Battery evaluation computer system and method based on multi-modal characteristics

Abstract

The invention discloses a battery evaluation computer system and method based on multi-modal characteristics, and belongs to the technical field of battery health management, and the system mainly comprises a multi-modal sensing module, specifically a grating sensor unit, which is used for collecting surface and internal deformation data of a battery, including wavelength offset (delta lambda), strain (epsilon) and deformation characteristics after temperature compensation; the electrochemical sensor unit is used for acquiring voltage, current, internal resistance and temperature data of the battery; and the acoustic sensor unit is used for detecting ultrasonic signals of gas separation or structure abnormity in the battery. By integrating the fiber grating sensor, the electrochemical sensor and the acoustic sensor, deformation, voltage / current, temperature and acoustic signals of the battery are acquired in real time, and high-precision battery state of health (SOH) evaluation and residual life (RUL) prediction are realized in combination with multi-modal data processing and an intelligent algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of battery status monitoring, and in particular to a battery evaluation computer system and method based on multi-modal features. Background Art

[0002] With the acceleration of global energy transformation, high energy density storage devices such as lithium-ion batteries and solid-state batteries are widely used in electric vehicles, renewable energy storage systems, aerospace and other fields. However, batteries undergo complex physical and chemical changes during cyclic use, including the shedding of electrode active materials, electrolyte decomposition, lithium dendrite growth, internal micro-short circuits, etc. These changes may lead to capacity decay, increased internal resistance, thermal runaway and even explosion. Therefore, high-precision real-time evaluation and early warning of battery status have become core issues that the industry needs to solve urgently.

[0003] Limitations of single-modality inspection methods Electrochemical parameter method: The traditional method evaluates the battery state of health (SOH) by monitoring parameters such as voltage, current, and internal resistance (EIS), and proposes a SOH calculation model based on the capacity decay rate. However, the change of electrochemical parameters usually lags behind the physical damage inside the battery. Experiments show that when the capacity decay reaches 5%, the lithium dendrites may have grown to a critical length (>50 μm), but at this time the internal resistance changes by only about 3%, which is difficult to trigger an effective warning.

[0004] Temperature monitoring method: Use an infrared thermal imager to monitor the temperature distribution on the battery surface. However, temperature changes are usually the last stage signal before thermal runaway occurs, and early intervention cannot be achieved.

[0005] Deformation detection method: Battery expansion is detected through pressure sensors, but traditional strain gauges have insufficient accuracy (±10 με) and anti-electromagnetic interference capabilities, and cannot distinguish between thermal expansion and mechanical deformation.

[0006] Disadvantages of multimodal fusion technology The few existing multimodal solutions have the following defects: Single sensor type: only integrates electrochemical and thermal data, lacks the capture of key features such as internal mechanical deformation and gas precipitation; The feature fusion algorithm is simple: it uses the linear weighted average method to allocate weights, ignoring the nonlinear correlation and information entropy differences between different features, resulting in low reliability of the fusion result; Poor real-time performance: The data acquisition frequency is low (usually ≤1 Hz), and it is impossible to capture transient abnormal signals during the battery charging and discharging process.

[0007] Application bottleneck of high-end detection technology Although laboratory-level detection methods can analyze the microscopic structural changes of batteries, they have significant drawbacks: Destructive testing: The battery needs to be disassembled, making it unsuitable for online monitoring; High cost: Difficult to apply on a large scale; High latency: Data processing takes several hours, unable to support real-time decision-making. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a battery evaluation computer system and method based on multi-modal features.

[0009] To solve the above technical problems, the present invention provides the following technical solutions: A battery evaluation computer system based on multi-modal features of the present invention includes: (1) A multi-modal sensing module, including: A grating sensor unit: used to collect deformation data on the surface and inside of the battery, including wavelength shift (Δλ), strain (ε), and deformation characteristics after temperature compensation; An electrochemical sensor unit: used to collect voltage, current, internal resistance, and temperature data of the battery; An acoustic sensor unit: used to detect ultrasonic signals of gas evolution or structural abnormalities inside the battery; (2) A data synchronous acquisition unit: Configure a multi-channel synchronous acquisition card to synchronously receive grating, electrochemical, and acoustic sensor data at different sampling frequencies, where the sampling frequency of grating data ≥ 1 kHz, and the sampling frequency of acoustic data ≥ 20 kHz; (3) A data processing module, including: A preprocessing unit: Perform temperature compensation and demodulation on the grating data, based on the formula:

[0010] Calculate the strain, and use Kalman filtering and wavelet denoising to eliminate noise; A feature extraction unit: Extract grating deformation energy and electrochemical capacity attenuation rate , as well as acoustic resonance frequency shift Δf; (4) A multi-modal fusion model: After dimensionality reduction of the features by principal component analysis (PCA), use the entropy weight method to assign weights for decision-level fusion, and the weight calculation formula is: where H j is the feature information entropy; (5) An evaluation and prediction module: Based on the fused features, predict the state of health (SOH) of the battery through a support vector regression (SVR) model, and predict the remaining useful life (RUL) through a long short-term memory network (LSTM); (6) Visualization and early warning module: It displays the 3D deformation thermogram and SOH trend curve of the battery in real time, and triggers multi-level early warnings when the deformation exceeds the preset threshold ( > 200 με) or the internal resistance growth rate (ΔR > 10%).

[0011] As a preferred technical solution of the present invention, the grating sensor unit is a fiber Bragg grating (FBG), with a wavelength range of 1520 - 1570 nm, deployed on the battery housing or electrode surface, 3 - 5 sensors are deployed for each battery, and dynamic temperature compensation is performed through the thermo-optic coefficient (ξ) and the coefficient of thermal expansion (α).

[0012] As a preferred technical solution of the present invention, the electrochemical sensor unit includes: (1) Constant current charge and discharge module, used to calibrate the reference capacity C nominal； (2) Electrochemical impedance spectroscopy (EIS) analysis module, obtaining the change curve of the internal resistance R with the aging time through frequency scanning; t with the aging time; (3) Infrared thermal imager, used to generate the battery surface temperature distribution matrix T(x, y, t).

[0013] As a preferred technical solution of the present invention, in the multi-modal fusion model: Principal component analysis (PCA) retains the principal components with a variance contribution rate ≥ 85%; The SVR model uses a radial basis kernel function , and the parameters are optimized to C = 10, γ = 0.1 through grid search; The input sequence of the LSTM network is , the number of hidden layer nodes is 64, and the output layer is a fully connected layer.

[0014] As a preferred technical solution of the present invention, the early warning module includes: First-level early warning: When the deformation energy E d exceeds 3 times the standard deviation of the historical average, a yellow early warning is triggered; Second-level early warning: When the SOH decline rate exceeds 0.5% / cycle and ΔR > 15%, an orange early warning is triggered; Third-level early warning: When the acoustic resonance frequency shift Δf > 5 kHz or the strain > 300 με, a red early warning is triggered and forced shutdown is performed.

[0015] As a preferred technical solution of the present invention, the system includes the following extended functions: Add a gas sensor module to detect the concentration of gas components inside the battery; Update the SVR and LSTM models through transfer learning to adapt to different battery types.

[0016] The present invention also discloses a battery evaluation method based on multi-modal features, including the following steps: (1) Synchronously collect grating deformation, electrochemical parameters and acoustic signals; (2) Demodulate and temperature compensate the grating data, and extract the deformation energy E d and the maximum strain rate max ; (3) Calculate the electrochemical SOH and the internal resistance growth rate ΔR, and extract the acoustic resonance frequency shift Δf; (4) Generate a comprehensive evaluation index by fusing multi-modal features through PCA dimensionality reduction and entropy weight method; (5) Use the SVR model to predict SOH and the LSTM model to predict RUL; (6) Trigger multi-level early warnings according to preset thresholds and generate a visual report.

[0017] As a preferred technical solution of the present invention, the calculation formula of the comprehensive evaluation index in step (4) is: Comprehensive aging index =

[0018] where w1, w2, and w3 are the weights assigned by the entropy weight method, and w1 + w 2+ w3 = 1.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The present invention has high-precision evaluation and prediction capabilities. By fusing grating deformation, electrochemical parameters and acoustic signals, it eliminates the noise interference of a single sensor, significantly improves the evaluation reliability, uses the entropy weight method to assign feature weights, solves the problem that the traditional fixed-weight model is sensitive to battery types, and captures the non-linear attenuation law of battery aging through the long short-term memory network. The RUL prediction error < 5%, and the error is reduced by 60% compared with traditional time series models such as ARIMA.

