A cloud-based collaborative multi-sensor battery fault detection method and system

Abstract

A cloud-based collaborative method and system for multi-sensor battery fault detection. The method includes: a vehicle-side terminal acquiring multi-sensor data from the battery system in real time, inputting it into a pre-built battery equivalent circuit model to obtain a model-predicted voltage value, and calculating a voltage residual sequence based on the measured voltage value and the model-predicted voltage value; the vehicle-side terminal uploading the voltage residual sequence to a cloud server; and the cloud server inputting the voltage residual sequence into a pre-trained fault diagnosis model to obtain fault classification results. The voltage residual sequence can uniformly characterize the fault features of voltage sensors, current sensors, or temperature sensors. This invention solves the problems of difficulty in multi-sensor joint diagnosis and the contradiction between vehicle-side computing power and cloud bandwidth, enabling continuous optimization of the diagnostic model.

Description

Technical Field

[0001] This application relates to the field of electric vehicle battery safety management, specifically to a cloud-based collaborative multi-sensor battery fault detection method and system. Background Technology

[0002] With the rapid development of the electric vehicle (EV) industry, the power battery system, as a core component, directly affects the safety and reliability of the vehicle and the safety of users. The battery management system (BMS) relies on various sensors, such as voltage, current, and temperature, to monitor the battery status in real time. The accuracy of these sensors is a prerequisite for state estimation (such as SOC and SOH) and fault diagnosis. However, under complex operating conditions and long-term operation, sensors are prone to bias, drift, accuracy degradation, or even complete failure. If these faults are not detected and isolated in time, they may lead to serious consequences such as overcharging, over-discharging, or thermal runaway.

[0003] In existing technologies, there are several main solutions for fault detection of battery sensors, but all of them have certain limitations: Hardware redundancy method: This method determines faults by adding extra sensors to compare signals. While simple and direct, it significantly increases the system's hardware cost, wiring complexity, and weight, which is inconsistent with the development trend of lightweighting and cost control in electric vehicles.

[0004] Single-sensor model diagnostic methods primarily rely on equivalent circuit models or electrochemical models, using the residual between the model-predicted voltage and the measured voltage to diagnose voltage sensor faults. However, for current and temperature sensors, due to the strong aperiodicity, randomness, and high susceptibility to operating conditions of their signals, it is difficult to establish accurate predictive models. Existing technologies typically treat current and temperature as input quantities to the model rather than the objects being diagnosed, making it difficult to achieve joint diagnosis of multiple sensors using the same model framework. This often necessitates designing independent diagnostic algorithms for different sensors, resulting in low system integration.

[0005] Embedded AI in Vehicles: Some existing technologies attempt to deploy deep learning models (such as CNN and RNN) on vehicle-mounted BMS chips for fault classification. However, vehicle-mounted computing resources are limited, making it difficult to support real-time inference of large-scale complex models; moreover, once the vehicle-mounted model is fixed, it is difficult to adapt to changes in aging characteristics and the emergence of new fault modes throughout the battery's entire life cycle, lacking online evolution capabilities, resulting in a decrease in diagnostic accuracy over time.

[0006] Cloud-based big data diagnostics: Another approach is to upload all raw sensor data from the vehicle to the cloud for analysis. While the cloud has ample computing power, uploading all raw data (especially high-frequency sampling data) puts significant strain on communication bandwidth, increasing operating costs. Alternatively, in areas with poor network signal, there is a risk of data upload delays or loss, making it difficult to meet the real-time requirements of fault diagnosis.

[0007] In summary, existing technologies suffer from problems such as high hardware redundancy costs, difficulties in multi-sensor joint diagnosis, high model complexity due to limited vehicle-side computing power, and inability of models to adapt to battery aging. Therefore, developing a low-cost, highly accurate sensor fault detection method that can adapt to changes throughout the battery's entire life cycle has become an urgent technical problem to be solved in the field of battery safety management. Summary of the Invention

[0008] This application provides a cloud-based collaborative battery multi-sensor fault detection method, device / system, equipment, and computer-readable storage medium, which can solve the technical problems existing in the prior art, such as the difficulty of multi-sensor joint diagnosis, the contradiction between vehicle-side computing resources and cloud communication bandwidth, and the inability of diagnostic models to adapt to changes throughout the battery's life cycle.

