A multi-material power equipment defect quantitative detection method, system and device based on infrared thermal imaging and a generation model and a storage medium

Abstract

The application discloses a kind of multi-material power equipment defect quantification detection method, system, equipment and storage medium based on infrared thermal imaging and generation model, it is related to power equipment nondestructive testing technical field, method includes: based on the establishment of finite element model of power equipment multi-material thermophysical parameter, simulation output thermal map and temperature curve are combined with defect parameter;Make test piece and experimentally collect real thermal map;Fusion simulation and experimental data construct standardized dataset;Train improved variational autoencoder to generate expanded defect samples;The application realizes the unified modeling of multi-material defect thermal response, can significantly improve the detection precision and generalization ability under small sample, generates image edge clear, texture real, and engineering applicability is strong.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing technology for power equipment, and in particular to a method, system, equipment, and storage medium for quantitative detection of defects in multi-material power equipment based on infrared thermal imaging and generative models. Background Technology

[0002] Currently, infrared thermal imaging nondestructive testing technology has been widely applied in fields such as power equipment condition monitoring, composite material defect identification, and industrial structural health diagnosis. Traditional infrared detection methods mainly identify local thermal anomaly areas by collecting data on the surface temperature field changes of the object under test, thereby achieving indirect detection of internal defects.

[0003] However, as power equipment develops towards larger scale, more composite materials, and higher reliability, more and more carbon fiber composite materials, epoxy resin insulation materials, and multi-material composite structures are being used in key equipment components. Traditional detection methods are gradually revealing the following shortcomings: Existing infrared detection methods are usually based on the assumption of a single material's thermal response to establish analytical models, which are difficult to apply to multi-material coupled structure scenarios. Different materials have significant differences in thermophysical parameters such as thermal conductivity, thermal diffusivity, specific heat capacity, and surface emissivity. Under the same thermal excitation conditions, the heat propagation laws within different materials are significantly different, resulting in inconsistent surface temperature response characteristics in defect regions. Traditional methods based on fixed thresholds or single thermal response models cannot effectively adapt to the complex thermal diffusion behavior in multi-material structures, and are prone to problems such as missed defects, false detections, and insufficient detection sensitivity. Traditional infrared thermal imaging detection methods lack the ability to quantitatively analyze the thermal response mechanisms of different materials. Existing research mostly relies on empirical experimental analysis and lacks systematic theoretical modeling of the defect thermal response formation mechanism, especially lacking research on the coupling relationship between defect size, defect depth, material thermal diffusivity, and the thermal conductivity of heterostructures. Therefore, it is difficult to accurately predict the thermal response of defects under complex operating conditions, resulting in a lack of theoretical basis for optimizing detection parameters; existing technologies lack a collaborative integration mechanism of simulation, experiment, and intelligent models. Most current research focuses only on single algorithm optimization, lacking a complete technical system from heat conduction mechanism analysis, finite element simulation, experimental verification to intelligent generative model construction, thus failing to form a unified solution applicable to defect detection in multi-material power equipment. Therefore, there is an urgent need to propose a data augmentation method that integrates multi-material thermal response mechanism analysis, finite element thermal simulation, infrared experimental verification, and improved generative models to achieve high-precision intelligent quantitative detection of defects in complex power equipment; existing infrared defect intelligent identification models heavily rely on a large number of labeled samples. Internal defects in power equipment are low-frequency events, and obtaining real defect samples is difficult, especially in multi-material scenarios where the data distribution of different defect types is extremely uneven. Traditional convolutional neural networks or classification models are prone to overfitting under small sample conditions, exhibiting poor model generalization ability and failing to meet the high-reliability detection requirements of complex environments in actual engineering; traditional generative models generate images of limited quality. While existing data augmentation methods based on generative adversarial networks or traditional variational autoencoders can expand the training data to some extent, the generated results generally suffer from problems such as blurred textures, missing edge details, and insufficient structural consistency. In particular, in infrared thermal images, the subtle temperature gradient information in the defect edge region is easily lost. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention provides a method, system, device and storage medium for quantitative detection of defects in multi-material power equipment based on infrared thermal imaging and generative models.

[0005] Therefore, the technical problem solved by the present invention is that existing infrared thermal imaging detection methods cannot adapt to the differences in thermal response of power equipment with multiple materials, and the model has insufficient generalization ability and lacks details in the generated images under small sample conditions.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for quantitative detection of defects in multi-material power equipment based on infrared thermal imaging and a generative model, comprising: Based on the thermophysical parameters of the multi-material structure of power equipment, a multi-material transient heat conduction finite element model is constructed, which incorporates the specimen geometry, material distribution and thermal excitation conditions. Based on the specimen geometry, material distribution and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, and combined with the set defect parameters, the numerical simulation outputs the simulated infrared thermogram sequence and temperature change curve under various defect conditions as a function of time. Based on the specimen geometry, material distribution and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, actual specimens were fabricated and an active infrared thermal imaging experimental platform was built to output real multi-material defect infrared thermal image sequences and temperature change data. The simulated infrared thermal image sequence is fused and preprocessed with the real multi-material defect infrared thermal image sequence to output a standardized multi-material defect infrared dataset. Based on a standardized multi-material defect infrared dataset, an improved variational autoencoder that integrates adaptive latent space optimization and perceptual loss constraints is trained to output expanded defect infrared image samples.

