Battery pack monitoring method and system based on random spectrum coding, device, storage medium

Abstract

The application provides a battery pack monitoring method and system based on random spectrum coding, equipment and a storage medium. The battery pack monitoring method based on random spectrum coding monitors the pressure between battery cells of a battery pack and the anti-private disassembly of a battery pack sensor based on the principle of random spectrum coding. The method can perform high-precision and high-reliability distributed monitoring on the pressure between the battery cells, and simultaneously performs physical-level identity recognition on the sensor, so that the purpose of anti-disassembly monitoring is achieved.

Description

Technical Field

[0001] This invention belongs to the fields of fiber optic sensing, artificial intelligence and security authentication technology, and specifically relates to a battery pack monitoring method based on random spectral coding, as well as a battery pack monitoring system, device and storage medium based on random spectral coding. Background Technology

[0002] Fiber Bragg grating (FBG) sensing technology, with its advantages of electromagnetic interference resistance, small size, and distributed measurement capabilities, has been widely applied in fields such as civil engineering and petrochemicals. In existing applications, wavelength division multiplexing (WDM) technology is typically used to construct sensor arrays, acquiring the measured physical quantity by monitoring the drift of the fiber Bragg grating's center wavelength. This approach primarily utilizes the "wavelength encoding" characteristic of the grating, while the rich morphological information in the grating's reflection spectrum (such as bandwidth, sidelobes, and symmetry)—these "spectral fingerprints" generated by manufacturing process differences—are often ignored as noise in traditional applications. Meanwhile, the safety monitoring of battery packs in new energy electric vehicles faces severe challenges: the "breathing effect" of the cells during charging and discharging leads to dynamically changing pressure between the cells, which is crucial for battery health assessment and thermal runaway early warning. However, the cells are in surface contact with extremely uneven pressure distribution, and the battery pack interior suffers from strong electromagnetic interference, limited space, and harsh environments, making it difficult for existing technologies to effectively address these challenges.

[0003] On the other hand, existing cell-to-cell pressure monitoring solutions mainly include: installing electronic pressure sensors inside the cells or on the battery pack casing; however, these solutions cannot directly measure the contact pressure between cells, and the electronic components are susceptible to electromagnetic interference and pose a risk of electrical sparks; sandwiching thin-film pressure sensors between cells, but poor surface contact uniformity leads to low measurement repeatability, complex leads are prone to aging after long-term use; and indirectly calculating pressure changes through parameters such as voltage and current, but this method has low accuracy, excessive model dependence, and cannot achieve real-time accurate monitoring. Therefore, how to achieve high-precision and high-reliability distributed online monitoring of cell-to-cell pressure without changing the existing battery pack structure, while overcoming the measurement uncertainties caused by non-uniform surface contact of the cells, is a technical challenge that urgently needs to be solved in this field.

[0004] To address the aforementioned issues, some studies have attempted to improve contact stability by optimizing the packaging structure of electronic sensors. However, repeatability is difficult to guarantee due to limitations in cell surface morphology and installation tolerances. Other solutions employ multi-point thin-film sensor arrays to acquire pressure distribution, but these suffer from complex leads, poor long-term reliability, and susceptibility to aging and failure in electrolyte environments. Existing technologies still suffer from common drawbacks such as indirect measurement, insufficient anti-interference capabilities, poor structural adaptability, and difficulties in system integration. Therefore, a novel sensing technology is urgently needed that can deeply integrate the inherent advantages of fiber optic sensing with the requirement for direct pressure measurement between cell cells. Summary of the Invention

[0005] The purpose of this invention is to provide a battery pack monitoring method based on random spectral coding, which solves the shortcomings of existing technologies such as low accuracy and reliability of sensor monitoring of battery pack cell pressure, and easy replacement or tampering of sensor identity authentication.

[0006] The purpose of this invention is also to provide a battery pack monitoring system, device, and storage medium based on random spectral coding.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A battery pack monitoring method based on random spectral coding includes the following steps: S1. Transmit the continuous spectrum incident light to the pressure sensing fiber grating in the fiber grating pressure sensor and collect the original reflection spectrum reflected by the pressure sensing fiber grating. S2. Preprocess the original reflectance spectrum, extract the reflectance peak region, and output a standardized spectral fragment; S3. Extract spectral features from standardized spectral fragments to form high-dimensional spectral feature vectors; S4. Perform pressure calculation on the extracted high-dimensional spectral feature vector to output the pressure value, or compare the extracted high-dimensional spectral feature vector with the initial spectral feature vector to determine whether the fiber optic pressure sensor has been disassembled or replaced.

[0008] In this invention, the preprocessing of the original spectrum in S2 includes filtering and denoising, limiting correction, and normalization of the original spectral data. This eliminates the influence of factors such as light source fluctuations and fiber bending loss on the results.

[0009] Furthermore, in S2, the reflection peak region is selected with a center wavelength of ±5 nm, resulting in 400 wavelength points.

[0010] In this invention, the spectral features extracted in S3 include multidimensional morphological parameters such as center wavelength, bandwidth of reflection peak, symmetry, sidelobe height, and ripple frequency.

[0011] In this invention, S3 extracts spectral features from standardized spectral fragments using a one-dimensional convolutional neural network with a network structure consisting of three convolutional layers and two fully connected layers.

[0012] In some embodiments of the present invention, the high-dimensional spectral feature vector v

[0013] in: λ B Center wavelength; Δλ 3dB 3 dB bandwidth; Skewness: spectral asymmetry; Kurtosis: spectral peak intensity; Sidelobe ratio: the ratio of sidelobe to main lobe.

[0014] In this invention, in step S4, the extracted high-dimensional spectral feature vector is used to calculate the pressure value, and the pressure value is obtained by solving the pressure value according to the established spectral feature-pressure relationship model.

[0015] Furthermore, the spectral feature-pressure relationship model is obtained by calibrating the fiber Bragg grating pressure sensor across its entire range on a standard pressure platform. The calibration method includes the following steps: performing a loading-unloading cycle on the fiber Bragg grating pressure sensor under different temperature conditions, simultaneously collecting standard pressure values ​​and spectral data, dividing the collected calibration data into training, validation, and test sets according to a ratio, extracting spectral feature vectors using a one-dimensional convolutional neural network, establishing a regression model from feature vectors to pressure values ​​using a BP neural network or random forest algorithm, and fixing the model parameters after the model training converges.