[0009] In a first aspect, embodiments of this application provide a cloud-based collaborative battery multi-sensor fault detection method, applied to a battery management system including a vehicle-side terminal and a cloud server, the method comprising: The vehicle-mounted terminal collects multi-sensor data of the battery system in real time, and the multi-sensor data includes at least voltage data, current data and temperature data. The vehicle terminal inputs the multi-sensor data into a pre-built battery equivalent circuit model to obtain the model predicted voltage value, and calculates and generates a voltage residual sequence based on the measured voltage value and the model predicted voltage value. The vehicle-mounted terminal uploads the voltage residual sequence to the cloud server; The cloud server inputs the received voltage residual sequence into a pre-trained fault diagnosis model to obtain fault classification results; The battery equivalent circuit model uses the current data and temperature data as input parameters to calculate the model's predicted voltage value, so that the voltage residual sequence is generated based on the coupling effect of multi-sensor data, which can uniformly characterize the fault characteristics of voltage sensors, current sensors, or temperature sensors.

[0010] In conjunction with the first aspect, in one embodiment, the battery equivalent circuit model is an RC equivalent circuit model; The model predicts a voltage value that is coupled with the current and temperature data. When the current or temperature sensor fails, the input fault current or temperature data causes a deviation in the model's predicted voltage value, thus making the voltage residual sequence contain the fault characteristics of the current or temperature sensor.

[0011] In conjunction with the first aspect, in one implementation, the pre-trained fault diagnosis model is a deep learning model based on the Transformer architecture.

[0012] In conjunction with the first aspect, in one implementation, the deep learning model based on the Transformer architecture includes an encoder part, removes the decoder part, and connects a fully connected layer and a Softmax classification layer at the output of the encoder part to form a sequence-to-label classification model.

[0013] In conjunction with the first aspect, in one implementation, the method further includes a model update step: The cloud server collects the voltage residual sequence and corresponding diagnostic results uploaded by the vehicle terminal as return data; When the returned data meets the preset conditions, the pre-trained fault diagnosis model is incrementally fine-tuned using the returned data, and the updated model parameters are deployed to the cloud server.

[0014] In conjunction with the first aspect, in one embodiment, before the vehicle-mounted terminal uploads the voltage residual sequence to the cloud server, it further includes: The voltage residual sequence is subjected to time window segmentation and normalization. The data uploaded to the cloud server only includes the processed voltage residual sequence and does not include the original current and temperature data.

[0015] In conjunction with the first aspect, in one implementation, the fault classification result includes sensor type and fault mode; The sensor types include voltage sensors, current sensors, and temperature sensors; the fault modes include at least one of the following: fixed bias fault, drift fault, accuracy degradation fault, and complete failure fault.

[0016] Secondly, embodiments of this application provide a cloud-coordinated multi-sensor battery fault detection system, including a vehicle-side unit and a cloud-side unit; The vehicle-side unit is used to collect multi-sensor data of the battery system in real time, generate a voltage residual sequence using the battery equivalent circuit model, and send the voltage residual sequence to the cloud unit. The voltage residual sequence can uniformly characterize the fault features of voltage sensors, current sensors, or temperature sensors. The cloud unit is used to receive the voltage residual sequence, analyze the voltage residual sequence using a pre-trained fault diagnosis model, and output fault classification results.

[0017] Thirdly, embodiments of this application provide a vehicle-side battery data processing method, applied to a vehicle-side terminal of a battery management system, the method comprising: Real-time acquisition of multi-sensor data from the battery system, wherein the multi-sensor data includes at least voltage data, current data, and temperature data; The multi-sensor data is input into a pre-constructed battery equivalent circuit model to obtain the model predicted voltage value, and a voltage residual sequence is generated based on the measured voltage value and the model predicted voltage value. The voltage residual sequence is uploaded to a cloud server so that the cloud server can perform multi-sensor fault diagnosis based on the voltage residual sequence; The voltage residual sequence is generated based on the coupled influence of the current data and temperature data on the model's predicted voltage value, and is used to uniformly characterize the fault characteristics of voltage sensors, current sensors, or temperature sensors.

[0018] Fourthly, embodiments of this application provide a cloud-based battery fault diagnosis method, applied to a cloud server of a battery management system, the method comprising: The vehicle terminal receives a voltage residual sequence uploaded by the vehicle terminal, the voltage residual sequence being calculated and generated by the vehicle terminal based on multi-sensor data of the battery system and the battery equivalent circuit model; The battery equivalent circuit model uses current data and temperature data as input parameters, so that the voltage residual sequence can uniformly characterize the fault characteristics of voltage sensor, current sensor or temperature sensor. The voltage residual sequence is input into a pre-trained fault diagnosis model to obtain fault classification results.

