Battery fault early-stage pre-detection method with multi-working-condition adaptation and multi-type identification

Abstract

The invention discloses a multi-working-condition adaptive and multi-type identification battery fault early-stage pre-detection method, and belongs to the field of battery safety management. Firstly, a data pool containing normal and various fault data is constructed, and then a double-layer combined diagnosis strategy is adopted: the first layer is based on adaptive threshold diagnosis of a data-driven model, and faults are identified through voltage prediction residual errors; in the second layer, mutual information statistical analysis is utilized to detect data outliers. And results of the two layers are jointly judged through OR logic, so that the detection sensitivity and the coverage rate are improved. And finally, fault data are classified in combination with an LOF algorithm, the problems that a traditional method is poor in adaptability to complex working conditions, insensitive to early-stage tiny faults and insufficient in classification capacity are effectively solved, and high-precision and low-false-alarm battery safety early warning is achieved.

Description

Technical Field

[0001] This invention belongs to the field of safety management technology for power batteries of new energy vehicles, and specifically relates to a data-driven non-invasive early fault diagnosis and classification method for lithium-ion batteries. Background Technology

[0002] With the rapid development of the electric vehicle industry, the reliability and safety of power battery systems have increasingly become a technological focus. Lithium-ion batteries, due to their high energy density, long cycle life, and low self-discharge rate, have become the mainstream choice for on-board power batteries. However, as a highly complex closed electrochemical system, its inherent time-varying characteristics and the complexity of electric vehicle operating conditions constantly challenge its reliability. Therefore, under non-invasive conditions, accurately identifying early battery faults and latent damage based on the battery's own signal characteristics is crucial for the safe operation of electric vehicles.

[0003] Currently, with the widespread application of lithium-ion batteries in electric vehicles and energy storage systems, their fault diagnosis technology has received significant attention due to its crucial role in system safety and reliability. Existing diagnostic methods are mainly divided into two categories: data-driven and statistical analysis-based. Data-driven methods use algorithms such as machine learning and neural networks to uncover hidden features in battery operating data, but they are highly dependent on the quality and quantity of training samples and may suffer from insufficient generalization ability under complex and variable operating conditions. Statistical analysis-based methods focus on the statistical regularities of parameters such as voltage and temperature. While computationally efficient, they are less sensitive to early-stage faults and struggle to address hidden failures caused by gradual aging within the battery. Therefore, balancing the accuracy and timeliness of diagnostic models under limited sensor data has become a core challenge in the field of lithium-ion battery fault diagnosis.

[0004] To address this issue, the research team proposed a hybrid fault diagnosis framework that integrates data-driven deep feature extraction capabilities with the real-time advantages of statistical analysis methods. Through the collaborative operation of a multi-dimensional parameter coupling analysis layer, a fault feature clustering layer, and a dynamic threshold determination layer, the framework significantly improves the early identification rate of faults while ensuring the system's rapid response capability under sudden faults, providing a new technical path for battery safety management. Summary of the Invention

[0005] This invention aims to overcome the shortcomings of existing technologies and provide an early detection method for battery faults that can adapt to complex operating conditions and accurately identify various fault types. The core of this method lies in a two-layer joint diagnostic architecture. First, this invention constructs a data pool containing normal and fault data, with fault data covering internal short circuits, poor terminal contact, and excessive aging faults. The diagnostic process is as follows: A data-driven method is used to predict the battery voltage. By calculating the residual between the predicted and actual values ​​and comparing it with an adaptive threshold, the first layer of rapid diagnosis based on the adaptive threshold is achieved. A statistical analysis method is used to identify statistical outliers as potential faults by analyzing the statistical dependence between the tested battery and normal batteries at the same temperature. The fault dataset is obtained by combining the adaptive threshold detection method and the statistical analysis method. Finally, a fault type identification model based on the LOF algorithm is used to distinguish between internal short circuit faults, terminal contact faults, and excessive aging faults.

