New energy automobile battery fault intelligent diagnosis method and system

Abstract

The invention relates to the technical field of battery fault detection, and particularly discloses a new energy automobile battery fault intelligent diagnosis method and system, and the method comprises the steps: collecting the three-dimensional time sequence data flow of the voltage, temperature and internal resistance of a battery pack through a distributed sensor array; the environment temperature, the number of charge and discharge cycles and the vehicle operation condition parameters are synchronously integrated as auxiliary diagnosis dimensions, and a multi-source heterogeneous data sensing system is constructed; the method comprises the following steps: preprocessing original data by adopting a dual-channel hybrid filtering architecture based on a multi-source heterogeneous data sensing system, and establishing an adaptive filtering parameter adjustment mechanism to generate multi-modal data; according to the method, multi-dimensional time sequence data and auxiliary diagnosis parameters of the battery pack are comprehensively collected through a constructed multi-source heterogeneous data sensing system, a rich and accurate data basis is provided for fault diagnosis, noise interference is reduced through a two-channel hybrid filtering architecture and a self-adaptive filtering parameter adjustment mechanism, and the fault diagnosis accuracy is improved. And a solid foundation is laid for subsequent feature extraction and model training.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery fault detection, and in particular relates to an intelligent diagnosis method and system for battery faults in new energy vehicles. Background Art

[0002] With the rapid development of new energy vehicles, batteries, as one of their core components, play a vital role. Battery performance directly impacts the vehicle's range, safety, and overall user experience. However, battery failure remains a challenge for the new energy vehicle industry. Battery failure can not only cause the vehicle to malfunction but also pose a potential safety hazard.

[0003] Currently, battery fault diagnosis technology for new energy vehicles primarily relies on traditional monitoring and analysis methods based on battery parameters (such as voltage, current, and temperature). While these methods can detect signs of battery failure to a certain extent, they have certain limitations. For example, traditional fault diagnosis methods often lack the ability to comprehensively judge multiple fault types, and under complex operating conditions, diagnostic accuracy and real-time performance are poor. With the continuous advancement of battery technology, the diversification of battery types, structures, and operating conditions has made traditional methods increasingly unable to meet the battery fault diagnosis requirements of modern new energy vehicles. Furthermore, most existing battery fault diagnosis technologies rely on centralized computing platforms, which cannot achieve rapid and real-time fault location and response. This is especially true when the vehicle is in motion, as it is impossible to rely on centralized servers for big data processing. This requires fault diagnosis systems to be able to respond quickly in an edge computing environment and perform real-time fault diagnosis and prediction.

[0004] Therefore, it is necessary to propose an intelligent diagnosis method and system for new energy vehicle battery faults to solve the problems of accurate positioning and dynamic optimization of battery faults in the existing technology.

[0005] The above information disclosed in this background technology is only for enhancing understanding of the background technology of the present invention and therefore it may contain information that does not constitute the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to provide a method and system for intelligent diagnosis of battery failures in new energy vehicles to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] Intelligent diagnosis method for new energy vehicle battery failure, including:

[0009] A distributed sensor array collects three-dimensional time-series data streams of battery pack voltage, temperature, and internal resistance, and simultaneously integrates ambient temperature, charge and discharge cycle counts, and vehicle operating parameters as auxiliary diagnostic dimensions to build a multi-source heterogeneous data perception system.

[0010] Based on the multi-source heterogeneous data perception system, a dual-channel hybrid filtering architecture is used to pre-process the raw data, and an adaptive filtering parameter adjustment mechanism is established to generate multimodal data;

[0011] Creating a multimodal deep learning architecture based on the multimodal data, performing joint spatial-temporal feature modeling through a tandem CNN-GRU network, and generating a feature vector set that fuses the spatial-temporal features;

[0012] Based on the feature vector set, a dynamic diagnosis model based on meta-learning is established to continuously optimize the model parameters and dynamically improve the fault diagnosis capability;

[0013] Based on the optimized dynamic diagnosis model, a three-level progressive fault diagnosis system is constructed to locate and trace the fault layer by layer;

[0014] Based on the three-level progressive fault diagnosis system, edge computing diagnosis nodes are deployed to perform localized real-time operation of the diagnosis algorithm.

[0015] Preferably, the multi-source heterogeneous data perception system adopts a dual-channel hybrid filtering architecture to pre-process the original data, and establishes an adaptive filtering parameter adjustment mechanism to generate multimodal data, including:

[0016] Dynamic noise elimination is performed on steady-state operating data through improved unscented Kalman filtering to remove random noise interference in a steady-state environment;

[0017] Adaptive wavelet packet decomposition and noise reduction method is used to process transient interference pulses of transient fluctuation data and eliminate the burst noise generated during the transient process of battery charging and discharging;

[0018] Establish an adaptive filter parameter adjustment mechanism based on ambient temperature to dynamically optimize filter parameters to adapt to data characteristics under different working conditions;

[0019] The confidence fusion of the dual-channel output results is performed through the Dempster-Shafer evidence theory to generate the multimodal data with confidence.