[0019] The beneficial effects of the technical solutions provided in this application include: Through a cloud-based collaborative architecture, the vehicle-side terminal uses a battery equivalent circuit model to convert multi-sensor data into voltage residual sequences. The cloud server then analyzes these sequences using a pre-trained model, resolving the technical challenges of joint multi-sensor diagnosis, limited vehicle-side computing resources, and the conflict between cloud communication bandwidth. Specifically, this application leverages the coupled influence of current and temperature data on the model's predicted voltage values ​​to achieve a unified characterization of voltage, current, and temperature sensor faults. While maintaining diagnostic accuracy, this significantly reduces the amount of data uploaded, alleviates communication bandwidth pressure, and supports continuous cloud-based optimization of the diagnostic model. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the architecture of an embodiment of the cloud-based collaborative multi-sensor battery fault detection system of this application; Figure 2 This is a schematic diagram of the equivalent circuit model of a second-order RC circuit. Figure 3 This is a schematic diagram illustrating the principle of generating normal voltage residual sequences. Figure 4 This is a schematic diagram illustrating the principle of voltage residual sequence generation caused by voltage sensor failure in this application. Figure 5 This is a schematic diagram illustrating the principle of voltage residual sequence caused by current sensor failure in this application. Figure 6 This is a schematic diagram illustrating the principle of the voltage residual sequence caused by a temperature sensor malfunction in this application. Figure 7 This is a schematic diagram of the traditional Transformer framework structure; Figure 8 This is a schematic diagram of the Transformer framework structure of the present invention. Figure 9 This is a flowchart of the cloud-based collaborative battery multi-sensor fault detection method of this application; Figure 10 This is a flowchart of the vehicle-side battery data processing method of this application; Figure 11 This is a flowchart of the cloud-based battery fault diagnosis method in this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0022] First, some of the technical terms used in this application will be explained to help those skilled in the art understand this application.

[0023] Voltage residual sequence: This refers to the sequence formed by arranging the differences between the measured voltage value of the battery and the predicted voltage value of the battery equivalent circuit model in chronological order. This sequence can reflect the working state of the sensor. When the sensor malfunctions, the voltage residual sequence will exhibit characteristic patterns related to the fault type.

[0024] Cloud-based collaboration: This refers to a technical architecture where vehicle-mounted terminals and cloud servers work together in a collaborative manner. The vehicle-mounted terminal is responsible for data acquisition and lightweight feature extraction (such as voltage residual calculation), while the cloud server handles complex model inference and updates. This architecture aims to balance the limitations of vehicle-mounted computing resources with the constraints of cloud communication bandwidth.

[0025] Coupling effect: This refers to the fact that changes in the values ​​of current and temperature data, which are used as input parameters for the battery equivalent circuit model, affect the calculated voltage prediction results. When a current or temperature sensor malfunctions, the fault data is transmitted to the voltage prediction value through this coupling relationship, leaving identifiable features in the voltage residual sequence.

[0026] Incremental fine-tuning refers to a machine learning method that uses newly collected data to make small-scale parameter adjustments to a trained model. Compared to retraining, incremental fine-tuning has lower computational costs and can preserve the model's existing knowledge, making it suitable for the continuous evolution of diagnostic models in the cloud.

[0027] Sequence-to-Label (Seq-to-Label): This refers to a type of machine learning task that maps an input sequence to a single-class label. The fault diagnosis task in this application belongs to this category, i.e., it takes a voltage residual sequence as input and outputs a fault class label.

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0029] In a first aspect, embodiments of this application provide a cloud-based collaborative battery multi-sensor fault detection system.

[0030] In one embodiment, such as Figure 1 As shown in the figure, this application provides a cloud-coordinated multi-sensor battery fault detection system, which includes a vehicle terminal 101 and a cloud server 102.

[0031] The vehicle-side terminal 101 is deployed in the battery management system (BMS) of the electric vehicle and mainly includes: Sensor module 1011: Used for real-time acquisition of multi-sensor data from the battery system, including at least a voltage sensor, a current sensor, and a temperature sensor. The voltage sensor measures the terminal voltage of a single battery cell or battery pack, the current sensor measures the charging and discharging current of the battery, and the temperature sensor measures the surface temperature of the battery or the ambient temperature.

