Quantum sensor-based power battery runaway early warning system and method

Abstract

The present application relates to the technical field of battery out-of-control early warning, in particular to a power battery out-of-control early warning system and method based on quantum sensors; the power battery out-of-control early warning system based on quantum sensors comprises: a multi-modal acquisition module, which uses traditional sensors and double-channel quantum sensors to jointly monitor a power battery pack and obtain monitoring data; a signal processing and feature extraction module, which pre-processes and extracts features from the monitoring data; a BMS deep fusion module, which performs anomaly detection, anomaly classification and anomaly prediction; a hierarchical collaborative execution module, which executes corresponding early warning actions according to the output results of the BMS deep fusion module; and a storage module, which stores all data and early warning records on a cloud platform upper computer for calling by the above-mentioned modules when needed. The present application uses double-channel quantum sensors to cooperatively capture full-band electromagnetic fingerprints, fuses traditional sensor multi-dimensional verification, suppresses vehicle-mounted interference, and realizes early and reliable early warning of battery thermal runaway.

Description

Technical Field

[0001] This invention relates to the field of battery runaway early warning technology, and specifically to a power battery runaway early warning system and method based on quantum sensors. Background Technology

[0002] With the rapid development of the global new energy vehicle industry, the safety of lithium-ion batteries, as their core power source, is becoming increasingly prominent. Thermal runaway in new energy batteries is one of the most serious safety incidents among battery runaway events. It is often triggered by multiple factors such as internal short circuits, overcharging, and mechanical damage, and can cause a chain reaction within a short period, leading to a rapid increase in battery pack temperature and even fire or explosion. Currently, early warning systems for battery thermal runaway mainly rely on traditional sensors to monitor macroscopic parameters such as temperature, voltage, current, and gas composition (e.g., CO, H2), and utilize big data and deep learning models for anomaly identification and risk prediction.

[0003] However, this seemingly mature monitoring system has revealed fundamental gaps in its ability to address early warning needs. First, traditional monitoring of temperature, voltage, and gas parameters suffers from a fundamental lag in battery safety early warning. These parameters are essentially "consequence indicators" that only emerge after thermal runaway has entered a rapid development phase—by the time significant, identifiable fluctuations in temperature and voltage occur, irreversible thermal collapse processes such as separator failure and exacerbated internal short circuits have often already occurred within the battery, leaving the effective intervention window for the control system virtually closed. Even more critically, thermal runaway does not occur suddenly; its incubation phase is accompanied by a series of weak precursory physical phenomena, such as micro-arc discharges caused by localized micro-short circuits, leakage current pulses caused by aging insulation materials, and electromagnetic radiation excited by the release of internal stress. These signals carry crucial information revealing the battery's "sub-healthy" state, but due to their weak amplitude and unique frequency domain, they are completely outside the sensing range of existing temperature, voltage, and gas sensors, constituting a natural "perception blind spot" for the traditional monitoring system. This prevents early warning capabilities from ever reaching the "post-event alarm" level, making it impossible to effectively identify and prevent thermal runaway in its early stages.

[0004] Furthermore, under complex vehicle operating conditions, road bumps and load switching generate strong background noise, contaminating sensor signals with numerous irrelevant fluctuations. Single-sensor methods acquire limited information dimensions, making it difficult to fully extract the subtle precursor features of thermal runaway from noise. Meanwhile, single-discrimination models lack the ability to cross-validate multi-source information, easily misjudging interference as anomalies or overlooking real risks, leading to both false alarms and missed warnings. This renders the warning system, while possessing alarm functionality, incapable of providing reliable early warnings. Overcoming this predicament hinges on using multi-sensor information fusion and multi-discrimination model collaboration as core methods: leveraging multi-dimensional acquisition of multiple physical signals and joint discrimination using deep learning models to form a redundant and complementary judgment mechanism. This effectively suppresses noise interference, accurately identifies early warnings, significantly reduces false alarms and missed warnings, and endows the system with the necessary early warning reliability. This is precisely the core dilemma currently facing battery safety warning technology: existing technologies do not lack alarm methods, but lack the ability to achieve "early warnings." Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to address the problems of insufficient early warning signal perception in existing power battery thermal runaway warning systems and insufficient reliability of single-sensor warning under complex vehicle operating conditions and noise interference. The present invention provides a power battery runaway warning system and method based on the joint discrimination of electric field characteristics, magnetic field characteristics and traditional monitoring characteristics, so as to improve the accuracy of internal anomaly identification of power batteries and the reliability of early warning of thermal runaway.

[0006] A quantum sensor-based power battery runaway early warning system includes: The multimodal acquisition module uses both traditional sensors and dual-channel quantum sensors to monitor the power battery pack and acquire monitoring data. The monitoring data includes traditional monitoring data, electric field data, and magnetic field data. The signal processing and feature extraction module preprocesses the monitoring data and extracts features from the preprocessed monitoring data. The BMS deep fusion module, based on a pre-built deep learning model, sequentially performs anomaly detection, anomaly classification, and anomaly prediction on all extracted features. The hierarchical collaborative execution module controls the execution of corresponding early warning actions based on the detection results, classification results, and prediction results obtained by the BMS deep fusion module, combined with preset early warning strategies. All data generated and acquired during the operation of the multimodal acquisition module, signal processing and feature extraction module, BMS deep fusion module and hierarchical collaborative execution module, as well as the early warning records of the early warning actions, are uploaded to the cloud platform host computer for storage. The multimodal acquisition module, signal processing and feature extraction module, BMS deep fusion module, and hierarchical collaborative execution module are connected to the cloud platform host computer and can retrieve data stored on the cloud platform host computer as needed.