[0016] Furthermore, the relevant calculation formulas for the calibration process are as follows: The equivalent stiffness of the arc-shaped sensing element in the fiber Bragg grating pressure sensor is ki (N / mm). The total pressure acting on the sensor panel is F, which is evenly distributed across N arc-shaped elements. The pressure on each arc-shaped element is:

[0017] F i The pressure exerted on the bow-shaped component; F: Total pressure acting on the sensor (i.e., pressure in the battery pack). N: Number of sensors; The horizontal displacement δi generated by the arc-shaped sensing element under vertical pressure Fi has an approximately linear relationship with the pressure F:

[0018] δ i : Horizontal displacement of the bow-shaped element; k i : The equivalent stiffness of the i-th arc-shaped element; The horizontal displacement δi applied to the fiber grating fixed on the arc-shaped sensing element results in the following axial strain:

[0019] Lg: Effective length of the fiber grating; ε i : Axial strain; Based on the wavelength drift formula, the relationship between pressure F and wavelength drift is as follows:

[0020] p e Effective elastic coefficient; K T Temperature sensitivity coefficient; ΔT: Temperature change; Δλ B,i : The measured wavelength drift of the i-th grating; Let the temperature drift measured by the temperature reference grating be Δλ. B,T =K T Δ T The net drift caused by pressure is:

[0021] Δλ B,i The measured wavelength drift of the i-th grating; Δλ B,T Temperature reference grating wavelength drift; For N pressure measuring points, the pressure measured at each point is P1, P2, ..., P N The overall equivalent expansion force of the battery module can be defined as

[0022] P i Pressure value at the i-th measuring point; P total Total equivalent expansion force; P avg Average expansion force; Based on the mechanical model of the battery cell-module, the equivalent pressure distribution coefficient on the side of the battery cell can be calculated:

[0023] P j Pressure value at the j-th measuring point; eta i is the equivalent pressure distribution coefficient for the i-th measuring point.

[0024] In this invention, a BP neural network machine learning algorithm is used to establish a mapping from spectral feature vectors to pressure values:

[0025] Where W and b are the network weights and biases, obtained through training with calibration data.

[0026] In this invention, in step S4, it is determined whether the sensor has been disassembled by calculating whether the cosine similarity between the extracted high-dimensional spectral feature vector and the initial spectral feature vector is greater than or equal to a threshold. If yes, it is determined that the sensor has not been disassembled and the verification is passed. If no, it is determined that the sensor has been disassembled or replaced, and an anti-tampering alarm is issued.

[0027] In some embodiments of the present invention, the runtime cosine similarity (sim) is calculated in the following manner:

[0028] Among them, v s v is the spectral feature vector of the fiber Bragg grating pressure sensor. s (0) is the initial spectral feature vector when the sensor is installed, which is stored as a template.

[0029] Furthermore, the threshold θ is set to 0.85~0.98.

[0030] Preferably, the threshold θ is taken as 0.90~0.95.

[0031] A battery pack monitoring system based on random spectral coding, comprising: A light source module, wherein the light source module is used to provide incident light of a continuous spectrum; A spectral acquisition module is used to acquire the complete reflection spectrum of incident light from the light source module transmitted to the pressure sensing fiber optic grating. The data processing module is used to process and analyze the reflection spectrum, perform pressure calculation to obtain the battery pack pressure value, or determine whether the fiber optic pressure sensor has been disassembled.

[0032] In this invention, the data processing module includes: The spectral preprocessing unit is used to filter, denoise, perform baseline correction and normalization on the acquired raw spectral data. The AI ​​feature extraction unit is used to extract the multi-dimensional spectral morphological feature vector of each pressure sensing fiber grating from the preprocessed spectrum using a deep neural network. The stress calculation unit is used to input the extracted spectral feature vector into a pre-trained spectral feature-stress regression model to calculate the stress value. The identification unit is used to store the initial spectral feature vector of the sensor as a template when it is installed, and compare the current spectral feature vector with the template during each measurement. If the similarity is lower than a set threshold, an anti-tampering alarm signal is output.

[0033] In this invention, the AI ​​feature extraction unit adopts a one-dimensional convolutional neural network. The network structure includes at least two convolutional layers and at least one fully connected layer. The input is a preprocessed sequence of spectral data points, and the output is a 64-dimensional or 128-dimensional spectral feature vector.

[0034] In this invention, the battery pack monitoring system based on random spectral coding further includes a fiber Bragg grating pressure sensor. The fiber Bragg grating pressure sensor includes a sensing probe, which comprises a base plate, a pressure contact panel, and an elastic sensitive layer. The base plate and the pressure contact panel are fixedly connected, and the elastic sensitive layer is disposed between the base plate and the pressure contact panel. The elastic sensitive layer includes multiple bow-shaped sensing elements, one end of which is connected to the base plate, and the other end is connected to the pressure contact panel. Each bow-shaped sensing element is provided with a pressure sensing fiber Bragg grating, such that the pressure sensing fiber Bragg gratings of different bow-shaped sensing elements are connected in series through optical fibers.

[0035] In this invention, the sensing probe further includes a temperature-compensated fiber Bragg grating, which is connected in series with a pressure-sensing fiber Bragg grating.

[0036] In this invention, the bow-shaped sensing element is a bow-shaped diaphragm structure, which is formed by stamping a beryllium copper alloy sheet, and the top of the bow-shaped sensing element has an optical fiber fixing groove.

[0037] Furthermore, the arc-shaped sensing element has a thickness of 0.3 mm, a chord length of 25 mm, an arc height of 1.5 mm, and an arch width of 3 mm.

[0038] In this invention, the base plate of the sensing probe is fixedly connected to the battery module; the pressure contact panel is in contact with the battery cell.

[0039] In this invention, the contact surface between the pressure contact panel and the battery cell is provided with a hemispherical protrusion structure, and the hemispherical protrusion structure is arranged in an array.