[0032] Data processing module 1012: This module runs the battery equivalent circuit model, calculates the predicted voltage value based on the collected multi-sensor data, and generates a voltage residual sequence. It also performs preprocessing on the voltage residual sequence, such as time window segmentation and normalization.

[0033] Communication module 1013: Used to upload the processed voltage residual sequence to the cloud server 102. Preferably, this module only uploads the voltage residual sequence and does not upload the original current and temperature data to reduce communication bandwidth usage.

[0034] Cloud server 102 is deployed in a remote data center or cloud platform, and mainly includes: Data receiving module 1021: used to receive voltage residual sequences uploaded from multiple vehicle-end terminals 101.

[0035] Diagnostic Inference Module 1022: Used to run a pre-trained fault diagnosis model, analyze the received voltage residual sequence, and output fault classification results. The fault classification results include sensor type (voltage sensor, current sensor, temperature sensor) and fault mode (fixed bias fault, drift fault, accuracy degradation fault, complete failure fault, etc.).

[0036] Model Management Module 1023: Used to store and manage versions of fault diagnosis models, collect feedback data, and perform incremental fine-tuning and update deployment of models.

[0037] Result Feedback Module 1024: Used to feed back the fault diagnosis results to the vehicle terminal 101 or the relevant operation and maintenance platform so that maintenance or safety measures can be taken in a timely manner.

[0038] It should be noted that, Figure 1 The system architecture shown is only one exemplary embodiment. In other embodiments of this application, the specific components of the vehicle terminal 101 and the cloud server 102 can be adjusted according to the actual application scenario. For example, some functions of the data processing module 1012 can be migrated to the cloud, or the diagnostic reasoning module 1022 can be partially moved to the vehicle terminal to achieve offline basic diagnosis.

[0039] It should also be noted that the communication between the vehicle terminal 101 and the cloud server 102 can use various wireless communication methods such as 4G / 5G cellular networks, Wi-Fi, and satellite communication. In areas with poor communication signals, the vehicle terminal 101 can locally cache the voltage residual sequence and upload it in batches after the network is restored; this application does not impose specific restrictions on this.

[0040] like Figure 2 As shown, in a preferred embodiment of this application, the battery equivalent circuit model adopts a second-order RC equivalent circuit model. This model can better balance computational complexity and voltage prediction accuracy, making it suitable for operation in vehicle-mounted embedded systems.

[0041] It should be noted that although this embodiment preferably uses a second-order RC model, in other embodiments of this application, the battery equivalent circuit model can also be a first-order RC model, a third-order RC model, a Thevenin model, or other physical models capable of describing the dynamic relationship between battery voltage, current, and temperature. As long as the model uses current and temperature data as input parameters to calculate the predicted voltage value, it falls within the protection scope of this application.

[0042] like Figure 3 As shown, under the condition that the battery system and all sensors are working normally, the generation process of the voltage residual sequence is as follows: The vehicle-mounted terminal collects the measured voltage of the battery in real time. Measured current and measured temperature .Will and The voltage is input into the battery equivalent circuit model described above, and the predicted voltage is calculated. Voltage residual . Composed of multiple moments Arranged in chronological order to form a voltage residual sequence. Under normal conditions, due to model errors and measurement noise, It typically fluctuates slightly around zero, with no obvious trend shift or specific waveform pattern.

[0043] This application utilizes the input-output coupling relationship of the battery equivalent circuit model to transform the faults of the current sensor and temperature sensor into characteristics of the voltage residual sequence.

[0044] Specifically, current and temperature It is a key input parameter for the battery equivalent circuit model.

[0045] Current Directly involved in ohmic voltage drop and polarization voltage Calculation; temperature Affecting model parameters For table lookup or calculation.

[0046] Therefore, when the current sensor or temperature sensor malfunctions, the input model's or Data distortion leads to model-predicted voltage This results in a deviation. At this point, even if the measured voltage... It is accurate; the calculated voltage residual is correct. It will also include this deviation information. This allows the voltage residual sequence to uniformly characterize the faults of three types of sensors: voltage, current, and temperature.

[0047] The following combination Figures 4 to 6 The specific characteristics of the three types of faults in the voltage residual sequence are explained respectively.

