A method and system for evaluating the performance of a damping thermal barrier material for a battery pack

Abstract

This invention discloses a method and system for evaluating the performance of damping and thermal insulation materials for battery packs, relating to the field of battery pack thermal management technology. The method includes: extracting a calibration sample set; applying transient thermal excitation and acquiring infrared thermographic sequences to extract thermal response features; performing industrial CT scans to obtain defect quantification indicators, which, together with the thermal response features, form a calibration dataset; establishing a quantitative correlation model based on the calibration dataset and determining batch quality thresholds according to the model output distribution; performing the same excitation and acquisition on online test samples to extract their features, inputting them into the model to obtain defect quantification prediction values, comparing them with the batch quality thresholds, and outputting the evaluation results; and supplementing the model and thresholds by extracting new samples and obtaining new calibration data when a preset period or trigger condition is met. This invention provides accurate material performance evaluation by establishing a quantitative model of thermal response features and defect quantification indicators, and enhances the consistency and stability of the evaluation through a dynamic update mechanism.

Description

Technical Field

[0001] This invention relates to the field of battery pack thermal management technology, specifically to a method and system for evaluating the performance of damping and heat insulation materials for battery packs. Background Technology

[0002] With the widespread adoption of energy-intensive equipment and the rapid development of the electric vehicle industry, battery packs, as a core component of electric vehicles, are receiving increasing attention. To ensure the safety and reliability of battery packs, performance evaluation of their internal materials has become a crucial aspect of battery design and manufacturing. In recent years, the thermal management function of damping and insulating materials in battery packs has gradually become a research focus. The main function of thermal management materials is to regulate internal temperature changes within the battery pack, prevent the impact of high temperatures on battery performance, and improve the efficiency of the battery system. With advancements in measurement and computer simulation technologies, industrial CT and infrared thermography have been widely applied to the performance evaluation of thermal management materials. These technologies provide relatively accurate means of acquiring performance data such as internal and external heat conduction and heat diffusion within the battery pack.

[0003] However, existing methods for evaluating the performance of battery pack damping and thermal insulation materials still have certain shortcomings. First, traditional thermal response feature extraction methods often rely on human experience, lacking efficient and reproducible techniques, resulting in low accuracy and consistency of evaluation results. Second, current performance evaluation methods mainly rely on single testing methods, such as infrared thermography or industrial CT scanning, lacking comprehensive analysis based on multiple data sources, thus limiting the ability to identify and predict material defects. Furthermore, most existing technologies fail to achieve dynamic adaptive updates and lack intelligent calibration capabilities for material performance differences between different production batches. This leads to insufficient stability and reliability of evaluation results, especially during mass production and long-term use, failing to adequately cope with complex and changing working environments. To address these issues, existing technologies cannot provide a systematic and iterative evaluation method; therefore, an evaluation system integrating multiple data sources and capable of dynamic updates is urgently needed to solve the aforementioned technical challenges. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing battery pack thermal management material evaluation methods have problems such as low accuracy and consistency of evaluation results, reliance on a single testing method, lack of dynamic update mechanism, and inability to effectively identify material defects. How to achieve more accurate and reliable performance evaluation based on multiple data sources and dynamic calibration mechanism?

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, embodiments of the present invention provide a method for evaluating the performance of damping and heat-insulating materials for battery packs, characterized by comprising the following steps:

[0008] S1: Extract a calibration sample set from the production batch corresponding to the material to be evaluated, apply a preset transient thermal excitation to each sample in the calibration sample set and collect an infrared thermal image sequence, and extract thermal response features based on the infrared thermal image sequence.

[0009] S2: Perform industrial CT scanning on the calibration sample set to obtain defect quantification indicators, and combine the defect quantification indicators with the thermal response characteristics of the corresponding samples to form a calibration dataset;

[0010] S3: Establish a quantitative correlation model between thermal response characteristics and defect quantification indicators based on the calibration dataset, and determine the batch quality threshold according to the output distribution of the quantitative correlation model;

[0011] S4: Apply the transient thermal excitation to the online sample to be tested and acquire an infrared thermal image sequence to extract the thermal response characteristics of the sample to be tested;

[0012] S5: Input the thermal response characteristics of the sample to be tested into the quantitative correlation model to obtain the defect quantification prediction value, compare the defect quantification prediction value with the batch quality threshold, and output the performance evaluation result of the sample to be tested;

[0013] S6: When the preset period or trigger condition is met, new calibration samples are extracted and S1 and S2 are executed to obtain new calibration data. The quantitative correlation model and the batch quality threshold are updated based on the new calibration data.

[0014] As a preferred embodiment of the method for evaluating the performance of damping and heat-insulating materials for battery packs according to the present invention, step S1, which involves applying a preset transient thermal excitation to each sample in the calibration sample set and acquiring an infrared thermal image sequence, includes:

[0015] Stratified sampling is used to draw samples from production batches to form a statistically representative calibration sample set;

[0016] A pulsed heat source was used to apply transient thermal excitation to each sample in the calibration sample set according to uniform excitation parameters;

[0017] While applying the transient thermal excitation, an infrared thermal imager is used to synchronously acquire infrared thermal image sequences of each sample surface with preset acquisition parameters, and a baseline thermal image is acquired before acquisition for background subtraction.

[0018] As a preferred embodiment of the performance evaluation method for damping and heat insulation materials used in battery packs according to the present invention, wherein: in step S1, the extraction of thermal response features based on the infrared thermal image sequence includes:

[0019] The infrared thermal image sequence is preprocessed, including bad pixel correction, noise reduction filtering, background subtraction, and baseline normalization.

[0020] From the preprocessed infrared thermal image sequence, temporal variation features, spatial distribution features, and spatiotemporal integrated features are extracted to form thermal response features.

[0021] As a preferred embodiment of the performance evaluation method for damping and heat insulation materials used in battery packs according to the present invention, wherein: in step S2, the step of forming a calibration dataset by combining the defect quantification index with the thermal response characteristics of the corresponding sample includes:

[0022] Industrial CT scans were performed on the calibration sample set to obtain the internal three-dimensional structural data of each sample.

[0023] Based on the three-dimensional structural data, defect quantification indicators are identified and calculated. These defect quantification indicators include the debonding area ratio, the delamination gap thickness, and the porosity, which characterizes the thermal resistance anomaly at the interface and the thermal conductivity heterogeneity of the bulk material.

[0024] The defect quantification index of each sample obtained by calculation is paired one-to-one with the thermal response characteristics of the same physical sample obtained in step S1 to form a calibration dataset.

[0025] As a preferred embodiment of the performance evaluation method for damping and heat insulation materials used in battery packs according to the present invention, wherein: in step S3, establishing a quantitative correlation model from thermal response characteristics to defect quantification indicators based on the calibration dataset includes:

[0026] The calibration dataset is divided into a training subset and a validation subset;

[0027] The quantitative correlation model is obtained by training the thermal response features in the training subset as input and the corresponding defect quantification index as output through a machine learning regression algorithm.

[0028] The trained model is validated using the validation subset to evaluate its prediction accuracy and generalization ability.

[0029] As a preferred embodiment of the performance evaluation method for damping and heat insulation materials used in battery packs according to the present invention, wherein: in step S3, determining the batch quality threshold based on the output distribution of the quantitative correlation model includes:

[0030] The thermal response characteristics of the calibration sample set are input into the quantitative correlation model to obtain the corresponding set of defect quantification prediction values.

[0031] Calculate a specific statistical quantile of the predicted value set, and determine the predicted value corresponding to the quantile as the batch quality threshold.