[0007] Preferably, the dual-channel quantum sensor has two channels: an EIT channel and a CPT channel; The EIT channel and the CPT channel share the same atomic gas chamber. The light sources from the EIT channel and the CPT channel are combined via a wavelength division multiplexer or a spatial optical path and then injected into the atomic gas cell.

[0008] Preferably, the working atom in the atomic gas cell is a cesium atom; The buffer gas filled in the atomic gas chamber is nitrogen or argon at a concentration of 10 to 50 Torr.

[0009] Preferably, the EIT channel uses a probe light with a wavelength of 852nm and a pump light with a wavelength of 509nm to form a Cs6S. 1 / 2 →6P 3 / 2 →50D 5 / 2 The Rydberg stepped energy level system serves as a light source for detecting high-frequency and transient electric fields inside a power battery pack.

[0010] Preferably, the CPT channel and the EIT channel use the same probe beam locked to the 852nm transition of the D2 line of the cesium atom; after the probe beam enters the CPT channel, coherent sidebands are generated by current modulation. The positive and negative first-order sideband frequency difference in the coherent sideband matches the hyperfine splitting frequency of the cesium atom's ground state, which is 9.192 GHz, corresponding to the modulation frequency f. mod ≈4.596GHz.

[0011] Preferably, the EIT channel and CPT channel in the dual-channel quantum sensor are configured to operate in two modes: simultaneous measurement mode or time-division multiplexing measurement mode; In the time-division multiplexing measurement mode, the CPT channel and the EIT channel work alternately, and the time interval between the alternations satisfies the acquisition of the transient electric field distribution.

[0012] Preferably, feature extraction is performed on the preprocessed monitoring data: Differential processing is used to perform feature processing on magnetic field data, and short-time Fourier transform is used to extract features from electric field data.

[0013] Preferably, the preset early warning strategy is a gradual strategy, specifically: Level 1 warning: Based on the anomaly detection phase, an anomaly is detected in the quantum sensing signal acquired by the dual-channel quantum sensor and triggered. Log is recorded and the sampling frequency of the signal processing and feature extraction modules is increased. Level 2 warning: Based on the classification results, if a stable leakage event caused by an internal short circuit or a rapid insulation degradation event is triggered, a signal suggesting power limitation is sent to the vehicle controller. The Level 3 warning is triggered when the quantum sensing signal reaches saturation and is accompanied by a voltage drop or the detection of trace gases. It then performs the final protective actions of cutting off the high-voltage relay, starting the liquid system at full speed, and notifying the occupants to evacuate.

[0014] Preferably, the electric field data includes: high-frequency electric field distribution and transient electric field distribution.

[0015] A quantum sensor-based early warning method for power battery runaway includes: S1. The power battery pack is monitored using both traditional sensors and dual-channel quantum sensors to obtain monitoring data; the monitoring data includes: traditional monitoring data, electric field data, and magnetic field data; S2. Preprocess the monitoring data and extract features from the preprocessed monitoring data; S3. Based on the pre-built deep learning model, all extracted features are sequentially subjected to anomaly detection, anomaly classification, and anomaly prediction, and the corresponding detection results, classification results, and prediction results are obtained. S4. Combine the detection results, classification results, and prediction results with the preset early warning strategy to control and execute the corresponding early warning action; S5. Upload all data generated and acquired during the monitoring data acquisition stage, the monitoring data preprocessing stage, the anomaly detection, anomaly classification and anomaly prediction stage of the extracted features, and the corresponding early warning action execution process, as well as the early warning records of the early warning actions, to the cloud platform host computer storage. The process includes the acquisition of monitoring data, the preprocessing of monitoring data, anomaly detection, anomaly classification, and anomaly prediction of extracted features, and the retrieval of data stored on the cloud platform's host computer as needed when executing corresponding early warning actions.

[0016] The technical solution of this invention has the following advantages: This invention is based on a dual-channel quantum sensor. Using the EIT channel and CPT channel in the same probe, it captures high-frequency arc radiation and DC / low-frequency magnetic field signals respectively, forming a broadband electromagnetic fingerprint recognition capability covering the quasi-static to terahertz frequency band, enabling early non-contact warning of battery thermal runaway. To address background noise and transient interference in the complex automotive electromagnetic environment, false alarms are suppressed through near-field placement, directional filtering, time-frequency analysis, and deep learning. Simultaneously, it integrates multi-dimensional traditional sensor information such as voltage, temperature, and gas, triggering warnings only when multiple dimensions are simultaneously abnormal, significantly improving system robustness and engineering reliability. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the overall framework of the power battery runaway early warning system based on quantum sensors of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0020] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0021] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0022] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0023] Example 1 This embodiment uses the power battery pack of the new energy vehicle as the monitored object; and proposes a power battery runaway early warning system based on quantum sensors, including: The multimodal acquisition module uses both traditional sensors and dual-channel quantum sensors to monitor the power battery pack and acquire monitoring data. The monitoring data includes traditional monitoring data, electric field data, and magnetic field data. The signal processing and feature extraction module preprocesses the monitoring data and extracts features from the preprocessed monitoring data. The BMS deep fusion module, based on a pre-built deep learning model, sequentially performs anomaly detection, anomaly classification, and anomaly prediction on all extracted features. The hierarchical collaborative execution module controls the execution of corresponding early warning actions based on the detection results, classification results, and prediction results obtained by the BMS deep fusion module, combined with preset early warning strategies. All data generated and acquired during the operation of the multimodal acquisition module, signal processing and feature extraction module, BMS deep fusion module and hierarchical collaborative execution module, as well as the early warning records of the early warning actions, are uploaded to the cloud platform host computer for storage. The multimodal acquisition module, signal processing and feature extraction module, BMS deep fusion module, and hierarchical collaborative execution module are connected to the cloud platform host computer and can retrieve data stored on the cloud platform host computer as needed.