[0048] like Figure 4 As shown, when the voltage sensor malfunctions (taking a fixed bias fault as an example), the measured voltage... ,in This is the fault bias. The model predicts the voltage at this time. Maintain accuracy based on normal current and temperature input.

[0049] The voltage residual becomes: ; Among them, the voltage residual sequence caused by voltage sensor failure mainly exhibits an overall amplitude shift characteristic. The mean of the residual sequence will deviate significantly from zero, and the shift is related to the fault bias. They are roughly equal. This characteristic is relatively stable under different operating conditions.

[0050] like Figure 5 As shown, when the current sensor malfunctions (taking drift fault as an example), the current input to the model... This will cause the model to predict the voltage. This produces a deviation.

[0051] The voltage residual becomes: ; In particular, because current faults often exhibit time-cumulative effects (such as drift) or are strongly correlated with operating conditions, the resulting voltage residual sequence typically shows a trend of drift or slope change. For example, the mean residual value gradually increases or decreases over time. This is significantly different from the fixed bias characteristics of voltage sensors, making it easier for fault diagnosis models to distinguish between them.

[0052] like Figure 6 As shown, when the temperature sensor malfunctions, the temperature input to the model... Errors caused by this to affect model parameters (such as...) , Table lookup error.

[0053] The voltage residual becomes: ; Because the effect of temperature on battery electrochemical parameters is nonlinear, and the sensitivity to temperature varies under different operating conditions, the voltage residual sequence caused by temperature sensor failure typically exhibits nonlinear fluctuations or operating condition-dependent characteristics. For example, the residual increases significantly under high-current discharge conditions, while it is smaller under quiescent conditions. This characteristic pattern distinguishes it from voltage and current faults.

[0054] To further reduce cloud communication bandwidth and improve the consistency of model input, the vehicle-side terminal performs the following preprocessing steps before uploading the voltage residual sequence: Time window segmentation: The continuous voltage residual sequence is divided into multiple data windows of fixed length. Each window serves as an independent diagnostic sample.

[0055] Normalization: Normalize the data within each window (e.g., Z-Score normalization) to eliminate dimensional differences between different battery cells.

[0056] Data filtering: Only upload the processed voltage residual sequence, not the original current and temperature data.

[0057] Through the above processing, the amount of uploaded data is significantly less than the original multi-sensor data, thereby reducing communication bandwidth usage and traffic costs. Simultaneously, since the voltage residual sequence already contains the fault characteristics of multiple sensors, the cloud model can complete multi-sensor joint diagnosis without requiring the original current / temperature data.

[0058] like Figure 7 As shown, the traditional Transformer model typically employs an encoder-decoder architecture, which mainly includes: Encoder section: Used for feature extraction and encoding of the input sequence; Decoder section: Used to generate the target sequence based on the encoder's output; Attention mechanisms include self-attention and cross-attention, which are used to capture dependencies within a sequence.

[0059] Traditional Transformer models are primarily designed for sequence-to-sequence (Seq-to-Seq) tasks, such as machine translation and text generation, and their output is a target sequence of similar length to the input sequence.

[0060] However, the battery sensor fault diagnosis task in this application belongs to the sequence-to-label classification task, that is, the input is a voltage residual sequence and the output is a single fault category label (such as "voltage sensor - bias fault"). If the traditional Transformer framework is used directly, the decoder part will generate unnecessary computational overhead, increasing cloud inference latency and resource consumption.

[0061] To address the specific characteristics of battery sensor fault diagnosis tasks, this application adaptively optimizes the traditional Transformer model. For example... Figure 8 As shown, the following is the main structure and processing procedure of the optimized model: The one-dimensional voltage residual sequence is used as the input to the model and mapped to a high-dimensional space through an embedding layer to capture richer features. At the same time, positional encoding is introduced to preserve the temporal information of the sequence, enabling the model to identify the relative positions of elements in the sequence and thus better understand temporal dependencies.

[0062] Multi-head self-attention is used to capture the correlation between different time steps in a sequence. This mechanism allows the model to learn the sequence representation in parallel across multiple subspaces.

[0063] For example, different attention heads can focus on different fault characteristic patterns: Some attention heads may focus on the overall bias (i.e., the residuals at all times deviate in the same direction); Some attention heads may focus on the drift slope (i.e., the linear change of the residual over time); Some attention heads may focus on increased high-frequency noise (i.e. fluctuations caused by decreased accuracy).

