A power battery pack screening test method based on multi-field coupling simulation

Abstract

The application relates to the technical field of battery testing, in particular to a power battery pack screening test method based on multi-field coupling simulation, which comprises the following steps: extracting a local overheating area based on infrared thermal imaging and image enhancement; adopting temperature gradient analysis and a support vector machine to identify a defect type; generating a diffusion trend index through pixel sequence analysis; predicting a defect evolution path under a force field by using a convolutional neural network; matching a historical risk mode to generate a probability distribution graph; evaluating key node weights by analyzing a conduction chain to generate an optimized screening point set; and finally comprehensively verifying output identification results and a risk report in multiple dimensions. The application can accurately identify internal defects and scientifically predict an evolution track, deeply reveals an energy conduction path through space-time sequence analysis under multi-field coupling, and improves the depth and accuracy of battery pack safety screening.

Description

Technical Field

[0001] This invention belongs to the field of battery testing technology, specifically a screening test method for power battery packs based on multi-field coupling simulation. Background Technology

[0002] The core objective of the power battery pack screening and testing technology field is to accurately assess and analyze the internal minute defects and thermal runaway evolution risks during the power battery pack manufacturing process by developing advanced methods and tools, thereby taking targeted quality control measures to ensure the operational safety of the power battery.

[0003] Battery pack testing technology is part of the manufacturing process of new energy vehicles. Its main task is to quantify and understand the impact of the microscopic defects inside the battery pack on the overall thermal stability, and to provide scientific risk prediction support for the design optimization and manufacturing of battery systems.

[0004] Among them, the power battery pack screening test method is a technical means for identifying internal defects under multi-field coupling environments. Its purpose is to accurately identify the impact of metallic foreign objects or separator damage on the safety performance of the battery pack. The concept of risk assessment includes extracting thermal distribution characteristics, simulating thermal runaway evolution paths, and matching risk probability patterns to minimize the safety risks caused by internal defects in the battery pack. Through this objective, manufacturers can more effectively reduce uncertainties in the production process, improve the safety quality of battery products, and enhance the end-user experience.

[0005] Traditional methods have several shortcomings in practical application. Previous methods often neglect comprehensive analysis of multi-dimensional data, such as the presence of metallic foreign objects inside the battery pack, the condition of the separator, and thermal diffusion under the coupling of electric and mechanical fields, leading to inaccurate assessments of potential safety risks within the battery pack. Traditional methods lack the application of thermal imaging spatiotemporal sequence analysis and nonlinear prediction models in simulating the correlation between local overheating areas and the overall conduction chain, limiting the accurate identification of risk evolution patterns. Furthermore, traditional methods typically lack a scientific predictive framework to support the evolution trend and risk probability distribution of defects under specific environments. In the interaction analysis between macroscopic performance testing and microscopic conduction chains, traditional methods fail to fully integrate historical thermal runaway risk data, resulting in an insufficiently in-depth assessment of the impact of manufacturing defects on the overall safety of the battery pack. Summary of the Invention

[0006] The purpose of this invention is to address the aforementioned shortcomings in the prior art by providing a power battery pack screening and testing method based on multi-field coupling simulation.

[0007] The objective of this invention is achieved through the following technical solution: a power battery pack screening test method based on multi-field coupling simulation, comprising the following steps: S1. Based on the multi-frame thermal distribution images acquired by the infrared thermal imaging device under the electric field and force field simulation environment, the contrast of the image sequence is improved and the contour feature is analyzed by grayscale conversion and edge enhancement processing methods, and the local overheated area is initially extracted to generate thermal imaging data containing the local overheated area. S2. Based on thermal imaging data, temperature gradient analysis and support vector machine algorithm are used to classify and identify the temperature change rate and abnormal hot spot coordinates, and generate preliminary defect location and type labels. S3. Based on the initial defect location and type label, the pixel sequence analysis method is used to monitor the temperature change amplitude of surrounding pixels, and when the temperature change sequence exceeds the preset threshold, the diffusion boundary area is marked to generate a diffusion trend index. S4. Based on the diffusion trend index, a convolutional neural network model and spatiotemporal sequence analysis method are used to predict the evolution path of defects under the influence of force field, and the evolution trajectory is vectorized to generate an evolution path vector. S5. Combining historical thermal runaway risk database and evolution path vectors, a probabilistic pattern matching method is used to analyze the matching degree between evolution paths and risk patterns, identify high-risk evolution patterns, and generate a risk probability distribution map. S6. Based on the risk probability distribution map, the energy conduction path from local overheating to overall thermal runaway is analyzed using image overlay and chain conduction analysis methods. The weights of key nodes in the conduction chain are evaluated to generate an optimized screening point set. S7. Based on the comprehensive optimization of the screening point set and evolution path verification results, the support vector machine algorithm and multi-dimensional feature verification are used to conduct a comprehensive verification and evaluation report design for the defect identification results, the accuracy of thermal runaway evolution and the depth of risk prediction, and generate the final defect identification results.

[0008] The beneficial effects of this invention are as follows: By applying grayscale conversion and edge enhancement processing methods, this invention achieves more accurate analysis of local overheated areas in multi-frame thermal distribution images of battery packs, improving the quality of thermal imaging data preprocessing. This method utilizes temperature gradient analysis and support vector machine algorithms to effectively explore the impact of metallic foreign objects and separator damage on the temperature field, helping to identify the location and type of initial defects. Furthermore, the application of convolutional neural network models in spatiotemporal sequence processing makes the prediction of defect evolution paths more scientific, providing a solid foundation for risk evolution analysis under the influence of force fields. By combining historical thermal runaway risk databases and risk probability distribution maps, this method can deeply analyze the conduction chain from local overheating to overall thermal runaway and evaluate key nodes in the conduction chain. Further validation using multi-dimensional feature vectors provides a comprehensive perspective for power battery pack defect identification and thermal runaway risk prediction, thereby promoting the formation of comprehensive assessment results for battery pack safety screening. Attached Figure Description

[0009] The invention will be further illustrated with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the invention. For those skilled in the art, other drawings can be obtained based on the following drawings without any creative effort.

[0010] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0011] The present invention will be further described in conjunction with the following embodiments.