[0040] Furthermore, the hemispherical protrusion structure has a protrusion diameter of 1.0 mm, a height of 0.3 mm, and a spacing of 1.5 mm.

[0041] Furthermore, the hemispherical protrusions are arranged in a hexagonal array.

[0042] A battery pack monitoring device based on random spectral coding is disclosed. The device includes a processor, a memory, and a computer program stored in the memory and executable by the processor. When the computer program is executed by the processor, it implements the steps of the battery pack monitoring method based on random spectral coding described above.

[0043] A storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of the above-described battery pack monitoring method based on random spectral coding.

[0044] The present invention has the following beneficial effects: (1) The present invention is a battery pack monitoring method based on random spectral coding. Based on the principle of random spectral coding, the method monitors the pressure between battery pack cells and the anti-tampering of battery pack sensors. This method can perform high-precision and high-reliability distributed monitoring of the pressure between cells, and at the same time perform physical-level identification of sensors to achieve the purpose of anti-tampering monitoring.

[0045] (2) The battery pack monitoring method based on random spectral coding of this invention uses an AI algorithm to extract multi-dimensional spectral morphological features determined by manufacturing randomness from the complete reflection spectrum of the fiber optic grating as a "spectral fingerprint" and establishes a "spectral feature-pressure" deep learning model to achieve high-precision pressure calculation. At the same time, the system uses the initial spectral features of the sensor at the time of installation as a unique identity template, and compares the current spectral features in real time through an AI similarity matching algorithm to achieve physical-level identity recognition and tamper-proof monitoring of the sensor.

[0046] (3) This invention transforms the individual differences of the grating, which are traditionally regarded as errors, into a dual technical advantage of measurement compensation and safety authentication. While improving the accuracy of pressure measurement, it provides the sensor itself with non-replicable anti-counterfeiting and anti-tampering functions, which is particularly suitable for application scenarios such as new energy battery packs that have strict requirements for safety and data credibility.

[0047] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0048] Figure 1 This is one of the structural schematic diagrams of the fiber optic pressure sensor probe of the present invention; Figure 2 This is the second schematic diagram of the structure of the fiber optic pressure sensor probe of the present invention; Figure 3 This is a schematic diagram of the CTP battery pack cell expansion pressure monitoring system of the present invention; Figure 4 This is a schematic diagram of the structure of the bow-shaped sensing element of the present invention; Figure 5 This is a flowchart of the battery pack monitoring method based on random spectral coding according to the present invention; Figure 6 These are detailed diagrams illustrating each step of the battery pack monitoring method based on random spectral coding of the present invention; The following are labeled in the attached diagram: 1. Base plate; 2. Pressure contact panel; 3. Bow-shaped sensitive element; 4. Pressure sensing fiber optic grating; 5. Temperature-compensated fiber optic grating; 6. Light source module; 7. Spectrum acquisition module; 8. Data processing module. Detailed Implementation

[0049] The technical solution of the present invention will be further described below with reference to specific embodiments, so that those skilled in the art can better understand and implement the technical solution of the present invention.

[0050] This invention provides a battery pack monitoring method, system, device, and storage medium based on random spectral coding. Based on the principle of random spectral coding, it monitors the pressure between battery cells and the anti-tampering capabilities of battery pack sensors. The method includes steps such as data acquisition, spectral preprocessing, AI feature extraction, feature vector branching, output, and application. It achieves high-precision and high-reliability distributed online monitoring of pressure between cells and identifies the sensors to prevent unauthorized tampering. A battery pack monitoring system based on random spectral coding is also provided, particularly suitable for CTP battery packs, featuring an ultra-thin design with high adaptability and high reliability.

[0051] This invention mainly addresses the following issues: 1. Traditional fiber optic grating sensing only utilizes the center wavelength, wasting rich spectral morphology information and limiting measurement accuracy; 2. Existing sensors lack physical-level authentication mechanisms, making them easily replaceable or tampered with; 3. It is difficult to simultaneously meet the dual requirements of direct pressure measurement and sensor anti-counterfeiting in complex environments such as battery packs.

[0052] The random spectral coding principle upon which this invention is based is as follows: During the manufacturing process of each fiber Bragg grating, due to factors such as imperfections in the phase mask, power fluctuations in the ultraviolet laser, and geometric tolerances of the fiber, its refractive index modulation profile is not ideally uniform, but rather exhibits random microscopic inhomogeneities. This inhomogeneity results in the reflection spectrum containing unique morphological characteristics in addition to the center wavelength: bandwidth, sidelobe structure, symmetry, ripple frequency, etc. These characteristics are determined by random factors in the manufacturing process and possess the following properties: Uniqueness: The spectral pattern of each grating is unique, similar to a human fingerprint; Non-reproducibility: Even using the same manufacturing process and parameters, it is impossible to reproduce the exact same spectral morphology; Stability: The spectral morphology remains stable throughout the sensor's lifespan (unless the grating is physically damaged). Stress sensitivity: The spectral morphology changes with strain / pressure, and can be used for sensing.

[0053] This invention refers to this naturally occurring random spectral difference as "random spectral coding" and uses it as the sensor's identification and measurement information carrier.

[0054] Example 1

[0055] like Figure 1-4 The battery pack monitoring system shown includes a fiber Bragg grating pressure sensor, which comprises a sensing probe and a demodulation system.

[0056] The sensing probe includes a base plate 1, a pressure contact panel 2, and an elastic sensitive layer. The base plate 1 and the pressure contact panel 2 are fixedly connected, and the elastic sensitive layer is disposed between the base plate 1 and the pressure contact panel 2. The base plate 1 of the sensing probe is fixedly connected to the battery module; the pressure contact panel 2 is in contact with the battery cell.