[0007] As a preferred scheme for a multi-material power equipment defect quantification detection method based on infrared thermal imaging and generative models, wherein: The thermophysical parameters of the multi-material structure based on power equipment are used to construct a multi-material transient heat conduction finite element model, which embeds the specimen geometry, material distribution, and thermal excitation conditions, including: Based on the thermophysical parameters of carbon fiber composite materials, epoxy resin insulation materials and heterogeneous inlay materials in power equipment, a three-dimensional transient heat conduction finite element model was established, which includes carbon fiber composite matrix, epoxy resin matrix, air pore defects, corroded metal inlay defects and heterogeneous material inlay defects. The geometric dimensions, material distribution and thermal excitation conditions of the specimen were embedded in the model.

[0008] As a preferred scheme for a multi-material power equipment defect quantification detection method based on infrared thermal imaging and generative models, wherein: The thermophysical parameters of the multi-material structure based on power equipment are used to construct a multi-material transient heat conduction finite element model, which embeds the specimen geometry, material distribution, and thermal excitation conditions. The model also includes: The transient heat conduction equation is used to describe the heat diffusion process; The model was discretized using free tetrahedral meshes, with local refinement applied to defect regions and material interfaces. An active pulsed thermal excitation method was adopted, and the heat flux density and duration uniformly applied to the surface of the specimen were set. The initial ambient temperature and convective heat transfer coefficient were also set. Natural convective heat transfer and radiative heat transfer on the model surface were considered, and a radiation model was used to describe the surface radiation process.

[0009] As a preferred scheme for a multi-material power equipment defect quantification detection method based on infrared thermal imaging and generative models, wherein: The specimen geometry, material distribution, and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, combined with the set defect parameters, are used to output simulated infrared thermogram sequences and temperature change curves under various defect conditions over time through numerical simulation, including: Based on the existing multi-material transient heat conduction finite element model, preset thermal excitation and boundary conditions are activated to conduct numerical simulations of defect parameters for different pore sizes, different depths and different inlay material types. The temperature field at each time step is calculated using a transient solver to obtain the temperature change curve of the defect region over time, and the peak temperature difference, peak time, and thermal decay characteristic parameters are extracted. By analyzing the temperature difference evolution curves under different defect conditions, a mapping relationship between defect thermal response and structural parameters is established. Output multiple sets of simulated infrared thermogram sequences that change over time, along with corresponding temperature change curves and quantized characteristic parameters.

[0010] As a preferred scheme for a multi-material power equipment defect quantification detection method based on infrared thermal imaging and generative models, wherein: The specimen geometry, material distribution, and thermal excitation conditions are embedded in the multi-material transient heat conduction finite element model. Actual specimens are fabricated, and an active infrared thermal imaging experimental platform is built. The system outputs a sequence of real multi-material defect infrared thermal images and temperature change data, including: An active infrared thermal imaging experimental platform was built, including a thermal excitation module, an infrared thermal imaging acquisition module, an automated mobile detection platform, and a data synchronization acquisition and control module; Actual specimens were fabricated, and the geometric dimensions, material distribution, and internal defect specifications of the actual specimens were consistent with the settings in the multi-material transient heat conduction finite element model. The same thermal excitation parameters as those in the multi-material transient heat conduction finite element model were used for heating. Infrared thermal imagers are used to acquire thermal response image sequences in real time and record the temperature change process in the defect area. Output infrared thermogram sequences and temperature change data under real experimental conditions.

[0011] As a preferred scheme for a multi-material power equipment defect quantification detection method based on infrared thermal imaging and generative models, wherein: The improved variational autoencoder, which integrates adaptive latent space optimization and perceptual loss constraints, is trained based on a standardized multi-material defect infrared dataset. The output includes augmented defect infrared image samples, including: An improved variational autoencoder that integrates adaptive latent space dimensionality optimization and perceptual loss constraint is constructed and trained, including an encoder network, a latent space distribution module, a decoder network, and a perceptual loss constraint module. The encoder network uses a convolutional neural network to extract features from the input heatmap and maps the features to the latent space. The latent variables are modeled using a Gaussian distribution. The optimization objectives of the improved variational autoencoder include reconstruction loss and KL divergence constraint.

[0012] As a preferred scheme for a multi-material power equipment defect quantification detection method based on infrared thermal imaging and generative models, wherein: The improved variational autoencoder, which integrates adaptive latent space optimization and perceptual loss constraints and is based on a standardized multi-material defect infrared dataset, outputs augmented defect infrared image samples, and further includes: An adaptive dimension selection mechanism based on the cumulative explained variance ratio is introduced to dynamically select the potential spatial dimension according to the data distribution characteristics; An improved variational autoencoder is trained using a standardized multi-material defect infrared dataset, and after training, it outputs augmented defect infrared image samples.