[0064] This enhances the model's ability to identify potential fault modes in the voltage residual sequence.

[0065] After each self-attention layer, features are further extracted through a feed-forward network, and the training process is stabilized using layer normalization. Layer normalization helps reduce internal covariate bias, accelerates convergence, and improves the model's generalization ability.

[0066] Since this task is a sequence-to-label classification problem, a decoder structure as in sequence-to-sequence tasks is not required. Instead, the model's output layer is designed as one or more fully connected layers followed by a softmax activation function to output the probability of each fault class. This design allows the model to predict individual labels directly from the sequence, simplifying the model structure and improving computational efficiency.

[0067] Through the adaptive optimizations described above, the Transformer model can better adapt to the needs of electric vehicle battery sensor fault diagnosis tasks. These optimizations not only improve the model's performance and accuracy but also enhance its feasibility and efficiency in practical applications.

[0068] The pre-trained fault diagnosis model is trained through the following steps: First, training data is constructed. Multi-sensor data of the battery system under various operating conditions are collected, including normal state data and various fault state data. For fault data, it can be obtained through real vehicle fault data, bench-injected faults (such as bias, drift, accuracy degradation, etc.), or simulation data. The above multi-sensor data is input into the battery equivalent circuit model to calculate and generate the corresponding voltage residual sequence, forming training samples.

[0069] Secondly, a supervised learning method is used to train the fault diagnosis model. The voltage residual sequence of the training samples is input into the model, the model outputs the fault category probability distribution, the loss between the output probability and the true label (such as cross-entropy loss) is calculated, and the model parameters are updated through the backpropagation algorithm to minimize the loss function.

[0070] It should be noted that training terminates when the loss function converges to a preset threshold, the accuracy of the validation set no longer improves, or the preset number of training epochs is reached.

[0071] The cloud server supports the online evolution of the fault diagnosis model. Voltage residual sequences and corresponding diagnostic results uploaded by the vehicle-side terminal are collected as feedback data. When the accumulated feedback data reaches a preset threshold or a new fault mode is detected, the existing model is incrementally fine-tuned using the feedback data. After the fine-tuned model parameters are tested and verified, they are deployed to the inference instance on the cloud server.

[0072] It should be noted that although this embodiment preferably uses the optimized Transformer architecture as the fault diagnosis model, in other embodiments of this application, the pre-trained fault diagnosis model can also be other types of deep learning models, such as Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), or One-Dimensional Convolutional Neural Network (1D-CNN). As long as the model can receive the voltage residual sequence as input and output the classification result of the sensor fault, it falls within the protection scope of this application.

[0073] In a complete embodiment of this application, as Figure 9 As shown, the execution flow of the cloud-based collaborative battery multi-sensor fault detection method is as follows: Step S1: Vehicle-side data acquisition and residual generation.

[0074] The vehicle-mounted terminal collects multi-sensor data from the battery system in real time, including voltage, current, and temperature data. The terminal inputs this multi-sensor data into a pre-built battery equivalent circuit model to obtain the model's predicted voltage value. Based on the measured voltage value and the model's predicted voltage value, the terminal calculates and generates a voltage residual sequence. The battery equivalent circuit model uses current and temperature data as input parameters, ensuring that the voltage residual sequence is generated based on the coupled effects of the multi-sensor data, thus providing a unified characterization of fault features from voltage, current, or temperature sensors.

[0075] Step S2: Vehicle-side data preprocessing and uploading.

[0076] The vehicle-mounted terminal performs time window segmentation and normalization on the voltage residual sequence. The terminal then uploads the processed voltage residual sequence to the cloud server. During this process, the uploaded data only includes the processed voltage residual sequence, excluding the original current and temperature data, to reduce communication bandwidth consumption.

[0077] Step S3: Cloud-based fault diagnosis.

[0078] The cloud server receives the voltage residual sequence uploaded by the vehicle-mounted terminal. The cloud server then inputs the voltage residual sequence into a pre-trained fault diagnosis model. Based on the aforementioned optimized Transformer architecture or other sequence modeling models, the fault diagnosis model analyzes the voltage residual sequence and outputs fault classification results. The fault classification results include sensor type and fault mode.

[0079] Step S4: Results feedback and model update.