[0012] Depend on Figure 1 It can be seen that the method described in this embodiment is... Figure 1 As can be seen, the present invention provides a power battery pack screening test method based on multi-field coupling simulation, including the following steps: S1, based on multi-frame thermal distribution images acquired by infrared thermal imaging equipment under electric field and force field simulation environment, grayscale conversion and edge enhancement processing methods are used to improve the contrast of the image sequence and analyze the contour features, and to perform preliminary extraction of local overheated areas, generating thermal imaging data containing local overheated areas; S2, based on the thermal imaging data, temperature gradient analysis and support vector machine algorithm are used to classify and identify the temperature change rate and abnormal hot spot coordinates, and to match the metal foreign object or diaphragm damage type, generating preliminary defect location and type labels; S3, based on the preliminary defect location and type labels, pixel sequence analysis method is used to monitor the temperature change amplitude of surrounding pixels, and when the temperature change sequence exceeds a preset threshold, the diffusion boundary region is marked, generating a diffusion trend index; S4, based on the... The following steps are taken: S5. Using the diffusion trend indicators, a convolutional neural network model and spatiotemporal sequence analysis method are employed to predict the evolution path of defects under the influence of a force field, and the evolution trajectory is vectorized to generate an evolution path vector; S6. Combining the historical thermal runaway risk database and the evolution path vector, a probabilistic pattern matching method is used to analyze the matching degree between the evolution path and the risk pattern, and high-risk evolution patterns are identified to generate a risk probability distribution map; S7. Based on the risk probability distribution map, image overlay and chain conduction analysis methods are used to analyze the energy conduction path from local overheating to overall thermal runaway, and the weights of key nodes in the conduction chain are evaluated to generate an optimized screening point set; S8. Combining the optimized screening point set and evolution path verification results, a support vector machine algorithm and multi-dimensional feature verification are used to conduct a comprehensive verification and evaluation report on the defect identification results, thermal runaway evolution accuracy, and risk prediction depth, generating the final defect identification result; S9. Based on the final defect identification result, a full life cycle risk assessment and safety benefit analysis are used to conduct a comprehensive assessment of the overall safety quality of the battery pack and production planning, generating a thermal runaway risk prediction report.

[0013] Specifically, the thermal imaging data containing the local overheated area includes grayscale histogram features, enhanced contour matrix, and candidate coordinates of overheated hot spots. The preliminary defect location and type labels include coordinates of metallic foreign objects, diaphragm damage probability values, and hot spot peak intensity distribution. The diffusion trend indicators are specifically pixel-level temperature rise rate, diffusion front movement vector, and thermal boundary stability parameters. The evolution path vector includes displacement trajectory under force field coupling, thermal conduction direction gradient, and evolution time scale parameters. The risk probability distribution map includes pattern matching confidence, historical high-risk evolution overlap, and evolution path risk weight. The optimized screening point set includes coordinates of key conduction nodes, node thermal resistance weight assessment, and screening density distribution suggestions. The final defect identification results include multi-dimensional feature verification scores, identification accuracy matrix, and thermal runaway evolution trajectory map. The thermal runaway risk prediction report includes safety threshold warning suggestions, quality control process adjustment planning, and full life cycle safety benefit prediction.

[0014] In the thermal imaging data generation step, which includes locally overheated areas, a sequence of images of the battery pack under multi-field coupling is acquired using an infrared thermal imaging device. Gray-scale conversion technology is used to map the RGB color spectrum into single-channel brightness information, including emissivity correction for high-sensitivity detectors, background noise suppression, and multi-frame averaging alignment. This data is processed into a standardized matrix format, such as the Kelvin temperature calibration value and spatial coordinate index for each pixel. Next, edge enhancement methods are applied using the Laplacian operator or the Sobel operator to analyze the boundary gradient of the hotspot region. This includes calculating the convolution kernel parameters of the gradient operator to identify hotspot distortion caused by small metallic foreign objects. During this process, enhancement coefficients are calculated as follows: The system is dynamically adjusted to adapt to different lens focal lengths to ensure the accuracy of local overheating area extraction. The analysis results reveal the geometric contour features of the overheated area, such as irregular hot spots and edge heat diffusion caused by local current concentration, providing accurate underlying data for subsequent classification and recognition. The final step generates a detailed dataset containing grayscale histograms, enhanced contours, and candidate coordinates.

[0015] In the initial defect location and type label generation step, temperature gradient analysis and support vector machine (SVM) algorithms are used to classify and identify abnormal hot spots based on thermal imaging data. Temperature gradient analysis involves identifying the first and second derivatives of the cooling curve radiating outward from the hot spot center, capturing differences in high-frequency hot spots caused by metallic foreign objects and low-frequency thermal distribution caused by diaphragm damage. The SVM model is used to perform hyperplane partitioning based on the extracted feature vectors, distinguishing between foreign object short circuits and physical structural damage. Refinement operations in this step include adjusting the kernel function parameters to accommodate nonlinear perturbations of the electric field intensity on the temperature field distribution. These models are used to lock the defect coordinates. Through these analyses, identification labels describing the defect coordinates, type (e.g., copper shavings, local diaphragm indentations), and peak intensity are generated, providing physical input for subsequent diffusion dynamics analysis.

[0016] In the diffusion trend index generation step, a pixel sequence analysis method is used to track the temperature evolution around the initial defect over time. This step involves comparing pixel-by-pixel thermal images of more than 100 consecutive frames to identify areas where the temperature rise exceeds a certain threshold. The mutation sequence was analyzed. By calculating the thermal coupling coefficient between pixels, a time-dependent model of heat wave diffusion was established. These models can predict the rate at which the overheated region spreads to surrounding individual cells at a fixed discharge rate. Furthermore, the steps included dynamically marking the diffusion boundary and calculating the curvature change of the diffusion front to predict the existence of a through-type heat conduction channel. The generated diffusion trend index not only helps engineers understand the dynamic evolution mechanism of defects but also provides a crucial temporal feature dimension for subsequent CNN path prediction.

