Battery thermal runaway arc signal acquisition system based on multi-parameter acquisition

Abstract

The present application relates to the technical field of battery management system and electrical fault diagnosis, and specifically discloses a battery thermal runaway arc signal acquisition system based on multi-parameter acquisition, which dynamically predicts the normal temperature response of the battery module under different working conditions by constructing a three-dimensional thermal-electric coupling simulation model, and realizes sensitive identification of early micro thermal abnormalities in combination with synchronous acquisition of multi-parameters such as wideband current noise, terminal voltage fluctuation and real-time temperature sequence; at the same time, the system utilizes adaptive filtering and time-frequency domain transformation technology to effectively suppress background noise interference, and improves the accuracy and reliability of fault identification through a multi-physical quantity collaborative verification mechanism; in addition, after identifying a suspected fault, the system can automatically trigger transient oversampling capture to generate a standardized fault data set, thereby providing a high-quality data basis for subsequent artificial intelligence algorithm training and optimization, and continuously improving the performance of the fault diagnosis system.

Description

Technical Field

[0001] This invention relates to the field of battery management systems and electrical fault diagnosis technology, and more specifically, to a battery thermal runaway arc signal acquisition system based on multi-parameter acquisition. Background Technology

[0002] Currently, high-energy-density lithium-ion batteries are increasingly used in electric vehicles and energy storage systems, and their safety has become a focus of industry attention. In battery systems, series fault arcing caused by loose connections, aging, corrosion, and other factors is a highly insidious and destructive early failure mode. This arcing not only causes localized overheating but can also act as an ignition source, ultimately inducing battery thermal runaway. Therefore, early and reliable identification of series fault arcing is crucial for ensuring the safe operation of battery systems.

[0003] Patent CN121324974A discloses a method and system for monitoring thermal runaway in lithium-sulfur batteries based on neural networks, including: S1: collecting individual cell state data, group-level heat dissipation data, and battery compartment temperature data and performing preprocessing; S2: constructing a dual-branch gating network, a multi-head attention enhancement mechanism, and a multi-scale residual enhancement network respectively, and sequentially extracting dual-modal battery features, battery heterogeneity features, and deep heterogeneity features; S3: extracting temperature field correlation features through a temperature field correlation gating fusion mechanism; and then extracting individual cell features.

[0004] S4: Extract cross-dimensional fusion features through a dynamic weighting mechanism based on the intrinsic correlation of dual features; calculate the abnormal heat value of each individual battery cell and the overall thermal runaway risk level; S5: Visualize the data on the monitoring terminal and trigger an early warning signal.

[0005] Patent CN119575219A discloses a method for monitoring thermal runaway of sodium-ion batteries based on multi-parameter hierarchical early warning. Specifically, it includes: inducing thermal runaway in sodium-ion batteries through overcharge experiments at different charging rates; during the experiments, using explosion-proof boxes, strain sensors, temperature sensors, and gas concentration sensors to monitor the battery status in real time to ensure experimental safety; collecting and recording data such as voltage, temperature, strain, and gas concentration of the sodium-ion battery during the thermal runaway process in real time through corresponding sensors; and integrating the multi-parameter hierarchical early warning method into a battery management system (BMS). The BMS system can monitor various parameters in the battery pack in real time, compare them with preset thresholds, and automatically trigger early warning signals.

[0006] Existing technologies have significant shortcomings in addressing early, insidious faults in lithium-ion batteries: 1) Insufficient sensitivity to early, minor thermal anomalies. Existing technologies mainly rely on fixed temperature or voltage thresholds for alarms, lacking dynamic benchmarks. Under normal high-rate discharge conditions, the battery itself will generate a significant global temperature rise. Fixed threshold methods are prone to confusing this temperature rise under normal operating conditions with minor, localized thermal anomalies caused by early faults, leading to severe detection lag or missed detections.

[0007] 2) The lack of a multi-physical quantity collaborative anti-interference mechanism means that existing technologies are often limited to single-dimensional electrical or thermal signals (or their simple superposition) in feature extraction, and have not established a deep collaborative judgment logic between thermodynamic and electrical high-frequency features. In actual complex electromagnetic environments, simple electrical monitoring is prone to high false alarm rates, while simple thermal monitoring is slow to react and cannot effectively distinguish between environmental interference and real fault signals.

[0008] 3) Lack of adaptive capture and high-fidelity dataset construction capabilities for transient high-frequency fault waveforms. Existing monitoring systems typically operate continuously at a fixed, low sampling rate, which cannot adaptively increase the sampling rate to capture microsecond-level transient high-frequency details with high fidelity at the moment a suspected arc is detected. At the same time, no mechanism is provided to automatically associate such scarce fault transient waveforms with the operating context and generate standardized high-quality datasets, making it difficult to support the continuous optimization and iteration of subsequent data-driven fault diagnosis models. Summary of the Invention

[0009] In view of this, in order to solve the problems mentioned in the background technology, a battery thermal runaway arc signal acquisition system based on multi-parameter acquisition is proposed.

[0010] The objective of this invention can be achieved through the following technical solution: This invention provides a battery thermal runaway arc signal acquisition system based on multi-parameter acquisition, including: a simulation modeling and reference parameter generation module, which acquires the physical parameters of the battery module, constructs a three-dimensional thermo-electric coupling simulation model, inputs preset operating parameters for solution, and generates a dynamic temperature response reference surface and specific thermo-electric characteristic parameters.

[0011] The raw data stream generation module collects broadband current noise signals, terminal voltage fluctuation signals, and real-time temperature sequences of the battery module through a sensor array, and performs timestamp synchronization to generate a multi-dimensional synchronized raw data stream.

[0012] The feature vector generation module inputs the multidimensional synchronous raw data stream into the dynamic temperature response reference surface to calculate the temperature residual change rate and obtain thermal anomaly features. It also performs time-frequency domain transformation on the broadband current noise signal to extract high-frequency band energy and pulse distribution features, and combines them to generate a multidimensional feature vector to be determined.

[0013] The trigger control signal generation module sets a dynamic alarm threshold based on specific thermo-electric characteristic parameters. It compares the multi-dimensional feature vector to be determined with the dynamic alarm threshold. When the values ​​of each dimension of the multi-dimensional feature vector to be determined exceed the dynamic alarm threshold at the same time, it generates a trigger control signal that includes the fault start time and the data interception window length.

[0014] The fault event data packet generation module, in response to the trigger control signal, locks historical and subsequent data based on the fault start time and data capture window length, increases the sampling frequency for oversampling storage, and packages them into independent fault event data packets.

[0015] The fault dataset generation module parses independent fault event data packets, aligns the sampling phase of data from each channel, adds classification labels based on preset experimental condition logs, and archives them to generate a standardized fault dataset.

[0016] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention predicts the normal temperature response of the battery under the current operating conditions by constructing a thermo-electric coupling model, and uses the residual change rate between the measured temperature and the model prediction value as the thermal anomaly feature, which can distinguish between the global temperature rise caused by normal high-rate discharge and the local and sudden temperature rise caused by fault arc. This mechanism uses the model to dynamically remove the influence of operating condition changes on the temperature reference, so that the system can focus on monitoring the dynamic process of temperature deviating from its normal behavior pattern, rather than its absolute value. This method improves the sensitivity to early small thermal anomalies, and compared with the traditional fixed temperature threshold method, it can identify the fault at an earlier stage of development, shortening the detection delay.