[0032] As a preferred embodiment of the method for evaluating the performance of damping and heat-insulating materials for battery packs according to the present invention, step S4, which involves applying the transient thermal excitation to the online sample to be tested and acquiring an infrared thermal image sequence, includes:

[0033] At the online inspection station on the production line, the transient thermal excitation is applied to the online test sample using the same excitation parameters as in step S1.

[0034] Using the same acquisition parameters as in step S1, the infrared thermal image sequence of the online sample to be tested is acquired synchronously.

[0035] Using the same feature extraction methods and procedures as in step S1, the thermal response features of the online sample to be tested are extracted from the infrared thermal imaging sequence.

[0036] As a preferred embodiment of the method for evaluating the performance of damping and heat-insulating materials for battery packs according to the present invention, step S5 specifically includes:

[0037] The thermal response characteristics of the sample to be tested are input into the quantitative correlation model, and the defect quantification prediction value of the sample is output.

[0038] The defect quantification prediction value is compared with the batch quality threshold, and the classification judgment result is obtained according to the preset judgment logic;

[0039] Generate and output a structured evaluation result that includes the sample identifier, the defect quantification prediction value, the batch quality threshold, and the classification judgment result.

[0040] As a preferred embodiment of the performance evaluation method for damping and heat insulation materials used in battery packs according to the present invention, step S6, which involves updating the quantitative correlation model and the batch quality threshold based on the newly added calibration data, includes:

[0041] When a preset cycle or triggering condition is met, new calibration samples are extracted from the current and subsequent production batches; wherein, the triggering condition includes abnormal conditions dynamically determined based on online detection data, and the abnormal conditions include: the defect quantification prediction values ​​of multiple consecutive test samples show a unidirectional changing trend and exceed the control limit, the proportion of samples with prediction confidence of model output below the threshold is continuously abnormal, and after CT sampling of samples judged to be in an edge state online, it is found that the deviation between the predicted value and the true value exceeds the tolerance.

[0042] Steps S1 and S2 are performed on the new calibration sample to obtain new calibration data, which is formed by pairing thermal response characteristics with corresponding defect quantification indicators.

[0043] The parameters of the quantitative correlation model and the batch quality threshold are updated using the newly added calibration data.

[0044] Secondly, embodiments of the present invention provide a performance evaluation system for damping and heat insulation materials in battery packs, comprising:

[0045] Sample extraction and thermal excitation module: Extracts a calibration sample set from the production batch corresponding to the material to be evaluated, applies a preset transient thermal excitation to each sample in the calibration sample set and acquires an infrared thermal image sequence, and extracts thermal response features based on the infrared thermal image sequence;

[0046] Industrial CT Scanning and Calibration Module: Performs industrial CT scanning on the calibration sample set to obtain defect quantification indicators, and combines the defect quantification indicators with the thermal response characteristics of the corresponding samples to form a calibration dataset;

[0047] Model building and threshold determination module: Based on the calibration dataset, a quantitative correlation model is established from thermal response characteristics to defect quantification indicators, and the batch quality threshold is determined according to the output distribution of the quantitative correlation model;

[0048] Online detection and thermal response acquisition module: Apply the transient thermal excitation to the online sample to be tested and acquire infrared thermal image sequences to extract the thermal response characteristics of the sample to be tested;

[0049] Performance evaluation and prediction module: Input the thermal response characteristics of the sample to be tested into the quantitative correlation model to obtain the defect quantification prediction value, compare the defect quantification prediction value with the batch quality threshold, and output the performance evaluation result of the sample to be tested;

[0050] Calibration data update module: When a preset period or trigger condition is met, new calibration samples are extracted and S1 and S2 are executed to obtain new calibration data. Based on the new calibration data, the quantitative correlation model and the batch quality threshold are updated.

[0051] The beneficial effects of this invention are as follows: By combining infrared thermography sequences with industrial CT scan data, this invention establishes a quantitative correlation model between thermal response characteristics and defect quantification indicators, achieving a more accurate and comprehensive performance evaluation. Furthermore, the introduction of a dynamic update mechanism enables the method to adaptively adjust the evaluation model, enhancing the consistency and stability of evaluation results across different production batches and overcoming the shortcomings of existing technologies. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein:

[0053] Figure 1 The first embodiment of the present invention provides an overall flowchart of a method for evaluating the performance of damping and heat insulation materials for battery packs;

[0054] Figure 2 This is a module connection diagram of a performance evaluation system for damping and heat insulation materials in a battery pack, provided as a third embodiment of the present invention. Detailed Implementation

[0055] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0056] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for evaluating the performance of damping and heat insulation materials for battery packs is provided.

[0057] S1: Extract a calibration sample set from the production batch corresponding to the material to be evaluated, apply a preset transient thermal excitation to each sample in the calibration sample set and acquire an infrared thermal image sequence, and extract thermal response features based on the infrared thermal image sequence.

[0058] In this embodiment, step S1 is used to generate reproducible thermal imaging response data and thermal response features within the same production batch, which serve as the input basis for subsequent industrial CT true value calibration and quantitative correlation model construction. It should be noted that the production batch can be identified by a batch number or process number assigned by the production system, and its corresponding production conditions should be consistent or traceable, including raw material formulation, curing or foaming conditions, lamination conditions, and thickness tolerance control conditions, to ensure the comparability and statistical representativeness of samples from the same batch.

[0059] Furthermore, when extracting the calibration sample set from the production batch, a combination of stratified sampling and representative sampling can be used to reduce sampling bias. For example, the sample can be stratified according to the first, middle, and last sections of the roll or sheet material, or according to the production time window, and several samples can be randomly selected from each stratum to form the calibration sample set. It should also be noted that when the material is a multi-layered composite structure, the layer sequence direction of the samples should be kept consistent, and the surface orientation, sampling location, and batch identification information of the samples should be recorded to avoid differences in thermal response caused by different layer sequence directions, which could interfere with subsequent calibration. Furthermore, to improve the consistency of infrared thermometry, the sample surface can be cleaned and placed in a stable environment to make the initial temperature of the samples tend to be consistent, thereby reducing the influence of dust, oil, and environmental drift on the infrared thermal imaging sequence.

[0060] In this embodiment, a preset transient thermal excitation is applied to each sample in the calibration sample set. The purpose is to introduce controllable heat input within a short period of time, causing a dynamic process of temperature rise, diffusion, and attenuation in the sample surface temperature field, thereby making the differences in internal structure observable in the infrared thermographic sequence. It should be noted that the transient thermal excitation uses a pulsed heat source, and transient thermal excitation is applied to each sample in the calibration sample set according to uniform excitation parameters. The pulsed heat source can be implemented in a contact or non-contact manner. The contact method can use a pulsed heating pad or heating plate to heat the sample surface in contact, while the non-contact method can use a short-time radiant heat source or pulsed hot air to heat the sample surface. The "preset" refers to the uniform setting of the excitation duration, excitation intensity or target temperature rise range, excitation area, and the relative positional relationship between the heat source and the sample, so as to ensure that the heat input conditions are consistent among different samples. It should also be noted that, in order to avoid artifacts caused by uneven heat input, a clamping structure can be used to ensure the stability of thermal contact in contact heating, or the distance, angle and coverage of the heat source and the sample can be controlled in non-contact heating. Furthermore, in order to reduce the interference of boundary heat dissipation differences on feature extraction, the sample can be fixed on a clamping stage with low thermal conductivity background, and a positioning frame or positioning pin can be used to ensure that the relative positions of the excitation areas are consistent.