[0024] In order to maximize detection efficiency, an improved sensor arrangement scheme is proposed in this embodiment: The positions of the battery modules and cell arrays in the power battery pack remain unchanged.

[0025] Dual-channel quantum sensors are installed at the main positive and negative buses of the battery module. In an array composed of multiple battery cells, every 12 cells form a unit, and only one dual-channel quantum sensor is installed to monitor the status of this group of cells. A daisy-chain or star topology is used to distribute the optical fiber. The light source and detector of the dual-channel quantum sensor are placed in a low-voltage area outside the power battery pack to avoid high temperatures affecting the laser's lifespan. The optical fiber passes through the sealed boundary of the battery pack, achieving intrinsically safe, power-free monitoring.

[0026] The cloud platform's host computer receives, stores, and transmits data. Through a 4G / 5G wireless communication link, the host computer receives multi-dimensional monitoring data in real time from the vehicle's signal processing and feature extraction module. This data includes high- and low-frequency partial discharge characteristic signals, weak magnetic field signals, thermal runaway risk levels, remaining safe time, and battery condition data provided by the BMS deep fusion module, such as cell voltage, current, SOC, temperature, and thermal management status. This data is not only used for remote real-time monitoring and historical data storage, but also, based on a large amount of actual vehicle operating data and thermal runaway cases, continuously iterates and optimizes the system's high- and low-frequency partial discharge characteristic library, weak magnetic field signal library, and parameters. Updated parameters are then sent to the vehicle via OTA (Over-The-Air) updates, thereby achieving dynamic improvement and adaptive evolution of the system's early warning capabilities.

[0027] Summarize: This embodiment proposes an early non-contact warning system. Utilizing a highly sensitive dual-channel quantum sensor attached to the exterior or gaps of the battery module, it can detect weak electromagnetic signals released during early stages of thermal runaway chain reactions within the battery pack—such as changes in material structure, abnormal ion migration, or intensified side reactions—without disassembling the battery cells or immersing them in electrolyte. This enables early diagnosis of the initiation stage of thermal runaway chain reactions. Compared to traditional temperature monitoring or gas composition analysis, electromagnetic signals possess physical characteristics such as light-speed propagation, strong penetration, and extremely low response delay. This allows the system to issue warnings within minutes to hours before thermal runaway occurs, significantly improving the timeliness and reliability of the warning. This innovation provides a new technical approach for preventative and proactive monitoring in the safety management of power battery packs, possessing significant scientific value and engineering application potential.

[0028] Example 2 Based on Example 1, this example discloses a power battery runaway early warning method based on quantum sensors, including: S1. The power battery pack is monitored using both traditional sensors and dual-channel quantum sensors to obtain monitoring data; the monitoring data includes: traditional monitoring data, electric field data, and magnetic field data; S2. Preprocess the monitoring data and extract features from the preprocessed monitoring data; S3. Based on the pre-built deep learning model, all extracted features are sequentially subjected to anomaly detection, anomaly classification, and anomaly prediction, and the corresponding detection results, classification results, and prediction results are obtained. S4. Combine the detection results, classification results, and prediction results with the preset early warning strategy to control and execute the corresponding early warning action; S5. Upload all data generated and acquired during the monitoring data acquisition stage, the monitoring data preprocessing stage, the anomaly detection, anomaly classification and anomaly prediction stage of the extracted features, and the corresponding early warning action execution process, as well as the early warning records of the early warning actions, to the cloud platform host computer storage. The process includes the acquisition of monitoring data, the preprocessing of monitoring data, anomaly detection, anomaly classification, and anomaly prediction of extracted features, and the retrieval of data stored on the cloud platform's host computer as needed when executing corresponding early warning actions.

[0029] Example 3 Based on Example 2, this example further describes how the monitoring data is acquired; in this example, there are two simultaneous monitoring methods for acquiring the monitoring data: the main monitoring method and the auxiliary monitoring method; Auxiliary monitoring methods: Traditional sensors are used to monitor changes in traditional targets such as temperature, voltage, current and gas inside the power battery pack. Since obtaining changes in temperature, voltage, current and gas inside the power battery is a common method in the prior art, it will not be described in detail in this embodiment.