[0006] To solve the above technical problems, the present invention is implemented using the following technical solution:

[0007] A method for early detection of battery faults that adapts to multiple operating conditions and identifies multiple types includes the following steps:

[0008] S1: Construct a data pool containing normal battery data and faulty battery data, wherein the fault types include internal short circuit faults, terminal contact faults, and excessive aging faults;

[0009] S2: A data-driven model is used to perform the first-level threshold diagnosis on the data in the data pool. The fault is determined by comparing the residual between the model-predicted voltage value and the actual voltage value to see if it exceeds the preset threshold.

[0010] S3: The mutual information algorithm is used to perform a second-level diagnosis on the data in the data pool. Faults are determined by calculating the mutual information value between data and identifying outliers.

[0011] S4: Based on the diagnostic results of S2 and S3, a joint judgment is made. If a fault is detected by either method, it is determined to be fault data.

[0012] S5: The fault type identification model based on the LOF algorithm is used to classify the fault data identified by S4, distinguishing between internal short circuit faults, terminal contact faults and excessive aging faults.

[0013] In a preferred embodiment, the construction of the data pool in step S1 includes:

[0014] S11: Internal short-circuit fault data is generated by introducing a short-circuit resistor in the second-order equivalent circuit model, as shown in formulas (1)-(4):

[0015] (1)

[0016] (2)

[0017] (3)

[0018] (4)

[0019] Wherein, OCV is an ideal voltage source. For ohm resistance, Polarization resistor, Polarized capacitor, For diffusion resistance, For diffusion capacitance, Terminal voltage, for Voltage across the resistor for Voltage across the resistor For flowing through The current through the resistor, For flowing through The current flowing through the short-circuited resistor;

[0020] S12: Terminal contact fault data is generated by introducing noise factors into the normal discharge curve;

[0021] S13: Excessive aging failure data is derived from actual aging test data.

[0022] In a preferred embodiment, the data-driven model in step S2 includes:

[0023] S21: The input layer receives timing data. The input variables include time, current, SOC, temperature, and SOH. The structure of the input data is shown in (5):

[0024] (5)

[0025] in express Battery current at any given moment express time, Indicates time The calculation of the battery management system data, Indicates time Battery temperature, Indicates time The battery health level, Indicates time Model input data;

[0026] S22: The first LSTM layer is used to capture local timing patterns. It consists of multiple LSTM units connected in series. Each unit regulates the information flow through an input gate, a forget gate, and an output gate. Output hidden state. and memory information ;

[0027] S23: The second LSTM layer is used to extract global temporal features and receives the hidden state output from the first layer. Further extract higher-order features and output ;

[0028] S24: The output layer outputs the predicted voltage value. The structure of the output data is as follows:

[0029] (6)

[0030] In the above formula Represents the model input data The prediction results.

[0031] In a preferred embodiment, the mutual information diagnosis in step S3 includes:

[0032] S31: Divide the data into different groups according to temperature;

[0033] S32: Select normal data within each temperature group as the baseline for mutual information calculation;

[0034] S33: Calculate the mutual information value between the data to be diagnosed and the reference data using the following formulas (7)-(10):

[0035] (7)

[0036] Another form of mutual information is:

[0037] (8)

[0038] (9)

[0039] (10)

[0040] in and They are respectively and Information entropy for and The joint entropy;

[0041] S34: Identify outliers using frequency distribution histograms, mutual information heatmaps, and quantile-quantile plots.

[0042] In a preferred embodiment, the joint determination and diagnosis in step S4 includes:

[0043] S41: If the threshold diagnosis or mutual information diagnosis result is a fault, it is determined to be faulty data.