[0061] Simultaneously or immediately after applying transient thermal excitation, infrared thermal image sequence data of each sample surface over a preset time series are acquired. It should be noted that the infrared thermal image acquisition device can be a fixedly installed infrared thermal imager or a thermal image acquisition module at the inspection station. Acquisition parameters include acquisition frame rate, acquisition duration, spatial resolution, and field of view. To ensure data reproducibility, a baseline thermal image can be acquired before the start of thermal excitation to characterize the initial temperature field of the sample, and continuous acquisition can be performed during and after the thermal excitation, covering the temperature rise and decay stages to ensure the complete recording of the thermal diffusion process. It should also be noted that the distance, angle, and focal length of the thermal imager relative to the sample should be kept fixed during thermal image acquisition, and parameters such as emissivity settings and background reflection compensation should be uniformly configured. If necessary, calibration can be performed using a standard reference film to reduce radiation temperature measurement errors caused by differences in surface conditions. Furthermore, the acquired infrared thermal image sequences can be named and stored according to batch number, sample number, and acquisition timestamp, and the corresponding excitation method, excitation parameters, and environmental conditions are recorded for subsequent traceability and verification.

[0062] Furthermore, when extracting thermal response features based on the infrared thermal image sequence, the original infrared thermal image sequence can be converted into structured characterization data that can be used for subsequent correlation modeling. It should be noted that the thermal response features may include temporal variation features, spatial distribution features, and spatiotemporal integrated features; wherein temporal variation features are used to characterize the process characteristics of temperature change over time within the area of ​​interest, including the temperature rise amplitude, time to reach peak value, decay rate, and trend; spatial distribution features are used to characterize hotspot area, hotspot distribution uniformity, temperature gradient morphology, and connectivity of abnormal areas; spatiotemporal integrated features are used to characterize the propagation trend of the heat diffusion front, the migration law of hotspot positions over time, and the duration of abnormal areas, thereby improving the sensitivity to internal defects causing heat flow detours or local thermal resistance anomalies. It should also be noted that, to enhance feature stability, the infrared thermal image sequence can be subjected to bad pixel correction, noise filtering, background subtraction, and baseline normalization before feature extraction to reduce the impact of environmental drift and equipment noise on the results; subsequently, the thermal response features corresponding to each calibration sample are stored in the form of a set of fields and associated with the sample number, batch information, and acquisition conditions.

[0063] Through the above steps, the output of step S1 includes: infrared thermal imaging sequence data corresponding to each sample in the calibration sample set, thermal response characteristics and their structured records corresponding to each sample, and metadata such as batch number, sample number, excitation parameters, and acquisition parameters related to the acquisition conditions. It should be noted that the above output provides a one-to-one, traceable, and verifiable data foundation for the subsequent industrial CT truth calibration in step S2 and the establishment of a quantitative correlation model in step S3, thereby ensuring the consistency of the evaluation method of this invention within batches and providing data support for subsequent batch updates and threshold adaptation.

[0064] S2: Perform industrial CT scans on the calibration sample set to obtain defect quantification indicators, and combine the defect quantification indicators with the thermal response characteristics of the corresponding samples to form a calibration dataset.

[0065] In this embodiment, step S2 involves performing an industrial CT scan on the calibration sample set to obtain traceable truth information about internal defects in the calibration samples. This truth information is then correlated one-to-one with the thermal response features obtained in step S1, thereby forming a calibration dataset that can be used for subsequent modeling and threshold determination. It should be noted that this step plays a low-frequency, high-precision calibration role in this invention, and its data accuracy and alignment consistency directly affect the predictive reliability of the subsequent quantitative correlation model and the stability of the batch quality threshold.

[0066] Furthermore, before performing industrial CT scanning on the calibration sample set, the samples can be numbered, verified, and their orientation prepared to ensure a unique correspondence with the thermal image acquisition records in step S1. It should be noted that a sample ledger containing batch number, sample number, sampling location, slice direction, and thermal image acquisition records can be established. Positioning grooves, clamps, or limiting surfaces can be installed on the scanning vehicle to fix the scanning posture of the samples and reduce reconstruction artifact differences caused by sample tilting, warping, or positional changes. It should also be noted that when the damping insulation material is a multi-layered composite structure, the slice direction of the samples should be kept consistent with the records in step S1, and the orientation of the upper surface of the samples should be fixed during scanning to provide a consistent reference system for subsequent analysis of defect distribution in the thickness direction and at interlayer interfaces. In some embodiments, identifiable positioning markers can be set without affecting the material itself to assist in subsequent spatial verification of the consistency between the CT area of ​​interest and the thermally excited area.

[0067] During industrial CT scanning, it should be noted that an appropriate scanning field of view and spatial resolution should be selected based on the material thickness, density, interlayer structure, and target defect scale. Stable X-ray source and detector acquisition parameters should also be set to ensure that the reconstructed volumetric data can distinguish structural details relevant to defect identification. Furthermore, the scan yields three-dimensional structural data containing voxel information, which can reflect structural anomalies such as internal pores, cracks, interface discontinuities, and interlayer separation regions. It should also be noted that to reduce the impact of scattering, hardening effects, and random noise on defect identification, necessary correction and filtering processes can be performed during or after reconstruction. The reconstruction quality should be checked and recorded, for example, to confirm the presence of streak artifacts, blurred boundaries, local saturation, or reconstruction gaps that affect segmentation stability. For samples with abnormal reconstruction quality, rescanning, parameter adjustment, or label removal can be performed to ensure the reliability of the calibration data.

[0068] Furthermore, when obtaining defect quantification indicators based on the aforementioned three-dimensional structural data, defect region identification and segmentation can be performed first, followed by calculation of defect quantification indicators under a unified statistical caliber to ensure comparability between different samples. It should be noted that defect region identification can employ a combination of grayscale threshold segmentation, region growing, and edge and morphological processing to distinguish between the material matrix region, interlayer interface region, and defect region. When the material has multiple interfaces, discontinuous or gap regions within the interface neighborhood can also be identified to characterize potential debonding or delamination defects. Further, the defect quantification indicators may include, but are not limited to, the total defect volume ratio, the degree of defect concentration in the thickness direction, the maximum connectivity scale of defects, the area ratio of defects at key interfaces, and the degree of defect distribution dispersion. The total defect volume ratio reflects the overall degree of porosity or voids, the area ratio at interfaces reflects the severity of debonding or delamination, the maximum connectivity scale reflects the length scale of the continuous thermal resistance barrier that defects may form, and the degree of defect distribution dispersion reflects the spatial characteristics of defect aggregation or dispersion, thereby providing a basis for the quantitative interpretation of subsequent thermal imaging anomaly responses. It should also be noted that, in order to improve the stability and repeatability of the indicators, noise removal and small connected component filtering can be performed on the segmentation results, and consistent statistics can be performed within a preset statistical range to avoid deviations in the indicators caused by edge clipping, local missing data or occasional artifacts.

[0069] After obtaining the defect quantification index for each calibration sample, it should be noted that the defect quantification index should be paired one-to-one with the thermal response features extracted from the same physical sample in step S1 to form a calibration dataset. Furthermore, the calibration dataset can be stored in a structured manner, including sample identification information, thermal response feature fields, defect quantification index fields, and scanning and acquisition condition information, to support the traceability of subsequent model training, cross-validation, and threshold determination. To avoid data mismatch, consistency checks can be performed before pairing, such as verifying whether the sample number, batch number, and acquisition record are consistent, and marking, removing, or retesting missing or abnormal items.

[0070] Through the above steps, the output of step S2 includes: industrial CT three-dimensional structural data corresponding to each calibration sample, defect quantification indicators calculated from the three-dimensional structural data, and a calibration dataset formed by pairing thermal response features with defect quantification indicators. It should be noted that the calibration dataset provides ground truth training samples for the subsequent step S3 to establish a quantitative correlation model from thermal response features to defect quantification indicators, enabling online thermal imaging assessment to be upgraded from qualitative screening to quantitative prediction, and providing an objective basis for determining batch quality thresholds.