[0017] (2) This invention fuses the dynamic thermal anomaly characteristics predicted by the model with the time-frequency domain characteristics of broadband current noise, and adaptively adjusts the judgment threshold according to the real-time operating conditions. Only when both thermodynamic and electrical characteristics show abnormalities is it judged as a suspected fault. This mechanism of multi-physical quantity collaborative verification can effectively suppress misjudgments that may be caused by a single information source. For example, the system can distinguish between electrical noise generated by the inverter but without abnormal temperature rise and signals generated by real electric arcs that have both thermo-electric characteristics. In this way, the false alarm rate caused by noise interference in complex electromagnetic environments is reduced, and the reliability of fault identification is improved.

[0018] (3) Upon identifying a suspected fault, this invention triggers transient oversampling capture of relevant signals and automatically labels the captured fault waveform data in conjunction with the operating log, ultimately forming a standardized fault dataset. This mechanism can transform each real or simulated fault event into a high-fidelity data sample rich in contextual information, solving the problem of difficulty in acquiring and organizing data on low-probability events such as battery arcs. This systematic construction of a high-quality dataset provides a data foundation for subsequent model training and optimization using artificial intelligence algorithms, enabling the performance of the fault diagnosis system to be continuously improved through data-driven iteration. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the system module structure connection of the present invention. Detailed Implementation

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

[0022] Please see Figure 1 This invention provides a battery thermal runaway arc signal acquisition system based on multi-parameter acquisition, including: a simulation modeling and benchmark parameter generation module, a raw data stream generation module, a feature vector to be determined generation module, a trigger control signal generation module, a fault event data packet generation module, and a fault dataset generation module.

[0023] Both the simulation modeling and baseline parameter generation module and the raw data stream generation module are connected to the feature vector generation module to be determined. Both the simulation modeling and baseline parameter generation module and the feature vector generation module to be determined are connected to the trigger control signal generation module. The trigger control signal generation module is connected to the fault event data packet generation module. The fault event data packet generation module is connected to the fault dataset generation module.

[0024] The simulation modeling and benchmark parameter generation module obtains the physical parameters of the battery module, constructs a three-dimensional thermo-electric coupling simulation model, inputs preset operating parameters for solution, and generates a dynamic temperature response benchmark surface and specific thermo-electric characteristic parameters.

[0025] In a specific embodiment of the present invention, a three-dimensional thermo-electric coupling simulation model is constructed, preset operating parameters are input for solution, and a dynamic temperature response reference surface and specific thermo-electric characteristic parameters are generated. This includes: inputting preset charge / discharge rates and ambient temperature into the three-dimensional thermo-electric coupling simulation model, calculating the steady-state temperature field distribution, and constructing a dynamic temperature response reference surface.

[0026] High-resistivity units are introduced into a three-dimensional thermo-electric coupling simulation model to simulate arc faults and obtain current waveform distortion data and local temperature field distribution data.

[0027] Extract high-frequency energy from current waveform distortion data and local hot spot temperature rise rate from local temperature field distribution data.

[0028] The ratio of high-frequency energy to the local hot spot temperature rise rate is calculated, and this ratio is confirmed as a specific thermo-electric characteristic parameter.

[0029] Specifically, in this embodiment of the invention, a data processing device first constructs and analyzes the thermo-electric coupling model. The data processing device first acquires the digital physical parameters of the battery module to be analyzed. These digital physical parameters include, but are not limited to, the three-dimensional geometric dimensions of each individual battery cell, the distance data between battery gaps, the material properties and geometric shape data of the positive and negative tabs, and the connection method and material data of the tab connectors. Based on the digital physical parameters, the data processing device uses the finite element analysis method to establish a three-dimensional thermo-electric coupling simulation model that includes the electrochemical heat generation mechanism and the Joule heating effect. In this three-dimensional thermo-electric coupling simulation model, the electrochemical heat generation part is calculated according to the Bernard-Newman-Tedman model, taking into account the irreversible heat generated by polarization overpotential and the reversible heat caused by entropy change; the Joule heating part is calculated by solving the current distribution, determining the resistive heat generated when the current flows through the battery internal resistance, tabs, and connectors.

[0030] Subsequently, the data processing equipment inputs a series of preset operating parameters as boundary conditions into the three-dimensional thermo-electric coupling simulation model for transient solution. These operating parameters include different charge / discharge rates, for example, ranging from 0.5C to 5C; different short-circuit currents, with amplitudes ranging from 500A to 3000A; and different ambient temperatures, for example, varying within the range of -20°C to 50°C. To simulate a series fault arc, a time-varying high-resistivity element is introduced at a specific connection node in the model. The resistivity and duration of this element are used to characterize the physical processes of the arc. Through simulation calculations, the data processing equipment obtains high-resolution time-series data under various simulated operating conditions. This data is specifically manifested as distorted current waveform data superimposed with high-frequency noise, and local temperature field distribution data showing a sharp increase in temperature near the arc point.

[0031] Finally, the data processing equipment performs quantitative analysis on the massive amount of time-series data generated by the simulation. By performing Fast Fourier Transform or Wavelet Transform on the current waveform distortion data, high-frequency band energy characteristics within different frequency bands are extracted. Simultaneously, for the local temperature field distribution data, the temperature change over time within the preset hotspot monitoring area is tracked, and its first derivative is calculated to obtain the local hotspot temperature rise rate. The data processing equipment repeats the above simulation and analysis process under normal operating conditions and fault arc conditions. Under normal operating conditions, the charge / discharge rate and ambient temperature are used as inputs, and the steady-state temperature field distribution is used as the output to construct a multi-dimensional dynamic temperature response reference surface. Under fault arc conditions, the strong correlation and anisotropy between the high-frequency components in the current waveform distortion data and the local hotspot temperature rise rate are systematically analyzed. The unique quantitative indicators exhibited in this correlation, such as the ratio or gradient of specific frequency band energy to temperature rise rate, are solidified into a set of specific thermo-electric characteristic parameters for use in subsequent fault identification steps.

[0032] The total heat generation power of the battery involved in this step can be calculated using the following formula:

[0033]

[0034] Among them, in the formula This indicates the total heat generation power of the battery module. This indicates the real-time current flowing through the battery. This is the open-circuit voltage of the battery. This is the battery's terminal voltage. (First term) It fully characterizes the irreversible heat generated inside the battery due to polarization resistance and the irreversible heat generated by pure ohms. This refers to the absolute temperature of the battery. (Second item) The portion is reversible heat caused by entropy change. This is the open-circuit voltage temperature coefficient, a value provided by the battery manufacturer. The representative model The equivalent DC resistance of each external connection component, including the tab resistance and the connector resistance, is already included in the preceding section due to internal battery heat generation. Only the heat generated by the series components outside the battery is taken into account.