[0057] In this embodiment, the base plate 1 serves as the fundamental support component for the sensor. It is made of 304 stainless steel through precision die casting, with a thickness of 0.5 mm and planar dimensions of 85 mm × 125 mm. Mounting holes are provided at the four corners of the base plate 1 for fixing the sensor to the side plate of the battery module or the cell support. Four positioning grooves are machined on the inner surface of the base plate 1, with a groove depth of 0.2 mm and a positional tolerance of ±0.05 mm. The pressure contact panel 2 is located on the top layer of the sensor and directly contacts the side of the cell. The pressure contact panel 2 is also made of 304 stainless steel through die casting, with a thickness of 0.5 mm and planar dimensions identical to the base plate 1. The contact surface between the pressure contact panel 2 and the cell has a hemispherical protrusion structure arranged in a hexagonal array. The outer surface of the pressure contact panel 2 has an array of micro-protrusion structures to form multi-point contact with the cell surface, transforming the random surface contact of the cell surface into multiple stable point contacts, eliminating the problem of uneven pressure distribution caused by uneven cell surface or installation angle deviation. The hemispherical protrusion structure has protrusions with a diameter of 1.0 mm, a height of 0.3 mm, and a spacing of 1.5 mm. The inner side of the pressure contact panel 2 is also machined with four positioning grooves, corresponding one-to-one with the grooves on the base plate 1. Guide sleeve holes with a diameter of Φ2.1 mm are also provided at the four corners of the pressure contact panel 2. The base plate 1 and the pressure contact panel 2 are laser-welded around their perimeter for a fully enclosed metal shell, preventing floating and loosening between mechanical parts.

[0058] The elastic sensitive layer includes multiple bow-shaped sensitive elements 3, one end of which is connected to the base plate 1, and the other end is connected to the pressure contact panel 2. Each bow-shaped sensitive element 3 is equipped with a pressure-sensing fiber grating 4, so that the pressure-sensing fiber gratings 4 of different bow-shaped sensitive elements 3 are connected in series through optical fibers. Under the action of vertical pressure, the bow-shaped sensitive element 3 undergoes elastic deformation, converting the vertical pressure into horizontal tension, and transmitting it to the pressure-sensing fiber grating 4 fixed on it, causing axial strain in the pressure-sensing fiber grating 4, resulting in changes in its center wavelength and spectral morphology.

[0059] In this embodiment, four bow-shaped sensing elements 3 are provided. Each bow-shaped sensing element 3 is a bow-shaped diaphragm structure, formed by stamping a beryllium copper alloy sheet. The arch of each bow-shaped sensing element 3 has an optical fiber fixing groove. The thickness a of the bow-shaped sensing element 3 is 0.3 mm, the chord length d is 25 mm, the arc height h is 1.5 mm, the arch width b is 3 mm, and the width of the optical fiber fixing groove is 0.5 mm. The four bow-shaped sensing elements 3 are symmetrically arranged at the four corners of the sensor, and their two ends are fixedly connected to the base plate 1 and the pressure contact panel 2 respectively by laser welding. The laser welding parameters are: power 150 W, welding speed 20 mm / s, weld width 0.3 mm, and weld depth 0.2 mm. Each bow-shaped sensing element 3 is fixed with a pressure-sensing fiber optic grating 4 through the optical fiber fixing groove. Different pressure-sensing fiber optic gratings 4 are connected by optical fibers, and different pressure-sensing fiber optic gratings 4 have different center wavelengths. In this embodiment, the center wavelengths of the four pressure-sensing fiber Bragg gratings 4 are 1530.5 nm, 1535.2 nm, 1540.8 nm, and 1545.3 nm, respectively. The sensitive area of ​​the grating is 4 cm long, the reflectivity is ≥90%, and the 3 dB bandwidth is ≤0.3 nm. Each grating is pre-strained with 0.2% (i.e., pre-stretched 10 μm) during installation, and then fixed in the fiber optic fixing groove of the bow-shaped sensitive element 3 using UV-curable adhesive with a dot diameter of 0.3 mm. It is cured by UV irradiation for 30 seconds. The temperature-compensated fiber Bragg grating 5 has a center wavelength of 1550.1 nm, is encapsulated in a stainless steel capillary tube with an outer diameter of 0.8 mm and a length of 10 mm, and is sealed at both ends with epoxy resin and fixed to the side of the pressure contact panel 2.

[0060] In this embodiment, the sensing probe also includes a temperature-compensated fiber Bragg grating 5, which is connected in series with the pressure-sensing fiber Bragg grating 4. The temperature-compensated fiber Bragg grating 5 is fixed to the non-stressed area of ​​the pressure contact panel 2. The temperature-compensated fiber Bragg grating 5 serves as a reference grating, using the wavelength drift of the pressure-sensing fiber Bragg grating 4 as a reference to calculate the relative drift between the two, thereby improving measurement accuracy. All five fiber Bragg gratings are connected in series on the same optical fiber via fusion splicing, with a splice loss ≤0.01 dB. The optical fiber exit is protected by a Φ2.0 mm armored optical cable and exits the sensor housing via an M5 threaded sealing connector; the sealing ring is made of fluororubber.

[0061] The demodulation system includes a light source module 6, a spectrum acquisition module 7, and a data processing module 8.

[0062] The light source module 6 is used to provide continuous spectrum incident light to the sensing probe. The light source module 6 is equipped with a broadband light source, which adopts a superluminescent diode with an output wavelength range of 1525 nm ~ 1565 nm, an output power of 15 mW, and a spectral flatness of ±2 dB.

[0063] The spectral acquisition module 7 is used to acquire the complete reflection spectrum reflected by the pressure-sensing fiber optic grating 4, rather than just the center wavelength value. The spectral acquisition module 7 uses an InGaAs linear array detector with a wavelength resolution of 0.5 pm, a sampling rate of 2 kHz, and a dynamic range of 60 dB.

[0064] The data processing module 8 is used to process and analyze the reflectance spectrum, calculate the battery pack pressure value, or determine whether the fiber Bragg grating pressure sensor has been disassembled. The data processing module 8 uses an ARM Cortex-A53 processor, is equipped with 2 GB of RAM, runs a Linux operating system, and has a built-in AI algorithm model. In this embodiment, the data processing module 8 extracts spectral feature vectors from multiple fiber Bragg gratings in the same sensing probe, constructs a joint feature matrix, and uses the joint features of the multiple gratings for identification and pressure calculation to improve identification robustness and measurement accuracy.

[0065] In this embodiment, the data processing module 8 includes: The spectral preprocessing unit is used to filter, denoise, perform baseline correction, and normalize the acquired raw spectral data.