[0013] Secondly, the present invention provides a multi-material power equipment defect quantification detection system based on infrared thermal imaging and a generative model, comprising: The multi-material heat conduction modeling module is used to construct a multi-material transient heat conduction finite element model based on the thermophysical parameters of the multi-material structure of power equipment, and embeds the specimen geometry, material distribution and thermal excitation conditions; The defect thermal response simulation module is used to output simulated infrared thermogram sequences and temperature change curves under various defect conditions through numerical simulation, based on the specimen geometry, material distribution and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model and the set defect parameters. The infrared thermal imaging experimental acquisition module is used to fabricate actual specimens and build an active infrared thermal imaging experimental platform based on the specimen geometry, material distribution and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, and output real multi-material defect infrared thermal image sequence and temperature change data. The multi-source data fusion preprocessing module is used to fuse and preprocess simulated infrared thermal image sequences with real multi-material defect infrared thermal image sequences, and output a standardized multi-material defect infrared dataset. An improved variational autoencoder generation module is used to train an improved variational autoencoder that integrates adaptive latent space optimization and perceptual loss constraints based on a standardized multi-material defect infrared dataset, and outputs expanded defect infrared image samples.

[0014] Thirdly, the present invention provides a computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model.

[0016] The beneficial effects of this invention are as follows: By establishing a multi-material thermal conduction finite element model, this invention achieves unified analysis of the thermal response of defects in carbon fiber composites, epoxy resins, and heterostructure inlays, improving the detection adaptability in complex material scenarios; it can systematically reveal the influence of defect size, depth, and material thermal diffusion characteristics on the thermal response law, realizing quantitative analysis of defect depth and thermal response characteristics; by improving the variational autoencoder to generate high-quality defect samples, it effectively alleviates the small sample problem and improves the generalization ability and stability of the intelligent detection model; after proposing the improved variational autoencoder model, the generated images are significantly superior to traditional generation methods in terms of edge sharpness, texture continuity, and structural realism; it establishes a complete technical system from thermal mechanism analysis, finite element simulation, experimental verification to generation enhancement and intelligent detection, which has strong engineering application value; this invention can be applied to online condition detection and intelligent operation and maintenance of power equipment, composite insulation structures, wind turbine blades, and large composite material structures, and has good industrialization value. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0018] Figure 1This is an overall flowchart of a multi-material power equipment defect quantification detection method based on infrared thermal imaging and generative model provided by the present invention.

[0019] Figure 2 This invention provides a schematic diagram of the construction principle of a multi-material transient heat conduction finite element model for a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model.

[0020] Figure 3 This invention provides an improved variational autoencoder (VAE) structure and training flowchart for a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model. Detailed Implementation

[0021] 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.

[0022] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model, including: S1: Based on the thermophysical parameters of the multi-material structure of power equipment, a multi-material transient heat conduction finite element model is constructed, which includes the specimen geometry, material distribution and thermal excitation conditions. S2: Based on the specimen geometry, material distribution and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, combined with the set defect parameters, the simulated infrared thermogram sequence and temperature change curve under various defect conditions are output through numerical simulation. S3: Based on the specimen geometry, material distribution and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, actual specimens are fabricated and an active infrared thermal imaging experimental platform is built to output real multi-material defect infrared thermal image sequences and temperature change data. S4: Fuse and preprocess the simulated infrared thermal image sequence with the real multi-material defect infrared thermal image sequence to output a standardized multi-material defect infrared dataset; S5: Based on a standardized multi-material defect infrared dataset, an improved variational autoencoder that integrates adaptive latent space optimization and perceptual loss constraints is trained to output expanded defect infrared image samples.

[0023] It should be noted that, through steps S1-S5, this invention constructs a complete technical closed loop from multi-material thermal conduction mechanism analysis, finite element simulation, experimental verification to data enhancement of the generated model, realizing intelligent quantitative detection of structural defects in multi-material structures such as carbon fiber composites, epoxy resin insulation materials, and heterogeneous metal inlays in power equipment. It effectively solves the problems of traditional methods being unable to adapt to the differences in thermal response of multiple materials, insufficient model generalization ability under small sample conditions, and lack of details in generated images, providing a systematic solution for high-reliability infrared non-destructive testing of complex power equipment.

[0024] Example 2, refer to Figures 1-3 As an embodiment of the present invention, based on the previous embodiment, a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model is provided, including: In this embodiment, the multi-material transient heat conduction finite element model constructed in step S1 above, based on the thermophysical parameters of the multi-material structure of the power equipment, includes embedded specimen geometry, material distribution, and thermal excitation conditions: Based on the thermophysical parameters of the multi-material structure of power equipment, a multi-material transient heat conduction finite element model is established through a finite element multiphysics coupling platform. The model includes the following components: carbon fiber composite matrix, epoxy resin matrix, air pore defects, corroded metal inlay defects, and heterogeneous material inlay defects.