[0030] Main monitoring method: In this embodiment, several dual-channel quantum sensors are arranged at key nodes of the power battery pack for monitoring; the dual-channel quantum sensors in this embodiment have two channels in total; the EIT channel and the CPT channel share the same atomic gas chamber; and the two channels are configured to operate simultaneously or in time-division multiplexing measurement modes. Among them, the EIT channel is based on Rydberg atoms and electromagnetically induced transparency effect. By detecting the transmission spectral characteristics of pump light and probe light in the atomic gas cell, it measures the high-frequency electric field distribution and transient electric field distribution inside the power battery pack, which can be used to identify micro-arc, partial discharge and insulation degradation signals; the CPT channel is based on coherent population trapping effect. By detecting the interaction between the modulated laser field and the hyperfine energy level of the atomic ground state, it measures the low-frequency magnetic field change inside the power battery pack, which can be used to identify local overcurrent, micro-short circuit and abnormal current distribution signals. Specifically: 1) Atomic gas chamber design: The atomic gas chamber is a miniature gas chamber made of borosilicate glass and silicon wafer anodic bonding, with an internal cavity size of 3mm×3mm×3mm.

[0031] The atomic chamber is filled with cesium with an isotopic purity of >98% as the working atom; Nitrogen or argon gas at a pressure of 10 to 50 Torr is filled into the atomic gas cell as a buffer gas, which mainly serves two purposes: First, the buffer gas can suppress the "dark state loss" in the optical pumping process through fine structure mixing or quenching, thereby maintaining the ground state coherence; Second, it restricts the mean free path of alkali metal atoms, prolongs the interaction time between alkali metal atoms and the optical field, and thus narrows the resonance linewidth of the CPT channel, improving the sensitivity of magnetic field measurement.

[0032] 2) Light source system configuration: To achieve two measurement modes, the dual-channel quantum sensor is equipped with two laser sources in practical applications, which are then combined into the atomic gas cell via a wavelength division multiplexer or a spatial optical path.

[0033] For the EIT channel: The light source is: probe light + pump light; The probe light has a wavelength of 852 nm and is used to excite cesium atoms from their ground state 6S. 1 / 2 Transition to the first excited state 6P 3 / 2 ; The pump light has a wavelength of 509 nm and is used to excite cesium atoms from the first excited state 6P. 3 / 2 It transitions to a Rydberg state with a principal quantum number greater than 50.

[0034] The probe light and pump light form a stepped energy level system. When the two-photon resonance condition is met, the population interference of particles in the intermediate state is suppressed, the transmittance of the probe light increases, and an EIT window is formed.

[0035] Using the saturable absorption spectroscopy (SAS) safety state technique, a portion of the probe light is split and directed to the atomic gas cell, locking the probe light frequency to induce cesium atoms to transition from their ground state (6S). 1 / 2 →First excited state 6P 3 / 2 The transition frequency of the pump light. In this embodiment, the wavelength of the pump light is 509 nm, and the principal quantum number of the Rydberg state is 50; part of the pump light is split into the atomic gas cell through electromagnetically induced transparency technology, and the pump light frequency is locked to enable the cesium atom to transition from the first excited state 6P. 3 / 2 →Rydberg 50D 5 / 2 The transition frequency.

[0036] Rydberg state polarizability characteristics, when a micro-arc or partial discharge occurs inside the power battery pack, the radiated radio frequency electric field E RF This will cause the Reedberg level to undergo Autler-Townes splitting or ACStark alternating Stark shift.

[0037] In this embodiment, the EIT channel can respond to wideband signals from 10MHz to 100GHz, with a sensitivity better than 1μV / cm / Hz. It is also a full-dielectric detector, unlike metal antennas which generate induced high voltage or tip discharge under strong electric fields.

[0038] For the CPT channel: The light source is: single-beam laser + microwave modulation; The single-beam laser is emitted using a vertical-cavity surface-emitting laser (VCSEL), with its center wavelength locked to the D2 line of cesium atoms. The D2 line refers to the characteristic transition spectral lines of cesium atoms from the ground state to two different excited states. The injection current of the VCSEL is driven by a microwave signal source, or an external electro-optic modulator is used, with a modulation frequency of f. mod =4.596 GHz, corresponding to half of the 9.192 GHz hyperfine splitting of the cesium atom's ground state. This will generate coherent first-order sidebands in the spectrum, which can then be used to form dark-state resonances in the CPT channel and detect low-frequency magnetic field variations inside the power battery pack.

[0039] When the frequency difference between the positive and negative first-order sidebands is exactly equal to the two hyperfine levels of the cesium atom's ground state: |F=1,mF and|F=2,mF When there is an energy difference, a coherent population trapping effect occurs, causing a sharp decrease in the absorption of light by the medium, resulting in a transmission peak on the detector used for receiving the light. An external magnetic field B causes the Zeeman sublevels to split. The resonance peak frequency f of the coherent population trapping effect is... CPT The relationship between the magnetic field and the magnetic field is as follows: ; Among them, the linear coefficient γ for the magnetically sensitive transition is approximately 3.5 MHz / T. The external magnetic field strength B can be accurately inverted by locking the resonant frequency or detecting changes in the transmitted light intensity. In the formula, CPT represents the coherent population trapping effect; Indicates the resonant frequency under zero magnetic field; denoted by the second-order Zeeman coefficient, it describes how quickly the resonant frequency changes with the square of the magnetic field.

[0040] In this embodiment, the CPT channel is mainly used to measure a constant magnetic field with a frequency of 0 to a low-frequency magnetic field of 1 kHz, with a resolution of up to 10 pT / Hz.