[0080] The cloud server feeds back the fault classification results to the vehicle terminal or maintenance platform. Simultaneously, the cloud server collects the voltage residual sequence and corresponding diagnostic results as feedback data. When the feedback data meets preset conditions, the cloud server uses the feedback data to incrementally fine-tune the fault diagnosis model and deploys the updated model parameters to the cloud server, enabling continuous model evolution.

[0081] In another embodiment of this application, a vehicle-side battery data processing method is provided. This method is applied to the vehicle-side terminal of the battery management system, independent of the specific diagnostic process on the cloud server. Figure 10 As shown, the method includes: Step S21: Data acquisition.

[0082] Real-time acquisition of multi-sensor data from the battery system, including at least voltage, current, and temperature data.

[0083] Step S22: Residual generation.

[0084] Multi-sensor data is input into a pre-built battery equivalent circuit model to obtain the model's predicted voltage value. A voltage residual sequence is then calculated based on the measured voltage value and the model's predicted voltage value. This voltage residual sequence is generated based on the coupled influence of current and temperature data on the model's predicted voltage value, and is used to uniformly characterize the fault features of voltage, current, or temperature sensors.

[0085] Step S23: Data upload.

[0086] The voltage residual sequence is uploaded to the cloud server so that the cloud server can perform multi-sensor fault diagnosis based on the voltage residual sequence.

[0087] In another embodiment of this application, a cloud-based battery fault diagnosis method is provided. This method is applied to the cloud server of the battery management system, independent of the specific data collection process of the vehicle-side terminal. Figure 11 As shown, the method includes: Step S31: Data reception.

[0088] The system receives the voltage residual sequence uploaded by the vehicle-mounted terminal. This voltage residual sequence is calculated by the vehicle-mounted terminal based on multi-sensor data from the battery system and a battery equivalent circuit model. The battery equivalent circuit model uses current and temperature data as input parameters, enabling the voltage residual sequence to uniformly characterize the fault features of voltage, current, or temperature sensors.

[0089] Step S32: Fault diagnosis.

[0090] The voltage residual sequence is input into a pre-trained fault diagnosis model to obtain fault classification results.

[0091] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the cloud-based collaborative multi-sensor battery fault detection method as described in any of the foregoing embodiments, or the vehicle-side battery data processing method as described in the foregoing embodiments, or the cloud-based battery fault diagnosis method as described in the foregoing embodiments.

[0092] The electronic device can be a battery management system controller in the vehicle terminal or a computing node in a cloud server.

[0093] This application also provides a computer-readable storage medium storing a computer program thereon. When the program is executed by a processor, it implements the cloud-coordinated multi-sensor battery fault detection method as described in any of the foregoing embodiments, or the vehicle-side battery data processing method as described in the foregoing embodiments, or the cloud-based battery fault diagnosis method as described in the foregoing embodiments.

[0094] Storage media can take many forms, such as ROM, RAM, magnetic disk, and optical disk.

[0095] In summary, the beneficial effects of the technical solutions provided in this application include: Multi-sensor joint diagnosis: By leveraging the coupling effect of the battery equivalent circuit model, current and temperature sensor faults are transformed into voltage residual sequence characteristics, achieving a unified characterization of multi-sensor faults using a single sequence, thus solving the problem of difficulty in diagnosing non-voltage sensors in related technologies.

[0096] Resource and bandwidth optimization: The vehicle only uploads the voltage residual sequence and not the original current and temperature data, which significantly reduces the communication bandwidth usage; the optimized model runs in the cloud, which reduces the computing pressure on the vehicle.

[0097] Model Continuous Evolution: Through cloud-based incremental fine-tuning mechanisms, the diagnostic model can adapt to battery aging and new fault modes, solving the problem of fixed vehicle-mounted models.

[0098] Rights protection friendly: It provides a variety of claim layouts, including system, complete method, vehicle-side single-sided method, cloud-side single-sided method, device and medium, to facilitate rights protection in different infringement scenarios.

[0099] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0100] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0101] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0102] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0103] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0104] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0105] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A cloud-based collaborative multi-sensor battery fault detection method, characterized in that, The method includes: After the vehicle-mounted terminal collects multi-sensor data of the battery system, it inputs the data into the battery equivalent circuit model to obtain the model predicted voltage value, and calculates and generates a voltage residual sequence based on the measured voltage value and the model predicted voltage value; the multi-sensor data includes at least voltage data, current data and temperature data; The vehicle-mounted terminal uploads the voltage residual sequence to the cloud server; The cloud server inputs the received voltage residual sequence into the fault diagnosis model to obtain the fault classification result; The battery equivalent circuit model uses the current data and temperature data as input parameters to calculate the model's predicted voltage value, so that the voltage residual sequence is generated based on the coupling effect of multi-sensor data, which can uniformly characterize the fault characteristics of voltage sensors, current sensors, or temperature sensors.