[0071] S3: Based on the calibration dataset, establish a quantitative correlation model between thermal response characteristics and defect quantification indicators, and determine the batch quality threshold according to the output distribution of the quantitative correlation model.

[0072] In this embodiment, step S3 is used to establish a quantitative correlation model between thermal response features and defect quantification indicators based on the calibration dataset formed in step S2. Under the premise that the model is reproducible and verifiable, the batch quality threshold is determined according to the output statistical distribution of the quantitative correlation model on the current batch, so that subsequent online detection can output defect quantification prediction values ​​and pass / fail judgments with a unified standard.

[0073] It should be noted that the key to this step is not to limit it to a certain algorithm, but to form an implementable and traceable modeling and threshold determination process, so as to avoid cross-batch misjudgment caused by using a fixed threshold.

[0074] Furthermore, before establishing the quantitative correlation model, consistency processing and quality control can be performed on the calibration dataset. It should be noted that each data point can be checked to ensure it simultaneously possesses both thermal response feature fields and corresponding defect quantification index fields. Entries with significant abnormalities in acquisition conditions or reconstruction quality can be marked, removed, or retested, such as missing frames in infrared thermal imaging sequences, failure to trigger thermal excitation, or unstable defect segmentation due to abnormal CT reconstruction quality. It should also be noted that uniform scaling and denoising processing can be performed on the thermal response features to reduce numerical drift under different acquisition conditions. A consistency check can be performed on the defect quantification index to ensure its statistical range, segmentation rules, and unit caliber remain consistent. Furthermore, redundancy or correlation screening can be performed on the thermal response features to reduce the impact of duplicate features on model stability. In some implementations, features can be grouped and managed according to time-varying, spatial distribution, and spatiotemporal joint categories to facilitate subsequent feature selection, model interpretation, and consistency verification.

[0075] In this embodiment, the quantitative correlation model can be implemented using machine learning algorithms. It should be noted that the calibration dataset can be divided into training data and validation data. The training data is used to learn the mapping relationship between thermal response features and defect quantification indicators, while the validation data is used to test the model's generalization ability on data not used in training, thus avoiding the model only memorizing and fitting calibration samples and failing to be applicable to online test samples. Furthermore, when the defect quantification indicators have multiple dimensions, a unified model can be established using a multi-output approach, or corresponding sub-models can be established for different defect quantification indicators. It should also be noted that extreme defect samples can be boundary-marked and their impact statistically analyzed during the training and validation phases to avoid a small number of extreme samples causing unreasonable stretching of the model output distribution, thereby affecting the stability of the threshold determination.

[0076] To enhance the robustness and generalization ability of the model, the training data needs to include sample data from multiple production batches under different process conditions. For example, during training, it is necessary to ensure that parameters such as raw material formulation, manufacturing process, and thickness tolerance of each sample are representative, ensuring that the model can adapt to differences in material properties under different batches or production conditions. Furthermore, cross-validation should be considered during model training to optimize hyperparameters and avoid overfitting.

[0077] During machine learning model training, feature selection techniques can be used to select the most useful features for predicting defect quantification indicators. For example, by calculating the correlation between each heat response feature and the defect quantification indicator, redundant or low-correlation features can be eliminated to improve model accuracy and training efficiency. Furthermore, the key to model tuning is optimizing hyperparameters such as the number of trees, tree depth, and sample weights to ensure the model can predict with optimal accuracy in real-world applications.

[0078] As an optional implementation, the quantitative correlation model can be established using a random forest regression algorithm. It should be noted that in this implementation, the thermal response characteristics of the calibration samples are used as input, and the corresponding defect quantification index is used as output. A predictive model is formed through ensemble learning of multiple regression decision trees. Furthermore, parameters such as the number of decision trees and the maximum depth can be set, and these parameters can be adaptively adjusted using the prediction deviation between training and validation data to obtain a model that meets the accuracy requirements of internal quality control. It should also be noted that the above optional implementation is only used to illustrate feasible paths for model construction and does not constitute a limitation on the model type.

[0079] After model training is complete, it should be noted that validation data can be used to verify the consistency of the model output. This verification includes checking whether the deviation level between the model's predicted defect quantification values ​​and the true CT values ​​meets internal quality control requirements, and whether the model exhibits systematic shifts under different sampling locations, thickness ranges, or slice order directions. Furthermore, the verification results can be linked to the sample register to identify issues such as sampling condition drift or insufficient sample representativeness, and to determine whether supplementary calibration samples, adjustment of feature extraction methods, or replacement of the model implementation method are necessary. When making improvements, consistent calibration methods should be maintained to ensure traceability.

[0080] When determining the batch quality threshold, it should be noted that the batch quality threshold is not set independently, but is derived from the statistical distribution of the set of defect quantification prediction values ​​output by the quantitative correlation model for calibration samples under the current batch material characteristics. This allows the threshold to adaptively change with slight differences in batch materials and process drift. Specifically, the quantitative correlation model can be applied to the thermal response characteristics of the current batch of calibration samples to obtain the set of defect quantification prediction values ​​for that batch of calibration samples, and the batch quality threshold can be determined based on the statistical distribution of this set of prediction values. Further, as a specific determination method, the predicted value corresponding to the higher quantile of the predicted value set can be calculated as the batch quality threshold; in some embodiments, the predicted value corresponding to 95% or 99% can be selected as the threshold to achieve a balance between release coverage and defect risk control. It should also be noted that, in order to avoid the threshold being dominated by model error and deviating from the true defect level, consistency verification can be performed after the threshold is generated by combining the true value of defect quantification of calibration samples to confirm that the degree of defect corresponding to the threshold meets the company's internal quality control standards. In some implementations, two levels of thresholds, namely warning threshold and rejection threshold, can be set to distinguish between slight deviation and significant abnormality and to provide a trigger basis for subsequent re-inspection strategies.

[0081] Through the above steps, the output of step S3 includes a trained and validated quantitative correlation model, and a batch quality threshold determined by the statistical distribution of the model's output. It should be noted that this output enables subsequent steps S4 and S5 to obtain quantifiable defect prediction results based solely on online thermal imaging features without performing CT scans, and to achieve consistent pass / fail determination using batch-specific thresholds, thereby improving determination reliability and batch consistency while ensuring detection efficiency.

[0082] S4: Apply the transient thermal excitation to the online sample to be tested and acquire an infrared thermal image sequence to extract the thermal response characteristics of the sample to be tested.

[0083] In this embodiment, step S4 involves applying the transient thermal excitation corresponding to step S1 to the sample under test at the online inspection station on the production line, and acquiring the infrared thermographic sequence of the sample to be tested, thereby extracting the thermal response characteristics of the sample under test as input for defect quantification prediction and performance evaluation in step S5. It should be noted that this step is a crucial transition from the offline calibration stage to the online application stage. The key is to ensure that the online acquisition calibrator is consistent with or within a controllable equivalent range of the offline calibration calibrator, so that the thermal response characteristics of the online sample can be reliably interpreted by the quantitative correlation model established in step S3, thereby avoiding systematic misjudgments caused by drift in excitation and acquisition conditions.

[0084] Furthermore, the online test sample can be a sheet, slice, or shaped part flowing during the production line transport process. The online testing station can be set at a fixed position on the main line or a branch testing channel, and is associated with upstream production information through sample number, batch number, or traceability code. It should be noted that, to ensure testing consistency, a positioning and clamping mechanism can be set at the station to allow the test sample to enter the preset testing area and maintain a stable posture. For example, positioning guards, clamping structures, or vacuum adsorption stages can be used to reduce the impact of warping, shaking, and displacement on thermal image acquisition. When the material is a multi-layer composite structure, it should also be noted that the upper surface direction of the test sample should be consistent with that of the calibration sample in step S1 to avoid the systematic offset of thermal response characteristics introduced by the difference in layer sequence direction.

