensor deployment scheme, redundant sensors are deployed on the electric vehicle battery.

5. The method for detecting power supply failure of an electric vehicle battery according to claim 1, characterized in that: The preprocessing and feature extraction of the multimodal data of the electric vehicle battery to obtain the multimodal feature vector of the electric vehicle battery includes: Preprocessing the multimodal data of the electric vehicle battery to obtain preprocessed voltage data, current data, temperature data, vibration data, acoustic data, and gas data; Obtaining a ratio of a standard deviation to an average value of the preprocessed voltage data, calculating a voltage stability of the electric vehicle battery based on the ratio, and obtaining a voltage fluctuation rate of the electric vehicle battery; By capturing the sudden change of the current data, the derivative of the pre-processed current data is obtained to obtain the current change rate of the electric vehicle battery; Based on the pre-processed temperature data, analyzing the uniformity of heat distribution inside the battery of the electric vehicle battery, calculating the difference of the pre-processed temperature data between adjacent batteries, and obtaining the temperature gradient of the electric vehicle battery; Based on the preprocessed vibration data, analyzing the frequency components of the preprocessed vibration data by Fourier transform to extract the main vibration frequency and intensity of the electric vehicle battery; Based on the preprocessed acoustic data, identifying abnormal acoustic modes inside the battery of the electric vehicle, and extracting features of the acoustic signal of the electric vehicle battery using Mel-frequency cepstral coefficients; Based on the pre-processed gas data, extracting the hydrogen concentration and carbon monoxide concentration change rate of the electric vehicle battery; The voltage fluctuation rate, current change rate, temperature gradient, main vibration frequency and intensity, acoustic signal characteristics, and hydrogen concentration and carbon monoxide concentration change rate of the electric vehicle battery are determined as the multimodal feature vector of the electric vehicle battery.

6. The method for detecting power supply failure of an electric vehicle battery according to claim 1, characterized in that: The multimodal feature vector of the electric vehicle battery is used to predict the potential fault propagation path of the electric vehicle battery through an intelligent diagnosis model, and the diffusion path and speed of the fault in the battery system of the electric vehicle battery are predicted. The potential fault propagation path prediction result is obtained, which includes: The battery system of the electric vehicle battery is modeled as a dynamic graph through the dynamic spatiotemporal graph convolutional network of the intelligent diagnosis model, the battery cells carrying the multimodal feature vectors are determined as nodes of the dynamic graph, and the edge weights of the dynamic graph are determined based on the multimodal feature vectors of each of the nodes and the multimodal feature vector-edge weight mapping set stored in the database; Based on the dynamic graph, the state evolution trend of the battery cell is captured through the temporal convolutional network of the intelligent diagnosis model, and the neighboring battery cell features are aggregated through the graph attention mechanism layer to identify the fault propagation path of the electric vehicle battery; Determine the fault category of the electric vehicle battery and the confidence value of the corresponding fault probability from the multimodal feature vector of the electric vehicle battery through the uncertainty quantification module of the intelligent diagnosis model based on the Bayesian deep learning layer; Predicting the diffusion path and speed of faults in the battery system of the electric vehicle battery based on a cellular automation model; The fault propagation path of the electric vehicle battery, the fault category and the corresponding confidence value of the fault probability, as well as the diffusion path and speed of the fault in the battery system of the electric vehicle battery are determined as potential fault propagation path prediction results.

7. A method for detecting power supply failure of an electric vehicle battery according to claim 6, characterized in that: The improved cellular automation model is used to predict the diffusion path and speed of electric vehicle battery failure in the battery system. The specific analysis process is as follows: Discretize the battery system of the electric vehicle battery into grid units, record the grid units as cells, and the cell state of each cell is determined by the previous iteration state of itself and its neighbors, where cell state = [temperature, SOC, internal resistance, fault flag]; In each iteration process, the cell state is updated. If the temperature or current of the cell exceeds a threshold, or an adjacent cell of the cell fails, the fault flag of the cell will be marked as a fault state. Repeating the iterative process until the battery system reaches a stable state preset in a database or reaches a preset maximum number of iterations; Record the position of the faulty cell in each time step, and draw a fault diffusion path diagram according to the position change of the faulty cell, wherein the faulty cell refers to the cell whose fault flag is marked as a fault state; The time and distance of the fault cell position corresponding to the initial position of the fault cell changing to at least one time step are obtained, and the fault diffusion speed is calculated according to the time and distance.

8. A method for detecting power supply failure of an electric vehicle battery according to claim 6, characterized in that: The method of locating and warning the electric vehicle battery fault based on the potential fault propagation path prediction result, and outputting the three-dimensional power supply fault heat map and hierarchical autonomous handling instructions of the electric vehicle battery includes: Based on the potential fault propagation path prediction result, obtaining the coordinates of the suspected fault point; The coordinates of the suspected fault point are input into the battery system digital twin of the electric vehicle battery to simulate the impact of the position abnormality on the global state in real time, and the physical rationality of the positioning result is verified. If the physical rationality of the positioning result is passed, the coordinates of the suspected fault point are retained. If the physical rationality of the positioning result is not passed, the coordinates of the suspected fault point are eliminated to obtain a credible set of suspected fault point positioning coordinates; Based on the fault category of the electric vehicle battery and the corresponding confidence value of the fault probability and in combination with the reliable suspected fault point location coordinate set, output a three-dimensional power supply fault heat map of the electric vehicle battery; Based on the fault category of the electric vehicle battery and the corresponding confidence value of the fault probability, a power supply fault warning signal of the electric vehicle battery is generated and a graded autonomous handling instruction is output.

9. A method for detecting power supply failure of an electric vehicle battery according to claim 8, characterized in that: The generating of the electric vehicle battery power supply failure warning signal and outputting the hierarchical autonomous handling instruction based on the fault type of the electric vehicle battery and the corresponding confidence value of the fault probability comprises: storing the fault category of the electric vehicle battery and the corresponding confidence value of the fault probability as a power supply fault label of the electric vehicle battery; Determine the power supply fault warning level that matches the power supply fault label according to the power supply fault label-power supply fault warning level mapping set stored in the database; When the power supply fault warning level is a minor fault, a yellow warning signal is issued to indicate the battery status of the electric vehicle, and an electric vehicle battery maintenance inspection instruction is issued; When the power supply fault warning level is a serious fault, a red warning signal is issued, and an instruction to cut off the connection between the battery and the vehicle system is issued to put the vehicle into a safe mode.

10. A power supply fault detection system for an electric vehicle battery, applied to a power supply fault detection method for an electric vehicle battery according to any one of claims 1 to 9, characterized in that: include: A multimodal data acquisition module, used to collect multimodal data of an electric vehicle battery through a sensor, and output multimodal data of the electric vehicle battery, wherein the multimodal data includes voltage data, current data, temperature data, vibration data, acoustic data, and gas data of the electric vehicle battery; A multimodal feature vector acquisition module, used for preprocessing and feature extraction of the multimodal data of the electric vehicle battery to obtain a multimodal feature vector of the electric vehicle battery; A fault propagation path prediction module is used to predict the potential fault propagation path of the electric vehicle battery through an intelligent diagnosis model based on the multimodal feature vector of the electric vehicle battery, predict the diffusion path and speed of the fault in the battery system of the electric vehicle battery, and obtain the potential fault propagation path prediction result; The hierarchical autonomous handling instruction output module is used to locate and warn the electric vehicle battery of the fault based on the potential fault propagation path prediction result, and output the three-dimensional power supply fault heat map of the electric vehicle battery and the hierarchical autonomous handling instructions.

Images