[0020] Preferably, the multimodal deep learning architecture is created based on the multimodal data, and spatial-temporal feature joint modeling is performed through a tandem CNN-GRU network to generate a feature vector set that integrates spatial-temporal features, including:

[0021] Using a deep separable convolutional neural network with a spatial pyramid pooling structure, multi-scale spatial distribution feature extraction is performed on the multimodal data to capture the distribution pattern of the battery status in the spatial dimension;

[0022] Among them, the spatial pyramid pooling structure is constructed in the following way:

[0023] A feature pyramid is constructed using a parallel arrangement of 3×3, 5×5, and 7×7 multi-scale convolution kernels. A channel attention mechanism consisting of a global average pooling layer and a fully connected excitation layer guides feature selection, imposes causal convolution constraints in the temporal dimension, and establishes a spatial pyramid pooling structure.

[0024] A gated recurrent unit network is used to capture the long-range temporal dependencies of the multimodal data and analyze the temporal evolution characteristics during the battery performance degradation process;

[0025] The multi-scale spatial distribution features and temporal evolution features are dynamically weighted and focused through the channel attention mechanism to generate a feature vector set that integrates spatial and temporal features.

[0026] Preferably, the method of establishing a dynamic diagnosis model based on meta-learning based on the feature vector set to continuously optimize model parameters and dynamically improve fault diagnosis capabilities includes:

[0027] Through an online incremental learning mechanism driven by a sliding time window, the model parameters corresponding to the feature vector set are updated in real time, and the model is continuously adapted to the dynamic changes of the battery state.

[0028] Utilizing a transfer learning adaptation unit to perform parameter migration and rapid adaptation for different battery models, and loading pre-trained model parameters based on device identification information acquired by the multi-source heterogeneous data perception system to optimize the convergence of the dynamic diagnostic model for new battery models;

[0029] An adversarial generative network is used to construct an anomaly detection enhancement mechanism, rare fault samples are generated based on the feature vector set, and the training data set is expanded to optimize the model's generalization ability to extreme working conditions.

[0030] Preferably, the online incremental learning mechanism is implemented in the following manner:

[0031] A short-term memory update mechanism driven by a sliding time window and a long-term knowledge retention strategy based on elastic weight consolidation are adopted to construct an incremental learning loss function that includes dynamic weight coefficients for memory retention and current task adaptation.

[0032] The transfer learning adaptation unit is implemented in the following way:

[0033] Contrastive learning is used to construct a model-invariant feature space for the battery model feature encoder. The cross-model statistics are aligned using an instance-normalized adaptive batch normalization layer. During the initial training phase, a fine-tuned suppressor is used to freeze the underlying feature extraction network parameters.

[0034] The anomaly detection enhancement mechanism improves the generative adversarial network by:

[0035] A fault sample generator based on Wasserstein distance is constructed, and a multi-discriminator collaborative training mechanism including time domain discriminator and frequency domain discriminator is adopted. The generated sample quality evaluator is constructed through the Fréchet starting distance evaluation index.

[0036] Preferably, the three-level progressive fault diagnosis system includes the following three aspects:

[0037] Level I: A real-time status classifier is established through an improved residual network to determine the battery health status based on the feature vector set to identify whether the battery is in a normal, warning or fault state;

[0038] Level II: Based on the Level I diagnosis, a fault pattern recognition engine is constructed through a time series convolutional network to locate the fault type based on the time series evolution characteristics and the dynamic diagnosis model.

[0039] Level III: Based on the Level II diagnosis, a fault tracing and analysis mechanism is established through the fault propagation path map. Based on the multi-source heterogeneous data perception system and the dynamic diagnosis model, the root cause of the fault is traced and maintenance decision support is provided.

[0040] Preferably, the fault tracing and analysis mechanism integrates:

[0041] (1) Based on the three-dimensional time series data stream and the dynamic diagnosis model parameters, an improved FTA analysis method is used to construct a fault propagation tree generator based on a physical model, the feature vector set is introduced as a constraint condition, the physical model is integrated with the data-driven features, and a composite fault propagation tree including the electrochemical side reaction path, the thermal runaway propagation path, and the mechanical stress diffusion path inside the battery is generated;

[0042] The fault propagation tree node weights are calibrated jointly by the output of Dempster-Shafer evidence theory and the generation results of rare fault samples to perform dynamic credibility assessment of the fault propagation path;

[0043] (2) Based on the fault propagation tree, a data-driven Bayesian network is constructed, in which the nodes correspond to the key physical events in the fault propagation tree, and the edge weights are initialized by the feature contribution output by the channel attention mechanism;

[0044] A dynamic conditional probability table update mechanism is introduced to adjust the inter-node conditional probabilities based on the real-time updated battery status parameters of the online incremental learning mechanism, and the causal reasoning engine is used to dynamically adapt the battery performance degradation process.

[0045] The Bayesian network reasoning results are fed back to the fault propagation tree generator through the transfer learning adaptation unit to correct the parameter uncertainty in the physical model and form a physical-data closed-loop iteration for fault tracing.

[0046] (3) Based on the fault propagation tree and the Bayesian network inference results, a spatiotemporal correlation visualization fault evolution graph interface is developed, which connects the following aspects:

[0047] ① Spatial dimension: Integrating multi-scale features extracted by the spatial pyramid pooling structure, the spatial diffusion path of the fault in the battery pack is displayed in the form of a heat map;

[0048] ② Time dimension: Superimposes time series data after steady-state and transient filtering to show the evolution of faults from early warning to final failure in the form of a time axis;

[0049] ③Causal dimension: Based on the Bayesian network reasoning results, the key causal relationships in the fault propagation path are marked with dynamic arrows, and the confidence of each link is output based on the Dempster-Shafer evidence theory.