[0035] Electrochemical heat generation refers to the heat generated during electrochemical reactions within the battery, consisting of irreversible and reversible heat. Joule heating refers to the heat energy converted from electrical energy when current flows through a resistive conductor. The dynamic temperature response reference surface is a multidimensional lookup table or fitting function stored in the memory of a data processing device. Its index is the battery's real-time operating current or charge / discharge rate and ambient temperature, and its value is the theoretically predicted temperature distribution at key points on the battery module surface under these normal operating conditions. Specific thermo-electric characteristic parameters are one or more sets of numerical pairs used to quantitatively describe the linkage between electrical and thermodynamic signals when a fault arc occurs. Assuming the application scenario is high-rate discharge of electric vehicles, this parameter can be set as the ratio of the high-frequency band energy of the current signal in the 300kHz to 1MHz frequency band to the corresponding temperature rise rate. The normal range of this ratio is based on a large amount of fault-free simulation data, for example, 0.01-0.1 J / (°C / s), while this value will jump to the range of 1-10 J / (°C / s) when a fault arc occurs.

[0036] For example, suppose the data processing device constructs a three-dimensional thermo-electric coupling simulation model of a module consisting of 10 square lithium-ion batteries connected in series. The battery gap is set to 2mm, the tabs are made of aluminum, and they are connected in series via laser-welded copper busbars. First, to establish a dynamic temperature response reference surface, the ambient temperature is set to 25℃, and a discharge simulation is performed at a 1C rate (50A). The simulation results show that, under thermal equilibrium, the temperature at the 5th connecting piece is the highest, at 38.5℃. This data point (operating condition: 1C, 25℃; result: temperature of connecting piece 5 38.5℃) is stored in the dynamic temperature response reference surface. Next, to establish specific thermo-electric characteristic parameters, at the same ambient temperature and discharge rate, a high-resistivity unit simulating a series fault arc is introduced at the 5th connecting piece. After the simulation began, the data processing equipment recorded oscillating noise with a peak value of 5A and a center frequency of 450kHz superimposed on the 50A current in the main circuit. Simultaneously, the temperature at the fifth connection point rose sharply from 28℃ to 68℃ within 10ms. The data processing equipment then calculated the quantization characteristics of this process, obtaining a high-frequency noise energy of approximately 0.2J and a local hotspot temperature rise rate of (68-28)℃ / 0.01s = 4000℃ / s. Finally, the system stored this pair of strongly correlated data (0.2J, 4000℃ / s) as a set of specific thermo-electric characteristic parameters characterizing this type of fault, for subsequent reference in constructing a fault identification baseline.

[0037] The raw data stream generation module collects broadband current noise signals, terminal voltage fluctuation signals, and real-time temperature sequences of the battery module through a sensor array, and performs timestamp synchronization to generate a multi-dimensional synchronized raw data stream.

[0038] In a specific embodiment of the present invention, generating a multidimensional synchronous raw data stream includes: acquiring a wideband current noise signal through a high-frequency current transformer and acquiring a terminal voltage fluctuation signal through a differential voltage sensor.

[0039] Based on the hotspot distribution prediction area determined by the three-dimensional thermal-electric coupling simulation model, real-time temperature sequences are collected through a multi-channel temperature sensor array.

[0040] A unified clock source is used to synchronously sample wideband current noise signals, terminal voltage fluctuation signals, and real-time temperature sequences.

[0041] A unified high-precision timestamp is added to the data points of each channel in the same sampling batch, and the data points are combined to generate a multi-dimensional synchronous raw data stream.

[0042] Specifically, to perform real-time monitoring of the battery module, this embodiment configures a high-speed data acquisition system based on multi-parameter sensing. First, a high-frequency current transformer is connected in series at the positive output terminal of the battery module's main series circuit. This transformer has sufficient bandwidth to capture the high-frequency components generated by potential electric arcs, thereby acquiring a wideband current noise signal. Simultaneously, a high-precision differential voltage sensor is connected in parallel between the positive and negative terminals of the battery module to acquire the terminal voltage fluctuation signal of the entire module.

[0043] Next, based on the predicted hotspot distribution area determined through simulation analysis using a thermo-electric coupling model, this embodiment executes precise deployment of temperature sensors. Specifically, for the battery tabs and connecting pieces where the simulation results show high temperature rise, a multi-channel temperature sensor array consisting of multiple independent temperature measuring units is deployed. The sensors in this array are closely attached to the surface of the target monitoring point to reduce heat conduction delay, thereby enabling the acquisition of real-time temperature sequences that reflect the spatial distribution and temporal evolution of local hotspots.

[0044] Finally, to ensure strict temporal alignment of different physical quantity signals, all sensors, including high-frequency current transformers, high-precision voltage sensors, and multi-channel temperature sensor arrays, are connected to a high-speed data acquisition controller with a unified clock source. This controller uses a highly stable internal crystal oscillator or external synchronization signal as a reference to synchronously sample and convert the analog signals of all channels to digital. At each sampling instant, the controller appends a uniform, high-precision timestamp to the data points of each channel within the same sampling batch. In this way, the system combines discrete, multi-source sensor readings into a strictly temporally synchronized data frame and continuously generates a multi-dimensional synchronous raw data stream, providing a high-quality data foundation for subsequent feature extraction and fusion analysis.

[0045] The broadband current noise signal refers to the high-frequency AC component superimposed on the main discharge current of the battery, with a frequency range typically covering from several kilohertz to several megahertz. In this embodiment, the high-frequency current transformer selected has a measurement bandwidth set to 1 kHz to 10 MHz. The multi-channel temperature sensor array consists of multiple K-type thermocouples, which feature fast response speed and wide temperature measurement range, meeting the monitoring requirements for rapid temperature rise processes caused by electric arcs. A unified clock source is provided by a temperature-compensated crystal oscillator inside the data acquisition controller, ensuring that the sampling time jitter between all channels is at the nanosecond level, thereby guaranteeing the accuracy of data synchronization. The multidimensional synchronous raw data stream is a structured data format in which each record contains a unified timestamp field and multiple data fields, corresponding to the current value, voltage value, and temperature value of each probe in the sensor array at the same moment.

[0046] Continuing with the example scenario above, we configure a data acquisition system for the module with 10 batteries connected in series. First, a 5MHz bandwidth Rogowski coil is installed as a high-frequency current transformer in the module's main circuit, and a differential voltage probe is connected to both ends of the module. Second, based on the prediction that the fifth connector is the main hotspot, three K-type thermocouples are arranged at the center of this connector and on the battery tabs on both sides, forming a three-channel temperature sensor array. All sensor signal lines are connected to a synchronous data acquisition card, which provides a unified clock source. The sampling rate for the current and voltage channels is set to 5MS / s, while the sampling rate for the three temperature channels is set to 1kS / s. All channel sampling actions are triggered by the same clock signal. During system operation, assuming at time t=15.000000s, the module is discharging stably at a current of 50A, the acquisition card generates and outputs a multi-dimensional synchronous raw data stream record with the following content: {timestamp: 15.000000s, current value: 50.08A, voltage value: 36.52V, temperature 1 (tab): 37.9℃, temperature 2 (connector 5): 38.5℃, temperature 3 (tab): 37.7℃}. This data stream is continuously generated and transmitted in real time to the subsequent data processing unit.