[0044] In the preferred technical solution, the fault type identification model in step S5 includes:

[0045] S51: Input the fault data identified in step S4 into the data proposed in step S5, and use the following LOF algorithm formula to classify different fault data, as shown in (11)-(14):

[0046] (11)

[0047] in, Given data points, For separation Point 1 Distance between the nearest data points for The first point The nearest point;

[0048] (12)

[0049] in, For reachable distance, the formula shows that: if Point Approaching Point, then Values ;like Point away Point, then The value is the actual distance between the two points. ;

[0050] (13)

[0051] in, for The local reachability density of a point is calculated as the reciprocal of the average reachability distances of points within its neighborhood. A higher density at a point indicates a higher reachability density. The larger the size, the lower the density. The smaller;

[0052] (14)

[0053] in, for Pointed Value. Normally, The value has the following meanings: When it is approximately 1, The density of a point is similar to that of its neighborhood. When it is greater than 1, The point density is much lower than the neighborhood density; When less than 1, The point density is higher, and it belongs to the core point.

[0054] Compared with the prior art, the advantages of the present invention are:

[0055] 1. A multi-fault generation model based on models and bench experiments for complex working conditions was developed.

[0056] 2. A neural network-based fault diagnosis model was developed that can adaptively adjust thresholds according to actual conditions.

[0057] 3. A power battery fault diagnosis method based on data and model hybrid drive, which integrates threshold method and statistical analysis, is proposed.

[0058] 4. This method can accurately identify abnormal signals of lithium-ion batteries with different aging levels under different temperatures and different rate conditions, and identify the types of faults. Attached Figure Description

[0059] Figure 1 A flowchart illustrating an early detection method for battery faults that adapts to multiple operating conditions and identifies multiple types, provided in an embodiment of the present invention.

[0060] Figure 2 This is a schematic diagram of voltage signals for internal short-circuit fault data, terminal contact fault data, excessive aging fault data, and normal data in an embodiment of the present invention.

[0061] Figure 3 A schematic diagram of a second-order equivalent circuit with an introduced short-circuit resistor used to construct internal short-circuit fault data in an embodiment of the present invention;

[0062] Figure 4 This is a schematic diagram illustrating the detection effect of the data-driven model for internal short-circuit faults in an embodiment of the present invention.

[0063] Figure 5 This is a schematic diagram illustrating the detection effect of the data-driven model for terminal contact faults in an embodiment of the present invention.

[0064] Figure 6 This is a schematic diagram illustrating the detection effect of the data-driven model of this invention on excessive aging faults.

[0065] Figure 7 This is a schematic diagram illustrating the detection effect of the mutual information method for internal short-circuit faults according to an embodiment of the present invention.

[0066] Figure 8This is a schematic diagram illustrating the detection effect of the mutual information method for terminal contact faults according to an embodiment of the present invention;

[0067] Figure 9 This is a schematic diagram illustrating the detection effect of the mutual information method for excessive aging faults according to an embodiment of the present invention;

[0068] Figure 10 This is a schematic diagram illustrating the effect of the fault type identification model based on the LOF algorithm in an embodiment of the present invention. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0070] The invention will now be further described with reference to the accompanying drawings.

[0071] See Figure 1 This invention provides an early detection method for battery faults that adapts to multiple operating conditions and identifies multiple types of faults, specifically including the following steps:

[0072] S1: Construct a data pool containing normal battery data and faulty battery data, wherein the fault types include internal short circuit faults, terminal contact faults, and excessive aging faults;

[0073] S2: A data-driven model is used to perform the first-level threshold diagnosis on the data in the data pool. The fault is determined by comparing the residual between the model-predicted voltage value and the actual voltage value to see if it exceeds the preset threshold.

[0074] S3: The mutual information algorithm is used to perform a second-level diagnosis on the data in the data pool. Faults are determined by calculating the mutual information value between data and identifying outliers.

[0075] S4: Based on the diagnostic results of S2 and S3, a joint judgment is made. If a fault is detected by either method, it is determined to be fault data.

[0076] S5: A fault type identification model based on the LOF algorithm is used to classify the fault data identified by S4, distinguishing between internal short circuit faults, terminal contact faults, and excessive aging faults.