[0050] Preferably, the three-level progressive fault diagnosis system is based on deploying edge computing diagnosis nodes to execute localized real-time operation of the diagnosis algorithm, including:

[0051] A three-stage distillation process is performed using model distillation technology to migrate the knowledge of the dynamic diagnosis model trained in the cloud to the diagnosis node model to adapt to the edge computing resource limitations;

[0052] The L1 norm criterion is used to design a channel pruning strategy to remove redundant network parameters. 8-bit fixed-point quantization combined with a dynamic weight sharing mechanism compresses the model storage space, lightweighting the diagnostic node model. The diagnostic algorithm performs millisecond-level localized real-time diagnosis in the battery management system.

[0053] Based on the multi-source heterogeneous data perception system and lightweight diagnosis node model deployment, a complete end-to-end power battery fault intelligent diagnosis process is formed.

[0054] New energy vehicle battery fault intelligent diagnosis system, including:

[0055] A multi-source heterogeneous data perception system is used to collect three-dimensional time-series data streams of battery pack voltage, temperature, and internal resistance in real time, and simultaneously integrate ambient temperature, number of charge and discharge cycles, and vehicle operating condition parameters as auxiliary diagnosis dimensions;

[0056] A dual-channel hybrid filtering architecture is used to filter the collected raw data and adaptively adjust the filtering parameters according to the ambient temperature to generate multimodal data;

[0057] A multimodal deep learning architecture for joint spatial-temporal feature modeling using a cascaded CNN-GRU network to generate a feature vector set that fuses spatial-temporal features.

[0058] A dynamic diagnosis model for continuous optimization and dynamic improvement of fault diagnosis capabilities based on meta-learning combined with the feature vector set;

[0059] A three-level progressive fault diagnosis system is used to locate and trace battery faults layer by layer and provide maintenance decision support;

[0060] Edge computing diagnostic nodes are used to execute localized real-time diagnostic algorithms and provide feedback on diagnostic results.

[0061] Preferably, the three-level progressive fault diagnosis system establishes a fault tracing and analysis mechanism, which constructs a composite fault propagation tree and Bayesian network by integrating physical models and data-driven features, and provides tracing of the root cause of the fault and maintenance decision support.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] The present invention constructs a multi-source heterogeneous data perception system to comprehensively collect multi-dimensional time series data and auxiliary diagnostic parameters of battery packs, providing a rich and accurate data basis for fault diagnosis. Through the dual-channel hybrid filtering architecture and adaptive filtering parameter adjustment mechanism, it reduces noise interference and lays a solid foundation for subsequent feature extraction and model training. Through the serial CNN-GRU network, space-time feature joint modeling is performed to deeply explore the spatial distribution law and time series evolution characteristics of the battery status, generate a highly integrated feature vector set, and enhance the accuracy and reliability of fault diagnosis.

[0064] In addition, the present invention establishes a dynamic diagnosis model based on meta-learning, realizes the continuous optimization of model parameters and the dynamic improvement of fault diagnosis capabilities, enables battery diagnosis to adapt to the changes in battery status under different working conditions, and significantly improves the flexibility and generalization ability of fault diagnosis; through the deployment of a three-level progressive fault diagnosis system and edge computing diagnostic nodes, the layer-by-layer positioning, tracing and localization of faults are realized in real time, which not only improves the efficiency and real-time performance of fault diagnosis, but also provides strong support for maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flow chart of the intelligent diagnosis method for new energy vehicle battery failure of the present invention;

[0066] Figure 2 This is a framework diagram of the new energy vehicle battery fault intelligent diagnosis system of the present invention. DETAILED DESCRIPTION

[0067] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0068] Example 1:

[0069] See also Figure 1 As shown, the intelligent diagnosis method for new energy vehicle battery failure includes:

[0070] A distributed sensor array collects three-dimensional time-series data streams of battery pack voltage, temperature, and internal resistance, and simultaneously integrates ambient temperature, charge and discharge cycle counts, and vehicle operating parameters as auxiliary diagnostic dimensions to build a multi-source heterogeneous data perception system.

[0071] Based on a multi-source heterogeneous data perception system, a dual-channel hybrid filtering architecture is used to pre-process the raw data, and an adaptive filtering parameter adjustment mechanism is established to generate multimodal data;

[0072] Dynamic noise elimination is performed on steady-state operating data through improved unscented Kalman filtering to remove random noise interference in a steady-state environment;

[0073] Adaptive wavelet packet decomposition and noise reduction method is used to process transient interference pulses of transient fluctuation data and eliminate the burst noise generated during the transient process of battery charging and discharging;

[0074] Establish an adaptive filter parameter adjustment mechanism based on ambient temperature to dynamically optimize filter parameters to adapt to data characteristics under different working conditions;

[0075] The Dempster-Shafer evidence theory is used to perform confidence fusion on the dual-channel output results to generate multimodal data with confidence.

[0076] Furthermore, by building a multi-source heterogeneous data perception system, combined with a dual-channel hybrid filtering architecture, an adaptive filter parameter adjustment mechanism, and an improved unscented Kalman filter, noise and interference during battery pack operation are effectively eliminated, improving data accuracy and stability. By dynamically optimizing filter parameters to adapt to changes in different operating conditions, accurate diagnosis and location of battery faults are ultimately achieved. Confidence fusion based on the Dempster-Shafer evidence theory ensures the reliability of diagnostic results and the comprehensive utilization of multimodal data, thereby improving the efficiency and accuracy of battery fault diagnosis.