[0047] The feature vector generation module inputs the multidimensional synchronous raw data stream into the dynamic temperature response reference surface to calculate the temperature residual change rate and obtain thermal anomaly features. It also performs time-frequency domain transformation on the broadband current noise signal to extract high-frequency band energy and pulse distribution features, and combines them to generate a multidimensional feature vector to be determined.

[0048] In a specific embodiment of the present invention, before inputting the multidimensional synchronous raw data stream into the dynamic temperature response reference surface to calculate the temperature residual change rate and obtain the thermal anomaly characteristics, the method further includes: acquiring the background electromagnetic noise signal of the environment where the battery module is located.

[0049] The weight coefficients of the adaptive filter are dynamically configured based on the high-frequency band energy of the background electromagnetic noise signal.

[0050] The broadband current noise signal in the multidimensional synchronous raw data stream is input into an adaptive filter for noise reduction processing to obtain the noise-reduced current noise signal.

[0051] The denoised current noise signal is used as a new broadband current noise signal to perform subsequent time-frequency domain transformation.

[0052] Specifically, the engineering purpose of introducing adaptive noise reduction processing before extracting thermal anomaly features in this embodiment is as follows: In the actual complex operating conditions of electric vehicles or energy storage systems, power electronic switching devices such as motor controllers, DC / DC converters, and on-board chargers generate a large amount of high-frequency electromagnetic switching noise during normal operation. The frequency bands of these background noises often overlap with the high-frequency radiation frequency bands generated by series fault arcs. If time-frequency domain transformation is performed directly, the energy of these normal switching noises will be mistakenly included in the arc features, resulting in a high false alarm rate.

[0053] The specific implementation and technical details of the noise reduction process are as follows: First, regarding the acquisition of background electromagnetic noise signals: The system acquires background high-frequency electromagnetic interference signals generated by other electrical equipment in the current space environment by arranging an independent broadband spatial electromagnetic antenna inside the battery module enclosure (or by using data collected by the main current sensor during the baseline verification phase when the system is just started and no electric arc is confirmed), and uses this as the reference signal for adaptive filtering. Second, regarding the dynamic configuration of the adaptive filter weight coefficients: A digital adaptive filter module (e.g., using the Least Mean Square algorithm, LMS) is deployed in the data processing equipment. The system uses the broadband current noise signal in the acquired multidimensional synchronous raw data stream as the primary input signal and the aforementioned acquired background electromagnetic noise signal as the reference input signal. Based on the current spectral characteristics of the reference signal, the filter algorithm calculates and updates the tap weight coefficients of the filter in real time and automatically according to the criterion of minimizing the mean square value of the system output error (primary input signal minus filter output signal) through an iterative algorithm. Finally, regarding noise reduction processing and synergistic effects: The adaptive filter with dynamically optimized weights can effectively "cancel out" interference components that are in phase and at the same frequency as the background electromagnetic noise in the main current signal. The noise-reduced current noise signal output after this subtraction process effectively eliminates normal switching pulse interference from inverters and other equipment, retaining only random high-frequency abrupt changes caused by localized arc discharge. Using this signal as input for subsequent short-time Fourier transforms ensures that the "high-frequency band energy" calculated by integration truly and uniquely reflects the physical energy intensity of the arc. This collaborative mechanism of "adaptive purification followed by time-frequency domain quantization" solves the anti-interference problem under complex electromagnetic conditions at the data source, improving the accuracy of subsequent feature comparison and fault triggering.

[0054] In a specific embodiment of the present invention, the thermal anomaly characteristics are obtained by inputting the multidimensional synchronous raw data stream into the dynamic temperature response reference surface to calculate the temperature residual change rate. This includes extracting the real-time current value and ambient temperature at the current moment from the multidimensional synchronous raw data stream.

[0055] Input the real-time current value and ambient temperature into the dynamic temperature response reference surface, and query to obtain the theoretically predicted temperature value.

[0056] Extract the real-time temperature sequence from the multidimensional synchronous raw data stream and calculate the temperature residual sequence between the real-time temperature sequence and the theoretically predicted temperature value.

[0057] The time derivative is calculated by applying a differential filter to the temperature residual sequence, and the time derivative is confirmed as a characteristic of thermal anomalies.

[0058] In a specific embodiment of the present invention, a time-frequency domain transformation is performed on the broadband current noise signal to extract high-frequency band energy and pulse distribution features, and a multi-dimensional feature vector to be determined is generated by combining them. This includes: performing a short-time Fourier transform on the broadband current noise signal to obtain the power spectral density distribution matrix in the time and frequency dimensions.

[0059] Within a preset arc characteristic frequency band and a set short time window, the power spectral density distribution matrix is ​​integrated to calculate the high-frequency band energy.

[0060] A bandpass filter is applied to a wideband current noise signal, and the kurtosis factor of the filtered signal is calculated as a pulse distribution characteristic.

[0061] The thermal anomaly features, high-frequency band energy, and pulse distribution features are concatenated into vectors to generate a multidimensional feature vector to be determined.

[0062] Specifically, in this embodiment of the invention, the data processing device performs signal preprocessing and feature extraction. First, the data processing device receives and parses the generated multi-dimensional synchronous raw data stream in real time. For each frame of data in the stream, the device extracts the real-time current value, real-time voltage value, and ambient temperature as input, and queries the established dynamic temperature response reference surface. Through interpolation or function calls, the reference surface outputs the theoretical predicted temperature values ​​at each temperature measurement point of the battery module under the current operating conditions.

[0063] Next, the data processing device extracts the real-time temperature sequence from the multidimensional synchronous raw data stream and calculates the residual between it and the theoretically predicted temperature value, generating a time-varying temperature residual sequence. To eliminate quasi-static bias caused by slow sensor drift or small fluctuations in ambient temperature, the device does not directly use the absolute value of the residual. Instead, it applies a differential filter to the temperature residual sequence or calculates its time derivative to extract the rate of change of the temperature residual, which is defined as the thermal anomaly characteristic. This allows the system to focus on the dynamic process of temperature deviation from the theoretical model, rather than its static deviation.

[0064] Simultaneously, the data processing equipment performs time-frequency domain transformation on the broadband current noise signal from the multidimensional synchronous raw data stream, such as using short-time Fourier transform or wavelet packet decomposition, to obtain the power spectral density distribution matrix of the signal in the time and frequency dimensions. Based on this time-frequency analysis result, the equipment calculates two key electrical characteristics. The first is the high-frequency band energy, which is achieved by doubly integrating the current power spectrum multiplied by the system's high-frequency equivalent impedance within a preset arc characteristic frequency band and a set short-time window. The second is the pulse distribution characteristics, such as calculating the kurtosis or kurtosis factor of the broadband current noise signal after bandpass filtering, to quantify the signal's impulsiveness or peak level. Finally, the data processing equipment combines the calculated thermal anomaly characteristics, high-frequency band energy, and pulse distribution characteristics to generate a multidimensional feature vector to be determined at each timestamp, and outputs it to the subsequent determination module.