2. The cloud-based collaborative multi-sensor battery fault detection method according to claim 1, characterized in that, The battery equivalent circuit model is an RC equivalent circuit model; The model predicts a voltage value that is coupled with the current and temperature data. When the current or temperature sensor fails, the input fault current or temperature data causes a deviation in the model's predicted voltage value, thus making the voltage residual sequence contain the fault characteristics of the current or temperature sensor.

3. The cloud-based collaborative multi-sensor battery fault detection method according to claim 1, characterized in that, The fault diagnosis model is a deep learning model based on the Transformer architecture.

4. The cloud-based collaborative multi-sensor battery fault detection method according to claim 3, characterized in that, The deep learning model based on the Transformer architecture includes an encoder part, removes the decoder part, and connects a fully connected layer and a Softmax classification layer at the output of the encoder part to form a sequence-to-label classification model.

5. The cloud-based collaborative multi-sensor battery fault detection method according to claim 1, characterized in that, The method also includes a model update step: The cloud server collects the voltage residual sequence and corresponding diagnostic results uploaded by the vehicle terminal as return data; When the returned data meets the preset conditions, the fault diagnosis model is incrementally fine-tuned using the returned data, and the updated model parameters are deployed to the cloud server.

6. The cloud-based collaborative multi-sensor battery fault detection method according to claim 1, characterized in that, Before the vehicle-mounted terminal uploads the voltage residual sequence to the cloud server, it also includes: The voltage residual sequence is subjected to time window segmentation and normalization. The data uploaded to the cloud server only includes the processed voltage residual sequence and does not include the original current and temperature data.

7. The cloud-based collaborative multi-sensor battery fault detection method according to claim 1, characterized in that, The fault classification results include sensor type and fault mode; The sensor types include voltage sensors, current sensors, and temperature sensors; the fault modes include at least one of the following: fixed bias fault, drift fault, accuracy degradation fault, and complete failure fault.

8. A cloud-based collaborative multi-sensor battery fault detection system, characterized in that, include: The vehicle-mounted terminal is used to collect multi-sensor data from the battery system, input the data into the battery equivalent circuit model, obtain the model predicted voltage value, calculate and generate a voltage residual sequence based on the measured voltage value and the model predicted voltage value, and upload the voltage residual sequence to the cloud server; the multi-sensor data includes at least voltage data, current data and temperature data; A cloud server is used to input the received voltage residual sequence into the fault diagnosis model to obtain fault classification results; The battery equivalent circuit model uses the current data and temperature data as input parameters to calculate the model's predicted voltage value, so that the voltage residual sequence is generated based on the coupling effect of multi-sensor data, which can uniformly characterize the fault characteristics of voltage sensors, current sensors, or temperature sensors.

9. A method for processing vehicle-side battery data, characterized in that, The method, applied to a vehicle-side terminal of a battery management system, includes: Collect multi-sensor data from the battery system, wherein the multi-sensor data includes at least voltage data, current data, and temperature data; The multi-sensor data is input into the battery equivalent circuit model to obtain the model predicted voltage value, and a voltage residual sequence is generated based on the measured voltage value and the model predicted voltage value. The voltage residual sequence is uploaded to a cloud server so that the cloud server can perform multi-sensor fault diagnosis based on the voltage residual sequence; The voltage residual sequence is generated based on the coupled influence of the current data and temperature data on the model's predicted voltage value, and is used to uniformly characterize the fault characteristics of voltage sensors, current sensors, or temperature sensors.

10. A cloud-based battery fault diagnosis method, characterized in that, A cloud server applied to a battery management system, the method comprising: The vehicle terminal receives a voltage residual sequence uploaded by the vehicle terminal, the voltage residual sequence being calculated and generated by the vehicle terminal based on multi-sensor data of the battery system and the battery equivalent circuit model; The battery equivalent circuit model uses current data and temperature data as input parameters, so that the voltage residual sequence can uniformly characterize the fault characteristics of voltage sensor, current sensor or temperature sensor. The voltage residual sequence is input into the fault diagnosis model to obtain the fault classification result.

Images