[0077] Create a multimodal deep learning architecture based on multimodal data, perform joint spatial-temporal feature modeling through a tandem CNN-GRU network, and generate a feature vector set that integrates spatial-temporal features;

[0078] Using a deep separable convolutional neural network with a spatial pyramid pooling structure, we extract multi-scale spatial distribution features from multimodal data and capture the distribution patterns of battery status in the spatial dimension.

[0079] Among them, the spatial pyramid pooling structure is constructed in the following way:

[0080] A feature pyramid is constructed using a parallel arrangement of 3×3, 5×5, and 7×7 multi-scale convolution kernels. A channel attention mechanism consisting of a global average pooling layer and a fully connected excitation layer guides feature selection, imposes causal convolution constraints in the temporal dimension, and establishes a spatial pyramid pooling structure.

[0081] A gated recurrent unit network is used to capture long-range temporal dependencies in multimodal data and analyze the temporal evolution characteristics during battery performance degradation.

[0082] The multi-scale spatial distribution features and temporal evolution features are dynamically weighted and focused through the channel attention mechanism to generate a feature vector set that integrates spatial and temporal features.

[0083] Furthermore, through a deep learning architecture based on multimodal data, combined with a cascaded CNN-GRU network and a spatial pyramid pooling structure, multi-scale features of the battery state in both spatial and temporal dimensions are effectively extracted. Spatial pyramid pooling and a channel attention mechanism capture the characteristics of the battery at different scales and temporal evolution, allowing for precise analysis of the changing patterns in battery performance degradation. Simultaneously, a gated recurrent unit network captures long-range temporal dependencies, improving the accuracy of predicting dynamic changes in the battery state. This feature vector set, which integrates spatial and temporal features, enhances the ability to monitor battery health and improves the accuracy of prediction and diagnosis.

[0084] Based on the feature vector set, a dynamic diagnosis model based on meta-learning is established to continuously optimize the model parameters and dynamically improve the fault diagnosis capability;

[0085] Through an online incremental learning mechanism driven by a sliding time window, the model parameters corresponding to the feature vector set are updated in real time, and the model continuously evolves and adapts to the dynamic changes in the battery state.

[0086] Utilize the transfer learning adaptation unit to perform parameter migration and rapid adaptation for different battery models, and load pre-trained model parameters based on device identification information obtained from a multi-source heterogeneous data perception system to optimize the convergence of the dynamic diagnosis model for new battery models.

[0087] A generative adversarial network is used to construct an enhanced anomaly detection mechanism, rare fault samples are generated based on the feature vector set, and the training data set is expanded to optimize the model's generalization ability to extreme working conditions.

[0088] Furthermore, a meta-learning-based dynamic diagnosis model, combined with online incremental learning and transfer learning, continuously optimizes the battery health diagnosis model, improving the real-time performance and accuracy of fault detection. A sliding time window-driven mechanism enables the model to adapt to dynamic changes in battery status in real time, while transfer learning ensures rapid adaptation and model parameter migration for different battery models. A generative adversarial network enhances anomaly detection by generating rare fault samples, expanding the training dataset, and optimizing the model's generalization to extreme operating conditions, thereby improving the accuracy and robustness of battery health diagnosis.

[0089] The online incremental learning mechanism is implemented in the following ways:

[0090] A short-term memory update mechanism driven by a sliding time window and a long-term knowledge retention strategy based on elastic weight consolidation are adopted to construct an incremental learning loss function that includes dynamic weight coefficients for memory retention and current task adaptation.

[0091] The transfer learning adaptation unit is implemented in the following ways:

[0092] Contrastive learning is used to construct a model-invariant feature space for the battery model feature encoder. The cross-model statistics are aligned using an instance-normalized adaptive batch normalization layer. During the initial training phase, a fine-tuned suppressor is used to freeze the underlying feature extraction network parameters.

[0093] The anomaly detection enhancement mechanism improves the generative adversarial network by:

[0094] A fault sample generator based on Wasserstein distance is constructed, and a multi-discriminator collaborative training mechanism including time domain discriminator and frequency domain discriminator is adopted. The generated sample quality evaluator is constructed through the Fréchet starting distance evaluation index.

[0095] Furthermore, an online incremental learning mechanism enables dynamic knowledge updating and long-term memory retention, enhancing the battery diagnostic model's adaptability to new data. The transfer learning adaptation unit achieves rapid adaptation between different battery models by constructing a battery model-invariant feature space and fine-tuning strategies, improving the model's cross-model diagnostic capabilities. The anomaly detection enhancement mechanism generates fault samples by improving the generative adversarial network, combining multi-discriminator collaborative training and Wasserstein distance, improving the quality of generated samples and the model's ability to detect extreme fault conditions, thereby enhancing the accuracy and robustness of battery fault diagnosis.

[0096] Based on the optimized dynamic diagnosis model, a three-level progressive fault diagnosis system is constructed to locate and trace faults layer by layer.

[0097] The three-level progressive fault diagnosis system includes the following three aspects:

[0098] Level I: A real-time status classifier is established through an improved residual network to determine the battery health status based on a feature vector set, identifying whether the battery is in a normal, warning, or fault state.