[0066] The AI ​​feature extraction unit is used to extract multi-dimensional spectral morphology feature vectors for each pressure-sensing fiber grating 4 from the preprocessed spectrum using a deep neural network. In the spectral feature vectors extracted by the AI ​​feature extraction unit, the spectral morphology features generated by the randomness in the fiber grating manufacturing process (including grating period inhomogeneity, refractive index modulation depth fluctuations, and fiber geometric tolerances) serve as the sensor's unique "spectral fingerprint." This fingerprint is bound to the sensor's physical entity and cannot be replicated. The AI ​​feature extraction unit employs a one-dimensional convolutional neural network. The network structure includes at least two convolutional layers and at least one fully connected layer. The input is a sequence of preprocessed spectral data points, and the output is a 64-dimensional or 128-dimensional spectral feature vector.

[0067] The stress calculation unit is used to input the extracted spectral feature vector into a pre-trained spectral feature-stress regression model to calculate the stress value. The identification unit stores the initial spectral feature vector of the sensor during installation as a template. During each measurement, it compares the current spectral feature vector with the template for similarity. If the similarity is lower than a set threshold, it outputs an anti-tampering alarm signal. The identification unit calculates the cosine similarity or Euclidean distance between the current spectral feature vector and the template feature vector. A mismatch is determined when the cosine similarity is lower than a preset threshold θ or the Euclidean distance is higher than a preset threshold. The threshold θ ranges from 0.85 to 0.98. Preferably, the threshold θ ranges from 0.90 to 0.95. The following calibration method is used to establish the mapping relationship between the spectral characteristics of the sensor and pressure. The specific calibration method includes the following steps: the fiber optic pressure sensor probe is subjected to loading-unloading cycles under different temperature conditions, and standard pressure values ​​and spectral data are collected simultaneously. The collected calibration data is divided into training set, validation set and test set according to the proportion. A one-dimensional convolutional neural network is used to extract spectral feature vectors. A BP neural network or random forest algorithm is used to establish a regression model from feature vectors to pressure values. After the model training converges, the model parameters are solidified into data processing module 8.

[0068] In this embodiment, the calibration equipment used is an Instron 5965 universal testing machine (range 0 ~ 10 kN, accuracy ±0.5% FS) and an ESPEC SH-241 constant temperature chamber (temperature range -40 ℃ ~ 150 ℃, temperature control accuracy ±0.3 ℃). During the calibration process, the pressure difference is zeroed under normal pressure conditions to eliminate the influence of inconsistencies in fiber grating fabrication on the consistency of the pressure sensor. The specific calibration steps are as follows: Step 1: Place the cell-cold plate simulation device with the sensor assembled under the indenter of the universal testing machine. Under 25℃ conditions, perform a loading-unloading cycle of 0 N → 5000 N → 0 N at a loading rate of 0.1 mm / min, repeating 5 times. Simultaneously collect the standard pressure value output by the universal testing machine and the spectral data output by the spectral acquisition module 7, with a sampling rate of 1Hz. Step 2: Repeat the loading-unloading cycle from Step 1 three times at each of the four temperature points: 0℃, 25℃, 45℃, and 60℃. Hold each temperature point for 30 minutes before starting the test. Step 3: Divide all collected calibration data into training, validation, and test sets in a 7:2:1 ratio. A one-dimensional convolutional neural network is used to extract spectral feature vectors. The network structure consists of three convolutional layers and two fully connected layers. The input is 1024-dimensional preprocessed spectral data, and the output is a 64-dimensional feature vector. A backpropagation (BP) neural network is used to establish a regression model from the feature vector to the pressure value. The network structure is 64-32-16-5, the activation function is ReLU, and the loss function is mean squared error. The model is trained using the training set, hyperparameters are adjusted using the validation set, and model accuracy is evaluated using the test set. After the model training converges, the model parameters are stored in data processing module 8. Training of stress regression model The trained CNN was used as a feature extractor (with its weights fixed), and the output 64-dimensional feature vector was input into a BP neural network for stress regression. The BP network structure was 64-32-16-1, with ReLU activation in the hidden layers and linear activation in the output layer. The loss function was mean squared error (MSE). During training, 70% of the data was used for training, 15% for validation, and 15% for testing.

[0069] After training, the weight parameters of the CNN and BP networks are merged and exported as an ONNX format model file, which is then deployed to data processing module 8. The single inference time is ≤5 ms, meeting real-time monitoring requirements.

[0070] The relevant calculation formulas are as follows: The equivalent stiffness of the arc-shaped sensing element in the fiber Bragg grating pressure sensor is ki (N / mm). The total pressure acting on the sensor panel is F, which is evenly distributed across N arc-shaped elements. The pressure on each arc-shaped element is:

[0071] F i The pressure exerted on the bow-shaped component; F: Total pressure acting on the sensor (i.e., pressure in the battery pack). N: Number of sensors; The horizontal displacement δi generated by the arc-shaped sensing element under vertical pressure Fi has an approximately linear relationship with the pressure F:

[0072] δ i : Horizontal displacement of the bow-shaped element; k i : The equivalent stiffness of the i-th arc-shaped element; The horizontal displacement δi applied to the fiber grating fixed on the arc-shaped sensing element results in the following axial strain:

[0073] Lg: Effective length of the fiber grating; ε i : Axial strain; Based on the wavelength drift formula, the relationship between pressure F and wavelength drift is as follows:

[0074] p e Effective elastic coefficient; K T Temperature sensitivity coefficient; ΔT: Temperature change; Δλ B,i : The measured wavelength drift of the i-th grating; Let the temperature drift measured by the temperature reference grating be Δλ. B,T =K T Δ T The net drift caused by pressure is: Δλ B,i The measured wavelength drift of the i-th grating; Δλ B,T Temperature reference grating wavelength drift; For N pressure measuring points, the pressure measured at each point is P1, P2, ..., P N The overall equivalent expansion force of the battery module can be defined as

[0075] P i Pressure value at the i-th measuring point; P total Total equivalent expansion force; P avg Average expansion force; Based on the mechanical model of the battery cell-module, the equivalent pressure distribution coefficient on the side of the battery cell can be calculated:

[0076] P j Pressure value at the j-th measuring point; eta i is the equivalent pressure distribution coefficient for the i-th measuring point.