[0085] When applying the transient thermal excitation to the online sample under test, it should be noted that the thermal excitation method and key parameters should be consistent with step S1 or kept within a preset tolerance range to ensure the comparability of thermal input. Furthermore, the transient thermal excitation can be implemented using contact or non-contact methods. Contact methods can use pulse heating elements or heating plates to briefly heat the surface of the sample under test, while non-contact methods can use short-term radiant heat sources or pulsed hot air to heat the surface of the sample under test. It should also be noted that, to adapt to the production line cycle time, the thermal excitation duration, excitation area, and target temperature rise range can be set to meet the requirements of rapid acquisition and attenuation observation. The heat source output can be controlled in a closed-loop or semi-closed-loop manner by a controller to reduce thermal input fluctuations caused by heat source aging, changes in ambient temperature, or changes in sample surface condition. Furthermore, output monitoring or reference temperature monitoring can be set near the heat source to verify the effectiveness of each thermal excitation. When the thermal input does not meet the preset tolerance or triggering conditions, the sample can be retested, marked as abnormal, or a re-inspection can be triggered.

[0086] When acquiring infrared thermal image sequences, it should be noted that the infrared thermal image acquisition device can be a fixedly installed infrared thermal imager or an integrated thermal image acquisition module. Its acquisition frame rate, acquisition duration, spatial resolution, and field of view should cover the temperature rise and decay process of the sample under test to ensure that the thermal diffusion behavior is completely recorded. Furthermore, a baseline thermal image can be acquired before the thermal excitation begins to characterize the initial temperature field of the sample under test, and continuous acquisition can be performed during and after the thermal excitation to form a chronologically arranged infrared thermal image sequence. It should also be noted that to ensure the synchronization of the thermal excitation event and image acquisition, the start-up time of the heat source and the thermal image acquisition device can be coordinated through hardware trigger signals or a timing controller, while maintaining a fixed distance, angle, and focal length of the thermal imager relative to the sample, and using uniform emissivity settings and background reflection compensation parameters. Furthermore, in some implementations, a standard reference sheet or reference area can be set up for online calibration or drift monitoring to reduce the impact of different surface conditions, background radiation changes and environmental disturbances on the temperature measurement results; a local shading or heat insulation shielding structure can also be set up to reduce the impact of on-site thermal radiation interference on the stability of infrared thermal image sequences.

[0087] Furthermore, when extracting thermal response features based on the infrared thermal image sequence, it should be noted that the same preprocessing and feature extraction methods as in step S1 can be used to transform the original infrared thermal image sequence into a structured set of feature fields. Specifically, bad pixel correction, noise filtering, background subtraction, and baseline normalization can be performed on the infrared thermal image sequence to reduce the impact of equipment noise and environmental drift. Subsequently, time-varying features, spatial distribution features, and spatiotemporal comprehensive features are extracted from the infrared thermal image sequence to characterize the temperature rise, diffusion, and attenuation behavior of the sample under test under thermal excitation. It should also be noted that the online feature extraction should maintain consistency with the feature field names, calculation methods, and statistical ranges in the calibration stage, and the extraction results should be bound and stored with the sample number, batch information, and acquisition conditions for subsequent traceability and verification. When the online acquisition conditions deviate significantly or the feature distribution shows abnormal drift, the sample under test can be marked as a sample requiring re-inspection, and the relevant information can be used as the data basis for the triggering conditions of step S6.

[0088] Through the above steps, the output of step S4 includes: the infrared thermographic sequence of the sample under test and the thermal response features extracted from the infrared thermographic sequence. It should be noted that the aforementioned thermal response features serve as the input to step S5, enabling the sample under test to achieve defect quantification prediction and performance evaluation without industrial CT scanning, thereby meeting the quality control requirements of high-throughput, traceability, and consistency determination for online production line inspection.

[0089] S5: Input the thermal response characteristics of the sample to be tested into the quantitative correlation model to obtain the defect quantification prediction value, compare the defect quantification prediction value with the batch quality threshold, and output the performance evaluation result of the sample to be tested.

[0090] In this embodiment, step S5 involves inputting the thermal response characteristics of the sample to be tested obtained in step S4 into the quantitative correlation model established in step S3 to obtain a defect quantification prediction value consistent with the defect quantification index described in step S2. The defect quantification prediction value is then compared with the batch quality threshold determined in step S3, and the performance evaluation result of the sample to be tested is output. It should be noted that this step achieves online conversion from thermal imaging response to defect quantification index and completes traceable quality decisions using a model-threshold integrated judgment mechanism, thereby supporting high-speed inspection and consistent release on the production line.

[0091] Furthermore, before inputting the thermal response features into the quantitative correlation model, the thermal response features of the sample to be tested can undergo input normalization processing consistent with the modeling stage to ensure consistent model input standards. It should be noted that the input normalization processing may include feature field integrity verification, outlier checking, and scaling according to the scaling rules determined in the modeling stage. When feature loss, abnormal acquisition, or exceeding a preset reasonable range is detected, the sample to be tested can be marked as requiring retesting or re-examination, and the triggering reason can be recorded to avoid abnormal input leading to model output distortion. It should also be noted that the thermal response features should be bound together with the batch number, sample number, and acquisition conditions of the sample to be tested to ensure traceability of the model output results.

[0092] After inputting the quantitative correlation model, it should be noted that the model output is the defect quantification prediction value of the sample to be tested. When the defect quantification index has multiple dimensions, the defect quantification prediction value can be a multi-dimensional prediction result, corresponding to different defect characterization dimensions, and can form a comprehensive evaluation caliber according to a preset risk priority to match internal quality control needs. Furthermore, in some embodiments, the model can also output prediction confidence or uncertainty indication information to characterize the reliability of the prediction result and whether the sample to be tested falls within the data distribution range covered by the model. As an optional implementation, the confidence or uncertainty indication information can be generated based on the dispersion of multiple prediction results of the integrated model, and a re-inspection process or increased sampling calibration frequency can be triggered when the confidence is lower than a preset threshold to reduce the risk of uncontrollable misjudgment caused by samples outside the distribution.

[0093] When comparing the defect quantification prediction value with the batch quality threshold, it should be noted that the batch quality threshold is not set independently, but is derived from the statistical distribution of the output of the current batch calibration samples by the quantitative correlation model, used to characterize the control boundary of the normal fluctuation range of the batch. Furthermore, the comparison judgment can be executed according to a preset judgment logic: in an implementation using a single threshold, when the defect quantification prediction value is not higher than the batch quality threshold, the output is "qualified"; when the defect quantification prediction value is higher than the batch quality threshold, the output is "unqualified" or requires re-inspection. In an implementation using two thresholds, a warning threshold and a rejection threshold can be set. When the defect quantification prediction value is not higher than the warning threshold, the output is "qualified"; when the defect quantification prediction value is higher than the warning threshold but not higher than the rejection threshold, the output is "warning" or requires re-inspection; when the defect quantification prediction value is higher than the rejection threshold, the output is "unqualified". It should also be noted that the judgment logic can be pre-configured and adjusted according to internal risk strategies, but its core lies in achieving consistent judgment based on the current batch-specific threshold, rather than relying on a fixed threshold.