[0077] See Figure 2 A schematic diagram of internal short-circuit fault data, terminal contact fault data, excessive aging data, and normal data. The construction of the data pool in step S1 includes:

[0078] S11: See also Figure 3Internal short-circuit fault data is generated by introducing a short-circuit resistor in the second-order equivalent circuit model, as shown in formulas (1)-(4):

[0079] (1)

[0080] (2)

[0081] (3)

[0082] (4)

[0083] Wherein, OCV is an ideal voltage source. For ohm resistance, Polarization resistor, Polarized capacitor, For diffusion resistance, For diffusion capacitance, Terminal voltage, for Voltage across the resistor for Voltage across the resistor For flowing through The current through the resistor, For flowing through The current flowing through the short-circuited resistor;

[0084] S12: Terminal contact fault data is generated by introducing noise factors into the normal discharge curve;

[0085] S13: Excessive aging failure data is derived from actual aging test data.

[0086] See Figure 4 , 5 6. Develop an adaptive threshold table. The data-driven model in step S2 includes:

[0087] S21: The input layer receives timing data. The input variables include time, current, SOC, temperature, and SOH. The structure of the input data is shown in (5):

[0088] (5)

[0089] in express Battery current at any given moment express time, Indicates time The calculation of the battery management system data, Indicates time Battery temperature, Indicates time The battery health level, Indicates time Model input data;

[0090] S22: The first LSTM layer is used to capture local timing patterns. It consists of multiple LSTM units connected in series. Each unit regulates the information flow through an input gate, a forget gate, and an output gate. Output hidden state. and memory information ;

[0091] S23: The second LSTM layer is used to extract global temporal features and receives the hidden state output from the first layer. Further extract higher-order features and output ;

[0092] S24: The output layer outputs the predicted voltage value. The structure of the output data is as follows:

[0093] (6)

[0094] In the above formula Indicates model input data The prediction results.

[0095] Table 1 Adaptive Threshold Table for Data-Driven Models

[0096]

[0097] See Figure 7 , 8 9. The mutual information diagnosis in step S3 includes:

[0098] S31: Divide the data into different groups according to temperature;

[0099] S32: Select normal data within each temperature group as the baseline for mutual information calculation;

[0100] S33: Calculate the mutual information value between the data to be diagnosed and the reference data using the following formulas (7)-(10):

[0101] (7)

[0102] Another form of mutual information is:

[0103] (8)

[0104] (9)

[0105] (10)

[0106] in and They are respectively and Information entropy for and The joint entropy;

[0107] S34: Identify outliers using frequency distribution histograms, mutual information heatmaps, and quantile-quantile plots.

[0108] The joint diagnosis in step S4 includes:

[0109] S41: If the threshold diagnosis or mutual information diagnosis result is a fault, it is determined to be faulty data.

[0110] See Figure 10 The fault type identification model in step S5 includes:

[0111] S51: Input the fault data identified in step S4 into the data proposed in step S5, and use the following LOF algorithm formula to classify different fault data, as shown in (11)-(14):

[0112] (11)

[0113] in, Given data points, For separation Point 1 Distance between the nearest data points for The first point The nearest point;

[0114] (12)

[0115] in, For reachable distance, the formula shows that: if Point Approaching Point, then Values ;like Point away Point, then The value is the actual distance between the two points. ;

[0116] (13)

[0117] in, for The local reachability density of a point is calculated as the reciprocal of the average reachability distances of points within its neighborhood. A higher density at a point indicates a higher reachability density. The larger the size, the lower the density. The smaller;

[0118] (14)

[0119] in, for Pointed Value. Normally, The value has the following meanings: When it is approximately 1, The density of a point is similar to that of its neighborhood. When it is greater than 1, The point density is much lower than the neighborhood density; When less than 1, The point density is higher, and it belongs to the core point.