[0099] Level II: Based on Level I diagnosis, a fault pattern recognition engine is built through a time series convolutional network to locate the fault type based on time series evolution characteristics and dynamic diagnosis models;

[0100] Level III: Based on the Level II diagnosis, a fault tracing and analysis mechanism is established through the fault propagation path map. Based on the multi-source heterogeneous data perception system and dynamic diagnosis model, the root cause of the fault is traced and maintenance decision support is provided.

[0101] Furthermore, this three-level progressive fault diagnosis system improves the accuracy and operability of battery fault diagnosis by gradually locating and tracing faults in a hierarchical manner. Level I uses an improved residual network to achieve real-time health status classification, promptly identifying whether the battery is in normal, warning, or faulty state. Level II uses a time-series convolutional network to accurately identify fault modes and locate the fault type. Level III uses fault propagation path maps to trace the root cause, combining multi-source data analysis to trace the root cause and provide maintenance decision support, ensuring the comprehensiveness, accuracy, and maintainability of battery fault diagnosis.

[0102] Fault tracing and analysis mechanism integration:

[0103] (1) Based on the three-dimensional time series data stream and dynamic diagnosis model parameters, an improved FTA analysis method is used to construct a fault propagation tree generator based on the physical model. The feature vector set is introduced as a constraint condition, and the physical model is integrated with the data-driven features to generate a composite fault propagation tree that includes the electrochemical side reaction path, thermal runaway propagation path, and mechanical stress diffusion path inside the battery.

[0104] The node weights of the fault propagation tree are jointly calibrated by the output of the Dempster-Shafer evidence theory and the generation results of rare fault samples to perform dynamic credibility assessment of the fault propagation path;

[0105] (2) Based on the fault propagation tree, a data-driven Bayesian network is constructed, where the nodes correspond to the key physical events in the fault propagation tree, and the edge weights are initialized by the feature contribution output by the channel attention mechanism;

[0106] A dynamic conditional probability table update mechanism is introduced to adjust the inter-node conditional probabilities based on the real-time updated battery status parameters of the online incremental learning mechanism, and the causal reasoning engine is used to dynamically adapt the battery performance degradation process.

[0107] The Bayesian network reasoning results are fed back to the fault propagation tree generator through the transfer learning adaptation unit to correct the parameter uncertainty in the physical model and form a physical-data closed-loop iteration for fault tracing.

[0108] (3) Based on the fault propagation tree and Bayesian network reasoning results, a spatiotemporal correlation visualization fault evolution map interface is developed. This interface connects the following aspects:

[0109] ① Spatial dimension: Integrating multi-scale features extracted by the spatial pyramid pooling structure, the spatial diffusion path of the fault in the battery pack is displayed in the form of a heat map;

[0110] ② Time dimension: Superimposes time series data after steady-state and transient filtering to show the evolution of faults from early warning to final failure in the form of a time axis;

[0111] ③Causal dimension: Based on the Bayesian network reasoning results, the key causal relationships in the fault propagation path are marked with dynamic arrows, and the confidence level of each link is output based on the Dempster-Shafer evidence theory.

[0112] Furthermore, this fault tracing and analysis mechanism integrates physical models with data-driven features to construct a multi-level fault propagation tree and Bayesian network. This dynamically assesses the propagation path and credibility of battery faults and updates battery status parameters in real time for accurate predictions. By visualizing fault evolution maps through spatiotemporal correlation, the spatial, temporal, and causal dimensions of faults are clearly displayed, helping users accurately track fault evolution, optimize fault diagnosis and early warning, and improve the accuracy and reliability of battery fault detection.

[0113] Based on a three-level progressive fault diagnosis system, edge computing diagnosis nodes are deployed to execute localized real-time operation of diagnostic algorithms;

[0114] A three-stage distillation process is performed using model distillation technology to migrate the knowledge of the dynamic diagnosis model trained in the cloud to the diagnosis node model to adapt to the limitations of edge computing resources;

[0115] The L1 norm criterion is used to design a channel pruning strategy to remove redundant network parameters. 8-bit fixed-point quantization combined with a dynamic weight sharing mechanism compresses the model storage space, lightweighting the diagnostic node model. The diagnostic algorithm performs millisecond-level localized real-time diagnosis in the battery management system.

[0116] Based on the multi-source heterogeneous data perception system and lightweight diagnosis node model deployment, a complete end-to-end power battery fault intelligent diagnosis process is formed.

[0117] Furthermore, this three-level progressive fault diagnosis system combines edge computing with lightweight models. Through model distillation and channel pruning, it optimizes the computational and storage efficiency of diagnostic nodes, enabling millisecond-level localized real-time diagnosis in the battery management system. By deploying a multi-source heterogeneous data perception system, it forms an efficient end-to-end intelligent power battery fault diagnosis process, enabling accurate and rapid fault detection and prediction on resource-limited edge devices, effectively improving the real-time and reliability of battery fault diagnosis.

[0118] Example 2:

[0119] See also Figure 2 As shown in the figure, the new energy vehicle battery fault intelligent diagnosis system includes:

[0120] A multi-source heterogeneous data perception system is used to collect three-dimensional time-series data streams of battery pack voltage, temperature, and internal resistance in real time, and simultaneously integrate ambient temperature, number of charge and discharge cycles, and vehicle operating condition parameters as auxiliary diagnosis dimensions;

[0121] A dual-channel hybrid filtering architecture is used to filter the collected raw data and adaptively adjust the filtering parameters according to the ambient temperature to generate multimodal data;

[0122] A multimodal deep learning architecture for joint spatial-temporal feature modeling using a cascaded CNN-GRU network to generate a feature vector set that fuses spatial-temporal features.