[0041] 3) Optical detection system: A photodetector is configured to receive laser signals passing through the atomic gas cell and convert them into electrical signals.

[0042] 4) To facilitate understanding, this embodiment further explains the acquisition of the transient electric field distribution: First, it's important to clarify that transient electric fields are strong electric field spikes that appear suddenly and disappear very quickly, on the order of microseconds or even nanoseconds. They are usually generated by phenomena such as micro-arcs or partial discharges. Although the duration is short, the voltage can reach several kilovolts, enough to break through insulation layers. Before a current power battery pack actually catches fire, there are often countless micro-arcs. At this point, traditional sensors, due to their slow sampling speed, simply cannot record the suddenly appearing and disappearing transient electric field.

[0043] To address this issue, this embodiment proposes a solution using the EIT channel: within the 10ms window when the EIT channel is open, the cesium atoms in the Rydberg state respond continuously in real time; when a nanosecond-level transient event occurs, such as a micro-arc event, the atomic energy level immediately shifts, causing the transmitted light intensity to change immediately. This response is a physically instantaneous and continuous process, which does not require electronic sampling, so it can capture nanosecond-level transient events and ultimately record the waveform completely.

[0044] Summarize: In this embodiment, working atoms are excited through the CPT channel and EIT channel respectively using lasers of different wavelengths and modulation methods in the probe of the same dual-channel quantum sensor.

[0045] EIT channel: Targeting broadband radio frequency radiation caused by diaphragm breakdown and micro-arcs. Rydberg state atoms, as natural broadband receivers, can capture weak nanosecond-level discharge pulses that are difficult to detect with traditional antennas. The Rydberg state atom sensing unit excites cesium atoms to a high Rydberg state using three photons and encodes the electric field information into an optical signal using the electromagnetically induced transparency effect; the signal pumping and enhancement unit uses a surface plasmon waveguide chip packaged on the surface of the atomic gas cell to enhance the pumping efficiency of the electric field and atoms; CPT channel: Targets DC / low-frequency magnetic field changes caused by internal short circuits. Compared to traditional Hall effect sensors, it offers advantages such as low zero drift, high sensitivity (<1nT), and no hysteresis effect. In the earliest stages of thermal runaway in a power battery pack, a micro-short circuit occurs inside the pack, resulting in a weak magnetic field. In an atomic gas chamber filled with cesium atoms and a buffer gas, the CPT channel detects the signal frequency difference Δf=γB of the EIT channel to invert the magnetic field strength B in real time. Continuous monitoring allows for the extraction of magnetic field change trends and abnormal fluctuations.

[0046] Traditional thermal runaway monitoring methods, such as voltage, temperature, or single-band electromagnetic detection, typically acquire signals that are indirect, one-dimensional, and bandwidth-limited mappings of specific physical processes. These signals have low information entropy and are insufficient to uniquely characterize complex internal battery failure mechanisms in their early stages. However, the EIT channel can directly measure electric field changes from quasi-static to terahertz frequencies with extremely high sensitivity, thus fully capturing the full-spectrum electric field signals excited by processes such as ion rearrangement, interface charge accumulation, and plasma generation from micro-short circuits within the battery pack. The CPT channel simultaneously acquires extremely weak magnetic field disturbance signals in the corresponding frequency band. This collaboratively acquired signal set not only covers a complete electromagnetic fingerprint, from low frequencies reflecting ion migration and SEI film rupture to high frequencies reflecting arc discharge and intense chemical reactions, but also constructs a more comprehensive electromagnetic profile of battery faults. Low frequencies reflect ion migration and SEI film rupture, while high frequencies reflect arc discharge and intense chemical reactions.

[0047] Example 4 Based on Example 3, this example further discloses the specific steps of signal processing and feature extraction: Step 1. Data Synchronization and Preprocessing: The high-speed signal demodulation unit, pre-built on FPGA or DSP, performs digital sampling of the quantum sensing signal acquired by the dual-channel quantum sensor through the built-in high-speed ADC. At the same time, it also receives traditional monitoring data from traditional sensors, such as current, voltage, temperature and gas data, through the CAN bus. Among them, data synchronization, noise reduction and demodulation processing are performed on traditional monitoring data; To avoid mutual interference between the microwave modulation of the CPT channel and the pump light of the EIT channel for quantum sensing signals, this embodiment uses a time-division multiplexing measurement mode to acquire data from the two channels separately. It should be noted that there are two measurement modes for the dual-channel quantum sensor: Mode 1: Simultaneous measurement mode; Operating mode: The light sources of the EIT channel and CPT channel are turned on simultaneously and irradiate the atomic gas cell at the same time; It should be noted that running the measurement mode simultaneously may pose a risk of crosstalk in practical applications.

[0048] Mode 2: Time-division multiplexing measurement mode; Operating mode: The light sources in the EIT channel and CPT channel are controlled to work alternately based on a preset time interval; It should be noted that the Mode 2 scheme improves the risk of crosstalk and successfully avoids mutual interference between the microwave modulation of the CPT channel and the pump light of the EIT channel, making it the optimal solution. Therefore, this embodiment will use the time-division multiplexing measurement mode as an example for specific introduction. Specifically: The system employs millisecond-level time-division multiplexing control. In each cycle, the CPT channel is activated for the first 10ms to measure the magnetic field, and then the EIT channel is switched for the next 10ms to measure the radio frequency electric field, thereby achieving high-precision acquisition of the two physical quantities alternately.