[0065] The thermal anomaly characteristics and high-frequency band energy calculations involved in this step are as follows:

[0066]

[0067]

[0068] Among them, in the formula The thermal anomaly characteristics at time t. and These are the measured temperature values ​​at the current time and the previous time, respectively. and It is based on the current at the current time and the current at the previous time respectively. The theoretically predicted temperature value obtained from the dynamic temperature response reference surface. for High-frequency band energy at any given moment. It is the result of a broadband current noise signal undergoing a short-time Fourier transform. For frequency. The equivalent high-frequency radiation and measured impedance constant of the main circuit within the characteristic frequency band. The set short-time window length, This is the time integration variable. Add... The time integral term makes the calculated It strictly corresponds to physical work or energy dissipation in terms of dimensions, and its unit is joule (J). and The lower and upper limits of the integration frequency are set based on specific thermo-electric characteristic parameters, for example, 300kHz and 1MHz.

[0069] The theoretically predicted temperature value is calculated based on the battery's electrothermal characteristics under healthy conditions using a dynamic temperature response reference surface, reflecting the battery's expected temperature state under current charge / discharge conditions. Thermal anomaly characteristics characterize the rate at which the measured temperature deviates from its theoretical healthy state; a rapidly increasing positive value suggests the presence of an abnormal heat source. Time-frequency domain transformation is a signal processing technique that reveals how the frequency components of a signal change over time. High-frequency band energy quantifies the intensity of electromagnetic radiation energy associated with arc discharge in the current signal. Pulse distribution characteristics are used to distinguish between stable Gaussian noise and non-Gaussian noise generated by the arc, which exhibits significant transient impact characteristics.

[0070] For example, the data processing device receives a multi-dimensional synchronous raw data stream at time t=15.010s, containing a current value of 50.15A and a measured temperature of 38.7℃ at the 5th connecting piece. First, the device uses the current value of 50.15A to query the dynamic temperature response reference surface, obtaining a theoretical predicted temperature of 38.52℃ for the 5th connecting piece under this condition. Next, the device calculates the temperature residual at the current moment as 0.18℃. The device also needs to retrieve data from the previous moment t=15.000s, where the current was 50.08A, the measured temperature was 38.5℃, and the theoretical predicted temperature was also 38.5℃; therefore, the residual at the previous moment was 0℃. Thus, the device calculates the thermal anomaly characteristic as (0.18℃-0℃) / 0.01s=18℃ / s. Simultaneously, the equipment performs a short-time Fourier transform on the broadband current noise signal near t=15.010s, and integrates the calculated power spectrum within a frequency band of 300kHz to 1MHz, obtaining a high-frequency band energy value of 0.05J. Furthermore, the kurtosis value of the signal within this frequency band is calculated to be 8.5. Finally, the data processing equipment combines these three features to generate a multidimensional feature vector to be determined at that moment as [18.0, 0.05, 8.5], and transmits it to the next step for fault determination.

[0071] The trigger control signal generation module sets a dynamic alarm threshold based on specific thermo-electric characteristic parameters. It compares the multi-dimensional feature vector to be determined with the dynamic alarm threshold. When the values ​​of each dimension of the multi-dimensional feature vector to be determined exceed the dynamic alarm threshold at the same time, it generates a trigger control signal that includes the fault start time and the data interception window length.

[0072] In a specific embodiment of the present invention, generating a trigger control signal that includes the fault initiation time and the data capture window length includes: generating a dynamic alarm threshold that includes an energy threshold and a thermal anomaly threshold based on the current charge / discharge rate and ambient temperature, combined with specific thermo-electric characteristic parameters.

[0073] Determine whether the high-frequency band energy in the multidimensional feature vector to be determined is greater than the energy threshold, and whether the thermal anomaly feature is greater than the thermal anomaly threshold.

[0074] If the high-frequency band energy is greater than the energy threshold and the thermal anomaly characteristics are greater than the thermal anomaly threshold, then a suspected electric arc event is determined to have occurred, and the timestamp of the current moment is extracted as the fault start time.

[0075] The length of the data capture window is calculated based on the difference between the high-frequency band energy and the energy threshold.

[0076] The fault start time and data capture window length are encapsulated to generate a trigger control signal.

[0077] Specifically, in this embodiment, the data processing device performs the dynamic matching and identification logic of fault characteristics. First, the data processing device adaptively sets a multi-dimensional dynamic alarm threshold based on the real-time operating conditions of the current battery module. Specifically, the device obtains the current charge / discharge rate or current value and ambient temperature in real time from the generated multi-dimensional synchronous raw data stream. Based on these operating condition parameters and calling the established specific thermo-electric feature parameter library, the device generates a dynamic alarm threshold vector that matches the dimensions of the multi-dimensional feature vector to be determined.

[0078] Subsequently, the data processing equipment compares the generated multi-dimensional feature vector to be determined for each frame with the currently valid dynamic alarm threshold vector in real time, element by element. This comparison process is a continuous logical judgment loop, ensuring continuous monitoring of the battery status.

[0079] During this comparison process, the data processing device classifies the current event as a suspected arcing event only if two key components in the multidimensional feature vector to be determined—namely, high-frequency band energy and thermal anomaly characteristics—simultaneously exceed their corresponding thresholds in the dynamic alarm threshold vector. This dual-condition judgment logic aims to ensure the accuracy of identification; that is, a genuine arcing fault must simultaneously exhibit anomalies in both electrical signals and thermodynamic responses.

[0080] Once a suspected arcing event is identified, the data processing equipment immediately generates a trigger control signal with a precise timestamp. This signal is a structured data packet that explicitly contains the fault initiation time T0 of the event, which is directly taken from the timestamp of the multidimensional feature vector to be determined in the frame that triggered the determination. In addition, the signal also includes a suggested data truncation window length, which can be dynamically adjusted based on the magnitude by which each component of the multidimensional feature vector exceeds a threshold, guiding subsequent data acquisition operations. This trigger control signal is immediately sent to the control unit of the high-speed data acquisition system.

[0081] The determination logic of this embodiment can be described by the following expression:

[0082]

[0083] Among them, in the formula for The result of the time determination, This indicates that the incident was suspected to be an electric arc event. and They represent the current respectively and ambient temperature The dynamic alarm thresholds for energy spectral density and thermal anomaly characteristics are as follows. and They represent High-frequency band energy and thermal anomaly characteristics at any given time.

[0084] The dynamic alarm threshold is a vector that is dynamically adjusted based on the real-time operating conditions of the battery. Its setting is based on the fact that the normal electrical noise baseline and thermal response characteristics of the battery differ under different operating conditions. In battery system fault arc detection, the typical value of the high-frequency band energy threshold is usually set at 0.02-0.05J (for the 300kHz to 1MHz band). This value is based on the statistical upper limit of background switching noise energy in numerous normal high-rate charge-discharge experiments, ensuring that only sudden arc high-frequency radiation exceeding this baseline by more than 3 standard deviations (3σ) can be detected. The typical value of the thermal anomaly threshold usually fluctuates dynamically between 10.0-20.0℃ / s (i.e., the temperature residual change rate). This value is based on the prediction of the thermo-electric coupling model, eliminating the extreme heat conduction rate limit caused by the instantaneous high temperature of several thousand degrees Celsius from a local arc after eliminating normal Joule heating. This threshold is much higher than the rate of normal battery overload or slow rise in ambient temperature (usually <1℃ / s). A suspected arc event is an internal system status flag used to initiate the subsequent transient fault detection process. The trigger control signal is an instruction sent by the data processing device to the data acquisition system to switch from passive continuous acquisition to active event capture mode. The fault initiation time T0 is the precise time point at which the precursor to the fault is detected. The data capture window length is typically preset to 100ms to 500ms to ensure complete recording of the waveforms before and after the arc occurs.