[0123] Dynamic diagnostic model, used for continuous optimization and dynamic improvement of fault diagnosis capabilities based on meta-learning combined with feature vector sets;

[0124] A three-level progressive fault diagnosis system is used to locate and trace battery faults layer by layer and provide maintenance decision support;

[0125] Establish a fault tracing and analysis mechanism. By integrating physical models with data-driven features, the fault tracing and analysis mechanism constructs a composite fault propagation tree and Bayesian network to provide root cause tracing and maintenance decision support.

[0126] Edge computing diagnostic nodes are used to execute localized real-time diagnostic algorithms and provide feedback on diagnostic results.

[0127] Application example: Real-time monitoring and fault warning of new energy vehicle battery fault intelligent diagnosis system

[0128] With the rapid development of the new energy vehicle market, users are placing higher demands on vehicle safety and reliability. As a core component of new energy vehicles, real-time monitoring of battery operating status and fault warnings are particularly important. A new energy vehicle manufacturer decided to integrate an intelligent battery fault diagnosis system into its new electric vehicles. This system comprehensively monitors the battery system's operating status and provides timely warnings, improving user experience and vehicle safety.

[0129] 1. System Integration and Deployment

[0130] (1) Multi-source heterogeneous data perception system:

[0131] A distributed sensor array, including high-precision voltage sensors, temperature sensors, and internal resistance sensors, is installed inside the battery pack and at key locations to ensure that the three-dimensional time series data stream of the battery pack can be collected in real time and accurately.

[0132] An ambient temperature sensor is integrated outside the vehicle to monitor changes in ambient temperature during vehicle operation. At the same time, the number of charge and discharge cycles and vehicle operating parameters (such as speed, acceleration, etc.) are obtained through the vehicle CAN bus as auxiliary diagnostic dimensions.

[0133] (2) Data transmission and processing:

[0134] The collected data is transmitted to the battery management system (BMS) in real time via the vehicle network. After preliminary processing of the data, the BMS uploads the key data to the cloud server for in-depth analysis and storage.

[0135] The cloud server has deployed an intelligent diagnosis system for new energy vehicle battery faults, which includes a data preprocessing module, a feature extraction and modeling module, a fault diagnosis and positioning module, and an edge computing and real-time response module.

[0136] 2. Real-time monitoring and fault warning process

[0137] (1) Data preprocessing:

[0138] After the cloud server receives the data uploaded by the BMS, it first preprocesses the raw data using a dual-channel hybrid filtering architecture. Through the improved unscented Kalman filtering and adaptive wavelet packet decomposition denoising method, it effectively removes steady-state and transient noise interference and improves data quality.

[0139] An adaptive filtering parameter adjustment mechanism is established based on the ambient temperature to ensure accurate preprocessing data under different working conditions.

[0140] (2) Feature extraction and modeling:

[0141] Utilizing a multimodal deep learning architecture, the preprocessed data is jointly modeled with spatial and temporal features through a tandem CNN-GRU network to generate a feature vector set that integrates spatial and temporal features.

[0142] These feature vector sets are input into a meta-learning-based dynamic diagnosis model, which can continuously optimize model parameters to adapt to the dynamic changes in battery status.

[0143] (3) Real-time monitoring and fault diagnosis:

[0144] Level I Diagnosis: The system uses an improved residual network to classify the battery's health status in real time, determining whether the battery is in a normal, warning, or fault state. If an anomaly is detected, an early warning mechanism is immediately triggered.

[0145] Level II diagnosis: Based on level I diagnosis, the system uses a temporal convolutional network to accurately identify fault modes and locate specific fault types, such as battery cell overheating and abnormal increase in internal resistance.

[0146] Level III diagnosis: The system performs fault tracing analysis through fault propagation path maps, traces the root cause of the fault, and provides maintenance decision support.

[0147] (4) Fault warning and notification:

[0148] When the system detects battery failure or potential risks, it immediately sends warning information to the driver and vehicle manufacturer through the on-board display, mobile phone APP and cloud management platform.

[0149] The warning information includes the fault type, severity, recommended maintenance measures and estimated maintenance time, ensuring that the driver can understand the battery status in a timely manner and take appropriate measures.

[0150] (5) Edge computing and real-time response:

[0151] To ensure rapid response while the vehicle is driving, the system deploys edge computing diagnostic nodes in the BMS.

[0152] Model distillation technology migrates the knowledge of dynamic diagnostic models trained in the cloud to diagnostic node models, enabling localized real-time operation of diagnostic algorithms. This ensures accurate and timely fault diagnosis even without an internet connection.

[0153] 3. Application Effect and User Feedback

[0154] Improved safety: Through real-time monitoring and fault warning, the system effectively reduces the risk of vehicle breakdown and safety accidents caused by battery failure, and enhances user confidence in vehicle safety.

[0155] Improve maintenance efficiency: The fault tracing analysis function provides maintenance personnel with detailed maintenance guides and spare parts lists, significantly shortening maintenance time and reducing maintenance costs.

[0156] Enhanced user experience: Users can check battery status and receive fault warning information at any time through the mobile phone app, which improves the convenience and comfort of use.