[0017] In the evolution path vector generation step, a path prediction scheme is designed based on diffusion trend indicators using convolutional neural networks (CNNs) and spatiotemporal sequence analysis. This includes analyzing the displacement response of the internal structure of the battery pack under the current force field environment (such as 10kN compression or high-frequency vibration) and designing a prediction model for thermal runaway propagation. For example, CNNs are used to extract spatial feature changes between different frames, and 3D convolutional kernels are used to capture the anisotropic diffusion path of heat flow under force field induction. Simultaneously, considering the influence of different cell arrangement structures on thermal resistance, such as using high sampling rate prediction in the liquid cooling plate contact area to optimize thermal path simulation, the result of this step is an evolution path vector that not only contains the geometric trajectory of thermal propagation but also quantifies the accelerating effect of the force field on the defect growth path.

[0018] In the risk probability distribution map generation step, a probabilistic pattern matching method is employed, combining a historical thermal runaway risk database and evolution path vectors. By calculating the cosine similarity between the current evolution vector and accident samples in the database, particularly considering the typical characteristics of diaphragm collapse caused by metal particle puncture, this method can accurately reflect the correlation between the evolution trajectory and known risk patterns. The model considers various factors, such as battery pack SOC state, ambient temperature fluctuations, and cooling system efficiency. By adjusting matching thresholds, such as confidence intervals... This ensures the scientific rigor of the identification results. Furthermore, the assessment includes the allocation of risk weights for evolutionary paths, such as the increased risk level near the electrolyte injection port or tab region. The generated charts detail the risk probability gradient under full coverage.

[0019] In the step of optimizing the screening point set generation, a chain transmission analysis method is used for weight evaluation based on the risk probability map. The system dynamics model can simulate the complex nonlinear feedback from microscopic defect initiation, mesoscopic thermal diffusion to macroscopic thermal runaway, and predict the contribution of different screening densities to reducing the false negative rate. These strategies include adjusting the key scanning area of ​​the infrared camera, setting the optimal spacing of the sensor layout, and introducing multi-dimensional field coupling compensation. The operability of the scheme is ensured by finely adjusting the model parameters, such as the electric field interference frequency and the image acquisition step size. The result of this step is an optimized screening point set, including a coordinate list, sampling priority, and threshold alarm triggering logic, providing specific guidance for achieving automated and efficient screening.

[0020] In the final defect identification result generation step, the screening point set and evolution path verification results are comprehensively optimized, and a combination of support vector machine (SVM) algorithm and multi-dimensional feature verification is used for a balanced approach. By simulating the feature responses under different defect evolution stages, the verification algorithm can eliminate false positive signals caused by fluctuations in the testing environment. In this process, the penalty parameter of SVM... and Carefully tuned to adapt to the highly dynamic coupled field search space, the results of the steps are a highly reliable defect identification matrix, including defect authenticity score, thermal runaway trajectory prediction accuracy, and risk depth prediction value, aiming to achieve global accuracy in power battery safety screening.

[0021] In the thermal runaway risk prediction report generation step, based on the final identification results, a comprehensive planning of the overall battery pack quality is conducted using a full life cycle assessment and safety benefit analysis. The life cycle assessment covers everything from impurity contamination during manufacturing and force field damage during transportation to fatigue accumulation during end-use, including the long-term impact of the separator's microscopic aging during cycling on thermal stability. The safety benefit analysis involves assessing the balance between the reduced accident rate and increased production costs after improving screening programs, ensuring the commercial viability of quality control. The generated report, including clearly defined risk warning levels, suggested process optimization paths, and expected safety benefit predictions, provides battery manufacturers with a closed-loop decision-making framework from defect detection to risk management.

[0022] The power battery pack screening test method based on multi-field coupling simulation described in this embodiment generates thermal imaging data containing local overheated areas in the following steps: S101, based on multiple frames of thermal distribution images under the electric field and force field simulation environment collected by infrared thermal imaging equipment, a data fusion algorithm is used to spatially synchronize the temperature matrix collected by the multi-view infrared camera. Then, the three-dimensional position of the battery pack corresponding to the data is calibrated through a geometric coordinate system, and a weighted average is applied. Differentiated weights are assigned according to the calibration accuracy of each camera in different temperature zones. Multi-source process parameters are integrated to generate a comprehensive thermal imaging data analysis set; S102, based on the comprehensive thermal imaging data analysis set, a time series analysis algorithm is used to analyze the temperature rise data within the test period through an autoregressive model to identify the long-term trend of background fluctuations and environmental interference, and spectral analysis is used to distinguish... The frequency components in the thermal imaging data reveal the periodic temperature fluctuations caused by the cooling cycle, generating thermal dynamic feature analysis results. S103: Based on the thermal dynamic feature analysis results, a machine learning classification algorithm is used to extract the shape factor and temperature center moment features of the overheated area through a support vector machine, and the feature space boundary is divided. Then, a decision tree algorithm is used to classify the causes of overheating (foreign objects / poor contact) according to the extracted features, generating an overheated area classification overview. S104: Based on the overheated area classification overview, an evolutionary situation analysis method is used. A Markov chain model is used to perform probabilistic analysis of multiple evolution levels of local overheating, predicting the thermal runaway conversion path. A stochastic process model is used to quantify the randomness of hot spot growth, assessing future safety risk changes, and generating thermal imaging data containing local overheated areas. In sub-step S101, a multi-camera acquisition system on the battery pack test line is used to achieve multi-view data fusion. First, the Celsius data collected by four infrared cameras are synchronized using a timestamp correction algorithm. Then, the 3D reconstruction technology of the vision system is used to map the depth coordinates of the hot spots within the battery pack. The data from each sensor is weighted according to its signal-to-noise ratio. For example, the weight of the long-wave infrared camera is set to... The weights of the mid-wave camera are set to The comprehensive analysis set generated in this process has a temperature resolution of 0.1K, providing a high-quality foundation for subsequent edge enhancement. In sub-step S102, the comprehensive analysis set is fed into the time-frequency domain processing unit. First, an autoregressive (AR) model is applied to identify the background temperature rise under steady-state current, which is typically related to external bus heating. Then, the thermal disturbance frequencies related to the pump frequency are identified through Fast Fourier Transform (FFT). These analyses reveal the noise patterns caused by non-defects, and the generated dynamic feature results can reflect the thermal flow characteristics inside the battery pack, playing a crucial role in distinguishing environmental noise from real defect points. In sub-step S103, the morphological features of the thermal imaging are processed. SVM projects multidimensional features (maximum temperature, average temperature, variance, and bias) into a high-dimensional space through the RBF kernel function. This enables the system to distinguish the subtle differences between bolt loosening and heating caused by vibration (large-area low-temperature rise) and short circuits caused by metallic foreign objects (point-like extremely high-temperature rise). Subsequently, the logic chain generated by the decision tree can quickly determine the defect category. The classification overview includes the defect type and its sensitivity to the force field, providing input for electric / force field optimization. In sub-step S104, a Markov chain model is used to define three risk levels: stable, slow evolution, and rapid propagation, and the state transition probability is calculated. For example, when the internal pressure exceeds... At this point, the probability of the system shifting to "rapid propagation" increases significantly. This stochastic process model can capture the risk of sporadic short circuits caused by force field disturbances, and the generated thermal imaging analysis data is the core basis for subsequent evolution path prediction. Assuming that in a module stress test, the sensor collects a hotspot temperature of... After synchronization and integration with a weight of 0.8, the autoregressive model revealed a nonlinear exponential increase in the temperature rise rate. SVM classification identified it as a 50μm stainless steel shaving foreign object. Markov chain predictions indicate that under a sustained force field, the probability of entering the pre-thermal runaway stage will increase from 2% to 28%.