[0085] For example, the data processing device receives a multi-dimensional feature vector [18.0, 0.05, 8.5] generated at time t=15.010s. First, the device obtains the operating conditions at that time, i.e., the current is approximately 50A and the ambient temperature is 25℃. Based on these operating conditions, the device queries the corresponding dynamic alarm threshold vector from its internally stored threshold model. Assume the query result is: thermal anomaly feature threshold. =15.0℃ / s, high-frequency band energy threshold =0.03J, pulse distribution characteristic threshold=7.0. Next, the device performs element-by-element comparison. Comparing thermal anomaly characteristics, 18.0℃ / s>15.0℃ / s, the condition is met. Comparing high-frequency band energy, 0.05J>0.03J, the condition is met. Since both main judgment conditions are met, the data processing device immediately classifies the event as a suspected arc event. Subsequently, the device generates a trigger control signal, in which the fault start time T0 is precisely marked as 15.010s. At the same time, since the high-frequency band energy of 0.05J exceeds the high-frequency band energy threshold of 0.03J by a large margin, the system dynamically sets the data interception window length to 200ms. Finally, this trigger control signal containing T0=15.010s and window length=200ms is sent to the circular buffer controller of the data acquisition system.

[0086] The fault event data packet generation module, in response to the trigger control signal, locks historical and subsequent data based on the fault start time and data capture window length, increases the sampling frequency for oversampling storage, and packages them into independent fault event data packets.

[0087] In a specific embodiment of the present invention, historical data and subsequent data are locked according to the fault start time and the data interception window length, and the sampling frequency is increased for oversampling storage, and an independent fault event data packet is generated, including: parsing the trigger control signal and obtaining the fault start time and the data interception window length.

[0088] Centered on the fault initiation time, the memory address range is calculated based on the data truncation window length, and historical data is locked in the circular buffer.

[0089] Send a command to the data acquisition hardware to increase the sampling clock frequency of the analog-to-digital converter to a preset multiple.

[0090] Subsequent data is acquired using the improved sampling clock frequency, and oversampling storage is completed.

[0091] The locked historical data and the oversampled subsequent data are copied from the circular buffer to the non-volatile storage area and encapsulated to generate independent fault event data packets.

[0092] Specifically, in this embodiment, the control unit of the high-speed data acquisition system responds to the generated trigger control signal to perform transient capture and storage of the fault waveform. First, upon receiving the trigger control signal, the circular buffer controller deployed in the system immediately parses out the fault start time T0 and the suggested data truncation window length contained therein. The circular buffer is a storage area in memory that continuously writes the latest data and overwrites the oldest data, naturally preserving historical data prior to the trigger time.

[0093] Based on the fault initiation time T0 and the suggested data capture window length, the circular buffer controller calculates the exact memory address range of the data that needs to be retained in the buffer, thereby locking in the data within a preset time window centered on the fault initiation time T0. This window simultaneously includes historical data before the trigger and subsequent data to be collected after the trigger.

[0094] While locking the data window, the control unit immediately sends instructions to the data acquisition hardware to dynamically adjust the sampling parameters. Specifically, for the voltage, current, and temperature signals within the locked window, the sampling clock frequency of the analog-to-digital converter is instantaneously increased for oversampling and storage. This ensures that transient high-frequency details caused by the electric arc, lasting on the order of microseconds, can be recorded with high fidelity.

[0095] Once all subsequent data within the preset time window has been collected, the control unit performs a data packaging operation. The entire locked, high-fidelity oversampled transient waveform data block is completely copied from the circular buffer to a separate, non-volatile storage area. This operation effectively prevents these critical fault data from being overwritten by subsequent continuous cyclic acquisition processes. Finally, this packaged data forms an independent fault event data packet, which is marked with a unique event identifier, awaiting subsequent parsing and archiving.

[0096] The circular buffer is a first-in, first-out (FIFO) data structure implemented in software or hardware. Its size is designed to hold data longer than the longest preset time window, ensuring a sufficient supply of pre-trigger data. The preset time window duration is defined by a suggested data truncation window length; for example, a 200ms window typically contains 100ms of pre-trigger data and 100ms of post-trigger data. Oversampling storage is a data acquisition technique that achieves higher temporal resolution and signal-to-noise ratio by sampling at frequencies far exceeding those required by the Nyquist theorem. In this embodiment, the sampling frequency can be instantaneously increased by 4 to 16 times. Microsecond-level high-frequency details refer to the steep rising and falling edges, as well as high-frequency oscillations, that appear on the current and voltage waveforms during the arc's formation, sustaining, and extinguishing processes. An independent fault event data packet is a self-contained data file or database record that contains not only the raw waveform data but also encapsulates all relevant metadata such as trigger time, operating parameters, sensor configuration, and sampling rate, facilitating subsequent independent analysis and tracing.

[0097] For example, the circular buffer controller of the high-speed data acquisition system receives a trigger control signal containing the fault start time T0 = 15.010s and the data capture window length = 200ms. First, the controller immediately calculates that the preset time window to be locked ranges from 14.910s (i.e., T0 - 100ms) to 15.110s (i.e., T0 + 100ms). The controller then locks the historical data corresponding to the time period from 14.910s to 15.010s in its internal circular buffer. Next, the controller sends a command to the analog-to-digital converter to instantly increase the sampling rate of the current and voltage channels, which were originally operating at 5ms / s, to 20ms / s, and to increase the sampling rate of the temperature channel from 1kS / s to 10kS / s, thereby configuring oversampling and storage of the subsequent data from 15.010s to 15.110s. At 15.110s, the entire 200ms high-resolution data window is now fully contained in the buffer. The controller immediately copies this data as a whole and encapsulates it together with metadata such as the trigger time T0=15.010s and the sampling rate used into an independent fault event data packet.

[0098] The fault dataset generation module parses independent fault event data packets, aligns the sampling phase of data from each channel, adds classification labels based on preset experimental condition logs, and archives them to generate a standardized fault dataset.

[0099] In a specific embodiment of the present invention, parsing independent fault event data packets and aligning the sampling phase of each channel data includes: extracting high-sampling-rate voltage and current signals and low-sampling-rate temperature signals from the independent fault event data packets.

[0100] A spline interpolation algorithm is used to upsample the low-sampling-rate temperature signal to generate an upsampled temperature signal.

[0101] The time resolution of the upsampled temperature signal is aligned with that of the high-sampling-rate voltage and current signal to generate a multi-channel synchronous waveform matrix.