[0094] Furthermore, the system generates and outputs structured performance evaluation results based on the above comparison results. It should be noted that the performance evaluation results include the sample identification information, defect quantification prediction value, the batch quality threshold used, the comparison judgment conclusion, and a timestamp. In some embodiments, prediction confidence level indication information, trigger reason records, model version number, and threshold version number may also be added to support subsequent auditing, review, and traceability. It should also be noted that the performance evaluation results can be output to downstream systems in real time, in forms including but not limited to displaying prompts on the human-machine interface, outputting sorting instructions to the sorting execution mechanism, writing the results to the quality database, and generating batch statistical reports. For samples judged as unqualified or subject to warnings, alarm prompts can be triggered simultaneously, and the samples can be guided to the re-inspection area or isolation area to reduce the risk of unqualified products being mixed in.

[0095] Through the above steps, the output of step S5 includes the quantified predicted value of defects in the sample to be tested and its corresponding performance evaluation results. It should be noted that the above output is not only used for the current online judgment, but also forms a continuously accumulating sequence of predicted values ​​and judgment results. This data, along with subsequently added true value calibration data, can be used to evaluate the long-term stability of the model and the applicability of the threshold, thus providing data support for updating the quantitative correlation model and batch quality threshold in step S6, thereby forming a continuously optimized intelligent quality control closed loop.

[0096] S6: When the preset period or trigger condition is met, new calibration samples are extracted and S1 and S2 are executed to obtain new calibration data. The quantitative correlation model and the batch quality threshold are updated based on the new calibration data.

[0097] In this embodiment, step S6 is a crucial step for the system's adaptive optimization and continuous learning. By periodically or based on specific triggering conditions, new calibration samples are extracted from the production batch, and the data from these new samples are merged with existing calibration data to update the quantitative correlation model and batch quality threshold. This ensures that the model always adapts to changes in material properties during production, maintaining the accuracy and stability of the detection. It should be noted that this process enables the system to continuously optimize through online data feedback, ensuring efficient and high-precision online quality assessment throughout different production cycles.

[0098] Furthermore, the preset cycle or triggering condition serves as the decision-making basis for initiating the update process. It should be noted that the preset cycle can be set based on production batch size, time span, or cumulative inspection quantity, such as "every 5000 products produced" or "every week of continuous operation." The triggering condition is dynamically generated based on the real-time quality data stream monitored during online inspection, and includes: a) the defect quantification prediction values ​​of several consecutive samples show a monotonically increasing trend or exceed the control limits of the control chart; b) the proportion of samples whose prediction confidence level output by the model is consistently lower than a preset threshold exceeds a certain threshold; c) CT sampling and re-inspection of samples judged as "warning" or in a borderline state online reveals a systematic deviation between their true values ​​and the model prediction values. It should also be noted that when any update condition is met, the system can automatically send an update prompt to the quality management personnel, who can then determine whether the triggering condition is met.

[0099] After triggering the update, new calibration samples need to be extracted from the current production batch or the immediately following batch. It should be noted that the sampling principle should be consistent with step S1, taking into account spatiotemporal representativeness. The newly extracted calibration samples are treated as a new set of "unknown samples," and steps S1 (applying the same transient thermal excitation and extracting thermal response characteristics) and S2 (performing industrial CT scans and obtaining the true value of defect quantification indicators) are strictly repeated. This yields a new set of paired "thermal response characteristics - defect quantification indicators" data that are homologous to the historical calibration data; this is the newly added calibration data.

[0100] Furthermore, the quantitative correlation model or batch quality threshold is updated based on the newly calibrated data. It should be noted that the update strategy can be selected based on the amount of data accumulated, the degree of performance degradation, and production line stability requirements, mainly including the following two modes:

[0101] Coordinated updating of model parameters and thresholds: New calibration data is merged with historical calibration datasets to form a new, larger training dataset. The quantitative correlation model is retrained using this dataset, and the batch quality threshold is redefined based on the output distribution of the new model across all merged calibration samples (or current batch samples). This mode is equivalent to performing a complete model retraining and is suitable for scenarios where there is sufficient data accumulation or significant changes in the process.

[0102] Independent threshold calibration and incremental model fine-tuning: Batch quality thresholds are primarily recalibrated based on newly added calibration data. For example, the thermal response characteristics of new samples are input into the existing model to obtain predicted values, which are then combined with their true CT values ​​to validate and adjust the original thresholds. Simultaneously, incremental learning algorithms can be used to fine-tune the parameters of the original model using only new data, without performing full retraining. This mode is computationally efficient and suitable for scenarios with stable production conditions requiring only minor adjustments.

[0103] It should also be noted that any update operation must be validated in a controlled "offline" or "simulated" environment before being deployed to the online system. Specifically, the updated model and thresholds must be validated using historical data or reserved test sets to ensure that their prediction accuracy is not lower than the level before the update and that no new systematic errors have been introduced. After the update is completed, the new model and thresholds must be versioned, recording the update date, triggering reason, data used, and performance validation report.

[0104] Through the steps described above, step S6 enables the self-iteration and evolution of the evaluation system. Its output is a new version of the updated quantitative correlation model or batch quality threshold. It should be noted that the system will smoothly switch to using the new version for online evaluation, thereby ensuring that the evaluation method can continuously adapt to the dynamic changes in the production line, maintain high accuracy and high reliability in quantitative detection capabilities over the long term, and ultimately form a complete, adaptive intelligent quality control lifecycle from "offline calibration" to "online detection" and then to "closed-loop update".

[0105] Example 2 is the second embodiment of the present invention, which differs from the previous embodiment in that:

[0106] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art or the current technical solution, can be embodied in the form of a software product. This current computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0107] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0108] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0109] Example 3, referring to Figure 2 As an embodiment of the present invention, a performance evaluation system for damping and heat insulation materials for battery packs is provided, which includes a sample extraction and thermal excitation module, an industrial CT scanning and calibration module, a model building and threshold determination module, an online detection and thermal response acquisition module, a performance evaluation and prediction module, and a calibration data update module.

[0110] Sample extraction and thermal excitation module: Extracts a calibration sample set from the production batch corresponding to the material to be evaluated, applies a preset transient thermal excitation to each sample in the calibration sample set and acquires an infrared thermal image sequence, and extracts thermal response features based on the infrared thermal image sequence;

[0111] Industrial CT Scanning and Calibration Module: Performs industrial CT scanning on the calibration sample set to obtain defect quantification indicators, and combines the defect quantification indicators with the thermal response characteristics of the corresponding samples to form a calibration dataset;

[0112] Model building and threshold determination module: Based on the calibration dataset, a quantitative correlation model is established from thermal response characteristics to defect quantification indicators, and the batch quality threshold is determined according to the output distribution of the quantitative correlation model;

[0113] Online detection and thermal response acquisition module: Apply the transient thermal excitation to the online sample to be tested and acquire infrared thermal image sequences to extract the thermal response characteristics of the sample to be tested;

[0114] Performance evaluation and prediction module: Input the thermal response characteristics of the sample to be tested into the quantitative correlation model to obtain the defect quantification prediction value, compare the defect quantification prediction value with the batch quality threshold, and output the performance evaluation result of the sample to be tested;

[0115] Calibration data update module: When a preset period or trigger condition is met, new calibration samples are extracted and S1 and S2 are executed to obtain new calibration data. Based on the new calibration data, the quantitative correlation model and the batch quality threshold are updated.

[0116] Example 4 is an embodiment of the present invention, which provides a method for evaluating the performance of damping and heat insulation materials for battery packs. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation / comparative experiments.