[0023] The method for screening and testing power battery packs based on multi-field coupling simulation described in this embodiment generates preliminary defect location and type labels in the following steps: S201, based on the thermal imaging data containing local overheated areas, an autoregressive moving average model is used to statistically model the hotspot time series data, analyze the autocorrelation and moving average characteristics in the historical temperature rise curves, reveal the periodic changes in temperature fluctuations, and generate a temperature rise trend analysis; S202, based on the temperature rise trend analysis, a system dynamics model is used to simulate the thermal field under the coupling force field by establishing and analyzing the differential equation describing the heat flow penetration rate in the multilayer cell structure. Dynamic changes, including conduction loss and radiation loss modes, generate thermal interference and loss mode analysis; S203, based on the thermal interference and loss mode analysis, apply network analysis methods to analyze the mutual thermal influence between differentiated cells by calculating the impedance connectivity and conduction coefficient of multiple nodes in the equivalent thermal circuit inside the battery pack, identify weak areas in the heat dissipation link, and generate thermal interaction analysis; S204, based on the thermal interaction analysis, use multivariate regression analysis, through statistical modeling, combined with historical thermal runaway test and real-time sensing data, to quantitatively predict the temperature rise evolution trend of the current defect in future discharge cycles, and generate preliminary defect location and type labels. In sub-step S201, the signal of the hot spot is modeled using an ARIMA model. First, real-time sampling data from charge and discharge tests are collected. The autoregressive part is used to capture the trend temperature rise caused by the increase in cell internal resistance, while the moving average part is used to filter out transient interference caused by electric field fluctuations. Parameters and Optimization was performed using the AIC criterion. The generated trend analysis revealed a deep quantitative relationship between energy accumulation and time series within the defect region. In sub-step S202, thermal field modeling employed a discrete form of the heat transfer differential equation. By establishing partial differential equations describing heat flow through the three-layer structure of aluminum shell-insulating film-electrolyte, the loss distribution under different states of charge (SOC) was simulated. The generated thermal interference pattern analysis indicated that under specific foreign object distributions, the thermal effect caused by local polarization current was the dominant factor in temperature rise, providing a physical basis for determining the foreign object size. In sub-step S203, the battery pack was abstracted as an impedance network. The equivalent thermal resistance between different nodes (cells) was calculated. High connectivity regions correspond to heat accumulation centers, typically heat dissipation blind spots. The thermal weak points identified by network analysis are potential thermal runaway points. In sub-step S204, a multivariate regression model integrated multiple variables such as foreign object material, contact pressure, and discharge current. By fitting a large dataset, a predictive equation for defect evolution was obtained. The model's predictions indicate which defects will be the first to exceed the safety threshold in future cycles. Assuming the data sequence is processed using ARIMA(2,1,1), a temperature peak is identified every 30 minutes. Simulation S202 shows that foreign body puncture leads to an increase in the local heat transfer coefficient. Impedance analysis revealed a coupling coefficient as high as 0.58 at the module edge. Multiple regression analysis predicted an additional temperature rise at this point under 2C discharge. .

[0024] The power battery pack screening test method based on multi-field coupling simulation described in this embodiment generates diffusion trend indicators in the following steps: S301, based on the preliminary defect location and type label, an autoregressive model is used to predict the change of temperature rise amplitude in the future multiple frames by calculating the autocorrelation in the historical temperature sequence, analyzing and simulating the time dependence of thermal diffusion, and generating a temperature rise intensity trend prediction; S302, based on the temperature rise intensity trend prediction, an exponential smoothing method is used to perform weighted averaging on the prediction data, highlighting the impact of recent force field changes on future thermal stability and weakening the impact of single-frame identification errors on diffusion path prediction, generating a short-term thermal risk prediction; S303, based on the short-term thermal risk prediction, a seasonal autoregressive comprehensive moving average model is used to analyze the interaction between ambient temperature and battery self-heating effect, predict the probability of thermal spread in a specific future test cycle, and generate a cyclic interference prediction analysis; S304, based on the cyclic interference prediction analysis, through cluster analysis, the internal region of the battery pack is divided into differentiated risk groups according to the similarity of thermal feature data, high-risk evolution frequency bands and key conduction paths are identified, and a diffusion trend indicator is generated. In substep S301, an autoregressive model quantifies the temperature rise evolution logic. For the key feature of the diaphragm rupture area, its autocorrelation function in past sampling sequences is analyzed. The electrochemical thermal productivity for the next 600 seconds is extrapolated. The generated trend prediction can anticipate the risk of thermal diffusion due to accelerated chemical reactions. In substep S302, exponential smoothing is used by setting a smoothing coefficient. The system focuses on temperature feedback from the most recent five frames, while eliminating random fluctuations in the infrared sensor during high-speed scanning. The generated short-term predictions guide the real-time monitoring system to dynamically adjust the sampling frequency. In substep S303, a seasonality model (SARIMA) is used to analyze the impact of ambient temperature on the test results. Since battery heat dissipation efficiency fluctuates non-linearly with ambient temperature, the model incorporates a periodic term. (Hourly) to predict hotspot detection bias caused by daytime temperature variations. This analysis can identify production risks during high-temperature periods. In sub-step S304, cluster analysis uses the K-means++ algorithm. Based on the highest temperature, temperature rise rate, and diffusion curvature, the battery pack is divided into safe zone, warning zone, and danger spread zone. This spatial clustering hierarchy makes risk management more targeted. It is assumed that the autoregressive model will predict a 15% increase in thermal productivity in the next stage. Exponential smoothing confirms that the defect propagation is induced by force field compression. The SARIMA model suggests that... During the above environmental tests, the interference probability was as high as 85%. Cluster analysis identified the central module as a "high-risk group".