[0102] The signal-to-noise ratio (SNR) of the multi-channel synchronous waveform matrix is ​​calculated using the sliding window method, and invalid data segments with an SNR lower than a preset threshold are removed.

[0103] In a specific embodiment of the present invention, classification labels are added to the preset experimental condition logs, and standardized fault datasets are generated by archiving them, including: extracting unique event identifiers from independent fault event data packets.

[0104] The corresponding fault type label is retrieved from the preset experimental condition log based on the unique event identifier.

[0105] Calculate the average current value of the multi-channel synchronous waveform matrix during the fault occurrence period, and map the average current value to a current level label.

[0106] Extract the ambient temperature metadata from the independent fault event data packet as the ambient temperature label.

[0107] The fault type label, current level label, and ambient temperature label are attached to the multi-channel synchronous waveform matrix and archived according to the preset data structure to generate a standardized fault dataset.

[0108] Specifically, in this embodiment, the data processing device performs the construction and standardization of the fault dataset. First, the data processing device reads the generated independent fault event data packets. The device parses the data packets, extracting the multi-channel raw waveform data contained therein, namely high-sampling-rate voltage and current signals and relatively low-sampling-rate temperature signals, while also reading metadata such as trigger time and operating parameters. To solve the problem of data point misalignment caused by inconsistent sampling rates of different channels, the device aligns the sampling phases of the data from each channel. Specifically, by using a spline interpolation algorithm, the temperature signal sequence with a lower sampling rate is upsampled to align its time resolution with that of the voltage and current signals, thereby generating a multi-channel synchronous waveform matrix with a unified time axis. Subsequently, to improve data quality, the device uses the sliding window method to calculate the signal-to-noise ratio (SNR) of the multi-channel synchronous waveform matrix, marking or removing invalid data segments with an SNR below a preset threshold that are significantly affected by environmental noise.

[0109] Next, the data processing equipment automatically adds labels to each cleaned data sample. Based on the timestamp or unique event identifier in the individual fault event data packet, the equipment searches a pre-established experimental condition log database to obtain the specific fault type corresponding to that fault event, as determined by the experimental design, such as "loose connecting bolts" or "oxidation and corrosion of connecting pieces," and uses this as the fault type label. Simultaneously, the equipment calculates the average current value during the fault occurrence period by analyzing waveform data and maps it to predefined current level labels, such as "low rate," "medium rate," or "high rate." Ambient temperature is directly extracted from the data packet's metadata and used as the ambient temperature label.

[0110] Finally, the data processing equipment archives the tagged multi-channel synchronization waveform matrix according to a preset data structure. Specifically, each fault event is stored as an independent data unit, such as an HDF5 file or database entry, containing a synchronization waveform data matrix, a tag vector containing fault type, current level, and ambient temperature tags, and complete metadata. By continuously repeating this process to process all the independent fault event data packets, a structured, standardized, high-quality lithium-ion battery series fault arc signal dataset is ultimately established, which can be directly used to train fault recognition AI models.

[0111] The signal-to-noise ratio calculations involved in this step are as follows:

[0112]

[0113] Among them, in the formula This represents the signal-to-noise ratio, and its unit is decibel (dB). The signal power is estimated in this embodiment by calculating the mean square value of the signal within a short time window centered at the fault initiation time T0. The noise power is estimated by calculating the mean square value of the signal during a period of steady-state operation prior to the fault. This signal-to-noise ratio threshold is empirically set based on the background noise level of the test environment in this field, and can be set to 3 dB.

[0114] In this system, sampling phase refers to the relative positions of data sampling points from different channels on the time axis in a multi-rate sampling system. Invalid data segments refer to data fragments whose signals are severely distorted due to external strong electromagnetic interference or other factors, failing to reflect the true state of the battery. The fault type label is a classification label that clearly indicates the physical cause of the arcing. The current level label is a classification obtained by discretizing continuous current values, facilitating the model's learning of fault characteristics under different loads. The pre-defined data structure is a standardized file or database schema, ensuring dataset consistency and scalability, and allowing direct use by machine learning frameworks.

[0115] For example, the data processing device begins processing the fault event data packet described above. First, the device parses the waveform data with a duration of 200ms, finding 4,000,000 data points (20MS / s) for the voltage and current signals, and 2,000 data points (10kS / s) for each of the three temperature signals. The device uses a cubic spline interpolation algorithm to sample the three temperature signals to 4,000,000 points respectively, precisely aligning them in time with the voltage and current signals to form a 4,000,000x5 multi-channel synchronous waveform matrix. Next, the device detects a strong interference segment with a signal-to-noise ratio of only 1.5dB in the first 20ms of the data matrix (corresponding to 14.910s to 14.930s), which is below the preset 3dB threshold. Therefore, the waveform data during this period is marked as invalid and excluded from subsequent processing. Subsequently, the equipment queried the experimental operating condition log database and found that the cause of the fault was "insufficient torque of M3 bolts," and generated a fault type label based on this: "Loose connecting bolts." Calculations showed that the average current during the fault was 50.1A, which falls within the "medium rate (1C-3C)" range, thus generating this current level label. Metadata showed the ambient temperature to be 25℃. Finally, the equipment created an HDF5 file, storing the cleaned and aligned waveform data into a dataset named "waveform_data," and storing the label vector composed of ["Loose connecting bolts," "Medium rate (1C-3C), "25℃"] into the dataset. This file was archived and became a new sample in the high-quality lithium-ion battery series fault arc signal dataset.

[0116] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A battery thermal runaway arc signal acquisition system based on multi-parameter acquisition, characterized in that, include: The simulation modeling and benchmark parameter generation module obtains the physical parameters of the battery module, constructs a three-dimensional thermo-electric coupling simulation model, inputs preset operating parameters for solution, and generates a dynamic temperature response benchmark surface and specific thermo-electric characteristic parameters. The raw data stream generation module collects the wideband current noise signal, terminal voltage fluctuation signal and real-time temperature sequence of the battery module through the sensor array, and performs timestamp synchronization to generate a multi-dimensional synchronized raw data stream; The feature vector generation module inputs the multidimensional synchronous raw data stream into the dynamic temperature response reference surface to calculate the temperature residual change rate to obtain thermal anomaly features, and performs time-frequency domain transformation on the broadband current noise signal to extract high-frequency band energy and pulse distribution features, and combines them to generate a multidimensional feature vector to be determined. The trigger control signal generation module sets a dynamic alarm threshold based on specific thermo-electric characteristic parameters, compares the multi-dimensional feature vector to be determined with the dynamic alarm threshold, and generates a trigger control signal containing the fault start time and the data interception window length when the values ​​of each dimension of the multi-dimensional feature vector to be determined exceed the dynamic alarm threshold simultaneously. The fault event data packet generation module, in response to the trigger control signal, locks historical data and subsequent data according to the fault start time and data capture window length, increases the sampling frequency for oversampling storage, and packages them to generate independent fault event data packets; The fault dataset generation module parses independent fault event data packets, aligns the sampling phase of data from each channel, adds classification labels based on preset experimental condition logs, and archives them to generate a standardized fault dataset.