[0077] Example 2

[0078] like Figure 5 , 6 The battery pack monitoring method based on random spectral coding shown includes the following steps: S1. Transmit the continuous spectrum incident light to the pressure sensing fiber grating in the fiber grating pressure sensor and collect the original reflection spectrum reflected by the pressure sensing fiber grating. The specific process is as follows: A fiber Bragg grating pressure sensor is constructed, which incorporates a pressure-sensing fiber Bragg grating. A broadband light source emits continuous-spectrum incident light, which is transmitted to the fiber Bragg grating of the pressure sensor. The fiber Bragg grating reflects narrowband light of a specific wavelength, and a spectral acquisition module acquires the complete reflection spectrum, ultimately outputting the raw reflection spectrum data.

[0079] S2. Preprocess the original reflectance spectrum, extract the reflectance peak region, and output a standardized spectral fragment; The specific process is as follows: After filtering and denoising the raw reflectance data (moving average filtering, window size 5), baseline correction (based on the unstressed state reference spectrum) is performed. Then, the corrected data is normalized, and the reflectance peak region (center wavelength ±5nm, about 400 wavelength points) is extracted. Finally, the standardized spectral fragment is output.

[0080] S3. Extract spectral features from standardized spectral fragments to form high-dimensional spectral feature vectors; The specific process is as follows: AI algorithms are used to extract the "fingerprint features" of each grating from the preprocessed spectrum. The extracted features include not only the center wavelength, but also multi-dimensional morphological parameters such as the bandwidth, symmetry, sidelobe height, and ripple frequency of the reflection peak, forming a high-dimensional spectral feature vector.

[0081] In this embodiment, a one-dimensional convolutional neural network is used to extract spectral features from standardized spectral fragments. The one-dimensional convolutional neural network structure consists of three convolutional layers and two fully connected layers. Specifically, the input layer contains 400-dimensional spectral data; convolutional layer 1 has 32 kernels, kernel size 11, stride 1, and ReLU activation; max pooling layer has pooling size 2 and stride 2; convolutional layer 2 has 64 kernels, kernel size 5, stride 1, and ReLU activation; max pooling layer has pooling size 2 and stride 2; convolutional layer 3 has 128 kernels, kernel size 3, stride 1, and ReLU activation; and a global average pooling layer. The fully connected layers transition from 128 neurons to 64 neurons. The final output is a 64-dimensional spectral feature vector (spectral fingerprint). See Table 1 for details.

[0082] Table 1. One-dimensional convolutional neural network structure and related parameters used in this invention to extract spectral feature vectors.

[0083] High-dimensional spectral eigenvector v

[0084] in: λ B Center wavelength; Δλ 3dB 3 dB bandwidth; Skewness: spectral asymmetry; Kurtosis: spectral peak intensity; Sidelobe ratio: the ratio of sidelobe to main lobe.

[0085] S4. Perform pressure calculation on the extracted high-dimensional spectral feature vector to output the pressure value, or compare the extracted high-dimensional spectral feature vector with the initial spectral feature vector to determine whether the fiber optic pressure sensor has been disassembled.

[0086] The specific process is as follows: (1) Pressure calculation The extracted spectral feature vectors are used to calculate the pressure value, which is obtained by solving the pressure value based on the established spectral feature-pressure relationship model.

[0087] The sensor probe was mounted on a standard pressure platform, and load-unload cycle tests were conducted at five temperature points: -20℃, 0℃, 25℃, 45℃, and 60℃. During the tests, the output pressure value of the standard pressure machine and the reflectance spectrum data of the sensor were simultaneously acquired. A pre-trained one-dimensional convolutional neural network (CNN) was used to extract feature vectors from the acquired spectra, and these feature vectors were used as input to a backpropagation neural network to regress and predict the current pressure value. The mean squared error (MSE) was used as the loss function during model training. n: total number of samples; y i : The true value (standard pressure value) of the i-th sample; The model prediction value for the i-th sample (the stress value output by the BP neural network), with Adam selected as the optimizer and an initial learning rate of 0.0005. This training strategy effectively compensates for individual differences in fiber gratings and the impact of environmental temperature changes on the measurement results.

[0088] Furthermore, for temperature-compensated gratings that are not subjected to external pressure, their spectral feature vectors are extracted separately and input into an auxiliary temperature regression network. This network outputs the offset ΔT of the current ambient temperature relative to the reference temperature. The temperature correction formula for the pressure value is: P_corrected = P_raw - α·ΔT, where α is the temperature coefficient obtained through calibration. Finally, the system outputs the temperature-compensated pressure value Pi corresponding to each pressure sensing grating.

[0089] The calculated pressure values ​​at each measuring point can be weighted or calculated according to the battery pack structure model. The weighted average method is suitable for scenarios with relatively uniform pressure distribution. , Number of pressure measurement points in the sensor : The temperature-compensated pressure value output by the i-th pressure sensing grating; Weighting coefficients, satisfying The sensitivity of the cell expansion response can be allocated based on the measurement point location. A structural mechanics distribution model (suitable for non-uniform pressure distribution scenarios) is used, combined with the mechanical structure of the cell-module, to define the equivalent pressure distribution coefficient on the side of the cell. The overall equivalent expansion force is: , : Pressure distribution coefficient at the i-th measuring point; P j : Pressure value at the j-th measuring point; The total equivalent expansion force on the side of the battery cell; : The mechanical transfer coefficient of the region where the i-th measuring point is located; The effective area corresponding to the i-th measuring point. This yields the overall pressure distribution between the cells. The output data is transmitted to the battery management system or host computer via a communication interface (such as RS485, CAN bus, or Ethernet) for battery health assessment, thermal runaway early warning, and charge / discharge strategy optimization.

[0090] (2) Identification (anti-tampering monitoring) To determine whether the sensor has been disassembled, the cosine similarity between the extracted high-dimensional spectral feature vector and the initial spectral feature vector is calculated to be greater than or equal to a threshold. If yes, the sensor is determined not to have been disassembled, and the verification is successful. If no, the sensor is determined to have been disassembled or replaced, and an anti-tampering alarm is issued.