[0025] The power battery pack screening test method based on multi-field coupling simulation described in this embodiment generates the evolution path vector in the following steps: S401, based on the diffusion trend index, a linear programming algorithm is used to set the charging and discharging current and environmental pressure in the test process as decision variables, and the battery pack structural strength limit and prediction accuracy index as constraints. The parameters of each test segment are allocated by solving an optimization problem to reduce the impact of environmental noise on the evolution path, generating an optimized test process design; S402, based on the optimized test process design, graph theory analysis is used, including constructing a heat flow distribution map of the battery pack surface, where feature points are nodes and heat conduction paths are edges, and calculating the equivalent thermal resistance between nodes. S403. Based on the optimized evolution trajectory configuration, a dynamic resource allocation model is applied to update the scanning step size and focal length of the infrared camera according to real-time sensor feedback. This allows for real-time adjustments to the test execution system to match changes in temperature rise rate, reduce image blurring and artifact interference during dynamic evolution, and generate vectorized evolution trajectory parameters. S404. Based on the vectorized evolution trajectory parameters, optimized test process design, and optimized evolution trajectory configuration, a comprehensive evolution management strategy is formulated to generate an evolution path vector. The current and pressure in the test process are set as decision variables, and the heat flux density conservation model formula is used to solve the optimization problem. in, Let be the local thermal energy density, and represent the coordinates. Location and Time Heat distribution, Let be the heat flux function, representing the product of the heat conduction rate and the temperature gradient. The internal heat source intensity is considered. In sub-step S401, a test-prediction optimization model is established using linear programming. The objective function is set to maximize the prediction confidence. The decision variables include current. and pressure The constraints include a pressure threshold that does not trigger the pressure relief valve and sensor sensitivity requirements. The optimal parameter combination is obtained by solving using the simplex method. The generated optimized design maximizes the excitation of defect thermal features while ensuring safety. In substep S402, graph theory analysis focuses on the continuity of heat flow. High thermal resistance nodes in the heat conduction path are identified by constructing a feature point impedance model. For these nodes, an asymmetric convolutional scan path is designed. The generated optimized configuration improves the capture efficiency of the evolution path by approximately 22%. In substep S403, the dynamic resource allocation model is based on feedback control. When the temperature rise rate exceeds... At that time, the controller automatically increases the frame rate of the infrared camera to [a higher value]. This closed-loop control effectively reduces image dynamic artifacts and ensures high fidelity of feature vectors at each evolution stage. In the S404 sub-step, a global evolution management strategy is formulated. "Multi-field redundancy recognition" is introduced, automatically enabling ultrasound-based assisted recognition when the force field is complex. The generated evolution path vector is a spatiotemporal trajectory report covering electrical, thermal, and mechanical dimensions. Assuming linear programming, the current is determined as... The pressure is To achieve optimal path clarity, the dynamic system automatically prioritizes allocating computational resources to the region upon detecting localized charge accumulation. This ultimately reduces the RMSE of evolution path predictions by 18%.

[0026] The method for screening and testing power battery packs based on multi-field coupling simulation described in this embodiment generates a risk probability distribution map in the following steps: S501, based on the historical thermal runaway risk database and the evolution path vector, an energy wave propagation simulation is performed using a lattice Boltzmann model. The internal thermal environment of the battery pack is simulated by simulating the oscillation and collision of thermal microparticles, thereby simulating the thermal risk of multiple regions along the evolution path and generating an initial risk simulation map; S502, based on the initial risk simulation map and combined with electrolyte physicochemical data, a cellular automata model is used to analyze the influence of foreign particle arrangement on local current distribution, and the electrochemical response rules of the microstructure are defined. The process involves simulating localized heat generation changes, assessing the direct impact of microstructure on thermal runaway trigger point distortion, and generating a defect thermal impact analysis map. S503: Based on the defect thermal impact analysis map, data fusion technology is used to combine temperature and force field stress data. By integrating multi-source parameters, the interaction between diaphragm displacement caused by mechanical stress and localized heating is analyzed, assessing its comprehensive impact on thermal propagation reliability, and generating an evolution assessment comprehensive map. S504: Based on the evolution assessment comprehensive map, a probabilistic assessment model is applied to comprehensively evaluate the matching degree. By quantitatively analyzing the evolution loss and its interaction with risk modes on the prediction depth, a risk probability distribution map is generated. In sub-step S501, the Lattice Boltzmann Method (LBM) is used for high-fidelity thermal simulation. By discretizing the energy evolution equation on the D3Q15 lattice, the propagation of heat flow between non-uniform cells is simulated. The initial map generated from the simulation results shows a heat flow intensity cloud map, intuitively reflecting the similarity of the current path to the thermal runaway mode. In substep S502, a cellular automata model is used to handle the microscopic heterogeneity within the foreign object. The defect layer is divided into millions of cellular units, each executing ohmic heating and chemical reaction rules. The generated analysis map reveals the microscopic root causes leading to premature thermal runaway. In substep S503, data fusion techniques overlay thermal and stress contour maps. Analysis shows that in the region of high stress at the tab root, slight changes in thermal resistance can lead to drastic temperature feedback. By establishing a thermo-mechanical coupling matrix, the risk of defect acceleration due to vibration is assessed. In substep S504, the risk probability is quantified. A matching score is calculated using a historical database. The evaluation model considers the contribution of SOC changes to the heat of reaction. The generated final distribution map includes risk confidence scores for different battery models. Assuming that LBM simulations show heat conduction refraction in the module gaps leading to localized heat accumulation, the final probability distribution map shows a thermal runaway matching degree of 88% under this condition.