2. The battery thermal runaway arc signal acquisition system based on multi-parameter acquisition according to claim 1, characterized in that, The process involves constructing a three-dimensional thermo-electric coupling simulation model, inputting preset operating parameters for solution, and generating a dynamic temperature response reference surface and specific thermo-electric characteristic parameters, including: Input the preset charge / discharge rate and ambient temperature into the three-dimensional thermo-electric coupling simulation model, calculate the steady-state temperature field distribution, and construct the dynamic temperature response reference surface; High-resistivity units are introduced into a three-dimensional thermo-electric coupling simulation model to simulate arc faults and obtain current waveform distortion data and local temperature field distribution data. Extract high-frequency energy from current waveform distortion data and local hot spot temperature rise rate from local temperature field distribution data; The ratio of high-frequency energy to the local hot spot temperature rise rate is calculated, and this ratio is confirmed as a specific thermo-electric characteristic parameter.

3. The battery thermal runaway arc signal acquisition system based on multi-parameter acquisition according to claim 1, characterized in that, The generation of the multidimensional synchronous raw data stream includes: Wideband current noise signals are acquired using a high-frequency current transformer, and terminal voltage fluctuation signals are acquired using a differential voltage sensor. Based on the hotspot distribution prediction area determined by the three-dimensional thermal-electric coupling simulation model, real-time temperature sequences are collected through a multi-channel temperature sensor array. A unified clock source is used to synchronously sample wideband current noise signals, terminal voltage fluctuation signals, and real-time temperature sequences. A unified high-precision timestamp is added to the data points of each channel in the same sampling batch, and the data points are combined to generate a multi-dimensional synchronous raw data stream.

4. The battery thermal runaway arc signal acquisition system based on multi-parameter acquisition according to claim 1, characterized in that, Before inputting the multidimensional synchronous raw data stream into the dynamic temperature response reference surface to calculate the rate of change of temperature residuals and obtain thermal anomaly characteristics, the method further includes: Acquire the background electromagnetic noise signal of the environment where the battery module is located; The weight coefficients of the adaptive filter are dynamically configured based on the spectral characteristics of the background electromagnetic noise signal. The broadband current noise signal in the multidimensional synchronous raw data stream is input into an adaptive filter for noise reduction processing to obtain the noise-reduced current noise signal. The denoised current noise signal is used as a new broadband current noise signal to perform subsequent time-frequency domain transformation.

5. The battery thermal runaway arc signal acquisition system based on multi-parameter acquisition according to claim 1, characterized in that, The process of inputting a multidimensional synchronous raw data stream into a dynamic temperature response reference surface to calculate the rate of change of temperature residuals to obtain thermal anomaly characteristics includes: Extract the real-time current value and ambient temperature at the current moment from the multidimensional synchronous raw data stream; Input the real-time current value and ambient temperature into the dynamic temperature response reference surface, and query to obtain the theoretically predicted temperature value; Extract the real-time temperature sequence from the multidimensional synchronous raw data stream, and calculate the temperature residual sequence between the real-time temperature sequence and the theoretically predicted temperature value; The time derivative is calculated by applying a differential filter to the temperature residual sequence, and the time derivative is confirmed as a characteristic of thermal anomalies.

6. The battery thermal runaway arc signal acquisition system based on multi-parameter acquisition according to claim 1, characterized in that, The process involves performing time-frequency domain transformation on the broadband current noise signal to extract high-frequency band energy and pulse distribution features, and combining these features to generate a multi-dimensional feature vector to be determined, including: Perform a short-time Fourier transform on the broadband current noise signal to obtain the power spectral density distribution matrix in the time and frequency dimensions; Within the preset characteristic frequency band of the electric arc and within the set short time window, the power spectral density distribution matrix is ​​integrated to calculate and obtain the high-frequency band energy. A bandpass filter is applied to a wideband current noise signal, and the kurtosis factor of the filtered signal is calculated as a pulse distribution characteristic. The thermal anomaly features, high-frequency band energy, and pulse distribution features are concatenated into vectors to generate a multidimensional feature vector to be determined.

7. The battery thermal runaway arc signal acquisition system based on multi-parameter acquisition according to claim 1, characterized in that, The generation of the trigger control signal, which includes the fault initiation time and the data truncation window length, includes: Based on the current charge / discharge rate and ambient temperature, combined with specific thermo-electric characteristic parameters, a dynamic alarm threshold including energy threshold and thermal anomaly threshold is generated. Determine whether the high-frequency band energy in the multidimensional feature vector to be determined is greater than the energy threshold, and whether the thermal anomaly feature is greater than the thermal anomaly threshold. If the high-frequency band energy is greater than the energy threshold and the thermal anomaly characteristics are greater than the thermal anomaly threshold, then a suspected electric arc event is determined to have occurred, and the timestamp of the current moment is extracted as the fault start time. The length of the data interception window is calculated based on the difference between the high-frequency band energy and the energy threshold. The fault start time and data capture window length are encapsulated to generate a trigger control signal.

8. The battery thermal runaway arc signal acquisition system based on multi-parameter acquisition according to claim 1, characterized in that, The process involves locking historical and subsequent data based on the fault initiation time and data capture window length, increasing the sampling frequency for oversampling storage, and packaging the data into independent fault event data packets, including: Analyze the trigger control signal to obtain the fault start time and data capture window length; Centered on the fault initiation time, calculate the memory address range based on the data truncation window length, and lock historical data in the circular buffer; Send a command to the data acquisition hardware to increase the sampling clock frequency of the analog-to-digital converter to a preset multiple; Subsequent data is acquired using the increased sampling clock frequency to complete oversampling storage; The locked historical data and the oversampled subsequent data are copied from the circular buffer to the non-volatile storage area and encapsulated to generate independent fault event data packets.

9. The battery thermal runaway arc signal acquisition system based on multi-parameter acquisition according to claim 1, characterized in that, The process of parsing independent fault event data packets and aligning the sampling phase of data from each channel includes: Extract high-sampling-rate voltage and current signals and low-sampling-rate temperature signals from independent fault event data packets; A spline interpolation algorithm is used to upsample the low-sampling-rate temperature signal to generate an upsampled temperature signal. Align the time resolution of the upsampled temperature signal with the high sampling rate voltage and current signal to generate a multi-channel synchronous waveform matrix; The signal-to-noise ratio (SNR) of the multi-channel synchronous waveform matrix is ​​calculated using the sliding window method, and invalid data segments with an SNR lower than a preset threshold are removed.

10. The battery thermal runaway arc signal acquisition system based on multi-parameter acquisition according to claim 9, characterized in that, The step of adding classification labels to the preset experimental condition logs and archiving them to generate a standardized fault dataset includes: Extract the unique event identifier from the individual fault event data packet; The corresponding fault type label is retrieved from the preset experimental condition log based on the unique event identifier. Calculate the average current value of the multi-channel synchronous waveform matrix during the fault occurrence, and map the average current value to a current level label; Extract the ambient temperature metadata from the independent fault event data packets as ambient temperature tags; The fault type label, current level label, and ambient temperature label are attached to the multi-channel synchronous waveform matrix and archived according to the preset data structure to generate a standardized fault dataset.

Images