[0025] Regarding the setting of material thermophysical parameters, the anisotropic thermal conductivity tensor is used to define the thermal conductivity characteristics of carbon fiber composites, with an in-plane thermal conductivity of 40 W / (m·K) and a thickness direction thermal conductivity of 3.3 W / (m·K); the thermal conductivity of epoxy resin is set to 0.25 W / (m·K); the thermal conductivity of air defects is set to 0.026 W / (m·K); and the thermal conductivity of the corroded metal inlay is set to 72 W / (m·K).

[0026] The model uses transient heat conduction equations to describe the heat diffusion process: in, Time (in seconds) represents the time variable in the heat conduction process. Indicates the density of the material; Indicates the specific heat capacity of the material; Represents the temperature field. Indicates the thermal diffusivity. For the Laplace operator, The intensity of the internal heat source.

[0027] The mesh is discretized using free tetrahedral meshes, with local refinement at defect regions and material interfaces to improve solution accuracy.

[0028] Regarding thermal excitation and boundary conditions, this invention employs an active pulsed thermal excitation method, utilizing a halogen lamp array to apply uniform heat flux density excitation to the specimen surface. Specific parameters include: heat flux density 9000 W / m³. 2 Excitation duration 3s, initial ambient temperature 20℃, convective heat transfer coefficient 10W / (m²) 2 • K). The model surface simultaneously considers natural convection and radiation heat transfer, and uses the Stefan-Boltzmann radiation model to describe the surface radiation process: .in: It represents the radiant power emitted per unit area of ​​an object's surface into the hemispherical space. Indicates the surface emissivity of a material; This represents the Stefan-Boltzmann constant; This indicates absolute temperature.

[0029] like Figure 2 The diagram shown is a schematic of the construction principle of the multi-material transient heat conduction finite element model provided by the present invention. It illustrates the entire process from inputting multi-material thermophysical parameters to the completion of the finite element model, including setting material parameters, transient heat conduction equations, mesh discretization and local refinement, and setting active pulse thermal excitation and convection-radiation boundary conditions.

[0030] In this embodiment, step S2 above, based on the specimen geometry, material distribution, and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, and combined with the set defect parameters, outputs simulated infrared thermogram sequences and temperature change curves under various defect conditions over time through numerical simulation, including: Based on the constructed multi-material transient heat conduction finite element model, the preset thermal excitation and boundary conditions are activated, and numerical simulations are performed for different defect parameters.

[0031] Specifically, by establishing defect models under different pore sizes, depths, and inlay materials, the thermal response characteristics are systematically analyzed.

[0032] The defect parameters are set as follows: aperture range 10mm~25mm, defect depth range 0.4mm~1.6mm, and insert types include air, corroded metal, and epoxy resin.

[0033] The temperature field at each time step was calculated using a transient solver to obtain the temperature variation curve of the defect region over time, and the peak temperature difference, peak time, and thermal decay characteristic parameters were extracted. By analyzing the temperature difference evolution curves under different defect conditions, a mapping relationship between the defect thermal response and structural parameters was established. Simulation results show that: the larger the defect aperture, the higher the peak temperature difference; the shallower the defect burial depth, the more pronounced the thermal response; the lower the thermal diffusivity, the longer the duration of the thermal signal; and the difference in thermal conductivity between heterostructures significantly alters the heat wave propagation path.

[0034] Furthermore, a model is established to model the relationship between defect depth and peak response time: .in: Indicates the peak time of the thermal response; Indicates the depth of the defect; This represents the thermal diffusivity of the material. This model can be used for quantitative estimation of defect depth.

[0035] The final output includes multiple sets of simulated infrared thermal image sequences (time resolution 0.1s, total duration 10s), along with corresponding temperature change curves and quantized characteristic parameters.

[0036] In this embodiment, step S3 above, based on the specimen geometry, material distribution, and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, involves fabricating an actual specimen and building an active infrared thermal imaging experimental platform to output a real multi-material defect infrared thermal image sequence and temperature change data, including: To verify the accuracy of the established multi-material transient heat conduction finite element model and simulation results, an active infrared thermal imaging experimental platform was constructed.

[0037] The experimental platform specifically includes: a halogen lamp thermal excitation module, an infrared thermal imaging acquisition module, an automated mobile detection platform, and a data synchronization acquisition and control module.

[0038] The geometric dimensions, material distribution (carbon fiber composite, epoxy resin, heterostructure), internal defect specifications (pore size, depth, inlay type), and thermal excitation parameters of the experimental specimens were all consistent with the settings in the multi-material transient heat conduction finite element model.

[0039] A heat flux density of 9000 W / m² was applied using a halogen lamp array. 2 It lasts for 3 seconds.