[0027] The power battery pack screening test method based on multi-field coupling simulation described in this embodiment generates an optimized screening point set in the following steps: S601, based on the risk probability distribution map, a system dynamics method is used for analysis. By establishing a dynamic feedback model of process variables, heat energy conduction, and their interactions, the impact of differentiated node layout on identification depth is simulated, generating dynamic analysis results of node adjustment; S602, based on the dynamic analysis results of node adjustment, a multi-criteria decision analysis is used, combined with image resolution and recognition parameters, to evaluate and compare the effectiveness of differentiated sensor combinations in improving identification accuracy, generating a screening optimization scheme to mitigate the risk of missed detection; S603, based on the screening optimization scheme to mitigate the risk of missed detection, Monte Carlo simulation is used. Through random sampling technology, multiple production scenarios after screening implementation are simulated to predict the performance of process adjustments under various environmental fluctuation conditions, generating screening implementation simulation prediction results; S604, based on the screening implementation simulation prediction results, a path optimization model is used to adjust and redesign each execution link, generating an optimized screening point set. In the S601 sub-step, system dynamics modeling describes the "heat energy accumulation-identification probability" feedback chain. Simulation is performed when the ambient temperature fluctuates. The analysis revealed how the node distribution affected the ability to detect minute defects. The results determined the node detection range as a key performance indicator for screening. In substep S602, multi-criteria decision analysis employed the Analytic Hierarchy Process (AHP). The criterion layer included accuracy, detection time, and hardware cost. The optimal path was selected through expert scoring and simulation comparison. In substep S603, Monte Carlo simulations were performed 20,000 times. Simulation parameters included camera shake error and the randomness of foreign object positions. Statistical results showed that, within a 99% confidence interval, the false negative rate under the new scheme decreased from 5.2% to 0.8%. In substep S604, the execution logic was redesigned. Based on the optimized weight distribution, processor computing resources were reallocated, and the machine vision's area of ​​interest was adjusted. The generated plan included the updated scan coordinate matrix.

[0028] The power battery pack screening and testing method based on multi-field coupling simulation described in this embodiment generates the final defect identification result through the following steps: S701, based on the optimized screening point set and evolution path verification results, a genetic algorithm is used to optimize the defect identification parameters and verification positions by simulating individual selection, gene crossover, and mutation, obtaining an optimized verification configuration; S702, based on the optimized verification configuration, a particle swarm optimization algorithm is used to simulate group cooperative search and fine-tune the classification hyperplane of the SVM model to improve identification efficiency and reduce overfitting risk, generating an optimized identification parameter scheme; S703, based on the optimized identification parameter scheme, a multi-objective genetic algorithm is used to capture the Pareto optimal solution among multiple objectives, including reducing false alarm noise, reducing detection time, and improving depth coverage, generating a comprehensively optimized multi-objective identification strategy; S704, based on the comprehensively optimized multi-objective identification strategy, a particle swarm optimization algorithm is used to refine the parameter tolerance range by iteratively updating the velocity and position of the scheme particles, generating the final defect identification result. In the S701 sub-step, the chromosomes of the genetic algorithm encode feature weights and spatial coordinates. Through 1000 generations of evolution, noise-sensitive recognition patterns were eliminated. The particle swarm optimization algorithm in S702 determined the penalty factor for SVM. The S703 addresses the trade-off between accuracy and time using NSGA-II. The final solution strikes an optimal balance between recognition accuracy (98%) and detection rate (30s / pcs).

[0029] The specific steps of the power battery pack screening and testing method based on multi-field coupling simulation described in this embodiment to generate a thermal runaway risk prediction report are as follows: S801, based on the final defect identification results, statistical analysis methods are used to perform trend analysis on defect data in batch production, followed by random forest evaluation of defect distribution, neural network mining of risk evolution patterns, and generation of quality status analysis results; S802, based on the quality status analysis results, a life cycle assessment model is used to analyze the safety losses in multiple stages of battery pack use, analyze the deep correlation between these losses and the thermal runaway risk caused by manufacturing defects, and generate a full life cycle quality assessment; S803, based on the full life cycle quality assessment, a cost-benefit analysis method is used to analyze the economic returns of differentiated improvement strategies through the present value method, and generate an optimized risk management strategy analysis; S804, based on the optimized risk management strategy analysis, combined with system dynamic simulation and multi-objective optimization models, a risk management plan matching specific battery pack characteristics is formulated, and a thermal runaway risk prediction report is generated. In S801, random forest identifies the membrane roll surface quality as the primary factor affecting subsequent risk fluctuations. S802 analysis shows that without current process optimization, the accident risk cost over the entire lifecycle is as high as RMB 120,000 per megawatt-hour. CBA analysis demonstrates that the internal rate of return (IRR) of this solution reaches 32%. The final report provides detailed predictive control measures, effectively reducing the probability of thermal runaway of the entire package.

[0030] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A power battery pack screening test method based on multi-field coupling simulation, characterized in that: Includes the following steps: S1. Based on the multi-frame thermal distribution images acquired by the infrared thermal imaging device under the electric field and force field simulation environment, the contrast of the image sequence is improved and the contour feature is analyzed by grayscale conversion and edge enhancement processing methods, and the local overheated area is initially extracted to generate thermal imaging data containing the local overheated area. S2. Based on thermal imaging data, temperature gradient analysis and support vector machine algorithm are used to classify and identify the temperature change rate and abnormal hot spot coordinates, and generate preliminary defect location and type labels. S3. Based on the initial defect location and type label, the pixel sequence analysis method is used to monitor the temperature change amplitude of surrounding pixels, and when the temperature change sequence exceeds the preset threshold, the diffusion boundary area is marked to generate a diffusion trend index. S4. Based on the diffusion trend index, a convolutional neural network model and spatiotemporal sequence analysis method are used to predict the evolution path of defects under the influence of force field, and the evolution trajectory is vectorized to generate an evolution path vector. S5. Combining historical thermal runaway risk database and evolution path vectors, a probabilistic pattern matching method is used to analyze the matching degree between evolution paths and risk patterns, identify high-risk evolution patterns, and generate a risk probability distribution map. S6. Based on the risk probability distribution map, the energy conduction path from local overheating to overall thermal runaway is analyzed using image overlay and chain conduction analysis methods. The weights of key nodes in the conduction chain are evaluated to generate an optimized screening point set. S7. Based on the comprehensive optimization of the screening point set and evolution path verification results, the support vector machine algorithm and multi-dimensional feature verification are used to conduct a comprehensive verification and evaluation report design for the defect identification results, the accuracy of thermal runaway evolution and the depth of risk prediction, and generate the final defect identification results.