[0091] Cosine similarity (sim) is calculated at runtime in the following way:

[0092] Among them, v s v is the spectral feature vector of the fiber Bragg grating pressure sensor. s (0) is the initial spectral feature vector when the sensor is installed, which is stored as a template.

[0093] The specific process is as follows: Under sensor free state (no external pressure, room temperature), reflectance spectra are continuously acquired 10 times. Ten 64-dimensional feature vectors (initial spectral feature vectors) are extracted using CNN, and their average value is calculated as the identity template v_template. The template is stored in the encrypted storage area of ​​data processing module 8 (or the secure storage area of ​​BMS). In normal operating mode (identification stage), the spectral feature vector v_current of the current measurement is extracted at each measurement.

[0094] Set a threshold θ, typically between 0.90 and 0.95. If sim ≥ θ: the sensor is determined not to have been disassembled, authentication is successful, and pressure data is output normally. If sim < θ: the sensor is determined to have been disassembled or replaced, and an anti-tampering alarm is issued.

[0095] In this embodiment, if sim < θ, the continuous mismatch counter is incremented by 1.

[0096] If the number of consecutive mismatches is less than 3: do not alarm temporarily, continue monitoring (to avoid false alarms due to momentary interference).

[0097] If the number of consecutive mismatches is ≥ 3: trigger the "anti-tampering alarm" and optionally lock the pressure data output.

[0098] For probes containing multiple pressure sensing gratings, the system can authenticate each grating separately. If a probe has four pressure sensing gratings, the sensor is considered valid only if at least three of the four gratings pass authentication; otherwise, an alarm is triggered.

[0099] Example 3

[0100] This embodiment integrates a battery pack monitoring system based on random spectral coding into a third-generation CTP battery pack. The system can be implemented according to the following steps: First, an ultra-thin fiber Bragg grating pressure sensor is installed in the gap between the large surface of each battery cell and the cold plate of the harmonica tube. The sensor is fixed to the surface of the cold plate by adhesive on the back and is naturally compressed after the battery cell is placed in the box. Second, multiple sensors in the battery pack are connected in series by optical fiber to form a link, which is led out to an external demodulation system through a sealed flange. Each demodulator is equipped with multiple optical channels, and each channel can connect to one sensor link. When the sensor is first installed in the battery pack, the system enters the registration mode: under room temperature and no external pressure conditions (sensor free state), the reflectance spectrum is continuously collected 10 times. The AI ​​feature extraction network obtains 10 64-dimensional feature vectors, and the average value is taken as the identity template v_template of the sensor and stored in the encrypted storage area of ​​the data processing module.

[0101] Then, the demodulation system's data processing module establishes communication with the battery management system via the CAN bus, sending the pressure value and authentication status of each measuring point in real time. After receiving the data, the BMS first performs authentication processing: if the status code indicates successful authentication, the pressure data is reliable and used for subsequent calculations; if the status code indicates identity mismatch or signal abnormality, continuous counting is started, and after multiple consecutive mismatches, a fault code is recorded and an anti-tampering alarm is triggered. At the same time, it is possible to choose to ignore the pressure data of that sensor or switch to a safe mode.

[0102] Subsequently, the BMS processes the reliable pressure data, calculates the average pressure of a single cell, the total expansion force of the entire package, and the pressure distribution uniformity index, and compares it with a preset threshold range. Based on the comparison results, it executes corresponding warning or protection actions. In this embodiment, under normal operating mode, the system extracts the current spectral feature vector v_current and calculates the cosine similarity sim for each measurement. If sim ≥ the threshold, authentication is successful, and the pressure calculation unit outputs the pressure value. If sim < the threshold θ, it counts once. After three consecutive mismatches, it outputs a "sensor authentication failed" alarm and stops outputting pressure data (or outputs a safe default value). For a probe containing four pressure sensing gratings, the system can authenticate each grating separately. Only when at least three of the four gratings pass authentication is the sensor identity deemed valid; otherwise, an alarm is triggered.

[0103] Through the above integration, this system can effectively protect against various scenarios: when a sensor's spectral characteristics change due to mechanical damage, the BMS automatically switches to secure mode and prompts for replacement after authentication failure; instantaneous spectral drift caused by drastic environmental changes is avoided by employing a continuous multiple-judgment mechanism. Therefore, this invention achieves a reliable binding between sensor identity and monitoring data, providing a reliable guarantee for the safe operation of the battery pack.

[0104] A battery pack monitoring device based on random spectral coding is provided. The device includes a processor, a memory, and a computer program stored in the memory and executable by the processor. When the computer program is executed by the processor, it implements the steps of the battery pack monitoring method based on random spectral coding of Embodiment 2.

[0105] A storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of the battery pack monitoring method based on random spectral coding in Embodiment 2.

[0106] Performance testing and effect verification (1) Verification of temperature compensation effect The sensor was placed in a constant temperature chamber without external pressure and stabilized for 30 minutes at temperatures of -20℃, 0℃, 25℃, 45℃, and 60℃. The raw pressure output Praw and the temperature compensation grating output ΔT were recorded at each temperature. The compensated pressure Pcorrected was calculated and compared with the true pressure. After temperature compensation, the pressure drift across the entire temperature range was ≤ ±0.2 N (<0.1% FS).

[0107] (2) Accuracy test of anti-tamper identification Ten different fiber grating pressure sensors were sequentially mounted on a test platform, and spectra were collected ten times for each sensor under no-pressure conditions at room temperature, generating their respective identity templates. template After randomly disassembling and replacing the gratings in 10 sensors, each sensor was subjected to multiple measurements under normal pressure loading (0~5000 N) and different temperatures (-20℃~60℃). The cosine similarity (sim) for each measurement was calculated, and the number of measurements below the threshold θ=0.92 was counted. If the similarity is ≥ 0.92: authentication is successful, the system determines that the sensor has not been disassembled, and outputs pressure data normally. If the similarity is < 0.92: authentication fails, the system determines that the sensor may have been disassembled, replaced, or damaged, immediately issues an "anti-tampering alarm," and can optionally lock the pressure data output.