[0020] 2. The early warning ability of the present invention is breakthrough, with a multi-level early warning mechanism, triggering yellow, orange, and red warnings at different levels to avoid over-intervention.

[0021] 3. The present invention is compatible with multiple types of batteries. It adapts to different battery systems through transfer learning, has flexible hardware expansion, supports plug-and-play sensor expansion, the software algorithm can be upgraded, and the fusion model parameters are updated through OTA to adapt to the aging law of new batteries. Description of the Drawings

[0022] The accompanying drawings are used to provide a further understanding of the present invention and form a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the accompanying drawings: Figure 1 It is a schematic structural diagram of the system of the present invention. Detailed implementation manners

[0023] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for explaining and illustrating the present invention and are not used to limit the present invention. Embodiment

[0024] As Figure 1 shown, a battery evaluation computer system based on multi-modal features of the present invention includes: (1) A multi-modal sensing module, including: A grating sensor unit: used to collect deformation data on the surface and inside of the battery, including wavelength offset (Δλ), strain (ε), and deformation characteristics after temperature compensation; An electrochemical sensor unit: used to collect voltage, current, internal resistance, and temperature data of the battery; An acoustic sensor unit: used to detect ultrasonic signals of gas evolution or structural anomalies inside the battery; (2) A data synchronous acquisition unit: Configure a multi-channel synchronous acquisition card to synchronously receive data from grating, electrochemical, and acoustic sensors at different sampling frequencies, where the sampling frequency of grating data ≥ 1 kHz, and the sampling frequency of acoustic data ≥ 20 kHz; (3) A data processing module, including: A preprocessing unit: perform temperature compensation and demodulation on the grating data, and calculate the strain based on the formula:

[0025] And use Kalman filtering and wavelet noise reduction to eliminate noise; A feature extraction unit: extract grating deformation energy 、electrochemical capacity attenuation rate , and acoustic resonance frequency offset Δf; (4) A multi-modal fusion model: After dimensionality reduction of the features by principal component analysis (PCA), use the entropy weight method to assign weights for decision-level fusion, and the weight calculation formula is: Where H j is the feature information entropy; An evaluation and prediction module: Based on the fused features, predict the state of health (SOH) of the battery through a support vector regression (SVR) model, and predict the remaining useful life (RUL) through a long short-term memory network (LSTM); Visualization and early warning module: It displays the 3D deformation thermal map and SOH trend curve of the battery in real time, and triggers multi-level early warnings when the deformation exceeds the preset threshold ( > 200 με) or the internal resistance growth rate (ΔR > 10%).

[0026] The grating sensor unit is a fiber Bragg grating (FBG) with a wavelength range of 1520 - 1570 nm. It is deployed on the battery shell or electrode surface. 3 - 5 sensors are deployed for each battery, and dynamic temperature compensation is performed through the thermo-optic coefficient (ξ) and the coefficient of thermal expansion (α).

[0027] The electrochemical sensor unit includes: (1) Constant current charge and discharge module, used to calibrate the reference capacity C nominal； (2) Electrochemical impedance spectroscopy (EIS) analysis module, which obtains the change curve of the internal resistance R with the aging time through frequency scanning; t (3) Infrared thermal imager, used to generate the battery surface temperature distribution matrix T(x, y, t).

[0028] As a preferred technical solution of the present invention, in the multi-modal fusion model: Principal component analysis (PCA) retains the principal components with a variance contribution rate ≥ 85%; The SVR model adopts a radial basis kernel function , and the parameters are optimized to C = 10, γ = 0.1 through grid search; The input sequence of the LSTM network is , the number of hidden layer nodes is 64, and the output layer is a fully connected layer.

[0029] The early warning module includes: Primary early warning: When the deformation energy E d exceeds 3 times the standard deviation of the historical average, a yellow early warning is triggered; Secondary early warning: When the SOH decline rate exceeds 0.5% / cycle and ΔR > 15%, an orange early warning is triggered; Tertiary early warning: When the acoustic resonance frequency deviation Δf > 5 kHz or the strain > 300 με, a red early warning is triggered and forced shutdown is performed.