[0157] To sum up, the new energy vehicle battery fault intelligent diagnosis system plays an important role in the specific use of the battery, realizes real-time monitoring of the battery system operating status and fault warning, and provides a strong guarantee for the safe operation of new energy vehicles.

[0158] Example 3:

[0159] The present invention also provides a computer-readable storage medium having a program for any of the above-described intelligent diagnostic methods for new energy vehicle batteries stored thereon. When executed by a processor, the program implements each of the above-described intelligent diagnostic methods and achieves the same technical effects. To avoid repetition, the details are omitted here. The computer-readable storage medium may be, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0160] In the description of this specification, the reference terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0161] The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to common designs. In the absence of conflicts, the same embodiment and different embodiments of the present invention may be combined with each other.

[0162] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0163] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent diagnostic method for new energy vehicle battery failure, characterized in that: include: A distributed sensor array collects three-dimensional time-series data streams of battery pack voltage, temperature, and internal resistance, and simultaneously integrates ambient temperature, charge and discharge cycle counts, and vehicle operating parameters as auxiliary diagnostic dimensions to build a multi-source heterogeneous data perception system. Based on the multi-source heterogeneous data perception system, a dual-channel hybrid filtering architecture is used to pre-process the raw data, and an adaptive filtering parameter adjustment mechanism is established to generate multimodal data; Creating a multimodal deep learning architecture based on the multimodal data, performing joint spatial-temporal feature modeling through a tandem CNN-GRU network, and generating a feature vector set that fuses the spatial-temporal features; Based on the feature vector set, a dynamic diagnosis model based on meta-learning is established to continuously optimize the model parameters and dynamically improve the fault diagnosis capability; Based on the optimized dynamic diagnosis model, a three-level progressive fault diagnosis system is constructed to locate and trace the fault layer by layer; Based on the three-level progressive fault diagnosis system, edge computing diagnosis nodes are deployed to perform localized real-time operation of the diagnosis algorithm.

2. The intelligent diagnosis method for new energy vehicle battery failure according to claim 1 is characterized in that: Based on the multi-source heterogeneous data perception system, a dual-channel hybrid filtering architecture is used to pre-process the original data, and an adaptive filtering parameter adjustment mechanism is established to generate multimodal data, including: Dynamic noise elimination is performed on steady-state operating data through improved unscented Kalman filtering to remove random noise interference in a steady-state environment; Adaptive wavelet packet decomposition and noise reduction method is used to process transient interference pulses of transient fluctuation data and eliminate the burst noise generated during the transient process of battery charging and discharging; Establish an adaptive filter parameter adjustment mechanism based on ambient temperature to dynamically optimize filter parameters to adapt to data characteristics under different working conditions; The confidence fusion of the dual-channel output results is performed through the Dempster-Shafer evidence theory to generate the multimodal data with confidence.

3. The intelligent diagnosis method for new energy vehicle battery failure according to claim 2 is characterized in that: The method of creating a multimodal deep learning architecture based on the multimodal data, performing joint spatial-temporal feature modeling through a tandem CNN-GRU network, and generating a feature vector set integrating spatial-temporal features includes: Using a deep separable convolutional neural network with a spatial pyramid pooling structure, multi-scale spatial distribution feature extraction is performed on the multimodal data to capture the distribution pattern of the battery status in the spatial dimension; Among them, the spatial pyramid pooling structure is constructed in the following way: A feature pyramid is constructed using a parallel arrangement of 3×3, 5×5, and 7×7 multi-scale convolution kernels. A channel attention mechanism consisting of a global average pooling layer and a fully connected excitation layer guides feature selection, imposes causal convolution constraints in the temporal dimension, and establishes a spatial pyramid pooling structure. A gated recurrent unit network is used to capture the long-range temporal dependencies of the multimodal data and analyze the temporal evolution characteristics during the battery performance degradation process; The multi-scale spatial distribution features and temporal evolution features are dynamically weighted and focused through the channel attention mechanism to generate a feature vector set that integrates spatial and temporal features.

4. The intelligent diagnosis method for new energy vehicle battery faults according to claim 3 is characterized in that: The method of establishing a dynamic diagnosis model based on meta-learning based on the feature vector set to continuously optimize model parameters and dynamically improve fault diagnosis capabilities includes: Through an online incremental learning mechanism driven by a sliding time window, the model parameters corresponding to the feature vector set are updated in real time, and the model is continuously adapted to the dynamic changes of the battery state. Utilizing a transfer learning adaptation unit to perform parameter migration and rapid adaptation for different battery models, and loading pre-trained model parameters based on device identification information acquired by the multi-source heterogeneous data perception system to optimize the convergence of the dynamic diagnostic model for new battery models; An adversarial generative network is used to construct an anomaly detection enhancement mechanism, rare fault samples are generated based on the feature vector set, and the training data set is expanded to optimize the model's generalization ability to extreme working conditions.

5. The intelligent diagnosis method for new energy vehicle battery fault according to claim 4 is characterized in that: The online incremental learning mechanism is implemented in the following ways: A short-term memory update mechanism driven by a sliding time window and a long-term knowledge retention strategy based on elastic weight consolidation are adopted to construct an incremental learning loss function that includes dynamic weight coefficients for memory retention and current task adaptation. The transfer learning adaptation unit is implemented in the following way: Contrastive learning is used to construct a model-invariant feature space for the battery model feature encoder. The cross-model statistics are aligned using an instance-normalized adaptive batch normalization layer. During the initial training phase, a fine-tuned suppressor is used to freeze the underlying feature extraction network parameters. The anomaly detection enhancement mechanism improves the generative adversarial network by: A fault sample generator based on Wasserstein distance is constructed, and a multi-discriminator collaborative training mechanism including time domain discriminator and frequency domain discriminator is adopted. The generated sample quality evaluator is constructed through the Fréchet starting distance evaluation index.