[0049] Step 2. Feature Extraction: For internal short circuit faults, differential processing is performed on the signals acquired by the CPT channel to eliminate the contribution of bus current and geomagnetic background, thereby identifying continuous abnormal magnetic leakage signals that are unrelated to the charging and discharging process of the power battery pack, which serve as the basis for determining internal short circuits. For micro-arc and insulation faults, a short-time Fourier transform is performed on the signal acquired by the EIT channel, and diagnosis is made based on the spectral characteristics: pink noise in the low-frequency band (5kHz–50kHz) indicates DC arc, and broadband radio frequency bursts in the high-frequency band (10MHz–500MHz) correspond to breakdown discharge or poor contact. At the same time, fixed-frequency interference such as motor controller switching noise needs to be filtered out, thereby achieving accurate identification of multiple types of electrical faults.

[0050] For feature extraction of traditional monitoring data, a variety of conventional techniques can be used, as long as they meet the extraction accuracy requirements of practical applications. Therefore, they will not be elaborated on further here.

[0051] It should be noted that during the evolution of thermal runaway in the power battery pack, electric field detection based on the EIT channel and magnetic field measurement based on the CPT channel can provide characteristic identification with clear physical correlation for faults at different stages: In the initial micro-internal short circuit stage, dendrite growth forms a weak leakage path, and the CPT channel can detect local DC magnetic field anomalies: magnetic field difference value ≠ 0; while the EIT channel has no significant radio frequency electric field signal; as the fault develops into an intermittent short circuit or micro-discharge, contact instability triggers micro-sparks, the magnetic field fluctuates and is accompanied by step changes, and broadband RF pulses in the 10MHz–1GHz frequency band appear in the spectrum obtained by the EIT channel; if a series arc occurs due to a loose connection, the magnetic field exhibits high-frequency ripple, and the electric field channel exhibits typical 1 / f noise of 5k–50kHz; During the severe internal short circuit phase, large-area failure of the diaphragm triggers thermal runaway, causing strong distortion of the magnetic field, and continuous high-energy radio frequency radiation is detected by the EIT channel. Based on the combined characteristics of multiple physical quantities obtained by the dual-channel quantum sensor, early and phased fault identification and warning can be achieved when traditional parameters such as voltage and temperature have not yet changed significantly.

[0052] It should be noted that f represents frequency; 1 / f noise represents a typical noise pattern with higher noise amplitude in the low-frequency range and lower noise amplitude in the high-frequency range; 5k–50kHz is the frequency range in which this noise characteristic is significantly present. Example 5 Based on Example 4, this example further discloses the BMS deep fusion module; The BMS deep fusion module includes a detection unit, a classification unit, and a prediction unit connected in sequence; Operating logic: Based on the pre-built fusion discrimination module, the extracted features are further subjected to anomaly detection, anomaly classification, and anomaly prediction. Specifically, the features extracted in Example 2 are input into the detection unit, and the detection result describing whether there is an abnormal signal is output. If the detection result indicates that there is an anomaly, several abnormal signals are input into the classification unit for anomaly classification, and the classification result is output. The classification result is output to the prediction unit for anomaly prediction, and the prediction result is output. The fusion discrimination module mentioned in this embodiment can realize the entire process diagnosis from early anomaly detection to thermal runaway time prediction. It should be noted that fusion discrimination is a common method in existing technologies. The specific network architecture used by the fusion discrimination module in actual applications is not an innovative aspect of this embodiment and can be selected according to actual accuracy requirements or cost. For example, the network architecture of the fusion discrimination module in this embodiment adopts a cascaded fusion network architecture; It should be noted that during anomaly detection, due to the high sensitivity of quantum sensing signals, this embodiment uses the determination of whether the electromagnetic environment deviates from a reference state to capture the earliest and weakest anomalies. The reference state is the electromagnetic field of the power battery pack during normal operation. It should also be noted that anomaly detection is not limited to quantum sensing signals; traditional monitoring data is also detected. However, compared to traditional monitoring data, quantum sensors can detect anomalies much earlier.

[0053] When classifying anomalies, the multi-source anomaly data output by the anomaly detection unit is fused with traditional monitoring data, and faults are identified through preset logical rules: for example, if the detected electric field signal is higher than the preset value but the voltage and current are normal and the magnetic field is stable, there is a possibility of external interference; if the electric field signal is accompanied by an abnormal differential magnetic field, it is confirmed as an internal battery fault, thus effectively distinguishing between internal short circuits and external interference.

[0054] When making anomaly predictions, the remaining time before thermal runaway is predicted based on the prediction training set composed of data stored in the host computer of the cloud platform, thereby completing the step-by-step reasoning and decision support from anomaly detection, anomaly classification to evolution warning.