[0030] The system includes the following extended functions: Add a gas sensor module to detect the concentration of gas components inside the battery; Update the SVR and LSTM models through transfer learning to adapt to different battery types.

[0031] The present invention also discloses a battery state evaluation method based on multi-modal sensing, including the following steps:​ (1) Synchronously collect the deformation of the grating, electrochemical parameters, and acoustic signals; (2) Demodulate the grating data and perform temperature compensation to extract the deformation energy E d and the maximum strain rate max ; (3) Calculate the electrochemical SOH and the internal resistance growth rate ΔR, and extract the acoustic resonance frequency shift Δf; (4) Fuse multi-modal features through PCA dimensionality reduction and the entropy weight method to generate a comprehensive evaluation index; (5) Use the SVR model to predict SOH and the LSTM model to predict RUL; (6) Trigger multi-level early warnings according to preset thresholds and generate a visual report.

[0032] In step (4), the calculation formula for the comprehensive evaluation index is: Comprehensive aging index =

[0033] where w1, w2, and w3 are the weights assigned by the entropy weight method, and w1 + w 2+ w3 = 1.

[0034] Specifically, it includes the following parts: System hardware deployment: Grating sensor deployment (1) Uniformly deploy 5 fiber Bragg gratings on the surface of the lithium-ion battery shell, with a wavelength range of 1520 - 1570 nm; (2) Fix the sensor with epoxy resin to ensure close contact with the battery surface; (3) The grating demodulator (Micron Optics SM130) collects the wavelength shift (Δλ) at a sampling frequency of 1 kHz.

[0035] Electrochemical sensor deployment: Connect voltage / current sensors to the positive and negative electrodes of the battery, with an accuracy of ±0.1 mV; Use an electrochemical workstation (Gamry Interface 5000) to perform constant current charge and discharge tests to calibrate the reference capacity C nominal ; An infrared thermal imager monitors the surface temperature distribution of the battery, with a sampling frequency of 10 Hz.

[0036] Acoustic sensor deployment: Install ultrasonic probes on the side of the battery, with a sampling frequency of 20 kHz; Collect internal gas evolution or structural anomaly signals through an acoustic emission instrument.

[0037] Data Processing and Feature Extraction: 1. Demodulate the grating data, based on the formula: Calculate the strain, where , Extract the deformation energy .

[0038] 2. Electrochemical feature extraction Calculate the capacity attenuation rate: ; Calculate the internal resistance growth rate: .

[0039] 3. Acoustic feature extraction Extract the resonant frequency shift .

[0040] Multi-modal Feature Fusion 1. PCA dimensionality reduction: Standardize the feature matrix .

[0041] Calculate the covariance matrix and extract the principal components, retaining the principal components with a variance contribution rate ≥ 85%.

[0042] Entropy weight method for weight assignment Calculate the feature information entropy: ; Assign weights: .

[0043] Evaluation and Prediction: SOH prediction Use the SVR model with the radial basis function as the kernel function: Optimize the parameters through grid search .

[0044] RUL prediction Use the LSTM network with the input sequence as ; The number of hidden layer nodes is 64, and the output layer is a fully connected layer.

[0045] Warning and Visualization Multi-level warning mechanism First-level warning: When the deformation energy Ed exceeds 3 times the standard deviation of the historical average, trigger a yellow warning; Second-level warning: When the SOH decline rate exceeds 0.5% / cycle and ΔR > 15%, trigger an orange warning; Third-level warning: When the acoustic resonance frequency shift Δf > 5 kHz or the strain ϵ > 300 με, trigger a red warning and force a shutdown.

[0046] Visual interface Real-time display of 3D deformation heat maps, SOH trend curves, and RUL prediction results; Generate an evaluation report in PDF format, including historical data and warning records.