[0117] This embodiment specifically demonstrates how the method described in this invention is applied to the quality control of a mass production line for an aerogel composite damping heat insulation pad used in a new energy vehicle battery pack. This material has a three-layer composite structure (ceramic fiber felt / aerogel layer / silicone rubber foam), and its internal debonding and pore uniformity are key factors affecting its heat insulation and buffering performance. Traditional methods rely on random CT scans, which are inefficient and cannot perform 100% inspection, while offline thermal imaging screening with fixed thresholds has a high misjudgment rate due to batch variations.

[0118] First, 30 pads were randomly selected from "Batch A" produced that day as the initial calibration sample set (S1) using stratified random sampling. Each sample was placed in a constant temperature (23±1°C) and constant humidity (50% RH) laboratory environment for 2 hours to stabilize its temperature. Subsequently, the sample surface was subjected to transient thermal excitation using a flash lamp array with a pulse width of 5 ms and an energy density of 1.8 J / cm². A mid-wave infrared thermal imager (frame rate 100 Hz, resolution 640×512) simultaneously acquired an infrared thermal image sequence within 5 seconds after excitation. Three key thermal response features were extracted from the sequence: F1 (cooling time constant τ, unit: seconds), characterizing the overall thermal insulation performance of the material; F2 (thermal diffusion uniformity index EU, dimensionless), reflecting the uniformity of the internal structure; and F3 (the percentage of the area of ​​the high-temperature zone above the average temperature rise by 20% in the first second after excitation, unit: %), which is sensitive to local debonding. All features were preprocessed with non-uniformity correction, temporal filtering, and background subtraction before extraction.

[0119] Subsequently, the 30 calibration samples were subjected to microfocus industrial CT scanning (15 μm resolution) to obtain their internal three-dimensional structure (S2). Using image segmentation algorithms, two core defect quantification indicators were quantitatively calculated: D1 (volume porosity, %), characterizing the overall material density; and D2 (interlayer debonding area ratio, %), specifically used to evaluate the bonding quality of the composite layer. The [F1, F2, F3] feature vectors of each sample were mapped one-to-one with the [D1, D2] ground truth indicators to form an initial calibration dataset containing 30 records.

[0120] Based on this dataset, a quantitative correlation model (S3) was established. The data was randomly divided into a training set (21 samples) and a validation set (9 samples) at a ratio of 70% / 30%. A random forest regression algorithm was used, with three thermal response features as input, to predict two defect indicators. After training, the model's R² for predicting porosity (D1) on the validation set was 0.96, and its R² for predicting debonding rate (D2) was 0.93, indicating good prediction accuracy. Subsequently, the trained model was applied to the feature data of all 30 calibration samples to obtain a set of predicted values. The 95th percentile of the predicted D2 (debonding rate) values ​​in this prediction set (i.e., 95% of the sample predictions were below this value) was taken as the online detection quality threshold Th_A for "batch A," which, after conversion, corresponds to a debonding area rate of approximately 1.72%.

[0121] At the online inspection station on the production line, 150 samples from "Batch A" produced consecutively were evaluated (S4 and S5). The online station replicated the excitation and acquisition parameters from the laboratory. The thermal image sequence of each sample was processed in real time, and the [F1, F2, F3] features were extracted and input into the above model to obtain its defect quantification prediction value. The predicted debonding rate was compared with the threshold Th_A (1.72%). The first 100 samples showed stable detection, but the system found that the predicted debonding rate of samples 101-105 showed a monotonically increasing trend (from 1.65% to 1.80%) for 5 consecutive samples, and the predicted value of sample 105 (1.80%) exceeded Th_A and was judged as "unqualified"; at the same time, samples 103 and 104 were judged as "warning". This triggered the preset update condition of "multiple consecutive sample predicted values ​​showing a monotonically increasing trend and exceeding the limit" (S6).

[0122] Based on the trigger, five new samples were urgently extracted from products 106-110 as additional calibration samples. These five samples underwent rigorous S1 and S2 procedures (i.e., the same thermal imaging and CT scans) to obtain new calibration data. The new data was merged with the original 30 data points, and the random forest model was fine-tuned using incremental learning. The threshold was recalculated based on the new model's predicted distribution across all 35 samples, resulting in an updated threshold Th_A' (95th percentile, corresponding to a debonding rate of 1.69%). The updated model and threshold were deployed back to the online system for subsequent testing. CT verification confirmed that the actual debonding rate of the 105th "non-conforming" sample was 2.01%, confirming the defect. Subsequent testing using the new threshold did not result in any further abnormal alarms, and production returned to stability.

[0123] The key data recorded is shown in Table 1:

[0124] Table 1: Experimental Data Recording Table

[0125]

[0126] First, the data demonstrates the high-precision predictive ability of the established quantitative correlation model. In the nine samples of the validation set (such as A-03, A-05, and A-08 in the table), the model-predicted debonding area ratio closely matches the true value measured by industrial CT. For example, the true CT value for sample A-05 is 1.87%, while the model-predicted value is 1.76%, with an absolute error of only 0.11 percentage points. Linear regression analysis of the predicted and true values ​​for all calibration samples yields a coefficient of determination (R²) exceeding 0.97. This overcomes the limitations of traditional thermal imaging inspection, which can only perform qualitative or rough grading, achieving a fundamental leap from "whether there is an anomaly" to "how much defect is present." This quantitative predictive capability enables the digitization and refinement of quality control standards, providing OEMs with clear material performance data, rather than just vague conclusions of "pass / fail."

[0127] Secondly, this embodiment highlights the scientific basis of batch-adaptive thresholds, effectively solving the problem of cross-batch misjudgment. Traditional methods, if using a fixed threshold (e.g., setting a debonding rate of 1.5% as unacceptable based on historical experience), would cause numerous misjudgments of "Batch A." ​​This is because calibration data shows that the overall material performance of this batch is good, with most samples having true values ​​below 1.0%, but its normal fluctuation upper limit (95th percentile) is around 1.72%. If a fixed threshold of 1.5% is used, while truly unacceptable samples such as A-08 (true value 2.10%) and A-05 (true value 1.87%) can be detected, acceptable samples such as A-03 (true value 1.23%), which fall within the normal fluctuation range of the batch, would also be "falsely rejected." Based on the threshold Th_A (1.72%) determined by the data distribution of the current batch, this invention successfully distinguishes between "common batch fluctuations" and "excessive actual defects". While ensuring the detection of actual non-conforming products (A-05, A-08), it minimizes the false positive rate of qualified products, directly improving production yield and economic benefits.

[0128] Furthermore, the dynamic triggering and adaptive update mechanism demonstrates the system's intelligence and reliability. During the online detection phase, for samples P-101 to P-105, the model's predicted values ​​showed a clear monotonically increasing trend (1.65% → 1.80%), eventually exceeding the threshold. This triggered the system's early warning and update process. CT re-examination of P-105 (judged as unqualified) and the newly added sample N-01 confirmed the existence of the actual defect (N-01 true value 2.01%), verifying the triggering condition's keen perception of potential quality slippage. Subsequently, the model and threshold were updated using the new data (N-01, N-02), with the threshold fine-tuned from Th_A (1.72%) to Th_A' (1.69%). This adjustment responded to slight quality fluctuations during production, making the control line more precise. After the update, the tests for P-106 and P-107 returned to "qualified" and the results were stable, indicating that the system quickly adapted to the new production state through self-learning, avoiding subsequent continuous misjudgments or omissions caused by a single occasional anomaly or slow drift. This closed-loop capability of "monitoring-diagnosis-optimization" is completely lacking in static inspection systems, and it significantly improves the foresight and long-term stability of quality control.