2. The power battery pack screening test method based on multi-field coupling simulation according to claim 1, characterized in that: Step S1 is as follows: S101. Based on the multi-frame thermal distribution images acquired by the infrared thermal imaging equipment under the electric field and force field simulation environment, the temperature matrix collected by the multi-view infrared camera is spatially synchronized using a data fusion algorithm. Then, the three-dimensional position of the battery pack corresponding to the data is calibrated through the geometric coordinate system. Then, a weighted average is applied, and differentiated weights are assigned according to the calibration accuracy of each camera in different temperature zones. The process parameters from multiple sources are integrated to generate a comprehensive thermal imaging data analysis set. S102. Based on the comprehensive thermal imaging data analysis set, a time series analysis algorithm is used to analyze the temperature rise data within the test cycle through an autoregressive model to identify the long-term trend of background fluctuations and environmental interference. The frequency components in the thermal imaging data are distinguished by spectral analysis to reveal the periodic temperature drop fluctuations caused by the cooling cycle and generate thermal dynamic characteristic analysis results. S103. Based on the thermal dynamic feature analysis results, a machine learning classification algorithm is used to extract the shape factor and temperature center moment features of the overheated region through a support vector machine, and the boundary of the feature space is divided. Then, a decision tree algorithm is used to classify the causes of overheating according to the extracted features, and an overview of the overheated region classification is generated. S104. Based on the overheated area classification overview, using the evolutionary situation analysis method, the Markov chain model is used to perform probabilistic analysis on multiple evolution levels of local overheating, predict the thermal runaway conversion path, use the stochastic process model to quantify the randomness of hot spot growth, assess future changes in safety risks, and generate thermal imaging data containing local overheated areas.

3. The power battery pack screening test method based on multi-field coupling simulation according to claim 1, characterized in that: Step S2 is as follows: S201. Based on the thermal imaging data containing the local overheated area, an autoregressive moving average model is adopted. By statistically modeling the hotspot time series data, the autocorrelation and moving average characteristics in the historical temperature rise curve are analyzed to reveal the periodic changes in temperature fluctuations and generate a temperature rise trend analysis. S202. Based on the temperature rise trend analysis, the system dynamics model is used to establish and analyze the differential equation describing the heat flow penetration rate in the multilayer cell structure, simulate the dynamic changes of the thermal field under the coupled force field, including conduction loss and radiation loss modes, and generate thermal interference and loss mode analysis. S203. Based on the thermal interference and loss mode analysis, network analysis method is applied to analyze the mutual thermal influence between different cells by calculating the impedance connectivity and conduction coefficient of multiple nodes in the equivalent thermal circuit inside the battery pack, identifying weak areas in the heat dissipation link, and generating thermal interaction analysis. S204. Based on the aforementioned thermal interaction analysis, multiple regression analysis is adopted. Through statistical modeling, combined with historical thermal runaway tests and real-time sensor data, the temperature rise evolution trend of the current defect in future discharge cycles is quantitatively predicted, and preliminary defect location and type labels are generated.

4. The power battery pack screening test method based on multi-field coupling simulation according to claim 1, characterized in that: Step S3 is as follows: S301. Based on the preliminary defect location and type label, an autoregressive model is used to predict the change of temperature rise amplitude in the future multiple frames by calculating the autocorrelation in the historical temperature sequence, analyzing and simulating the time dependence of heat diffusion, and generating a temperature rise intensity trend prediction. S302. Based on the predicted temperature rise intensity trend, the predicted data is weighted and averaged using the exponential smoothing method to highlight the impact of recent force field changes on future thermal stability and reduce the impact of single-frame recognition error on diffusion path prediction, thereby generating a short-term thermal risk prediction. S303. Based on the aforementioned short-term thermal risk prediction, a seasonal autoregressive moving average model is used to analyze the interaction between ambient temperature and battery self-heating effect, predict the probability of thermal spread within a specific test cycle in the future, and generate a cyclical interference prediction analysis by referring to environmental factors. S304. Based on the cyclic interference prediction analysis, the internal region of the battery pack is divided into differentiated risk groups according to the similarity of thermal feature data through cluster analysis, high-risk evolution frequency bands and key conduction paths are identified, and key risk characteristics are highlighted by quantifying the distribution differences of process parameters, and diffusion trend indicators are generated.

5. The power battery pack screening test method based on multi-field coupling simulation according to claim 1, characterized in that: Step S4 is as follows: S401. Based on the diffusion trend index, a linear programming algorithm is used to set the charging and discharging current and environmental pressure in the test process as decision variables, and the battery pack structural strength limit and prediction accuracy index as constraints. The parameters of each test segment are allocated by solving the optimization problem, reducing the impact of environmental noise on the evolution path, and generating an optimized test process design. S402. Based on the optimized design of the test process, graph theory analysis is used, including constructing a heat flow distribution map of the battery pack surface, where feature points are used as nodes and heat conduction paths are used as edges, calculating the equivalent thermal resistance distribution between nodes, adjusting the path prediction accuracy according to the heat flow distribution, and generating an optimized configuration of the evolution trajectory. S403. Based on the optimized configuration of the evolution trajectory, apply the dynamic resource allocation model, update the scanning step size and focal length of the infrared camera according to real-time sensor feedback, adjust the test execution system in real time, match the changes in temperature rise rate, reduce image blurring and artifact interference in dynamic evolution, and generate vectorized parameters of the evolution trajectory. S404. Based on the vectorized parameters of the evolution trajectory, the optimized design of the test process, and the optimized configuration of the evolution trajectory, formulate a comprehensive evolution management strategy, adjust the production line layout and calibration frequency, and generate an evolution path vector.