[0040] An infrared thermal imager acquires a sequence of thermal response images in real time and records the temperature change process in the defect area. The consistency of the thermal response pattern is verified by comparing and analyzing the experimental data with the simulation data generated by S2.

[0041] Experimental results show that the simulation and experiment have high consistency in terms of temperature change trends, consistent hot spot evolution patterns in the defect region, and small errors in peak time and thermal decay patterns. This demonstrates that the established multi-material transient heat conduction finite element model and simulation results can effectively reflect the real defect thermal response process.

[0042] The final output is an infrared thermogram sequence and temperature change data under real experimental conditions.

[0043] In this embodiment, step S4 above involves fusing and preprocessing the simulated infrared thermal image sequence with the real multi-material defect infrared thermal image sequence to output a standardized multi-material defect infrared dataset, including: By integrating finite element simulation infrared thermal image sequences with real experimental infrared thermal image sequences, a multi-material defect infrared dataset is established.

[0044] The dataset includes: different material types; different defect sizes; different defect depths; different thermal excitation conditions; and different time-series thermal images.

[0045] The original heatmap was preprocessed as follows: (1) grayscale normalization; (2) noise filtering; (3) time series truncation; (4) image size unification; (5) data augmentation.

[0046] After the above processing, a standardized multi-material defect infrared dataset is output, which can be directly used for subsequent generative model training and intelligent recognition model optimization.

[0047] In this embodiment, step S5 above, based on a standardized multi-material defect infrared dataset, trains an improved variational autoencoder that fuses adaptive latent space optimization and perceptual loss constraints, outputting augmented defect infrared image samples, including: An improved variational autoencoder model that integrates adaptive latent space dimensionality optimization and perceptual loss constraints is constructed and trained. The model consists of an encoder network, a latent space distribution module, a decoder network, and a perceptual loss constraint module.

[0048] The encoder utilizes a convolutional neural network to extract features from the input heatmap and maps them to the latent space. The latent variables are modeled using a Gaussian distribution. .in: The latent variable is a low-dimensional feature representation of the input infrared heatmap after the encoder maps it to the latent space. This represents the mean of the latent variables; Indicates standard deviation; This represents a randomly sampled variable. The model optimization objectives include reconstruction loss and KL divergence constraints: ,in, This represents the total loss function.

[0049] To address the information redundancy problem caused by the fixed latent dimensions in traditional variational autoencoders, this invention introduces an adaptive dimension selection mechanism based on the cumulative explained variance ratio. This mechanism dynamically selects the latent spatial dimension according to the data distribution characteristics, thereby improving the model's expressive power and generation stability. Simultaneously, a perceptual loss constraint based on a VGG16 network is introduced, making the generated images more closely resemble real infrared images in terms of structural similarity and texture details.

[0050] like Figure 3 The diagram shown illustrates the improved variational autoencoder structure and training flowchart provided by this invention, describing the encoder convolutional feature extraction, Gaussian latent space modeling, adaptive dimension selection, decoder generation, and total loss optimization process including reconstruction loss, KL divergence, and VGG16 perceptual loss.

[0051] An improved variational autoencoder that integrates adaptive latent space optimization and perceptual loss constraints was trained using a standardized multi-material defect infrared dataset.

[0052] After training, high-quality, sharp-edged, and realistically textured augmented defect infrared image samples are output.

[0053] In one possible implementation, augmented defect infrared image samples can be merged with the original multi-material defect infrared dataset to form an enhanced training set. Based on this enhanced training set, a deep learning regression or classification network (such as a convolutional neural network) can be used for training to learn the mapping from infrared thermogram sequences to quantification parameters such as defect size and depth. Alternatively, the relational model in S2 can be directly utilized. This indicates that the peak thermal response time is directly proportional to the square of the defect depth and inversely proportional to the material's thermal diffusivity. The peak thermal response time is extracted from the measured infrared thermal image of the electrical equipment under test. Substituting the values ​​into the model allows us to inversely deduce the defect depth. This enables quantitative estimation of defect depth.

[0054] Meanwhile, by analyzing the hot spot morphology and temperature difference distribution in the defect area, the pore size and location information of the defect can be further quantified.

[0055] Therefore, the expanded samples generated by this invention effectively alleviate the training difficulties of the quantitative detection model under small sample conditions and improve the estimation accuracy of parameters such as defect depth and size.

[0056] Example 3: The above is an illustrative scheme of a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model, according to this embodiment. It should be noted that the technical solution of a multi-material power equipment defect quantification detection system based on infrared thermal imaging and a generative model belongs to the same concept as the above-described multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model. Details not described in detail in the technical solution of the multi-material power equipment defect quantification detection system based on infrared thermal imaging and a generative model in this embodiment can be found in the description of the above-described multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model.