[0047] Experimental data

[0048] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A battery evaluation computer system based on multimodal features, characterized in that: include: (1) Multimodal sensing module, including: Grating sensor unit: used to collect battery surface and internal deformation data, including wavelength offset (Δλ), strain (ε) and deformation characteristics after temperature compensation; Electrochemical sensor unit: used to collect battery voltage, current, internal resistance and temperature data; Acoustic sensor unit: ultrasonic signals used to detect gas precipitation or structural abnormalities inside the battery; (2) Data synchronization acquisition unit: equipped with a multi-channel synchronous acquisition card to synchronously receive grating, electrochemical and acoustic sensor data at different sampling frequencies, where the grating data sampling frequency is ≥1 kHz and the acoustic data sampling frequency is ≥20 kHz; (3) Data processing module: including: Preprocessing unit: Temperature compensation and demodulation of grating data, based on the formula: Calculate strain and eliminate noise using Kalman filtering and wavelet denoising; Feature extraction unit: extract grating deformation energy , electrochemical capacity decay rate , and the acoustic resonance frequency shift Δf; (4) Multimodal fusion model: After reducing the feature dimension through principal component analysis (PCA), the entropy weight method is used to assign weights for decision-level fusion. The weight calculation formula is: Among them, H j is the feature information entropy; (5) Evaluation and prediction module: Based on the fused features, the support vector regression (SVR) model is used to predict the battery state of health (SOH), and the long short-term memory network (LSTM) is used to predict the remaining life (RUL); (6) Visualization and early warning module: Real-time display of battery 3D deformation thermal map and SOH trend curve, and when the deformation exceeds the preset threshold ( >200μϵ) or internal resistance growth rate (ΔR>10%) triggers multi-level warning.

2. A battery evaluation computer system based on multimodal features according to claim 1, characterized in that: The grating sensor unit is a fiber Bragg grating (FBG) with a wavelength range of 1520-1570 nm. It is deployed on the battery casing or electrode surface. Each battery is deployed with 3-5 sensors, and dynamic temperature compensation is performed through the thermo-optical coefficient (ξ) and thermal expansion coefficient (α).

3. A battery evaluation computer system based on multimodal features according to claim 1, characterized in that: The electrochemical sensor unit comprises: (1) Constant current charge and discharge module, used to calibrate the reference capacity C nominal； (2) Electrochemical impedance spectroscopy (EIS) analysis module, which obtains the internal resistance R by frequency scanning t Curve of change with aging time; (3) Infrared thermal imager, used to generate the battery surface temperature distribution matrix T (x, y, t).

4. A battery evaluation computer system based on multimodal features according to claim 1, characterized in that: In the multimodal fusion model: Principal component analysis (PCA) retained principal components with variance contribution ≥ 85%; The SVR model uses the radial basis kernel function , the parameters are optimized by grid search to C = 10, γ = 0.1; The LSTM network input sequence is , the number of hidden layer nodes is 64, and the output layer is a fully connected layer.

5. The battery evaluation computer system based on multimodal features according to claim 1, characterized in that: The early warning module comprises: Level 1 warning: When the deformation energy E d When the standard deviation exceeds 3 times of the historical average, a yellow warning is triggered; Level 2 warning: When the SOH decrease rate exceeds 0.5% / cycle and ΔR>15%, an orange warning is triggered; Level 3 warning: When the acoustic resonance frequency shifts Δf>5kHz or the strain When >300με, a red warning is triggered and the machine is forced to shut down.

6. A battery evaluation computer system based on multimodal features according to claim 1, characterized in that: The system includes the following extended functions: Add a gas sensor module to detect the concentration of gas components inside the battery; Update the SVR and LSTM models through transfer learning to adapt to different battery types.

7. A battery evaluation method based on multimodal features, characterized in that: The following steps are involved: (1) Synchronously collect grating deformation, electrochemical parameters and acoustic signals; (2) Demodulate and temperature compensate the grating data to extract the deformation energy E d and maximum strain rate max ; (3) Calculate the electrochemical SOH and internal resistance growth rate ΔR, and extract the acoustic resonance frequency shift Δf; (4) Generate comprehensive evaluation indicators by fusing multimodal features through PCA dimensionality reduction and entropy weight method; (5) Use the SVR model to predict SOH and the LSTM model to predict RUL; (6) Trigger multi-level warnings based on preset thresholds and generate visual reports.

8. The battery evaluation method based on multimodal features according to claim 7, characterized in that: The calculation formula of the comprehensive evaluation index in step (4) is: Comprehensive aging index = Among them, w1, w2, and w3 are the weights assigned by the entropy weight method, and w1+w 2+ w3=1.

Images