6. The intelligent diagnosis method for new energy vehicle battery failure according to claim 5 is characterized in that: The three-level progressive fault diagnosis system includes the following three aspects: Level I: A real-time status classifier is established through an improved residual network to determine the battery health status based on the feature vector set to identify whether the battery is in a normal, warning or fault state; Level II: Based on the Level I diagnosis, a fault pattern recognition engine is constructed through a time series convolutional network to locate the fault type based on the time series evolution characteristics and the dynamic diagnosis model. Level III: Based on the Level II diagnosis, a fault tracing and analysis mechanism is established through the fault propagation path map. Based on the multi-source heterogeneous data perception system and the dynamic diagnosis model, the root cause of the fault is traced and maintenance decision support is provided.

7. The intelligent diagnosis method for new energy vehicle battery failure according to claim 6 is characterized in that: The fault tracing and analysis mechanism integrates: (1) Based on the three-dimensional time series data stream and the dynamic diagnosis model parameters, an improved FTA analysis method is used to construct a fault propagation tree generator based on a physical model, the feature vector set is introduced as a constraint condition, the physical model is integrated with the data-driven features, and a composite fault propagation tree including the electrochemical side reaction path, the thermal runaway propagation path, and the mechanical stress diffusion path inside the battery is generated; The fault propagation tree node weights are calibrated jointly by the output of Dempster-Shafer evidence theory and the generation results of rare fault samples to perform dynamic credibility assessment of the fault propagation path; (2) Based on the fault propagation tree, a data-driven Bayesian network is constructed, in which the nodes correspond to the key physical events in the fault propagation tree, and the edge weights are initialized by the feature contribution output by the channel attention mechanism; A dynamic conditional probability table update mechanism is introduced to adjust the inter-node conditional probabilities based on the real-time updated battery status parameters of the online incremental learning mechanism, and the causal reasoning engine is used to dynamically adapt the battery performance degradation process. The Bayesian network inference results are fed back to the fault propagation tree generator through the transfer learning adaptation unit to correct the parameter uncertainty in the physical model and form a physical-data closed-loop iteration for fault tracing. (3) Based on the fault propagation tree and the Bayesian network inference results, a spatiotemporal correlation visualization fault evolution graph interface is developed, which connects the following aspects: ① Spatial dimension: The multi-scale features extracted by the spatial pyramid pooling structure are integrated to display the spatial diffusion path of the fault in the battery pack in the form of a heat map; ② Time dimension: Superimposes time series data after steady-state and transient filtering to show the evolution of faults from early warning to final failure in the form of a time axis; ③Causal dimension: Based on the Bayesian network reasoning results, the key causal relationships in the fault propagation path are marked with dynamic arrows, and the confidence of each link is output based on the Dempster-Shafer evidence theory.

8. The intelligent diagnosis method for new energy vehicle battery faults according to claim 7 is characterized in that: Based on the three-level progressive fault diagnosis system, edge computing diagnosis nodes are deployed to perform localized real-time operation of the diagnosis algorithm, including: A three-stage distillation process is performed using model distillation technology to migrate the knowledge of the dynamic diagnosis model trained in the cloud to the diagnosis node model to adapt to the edge computing resource limitations; The L1 norm criterion is used to design a channel pruning strategy to remove redundant network parameters. 8-bit fixed-point quantization combined with a dynamic weight sharing mechanism compresses the model storage space, lightweighting the diagnostic node model. The diagnostic algorithm performs millisecond-level localized real-time diagnosis in the battery management system. Based on the multi-source heterogeneous data perception system and lightweight diagnosis node model deployment, a complete end-to-end power battery fault intelligent diagnosis process is formed.

9. New energy vehicle battery fault intelligent diagnosis system, characterized by: include: A multi-source heterogeneous data perception system is used to collect three-dimensional time-series data streams of battery pack voltage, temperature, and internal resistance in real time, and simultaneously integrate ambient temperature, number of charge and discharge cycles, and vehicle operating condition parameters as auxiliary diagnosis dimensions; A dual-channel hybrid filtering architecture is used to filter the collected raw data and adaptively adjust the filtering parameters according to the ambient temperature to generate multimodal data; A multimodal deep learning architecture for joint spatial-temporal feature modeling using a cascaded CNN-GRU network to generate a feature vector set that fuses spatial-temporal features. A dynamic diagnosis model for continuous optimization and dynamic improvement of fault diagnosis capabilities based on meta-learning combined with the feature vector set; A three-level progressive fault diagnosis system is used to locate and trace battery faults layer by layer and provide maintenance decision support; Edge computing diagnostic nodes are used to execute localized real-time diagnostic algorithms and provide feedback on diagnostic results.

10. The intelligent diagnosis system for new energy vehicle battery faults according to claim 9 is characterized in that: The three-level progressive fault diagnosis system establishes a fault tracing and analysis mechanism. The fault tracing and analysis mechanism integrates physical models and data-driven features to construct a composite fault propagation tree and Bayesian network, providing tracing of the root cause of the fault and maintenance decision support.

Images