[0057] This embodiment also provides a multi-material power equipment defect quantification detection system based on infrared thermal imaging and a generative model, including: The multi-material heat conduction modeling module is used to construct a multi-material transient heat conduction finite element model based on the thermophysical parameters of the multi-material structure of power equipment, and embeds the specimen geometry, material distribution and thermal excitation conditions; The defect thermal response simulation module is used to output simulated infrared thermogram sequences and temperature change curves under various defect conditions through numerical simulation, based on the specimen geometry, material distribution and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model and the set defect parameters. The infrared thermal imaging experimental acquisition module is used to fabricate actual specimens and build an active infrared thermal imaging experimental platform based on the specimen geometry, material distribution and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, and output real multi-material defect infrared thermal image sequence and temperature change data. The multi-source data fusion preprocessing module is used to fuse and preprocess simulated infrared thermal image sequences with real multi-material defect infrared thermal image sequences, and output a standardized multi-material defect infrared dataset. An improved variational autoencoder generation module is used to train an improved variational autoencoder that integrates adaptive latent space optimization and perceptual loss constraints based on a standardized multi-material defect infrared dataset, and outputs expanded defect infrared image samples.

[0058] This embodiment also provides a computer device applicable to a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model, as proposed in the above embodiments.

[0059] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model as proposed in the above embodiment.

[0060] The storage medium proposed in this embodiment belongs to the same inventive concept as the multi-material power equipment defect quantification detection method based on infrared thermal imaging and generative model proposed in the above embodiment. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0061] Example 4, an embodiment of the present invention, provides a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model. To verify the effectiveness of the proposed method and the accuracy of the simulation model, this example uses a carbon fiber composite material specimen as the object and conducts an active infrared thermal imaging detection experiment. The specific experimental process and results are as follows: A carbon fiber composite specimen with dimensions of 300mm×300mm×2mm was prepared, and a matrix of air-pore defects was set inside the specimen. The defect diameters were set to 10mm, 15mm, 20mm and 25mm, respectively, and the corresponding defect depths were 0.4mm, 0.8mm, 1.2mm and 1.6mm.

[0062] A halogen lamp array was used to apply 18KW pulsed thermal excitation to the surface of the specimen for 3s. The thermal response image sequence was continuously acquired within a time window of 0 to 10s using an infrared thermal imager.

[0063] Experimental results show that: (1) All defective areas showed significantly higher temperature response characteristics than non-defective areas, and the temperature difference curves showed a rapid heating-slow decay pattern. (2) Under the same defect depth conditions, the larger the pore size, the more obvious the peak thermal response. For example, when the defect depth is the same at 0.4 mm, the peak temperature difference of the 25 mm pore diameter defect is significantly higher than that of the 10 mm pore diameter defect, indicating that the larger size defect has a stronger thermal resistance effect; (3) Under the same aperture conditions, the thermal response of shallow defects appears earlier and the defects are more identifiable; the peak time of deep defects is significantly delayed due to the increased heat wave propagation path. (4) In the experimental thermal image, the defect area reaches the best thermal contrast within a period of about 2 to 3 seconds after the heating is completed, at which time the defect edge is the clearest. (5) The experimental results are basically consistent with the temperature change trend obtained by finite element simulation, which verifies the accuracy of the heat conduction model.

[0064] Furthermore, analysis of thermal field cloud maps at different time points revealed that shallow defects first form high-temperature hot spots as time progresses, while deep defects gradually appear in the subsequent cooling stage, further verifying the coupling mechanism of thermal wave reflection and thermal diffusion.

[0065] Example 5, an embodiment of the present invention, provides a method for quantitative detection of defects in multi-material power equipment based on infrared thermal imaging and a generative model. To further verify the applicability of the present invention in the detection of defects in multi-material inlays, this embodiment sets corrosion metal inlay defects in a carbon fiber composite matrix and conducts experiments under the same thermal excitation conditions. Details are as follows: A defect model of carbon fiber composite inlay was established. A 150mm×150mm×2mm metal inlay area was set inside the matrix, and the corrosion defects of the metal in the actual equipment were simulated by the corrosion of the metal material with degraded thermal conductivity.

[0066] Carbon fiber composite cover plates with thicknesses of 0.5 mm, 1 mm, and 2 mm were placed on top of the inlay to simulate the internal inlay defect structure under different burial depth conditions.

[0067] During the experiment, active pulse thermal excitation was used for heating, and infrared thermogram sequences were acquired simultaneously.

[0068] Experimental results show that: (1) Due to the significant difference in thermal conductivity between the corroded metal inlay region and the substrate material, the surface thermal response of the corroded metal inlay region shows a significant temperature difference compared with the normal region; (2) As the thickness of the cover plate increases, the peak value of the defect thermal response gradually decreases, and the edge of the thermal anomaly area gradually becomes blurred; (3) The defect boundary is clearest under the 0.5mm cover plate condition, while the thermal response is significantly weakened under the 2mm cover plate condition, indicating that the burial depth of the defect will significantly reduce the infrared detection sensitivity. (4) The peak time of the thermal response is delayed with the increase of the capping layer thickness, indicating that the thermal wave propagation time is positively correlated with the defect burial depth; (5) The defect area appears as a local temperature anomaly area in the thermal field cloud map, and the thermal anomaly morphology is basically consistent with the actual size of the inlay.