[0120] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-condition adaptive and multi-class identification battery fault early detection method, characterized in that, The method comprises the following steps: S1: constructing a data pool containing normal battery data and fault battery data, the fault types including internal short circuit fault, terminal contact fault and over-aging fault; S2: performing first layer threshold diagnosis on the data in the data pool by using a data-driven model, and determining the fault by comparing whether the residual error between the model predicted voltage value and the actual voltage value exceeds a preset threshold; S3: performing second layer diagnosis on the data in the data pool by using a mutual information algorithm, and determining the fault by calculating the mutual information value between the data and identifying outliers; S4: jointly determining based on the diagnosis results of S2 and S3, and determining as fault data if any method detects a fault; S5: performing classification on the fault data identified in S4 by using a fault type identification model based on the LOF algorithm, and distinguishing the internal short circuit fault, the terminal contact fault and the over-aging fault.

2. The multi-working-condition-adapting and multi-class-identifying battery fault early pre-checking method according to claim 1, characterized in that, The construction of the data pool in the step S1 comprises: S11: the internal short circuit fault data is generated by introducing a short circuit resistance in a second-order equivalent circuit model, and the specific formula is shown as (1)-(4): （1） （2） （3） （4） where OCV is the ideal voltage source, is the ohmic resistance, is the polarization resistance, is the polarization capacitance, is the diffusion resistance, is the diffusion capacitance, is the terminal voltage, is the voltage across the resistance, is the voltage across the resistance, is the current through the resistance, is the current through the shorted resistance; S12: the terminal contact fault data is generated by introducing a noise factor in a normal discharge curve; S13: the over-aging fault data is derived from actual aging experiment data.

3. The multi-working-condition-adapting and multi-class-identifying battery fault early pre-checking method according to claim 1, characterized in that, The data-driven model in the step S2 comprises: S21: the input layer receives time series data, the input variables including time, current, SOC, temperature and SOH, and the structure of the input data is shown as (5): （5） in express Battery current at any given moment express time, Indicates time The calculation of the battery management system data, Indicates time Battery temperature, Indicates time The battery health level, Indicates time Model input data; S22: the first layer LSTM is used for capturing local timing patterns, which is connected by multiple LSTM units, each of which adjusts information flow through an input gate, a forget gate, and an output gate; output hidden state and memory information ; S23: the second layer LSTM is used to extract global timing features, receives the hidden state of the first layer output , further extracts high-order features, and outputs ; S24: the output layer outputs the predicted voltage value, and the structure of the output data is as follows: （6） In the above formula, the represents the prediction result of the model input data .

4. The multi-working-condition-adapting and multi-class-identifying battery fault early pre-checking method according to claim 1, characterized in that, The mutual information diagnosis in the step S3 comprises: S31: dividing the data into different groups according to temperature; S32: selecting normal data as the mutual information calculation baseline in each temperature group; S33: calculating the mutual information value between the data to be diagnosed and the reference data by the following formulas (7)-(10): （7） Another representation of mutual information is: （8） （9） （10） wherein and respectively and the information entropy of is and the joint entropy; S34: identifying outliers by frequency distribution histogram, mutual information thermodynamic diagram and quantile-quantile diagram.

5. The multi-working-condition-adapting and multi-class-identifying battery fault early pre-checking method according to claim 1, characterized in that, The joint determination diagnosis in the step S4 comprises: S41: determining as fault data if the threshold diagnosis or the mutual information diagnosis result is fault.

6. The multi-working-condition-adapting and multi-class-identifying battery early fault detection method according to claim 1, characterized in that, The fault type identification model in the step S5 comprises: S51: inputting the fault data identified in the step S4 into the step S5, and classifying different fault data by using the following LOF algorithm formulas, the formulas being shown as (11)-(14): （11） wherein, is the distance of the data point, is the distance of the data point, is the distance of the data point, is the distance of the data point, is the distance of the data point, is the distance of the data point, is the distance of the data point, （12） wherein, is the reachable distance, which indicates that if is close to , then takes the value ; if is far from , then takes the value of the real distance between the two points ; （13） wherein, is the local reachable density of a point, calculated as the average inverse of the reachable distance of points in the neighborhood; the higher the density of the point the greater, the lower the density the smaller; （14） wherein, is the value; usually, the value has the following meanings: when is greater than 1, is less than 1.

Images