[0108] (3) Long-term stability test The sensor was mounted on a test fixture, subjected to a constant pressure of 3000 N, and placed in a constant temperature environment of 45°C.

[0109] The system operated continuously for 30 days, recording pressure output values ​​and spectral feature vectors daily. A full-range (0~5000 N) calibration check was performed weekly, recording changes in the calibration curve. After 30 days of long-term operation, pressure drift was <0.1% FS / month, and the spectral feature similarity remained above 0.95 (far exceeding the threshold of 0.92), indicating stable sensor performance and identification capabilities.

[0110] The above embodiments of the present invention are not intended to limit the scope of protection of the present invention. The implementation of the present invention is not limited thereto. All other modifications, substitutions or alterations made by those skilled in the art based on the above content of the present invention and in accordance with ordinary technical knowledge and common practices in the art, without departing from the basic technical concept of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A battery pack monitoring method based on random spectral coding, characterized in that, Includes the following steps: S1. Continuous spectral incident light is transmitted to the pressure-sensing fiber grating in the fiber grating pressure sensor, and the original reflection spectrum reflected by the pressure-sensing fiber grating is collected. The fiber grating pressure sensor includes a sensing probe, which includes a base plate, a pressure contact panel, and an elastic sensitive layer. The base plate of the sensing probe is fixedly connected to the battery module. The pressure contact panel is in contact with the battery cell. The base plate and the pressure contact panel are fixedly connected, and the elastic sensitive layer is disposed between the base plate and the pressure contact panel. The elastic sensitive layer includes multiple bow-shaped sensing elements, one end of which is connected to the base plate and the other end of which is connected to the pressure contact panel. Each bow-shaped sensing element is provided with a pressure-sensing fiber grating, so that the pressure-sensing fiber gratings of different bow-shaped sensing elements are connected in series through optical fibers. S2. Preprocess the original reflectance spectrum, extract the reflectance peak region, and output a standardized spectral fragment; S3. Extract spectral features from standardized spectral fragments to form high-dimensional spectral feature vectors; S4. Perform pressure calculation on the extracted high-dimensional spectral feature vector to output the pressure value, or compare the extracted high-dimensional spectral feature vector with the initial spectral feature vector to determine whether the fiber optic pressure sensor has been disassembled or replaced.

2. The battery pack monitoring method based on random spectral coding according to claim 1, characterized in that, The spectral features extracted in S3 include multidimensional morphological parameters such as center wavelength, bandwidth of reflection peak, symmetry, sidelobe height, and ripple frequency. The spectral features extracted from the standardized spectral fragments in S3 are obtained using a one-dimensional convolutional neural network with a network structure of three convolutional layers and two fully connected layers.

3. The battery pack monitoring method based on random spectral coding according to any one of claims 1-2, characterized in that, The high-dimensional spectral feature vector v in: λ B Center wavelength; Δλ 3dB 3 dB bandwidth; Skewness: spectral asymmetry; Kurtosis: spectral peak intensity; Sidelobe ratio: the ratio of sidelobe to main lobe.

4. The battery pack monitoring method based on random spectral coding according to claim 3, characterized in that, In S4, the extracted high-dimensional spectral feature vectors are used to calculate the pressure value, and the pressure value is obtained according to the established spectral feature-pressure relationship model. The spectral feature-pressure relationship model is obtained by calibrating the fiber optic grating pressure sensor across its full range on a standard pressure platform. The calibration method includes the following steps: performing a loading-unloading cycle on the sensor probe under different temperature conditions, simultaneously collecting standard pressure values ​​and spectral data, dividing the collected calibration data into training set, validation set and test set according to the proportion, extracting spectral feature vectors using a one-dimensional convolutional neural network, and establishing a regression model from feature vectors to pressure values ​​using a BP neural network or random forest algorithm. After the model training converges, the model parameters are fixed.

5. The battery pack monitoring method based on random spectral coding according to claim 4, characterized in that, A BP neural network machine learning algorithm is used to establish the mapping from spectral feature vectors to pressure values: Where W and b are the network weights and biases, obtained through training with calibration data.

6. The battery pack monitoring method based on random spectral coding according to claim 5, characterized in that, In S4, it is determined whether the sensor has been disassembled by calculating whether the cosine similarity between the extracted high-dimensional spectral feature vector and the initial spectral feature vector is greater than or equal to a threshold. If yes, the sensor is determined not to have been disassembled and the verification is passed. If no, the sensor is determined to have been disassembled or replaced, and an anti-tampering alarm is issued.

7. The battery pack monitoring method based on random spectral coding according to claim 6, characterized in that, Runtime cosine similarity (sim) is calculated as follows: Among them, v s v is the spectral feature vector of the fiber Bragg grating pressure sensor. s (0) is the initial spectral feature vector when the sensor is installed, which is stored as a template; the threshold θ is 0.90~0.

95.

8. A battery pack monitoring system based on random spectral coding, characterized in that, include: A light source module, wherein the light source module is used to provide incident light of a continuous spectrum; A fiber Bragg grating pressure sensor includes a sensing probe, which comprises a base plate, a pressure contact panel, and an elastic sensitive layer. The base plate is fixedly connected to a battery module. The pressure contact panel is in contact with a battery cell. The base plate and the pressure contact panel are fixedly connected, and the elastic sensitive layer is disposed between the base plate and the pressure contact panel. The elastic sensitive layer includes multiple arc-shaped sensing elements, one end of which is connected to the base plate and the other end of which is connected to the pressure contact panel. Each arc-shaped sensing element is provided with a pressure-sensing fiber Bragg grating, such that the pressure-sensing fiber Bragg gratings of different arc-shaped sensing elements are connected in series via optical fibers. A spectral acquisition module is used to acquire the complete reflection spectrum of incident light from the light source module transmitted to the pressure sensing fiber optic grating. The data processing module is used to process and analyze the reflection spectrum, perform pressure calculation to obtain the battery pack pressure value, or determine whether the fiber optic pressure sensor has been disassembled.

9. A battery pack monitoring device based on random spectral coding, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, it implements the steps of the battery pack monitoring method based on random spectral coding as described in any one of claims 1-7.

Images