6. The power battery pack screening test method based on multi-field coupling simulation according to claim 1, characterized in that: Step S5 is as follows: S501. Based on the historical thermal runaway risk database and the evolution path vector, the energy wave propagation simulation is carried out using the lattice Boltzmann model. The internal thermal environment of the battery pack is simulated by simulating the oscillation and collision of thermal microparticles. The thermal risk simulation of multiple regions of the evolution path is realized, the potential improvement of the implementation of the scheme on risk prediction is analyzed, and an initial risk simulation diagram is generated. S502. Based on the initial risk simulation diagram and combined with the electrolyte physicochemical data, the influence of foreign particle arrangement on local current distribution is analyzed using a cellular automata model. Local heat generation changes are simulated by defining the electrochemical response rules of the microstructure, the direct influence of the microstructure on the distortion of the thermal runaway trigger point is evaluated, and a defect thermal influence analysis diagram is generated. S503. Based on the aforementioned defect thermal impact analysis diagram, the temperature field and force field stress data are combined using data fusion technology. By integrating multi-source parameters, the interaction between diaphragm displacement and local heating caused by mechanical stress is analyzed, its comprehensive impact on thermal propagation reliability is evaluated, and an evolution evaluation comprehensive diagram is generated. S504. Based on the aforementioned evolution assessment comprehensive map, a probability assessment model is applied to comprehensively evaluate the matching degree. By quantitatively analyzing the evolution loss and its interaction with risk patterns on the prediction depth, a compensation scheme is provided to conduct a comprehensive assessment of the battery pack safety performance and generate a risk probability distribution map.

7. The power battery pack screening test method based on multi-field coupling simulation according to claim 1, characterized in that: Step S6 is as follows: S601. Based on the risk probability distribution map, the system dynamics method is used for analysis. By establishing a dynamic feedback model of process variables, heat energy conduction and their interaction, the influence of differentiated node layout on recognition depth is simulated. The dynamic analysis results of node adjustment are generated with reference to electric field interference frequency and image acquisition step size. S602. Based on the dynamic analysis results of the node adjustment, adopt multi-criteria decision analysis, combine image resolution and recognition parameters, evaluate and compare the effectiveness of differentiated sensor combinations in improving recognition accuracy, and generate a screening optimization scheme to mitigate the risk of missed detection. S603. Based on the screening optimization scheme to mitigate the risk of missed detection, Monte Carlo simulation is used to simulate multiple production scenarios after the screening is implemented through random sampling technology, predict the performance of process adjustment under various environmental fluctuation conditions, evaluate the robustness of the scheme, and generate screening implementation simulation prediction results. S604. Based on the simulation prediction results of the screening implementation, the path optimization model is adopted to adjust and redesign each execution link, and the optimized screening point set is generated by referring to the coordinate list and sampling priority optimization.

8. The power battery pack screening test method based on multi-field coupling simulation according to claim 1, characterized in that: Step S7 is as follows: S701. Based on the optimized screening point set and evolution path verification results, a genetic algorithm is used to optimize the defect identification parameters and verification locations by simulating individual selection, gene crossover and mutation, and to obtain the optimized verification configuration. S702. Based on the optimized verification configuration, the particle swarm optimization algorithm is used to simulate group cooperative search and optimize the classification hyperplane and penalty parameters of the support vector machine model to improve recognition efficiency and reduce the risk of overfitting, thereby generating an optimized recognition parameter scheme. S703. Based on the optimized identification parameter scheme, a multi-objective genetic algorithm is adopted to capture the Pareto optimal solution by simulating the natural selection mechanism among multiple objectives, including reducing false alarm noise, reducing detection time and improving depth coverage, balancing multi-dimensional indicators, and generating a comprehensive optimized multi-objective identification strategy. S704. Based on the comprehensive optimized multi-target identification strategy, a particle swarm optimization algorithm is adopted. By iteratively updating the velocity and position of the particles representing potential solutions, the particles are made to move closer to the individual and global optimal identification points. The strategy is then finely adjusted, the parameter tolerance range is refined, and the final defect identification result is generated.

9. The power battery pack screening test method based on multi-field coupling simulation according to claim 1, characterized in that: The power battery pack screening test method based on multi-field coupling simulation also includes the following steps: S8. Based on the final defect identification results, a comprehensive assessment of the overall safety and quality of the battery pack and production planning are conducted using full life cycle risk assessment and safety benefit analysis, and a thermal runaway risk prediction report is generated.

10. The power battery pack screening test method based on multi-field coupling simulation according to claim 9, characterized in that: Step S8 is as follows: S801. Based on the final defect identification result, statistical analysis methods are used to conduct trend analysis and fluctuation exploration of defect data in batch production. Subsequently, a machine learning prediction model is used to evaluate the defect distribution through random forest, and neural network is used to mine the risk evolution law to generate quality status analysis results. S802. Based on the quality status analysis results, a life cycle assessment model is adopted to analyze the safety losses in multiple stages of battery pack use and to analyze the deep correlation between them and the risk of thermal runaway caused by manufacturing defects. In this way, a reliability life cycle assessment is carried out to generate a battery pack system life cycle quality assessment. S803. Based on the life cycle quality assessment of the battery pack system, the cost-benefit analysis method is used to analyze the economic returns of the differentiated improvement strategy through the present value method, the internal rate of return method is used to assess the financial feasibility of the process upgrade, the most efficient risk management strategy is selected, and an optimized risk management strategy analysis is generated. S804. Based on the analysis of the optimized risk management strategy, combined with the system dynamic simulation method and multi-objective optimization model, the evolution behavior of the battery pack process system is simulated through system dynamic theory, the optimal quality balance point is searched using a genetic algorithm, a risk management plan matching the specific characteristics of the battery pack is formulated, and a thermal runaway risk prediction report is generated.

Images