[0055] Example 6 Based on Example 5, this example further discloses and introduces the hierarchical collaborative execution module: In this embodiment, the hierarchical collaborative execution module controls the execution of corresponding early warning actions based on the detection results, classification results, and prediction results obtained by the BMS deep fusion module, combined with a preset early warning strategy. Specifically: The preset early warning strategy is a gradual strategy; When no abnormal data is found during anomaly detection, the preset early warning strategy will not be activated; When an anomaly is detected, the initial data that can usually be detected is only the quantum sensing signal. Therefore, the setting here is that the first detected anomaly is an anomaly in the quantum sensing signal, specifically manifested as an electromagnetic environment deviation from the reference state. At this time, the first-level warning in the preset warning strategy is triggered, logs are recorded, and the sampling frequency of the signal processing and feature extraction modules is increased. When the classification results indicate the presence of a stable leakage current event caused by an internal short circuit or a rapid insulation degradation event, the secondary warning in the preset warning strategy is triggered. It should be noted that even if the temperature does not change in the traditional monitoring data, the secondary warning is still triggered, and a signal suggesting power limiting is sent to the vehicle controller.

[0056] During the anomaly prediction phase, the classification results are used as input to predict the remaining time before thermal runaway occurs. It should be noted that when the quantum sensing signal reaches saturation, accompanied by a voltage drop or the detection of trace gases, thermal runaway is inevitable. In this case, a level-three warning must be triggered immediately, and final protective measures must be implemented, including disconnecting the high-voltage relay, activating the liquid system at full speed, and notifying the occupants to evacuate. A strong electric arc is generated when the quantum sensing signal reaches saturation.

[0057] Summarize: The innovative effects of combining traditional and quantum sensors: The dual-channel quantum sensor used in this embodiment measures the electric field based on the fact that when an internal short circuit involves metal melting and vaporization, such as when the diaphragm collapses and the positive and negative electrodes come into direct contact, a tiny local electric arc may be generated. This arc has a strong signal and is easy to detect, but it is sometimes already a manifestation of the middle and late stages of thermal runaway. The magnetic field measurement is based on the fact that the internal short circuit point generates material vaporization and ionization, spark discharge, etc., which may occur before significant changes in temperature and voltage, providing a valuable early warning window.

[0058] However, while this technology boasts significant advantages in early fault signal capture due to its high sensitivity and wideband response, various interferences in the complex automotive electromagnetic environment can severely impact the accurate extraction and identification of signals. These electromagnetic signals not only arise during battery thermal runaway but also occur within the vehicle itself, such as in motor drives, high-frequency PWM noise generated by IGBT / MOSFET switches, onboard DC-DC converters, air conditioning compressors, and various electronic control units; as well as in the external environment, such as electromagnetic radiation from high-voltage power lines, substations, communication base stations, wireless devices from other vehicles, and charging piles. Significant interference occurs in this complex automotive electromagnetic environment. First, strong background electromagnetic noise, such as wideband noise generated by motor drives and DC-DC converters, can completely drown out the weak early electromagnetic radiation signals excited by the breaking of molecular bonds within the battery or localized overheating, preventing the system from effectively detecting them and leading to missed alarms. Second, external transient electromagnetic pulse interference, such as switching actions in the vehicle's electrical system, relay on / off states, or electrostatic discharge events in the environment, can produce time-domain or frequency-domain characteristics similar to the initial stage of thermal runaway, easily misinterpreted by the system as fault signals, thus triggering false alarms and reducing the reliability of the warning system. Furthermore, the most common impact is that interference significantly reduces the signal-to-noise ratio of the signal, making it extremely difficult to extract weak fault features, which often have specific modulation characteristics, from the mixed electromagnetic background. This not only increases the complexity of signal processing, but may also lead to incomplete or distorted feature extraction, thereby affecting the accuracy of subsequent pattern recognition and early warning decisions.

[0059] However, by using traditional sensors and dual-channel quantum sensors for collaborative monitoring, the system no longer relies solely on electromagnetic signals or traditional sensing. Instead, it comprehensively assesses information from multiple sensors, including voltage, temperature, gas, and pressure, and can also add smoke and electromagnetic / acoustic emission data as needed. The system only determines thermal runaway when multiple sensors simultaneously malfunction, thus fundamentally eliminating false alarms from a single sensor and improving system robustness. Firstly, at the spatial and hardware level, near-field probes are deployed to enhance signal strength, combined with directional shielding and characteristic frequency filtering techniques to suppress environmental interference. Secondly, at the signal level, time-frequency analysis methods are used to capture transient characteristics, and machine learning is further employed to intelligently identify thermal runaway arcs and external interference. Finally, at the system integration and verification level, sensor environmental adaptability selection, multi-level early warning strategy design, and testing and calibration in the complex electromagnetic environment of a real vehicle ensure the system's reliability and engineering feasibility in practical applications.