[0069] Experimental results further demonstrate that the difference in thermal diffusion between materials with different thermal conductivity is an important reason for the formation of infrared thermal anomalies.

[0070] Example 6, an embodiment of the present invention, provides a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model. To verify the detection effect of the present invention in different matrix materials, this embodiment uses epoxy resin as the matrix, with air defects and carbon fiber composite inlay defects inside. The experimental process and results are as follows: An epoxy resin matrix model was established, and air defects and carbon fiber composite inlay defects were set inside it.

[0071] The experiment was conducted under the same thermal excitation conditions as in the aforementioned embodiments.

[0072] Experimental results show that: (1) Due to the low thermal diffusivity of epoxy resin, the heat spreads slowly inside the material, so the duration of the defect thermal response is significantly longer than that of carbon fiber composites. (2) The air defect region forms obvious high-temperature hot spots, while the carbon fiber inlay region has a higher thermal conductivity and the local heat diffusion rate is faster, showing different thermal field distribution characteristics. (3) Compared with carbon fiber composite matrix, the thermal anomaly region in epoxy resin specimen lasts longer, which facilitates subsequent thermal image acquisition and feature extraction. (4) The experimental heat map and simulation results show high consistency in terms of hot spot location, temperature difference change trend and peak response time.

[0073] The above results show that the differences in thermal diffusivity and thermal conductivity of different materials can significantly affect the infrared defect imaging effect.

[0074] Example 7, an embodiment of the present invention, provides a multi-material power equipment defect quantification detection method based on infrared thermal imaging and a generative model. To verify the superiority of the improved variational autoencoder proposed in this invention in defect image generation, this embodiment uses a constructed multi-material infrared defect dataset to train a traditional variational autoencoder and the improved model of this invention, respectively, and performs comparative analysis. Details are as follows: An improved variational autoencoder model was trained using a constructed multi-material infrared defect dataset.

[0075] The model uses a convolutional neural network as the main structure of the encoder and decoder, and introduces a perceptual loss constraint mechanism based on the VGG16 network. At the same time, it adopts an adaptive latent space dimension optimization method to automatically determine the dimension of latent variables.

[0076] In the experiment, the traditional variational autoencoder model and the improved model proposed in this invention were compared and analyzed.

[0077] Experimental results show that: (1) The heatmaps generated by traditional variational autoencoders have problems with blurred edges and missing textures, while the defect edges in the images generated by this invention are clearer; (2) The heat maps generated by the improved model are closer to real infrared images in terms of temperature gradient continuity, local texture structure and hot spot morphology; (3) By introducing perceptual loss, the structural similarity index of the generated samples is significantly improved, and the key thermal features of the defect area can be effectively preserved.

[0078] 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 quantitative detection of defects in multi-material power equipment based on infrared thermal imaging and a generative model, characterized in that, include: Based on the thermophysical parameters of the multi-material structure of power equipment, a multi-material transient heat conduction finite element model is constructed, which incorporates the specimen geometry, material distribution and thermal excitation conditions. Based on the specimen geometry, material distribution and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, and combined with the set defect parameters, the numerical simulation outputs the simulated infrared thermogram sequence and temperature change curve under various defect conditions as a function of time. Based on the specimen geometry, material distribution and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, actual specimens were fabricated and an active infrared thermal imaging experimental platform was built to output real multi-material defect infrared thermal image sequences and temperature change data. The simulated infrared thermal image sequence is fused and preprocessed with the real multi-material defect infrared thermal image sequence to output a standardized multi-material defect infrared dataset. Based on a standardized multi-material defect infrared dataset, an improved variational autoencoder that integrates adaptive latent space optimization and perceptual loss constraints is trained to output expanded defect infrared image samples.

2. The method for quantitative detection of defects in multi-material power equipment based on infrared thermal imaging and a generative model as described in claim 1, characterized in that, The thermophysical parameters of the multi-material structure based on power equipment are used to construct a multi-material transient heat conduction finite element model, which embeds the specimen geometry, material distribution, and thermal excitation conditions, including: Based on the thermophysical parameters of carbon fiber composite materials, epoxy resin insulation materials and heterogeneous inlay materials in power equipment, a three-dimensional transient heat conduction finite element model was established, which includes carbon fiber composite matrix, epoxy resin matrix, air pore defects, corroded metal inlay defects and heterogeneous material inlay defects. The geometric dimensions, material distribution and thermal excitation conditions of the specimen were embedded in the model.

3. The method for quantitative detection of defects in multi-material power equipment based on infrared thermal imaging and a generative model as described in claim 2, characterized in that, The thermophysical parameters of the multi-material structure based on power equipment are used to construct a multi-material transient heat conduction finite element model, which embeds the specimen geometry, material distribution, and thermal excitation conditions. The model also includes: The transient heat conduction equation is used to describe the heat diffusion process; The model was discretized using free tetrahedral meshes, with local refinement applied to defect regions and material interfaces. An active pulsed thermal excitation method was adopted, and the heat flux density and duration uniformly applied to the surface of the specimen were set. The initial ambient temperature and convective heat transfer coefficient were also set. Natural convective heat transfer and radiative heat transfer on the model surface were considered, and a radiation model was used to describe the surface radiation process.