[0129] In summary, this embodiment demonstrates through specific data and a complete process that the present invention deeply integrates offline, destructive, and high-cost CT detection ground truth with online, non-destructive, and low-cost thermal imaging detection. Through machine learning models and adaptive frameworks, it derives a real-time, quantitative, and self-evolving intelligent quality assessment system, effectively resolving the core contradiction between the extremely high requirements for material consistency in high-end manufacturing and the prohibitive cost of full inspection.

[0130] 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 it. 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 spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for evaluating the performance of damping and heat-insulating materials for battery packs, characterized in that, include: S1: Extract a calibration sample set from the production batch corresponding to the material to be evaluated, apply a preset transient thermal excitation to each sample in the calibration sample set and collect an infrared thermal image sequence, and extract thermal response features based on the infrared thermal image sequence. S2: Perform industrial CT scanning on the calibration sample set to obtain defect quantification indicators, and combine the defect quantification indicators with the thermal response characteristics of the corresponding samples to form a calibration dataset; S3: Establish a quantitative correlation model between thermal response characteristics and defect quantification indicators based on the calibration dataset, and determine the batch quality threshold according to the output distribution of the quantitative correlation model; S4: Apply the transient thermal excitation to the online sample to be tested and acquire an infrared thermal image sequence to extract the thermal response characteristics of the sample to be tested; S5: Input the thermal response characteristics of the sample to be tested into the quantitative correlation model to obtain the defect quantification prediction value, compare the defect quantification prediction value with the batch quality threshold, and output the performance evaluation result of the sample to be tested; S6: When the preset period or triggering condition is met, new calibration samples are extracted and S1 and S2 are executed to obtain new calibration data. The quantitative correlation model and the batch quality threshold are updated based on the new calibration data. The process of determining the batch quality threshold based on the output distribution of the quantitative correlation model includes: The thermal response characteristics of the calibration sample set are input into the quantitative correlation model to obtain the corresponding set of defect quantification prediction values. Calculate a specific statistical quantile of the predicted value set, and determine the predicted value corresponding to the quantile as the batch quality threshold; The step of updating the quantitative correlation model and the batch quality threshold based on the newly added calibration data includes: When a preset cycle or triggering condition is met, new calibration samples are extracted from the current and subsequent production batches; wherein, the triggering condition includes abnormal conditions dynamically determined based on online detection data, and the abnormal conditions include: the defect quantification prediction values ​​of multiple consecutive test samples show a unidirectional changing trend and exceed the control limit, the proportion of samples with prediction confidence of model output below the threshold is continuously abnormal, and after CT sampling of samples judged to be in an edge state online, it is found that the deviation between the predicted value and the true value exceeds the tolerance. Steps S1 and S2 are performed on the new calibration sample to obtain new calibration data, which is formed by pairing thermal response characteristics with corresponding defect quantification indicators. The parameters of the quantitative correlation model and the batch quality threshold are updated using the newly added calibration data.

2. The method for evaluating the performance of damping and heat-insulating materials for battery packs as described in claim 1, characterized in that, In step S1, applying a preset transient thermal excitation to each sample in the calibration sample set and acquiring an infrared thermographic sequence includes: Stratified sampling is used to draw samples from production batches to form a statistically representative calibration sample set; A pulsed heat source was used to apply transient thermal excitation to each sample in the calibration sample set according to uniform excitation parameters; While applying the transient thermal excitation, an infrared thermal imager is used to synchronously acquire infrared thermal image sequences of each sample surface with preset acquisition parameters, and a baseline thermal image is acquired before acquisition for background subtraction.

3. The method for evaluating the performance of damping and heat-insulating materials for battery packs as described in claim 2, characterized in that, In step S1, the extraction of thermal response features based on the infrared thermal image sequence includes: The infrared thermal image sequence is preprocessed, including bad pixel correction, noise reduction filtering, background subtraction, and baseline normalization. From the preprocessed infrared thermal image sequence, temporal variation features, spatial distribution features, and spatiotemporal integrated features are extracted to form thermal response features.

4. The method for evaluating the performance of damping and heat-insulating materials for battery packs as described in claim 3, characterized in that, In step S2, the step of assembling the calibration dataset by combining the defect quantification index with the thermal response characteristics of the corresponding sample includes: Industrial CT scans were performed on the calibration sample set to obtain the internal three-dimensional structural data of each sample. Based on the three-dimensional structural data, defect quantification indicators are identified and calculated. These defect quantification indicators include the debonding area ratio, the delamination gap thickness, and the porosity, which characterizes the thermal resistance anomaly at the interface and the thermal conductivity heterogeneity of the bulk material. The defect quantification index of each sample obtained by calculation is paired one-to-one with the thermal response characteristics of the same physical sample obtained in step S1 to form a calibration dataset.

5. The method for evaluating the performance of damping and heat-insulating materials for battery packs as described in claim 4, characterized in that, In step S3, establishing a quantitative correlation model from thermal response features to defect quantification indicators based on the calibration dataset includes: The calibration dataset is divided into a training subset and a validation subset; The quantitative correlation model is obtained by training the thermal response features in the training subset as input and the corresponding defect quantification index as output through a machine learning regression algorithm. The trained model is validated using the validation subset to evaluate its prediction accuracy and generalization ability.

6. The method for evaluating the performance of damping and heat-insulating materials for battery packs as described in claim 5, characterized in that, In step S4, applying the transient thermal excitation to the online sample to be tested and acquiring the infrared thermographic sequence includes: At the online inspection station on the production line, the transient thermal excitation is applied to the online test sample using the same excitation parameters as in step S1. Using the same acquisition parameters as in step S1, the infrared thermal image sequence of the online sample to be tested is acquired synchronously. Using the same feature extraction methods and procedures as in step S1, the thermal response features of the online sample to be tested are extracted from the infrared thermal imaging sequence.

7. The method for evaluating the performance of damping and heat-insulating materials for battery packs as described in claim 6, characterized in that, Step S5 specifically includes: The thermal response characteristics of the sample to be tested are input into the quantitative correlation model, and the defect quantification prediction value of the sample is output. The defect quantification prediction value is compared with the batch quality threshold, and the classification judgment result is obtained according to the preset judgment logic; Generate and output a structured evaluation result that includes the sample identifier, the defect quantification prediction value, the batch quality threshold, and the classification judgment result.

8. A performance evaluation system for damping and heat insulation materials for battery packs, used to implement the performance evaluation method for damping and heat insulation materials for battery packs as described in any one of claims 1 to 7, characterized in that, include: Sample extraction and thermal excitation module: Extracts a calibration sample set from the production batch corresponding to the material to be evaluated, applies a preset transient thermal excitation to each sample in the calibration sample set and acquires an infrared thermal image sequence, and extracts thermal response features based on the infrared thermal image sequence; Industrial CT Scanning and Calibration Module: Performs industrial CT scanning on the calibration sample set to obtain defect quantification indicators, and combines the defect quantification indicators with the thermal response characteristics of the corresponding samples to form a calibration dataset; Model building and threshold determination module: Based on the calibration dataset, a quantitative correlation model is established from thermal response characteristics to defect quantification indicators, and the batch quality threshold is determined according to the output distribution of the quantitative correlation model; Online detection and thermal response acquisition module: applies the transient thermal excitation to the online sample to be tested and acquires infrared thermal image sequences, and extracts the thermal response characteristics of the sample to be tested; Performance evaluation and prediction module: Input the thermal response characteristics of the sample to be tested into the quantitative correlation model to obtain the defect quantification prediction value, compare the defect quantification prediction value with the batch quality threshold, and output the performance evaluation result of the sample to be tested; Calibration data update module: When a preset period or trigger condition is met, new calibration samples are extracted and S1 and S2 are executed to obtain new calibration data. Based on the new calibration data, the quantitative correlation model and the batch quality threshold are updated.

Images