[0060] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A power battery runaway early warning system based on quantum sensors, characterized in that, include: The multimodal acquisition module uses both traditional sensors and dual-channel quantum sensors to monitor the power battery pack and acquire monitoring data. The monitoring data includes: traditional monitoring data, electric field data, and magnetic field data; traditional sensors include: current sensors, voltage sensors, temperature sensors, and gas sensors; and traditional monitoring data includes: current, voltage, temperature, and gas data. The signal processing and feature extraction module preprocesses the monitoring data and extracts features from the preprocessed monitoring data. The BMS deep fusion module, based on a pre-built deep learning model, sequentially performs anomaly detection, anomaly classification, and anomaly prediction on all extracted features. The hierarchical collaborative execution module controls the execution of corresponding early warning actions based on the detection results, classification results, and prediction results obtained by the BMS deep fusion module, combined with preset early warning strategies. All data generated and acquired during the operation of the multimodal acquisition module, signal processing and feature extraction module, BMS deep fusion module, and hierarchical collaborative execution module, as well as the early warning records of the early warning actions, are uploaded to the cloud platform's host computer storage; each module retrieves the data stored on the cloud platform's host computer as needed. The dual-channel quantum sensor has two channels: the EIT channel and the CPT channel; The EIT channel and the CPT channel share the same atomic gas chamber. The light sources of the EIT channel and the CPT channel are combined by a wavelength division multiplexer or a spatial optical path and then injected into the atomic gas cell. The EIT channel uses an 852nm probe light and a 509nm pump light to form the Cs6S. 1 / 2 →6P 3 / 2 →50D 5 / 2 The Rydberg stepped energy level system serves as a light source for detecting high-frequency and transient electric fields inside a power battery pack. The CPT channel and the EIT channel use the same probe beam locked to the 852nm transition of the D2 line of cesium atoms; after the probe beam is split and enters the CPT channel, coherent sidebands are generated by current modulation. The positive and negative first-order sideband frequency difference in the coherent sideband matches the hyperfine splitting frequency of the cesium atom's ground state, which is 9.192 GHz, corresponding to the modulation frequency f. mod ≈4.596GHz; In the dual-channel quantum sensor, the EIT channel and CPT channel are configured to operate in two modes: simultaneous measurement mode or time-division multiplexing measurement mode. In the time-division multiplexing measurement mode, the CPT channel and the EIT channel work alternately, and the time interval between the alternations satisfies the acquisition of the transient electric field distribution.

2. The power battery runaway early warning system based on quantum sensors according to claim 1, characterized in that, The working atom in the atomic gas cell is a cesium atom; The buffer gas filled in the atomic gas chamber is nitrogen or argon at a concentration of 10 to 50 Torr.

3. The power battery runaway early warning system based on quantum sensors according to claim 2, characterized in that, Feature extraction is performed on the preprocessed monitoring data: Differential processing is used to perform feature processing on magnetic field data, and short-time Fourier transform is used to extract features from electric field data.

4. The power battery runaway early warning system based on quantum sensors according to claim 3, characterized in that, The preset early warning strategy is a gradual strategy, specifically: Level 1 warning: Based on the anomaly detection phase, an anomaly is detected in the quantum sensing signal acquired by the dual-channel quantum sensor and triggered. Log is recorded and the sampling frequency of the signal processing and feature extraction modules is increased. Level 2 warning: Based on the classification results, if a stable leakage event caused by an internal short circuit or a rapid insulation degradation event is triggered, a signal suggesting power limitation is sent to the vehicle controller. The Level 3 warning is triggered when the quantum sensing signal reaches saturation and is accompanied by a voltage drop or the detection of trace gases. It then performs the final protective actions of cutting off the high-voltage relay, starting the liquid system at full speed, and notifying the occupants to evacuate.

5. The power battery runaway early warning system based on quantum sensors according to claim 1, characterized in that, The electric field data includes: high-frequency electric field distribution and transient electric field distribution.

6. A power battery runaway early warning method based on quantum sensors, characterized in that, include: S1. The power battery pack is monitored using both traditional sensors and a dual-channel quantum sensor to acquire monitoring data. The monitoring data includes: traditional monitoring data, electric field data, and magnetic field data. The traditional sensors are: current sensor, voltage sensor, temperature sensor, and gas sensor. The traditional monitoring data includes: current, voltage, temperature, and gas data. S2. Preprocess the monitoring data and extract features from the preprocessed monitoring data; S3. Based on the pre-built deep learning model, all extracted features are sequentially subjected to anomaly detection, anomaly classification, and anomaly prediction, and the corresponding detection results, classification results, and prediction results are obtained. S4. Combine the detection results, classification results, and prediction results with the preset early warning strategy to control and execute the corresponding early warning action; S5. Upload all data generated and acquired during the monitoring data acquisition stage, the monitoring data preprocessing stage, the anomaly detection, anomaly classification and anomaly prediction stage of the extracted features, and the corresponding early warning action execution process, as well as the early warning records of the early warning actions, to the cloud platform host computer storage. The dual-channel quantum sensor has two channels: the EIT channel and the CPT channel; The EIT channel and the CPT channel share the same atomic gas chamber. The light sources of the EIT channel and the CPT channel are combined by a wavelength division multiplexer or a spatial optical path and then injected into the atomic gas cell. The EIT channel uses an 852nm probe light and a 509nm pump light to form the Cs6S. 1 / 2 →6P 3 / 2 →50D 5 / 2 The Rydberg stepped energy level system serves as a light source for detecting high-frequency and transient electric fields inside a power battery pack. The CPT channel and the EIT channel use the same probe beam locked to the 852nm transition of the D2 line of cesium atoms; after the probe beam is split and enters the CPT channel, coherent sidebands are generated by current modulation. The positive and negative first-order sideband frequency difference in the coherent sideband matches the hyperfine splitting frequency of the cesium atom's ground state, which is 9.192 GHz, corresponding to the modulation frequency f. mod ≈4.596GHz; In the dual-channel quantum sensor, the EIT channel and CPT channel are configured to operate in two modes: simultaneous measurement mode or time-division multiplexing measurement mode. In the time-division multiplexing measurement mode, the CPT channel and the EIT channel work alternately, and the time interval between the alternations satisfies the acquisition of the transient electric field distribution.

Images