4. The multi-material power equipment defect quantitative detection method based on infrared thermal imaging and generative model as described in claim 3, characterized in that, The specimen geometry, material distribution, and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, combined with the set defect parameters, are used to output simulated infrared thermogram sequences and temperature change curves under various defect conditions over time through numerical simulation, including: Based on the existing multi-material transient heat conduction finite element model, preset thermal excitation and boundary conditions are activated to conduct numerical simulations of defect parameters for different pore sizes, different depths and different inlay material types. The temperature field at each time step is calculated using a transient solver to obtain the temperature change curve of the defect region over time, and the peak temperature difference, peak time, and thermal decay characteristic parameters are extracted. By analyzing the temperature difference evolution curves under different defect conditions, a mapping relationship between defect thermal response and structural parameters is established. Output multiple sets of simulated infrared thermogram sequences that change over time, along with corresponding temperature change curves and quantized characteristic parameters.

5. The multi-material power equipment defect quantitative detection method based on infrared thermal imaging and generative model as described in claim 4, characterized in that, The specimen geometry, material distribution, and thermal excitation conditions are embedded in the multi-material transient heat conduction finite element model. Actual specimens are fabricated, and an active infrared thermal imaging experimental platform is built. The system outputs a sequence of real multi-material defect infrared thermal images and temperature change data, including: An active infrared thermal imaging experimental platform was built, including a thermal excitation module, an infrared thermal imaging acquisition module, an automated mobile detection platform, and a data synchronization acquisition and control module; Actual specimens were fabricated, and the geometric dimensions, material distribution, and internal defect specifications of the actual specimens were consistent with the settings in the multi-material transient heat conduction finite element model. The same thermal excitation parameters as those in the multi-material transient heat conduction finite element model were used for heating. Infrared thermal imagers are used to acquire thermal response image sequences in real time and record the temperature change process in the defect area. Output infrared thermogram sequences and temperature change data under real experimental conditions.

6. The method for quantitative detection of defects in multi-material power equipment based on infrared thermal imaging and a generative model as described in claim 5, characterized in that, The improved variational autoencoder, which integrates adaptive latent space optimization and perceptual loss constraints, is trained based on a standardized multi-material defect infrared dataset. The output includes augmented defect infrared image samples, including: An improved variational autoencoder that integrates adaptive latent space dimensionality optimization and perceptual loss constraint is constructed and trained, including an encoder network, a latent space distribution module, a decoder network, and a perceptual loss constraint module. The encoder network uses a convolutional neural network to extract features from the input heatmap and maps the features to the latent space. The latent variables are modeled using a Gaussian distribution. The optimization objectives of the improved variational autoencoder include reconstruction loss and KL divergence constraint.

7. The multi-material power equipment defect quantification detection method based on infrared thermal imaging and generative model as described in claim 6, characterized in that, The improved variational autoencoder, which integrates adaptive latent space optimization and perceptual loss constraints and is based on a standardized multi-material defect infrared dataset, outputs augmented defect infrared image samples, and further includes: An adaptive dimension selection mechanism based on the cumulative explained variance ratio is introduced to dynamically select the potential spatial dimension according to the data distribution characteristics; An improved variational autoencoder is trained using a standardized multi-material defect infrared dataset, and after training, it outputs augmented defect infrared image samples.

8. A multi-material power equipment defect quantification detection system based on infrared thermal imaging and generative models, using the method described in any one of claims 1 to 7, characterized in that, include: The multi-material heat conduction modeling module is used to construct a multi-material transient heat conduction finite element model based on the thermophysical parameters of the multi-material structure of power equipment, and embeds the specimen geometry, material distribution and thermal excitation conditions; The defect thermal response simulation module is used to output simulated infrared thermogram sequences and temperature change curves under various defect conditions through numerical simulation, based on the specimen geometry, material distribution and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model and the set defect parameters. The infrared thermal imaging experimental acquisition module is used to fabricate actual specimens and build an active infrared thermal imaging experimental platform based on the specimen geometry, material distribution and thermal excitation conditions embedded in the multi-material transient heat conduction finite element model, and output real multi-material defect infrared thermal image sequence and temperature change data. The multi-source data fusion preprocessing module is used to fuse and preprocess simulated infrared thermal image sequences with real multi-material defect infrared thermal image sequences, and output a standardized multi-material defect infrared dataset. An improved variational autoencoder generation module is used to train an improved variational autoencoder that integrates adaptive latent space optimization and perceptual loss constraints based on a standardized multi-material defect infrared dataset, and outputs expanded defect infrared image samples.

9. A computer device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images