A defect depth prediction method based on a sequence of pulsed infrared thermal images

Abstract

A defect depth prediction method based on pulse infrared thermal image sequence relates to the field of non-contact nondestructive testing, solves the problems of insufficient defect depth quantitative evaluation precision, dependence on empirical model and difficulty in adapting to different materials in existing active infrared thermal imaging detection, fully utilizes time information of the thermal image sequence, automatically extracts features through a deep learning model, realizes accurate prediction of the buried depth of internal defects of the material, and improves the reliability and generalization ability of the detection results under different material conditions. The present application utilizes pulse thermal excitation combined with infrared thermal imaging to obtain the surface temperature response of the material, compares and analyzes the temperature signals of the defect area and the non-defect area, introduces a long short-term memory neural network (LSTM) to model the time sequence features, and realizes accurate prediction of the hidden defect depth.

Description

Technical Field

[0001] This invention relates to the field of non-contact nondestructive testing technology, and in particular to a method for predicting defect depth based on the relative norm ratio of defect regions using infrared thermographic sequences and a long short-term memory neural network. Background Technology

[0002] Infrared thermography is an important nondestructive testing method for detecting internal defects in materials. Active infrared thermography introduces heat flow through external thermal excitation, causing transient changes in the surface temperature of the sample. If discontinuities or defects (such as voids or delamination) exist within the sample, anomalies will occur in the surface temperature field distribution. These anomalies manifest as differences in temperature response between the region above the defect and the defect-free region, thus providing the possibility of quantitatively assessing defect depth based on surface temperature response.

[0003] Currently, quantitative assessment of defect depth remains a key focus and challenge in the field of nondestructive testing. Traditionally, researchers often use characteristic parameters from thermal imaging sequences to infer defect depth, such as the highest temperature rise (peak temperature) of the defect region relative to the background or the delay time to reach the peak. Simultaneously, various thermal imaging data processing methods have been developed, such as independent component thermal imaging (ICT), pulsed phase thermal imaging (PPT), thermal signal reconstruction (TSR), and principal component thermal imaging (PCT), to extract defect features from thermal imaging sequences. These methods have improved defect detection and characterization capabilities to some extent, but limitations remain: on the one hand, they often rely on accurate material thermophysical parameters or manually selected features, leading to insufficient adaptability to different materials or complex conditions; on the other hand, single features (such as peak temperature or time) are easily affected by noise and experimental conditions, resulting in limited depth prediction accuracy.

[0004] With the development of artificial intelligence and deep learning technologies, researchers have begun to explore the application of neural networks to data analysis in thermal imaging nondestructive testing. Deep learning models can automatically extract complex features from large amounts of data and have been used for tasks such as defect identification and quantitative assessment. For example, some studies have used finite element simulation to generate data to train neural networks to assess defect depth, achieving some success. However, several problems remain unresolved in predicting defect depth for thermal imaging sequences: some methods require manual analysis of temperature curves or rely on the experience of inspectors, resulting in high subjectivity and low efficiency. In recent years, although methods using machine learning techniques such as neural networks have emerged to attempt to predict defect depth, these often require a large amount of representative data to train the model and lack generalization ability for different types of workpieces. Existing technologies struggle to simultaneously achieve both accuracy and applicability in prediction. Therefore, a new technical solution is urgently needed to address the aforementioned problem of inaccurate defect depth prediction. Summary of the Invention

[0005] This invention addresses the problems of insufficient accuracy in quantitative assessment of defect depth in existing active infrared thermal imaging detection, reliance on empirical models, and difficulty in adapting to different materials. It fully utilizes the temporal information of thermal imaging sequences and automatically extracts features through a deep learning model to achieve accurate prediction of the burial depth of internal defects in materials, thereby improving the reliability and generalization ability of detection results under different material conditions.

[0006] To address the aforementioned problems, this invention provides a defect depth prediction method based on the ratio of time-domain thermal signals. This method utilizes pulsed thermal excitation combined with infrared thermal imaging to acquire the material surface temperature response. By comparing and analyzing the temperature signals of defective and defect-free regions, and introducing a Long Short-Term Memory (LSTM) neural network to model the time-series features, it achieves accurate prediction of the hidden defect depth. Specifically, this invention includes the following steps:

[0007] Step 1: Instantly heat the sample under test using a pulsed thermal excitation device, and non-contactly acquire the infrared thermal image sequence of the sample surface using an infrared imaging device;

[0008] Step 2: During the cooling process, select a frame with a strong thermal contrast between the defect and the background when the defect outline is clear during the cooling process after pulse excitation (i.e., it can be clearly seen with the naked eye). Use image processing algorithms to identify the outline of the defect area in the thermal image and determine the corresponding defect-free reference area.

[0009] Step 3: Extract the temperature-time signal curves of the defect area and the reference area over the entire time range. For the two sets of time-domain temperature signals obtained, calculate the L1 norm ratio of the defect area to the non-defect area within the heating range under different defect depth conditions (this is the process of predicting the defect depth, so any depth of defect that can be detected by infrared technology can be used). That is, the ratio of the sum of the absolute values ​​of the temperature signals of the defect area and the non-defect area at each time moment, thus forming a time series of ratios reflecting the difference in thermal response between the two over time.

[0010] Step 4: Input the above ratio time series into a pre-trained LSTM (Long Short-Term Memory) network regression model to perform dynamic feature analysis on the defect and predict the first visible moment of the defect. This first visible moment is defined as the time point when the defect sign is first clearly detectable in the thermal image sequence, reflecting the time delay experienced by the defect from being heated to its appearance, thus avoiding large errors in the manual intervention process.

[0011] Step 5: Perform extensive 3D finite element simulation analysis on typical defect depths (3D finite element simulation is a common technique in this field, easily understood and implemented by those skilled in the art, and is a standard skill). Model the curve of the "first visible moment of the defect" versus the defect depth, as output by the network in Step 4, using a parametric scanning method for the defect depth. Calculate (those skilled in the art can easily perform this calculation using the aforementioned description; it only requires parametric scanning of the depth information from the results output in Step 4) the result determination coefficient R. 2 Find the depth d and the first visible time t d The mapping relationship between two variables (usually an exponential relationship).

[0012] Step 6: Using the mapping relationship from Step 5 and several defects of known depth, establish and calibrate the mapping relationship curve between the "first visible time of the defect" and the defect depth in the current material. To ensure accuracy, the depths of the selected defects should cover the detection range as much as possible (the range varies under different defect conditions; generally, the minimum range is the depth at which the defect is detected when the pulse excitation time is greater than one order of magnitude, and the maximum range is the maximum depth assuming the defect outline remains clearly visible after pulse excitation), and the defects should be evenly distributed. Generally, selecting three defects of known depth is sufficient; the denser the number, the more accurate the prediction result. Data is collected from defects of unknown depth using pulse excitation. The "first visible time of the defect" output from the infrared time series is input into the LSTM network and then input into the mapping relationship curve between the "first visible time of the defect" and the defect depth to quickly and accurately predict the defect depth.

[0013] As a preferred embodiment, this invention addresses isotropic and homogeneous materials with similar material parameters. Step 6 only requires three or more sets of known-depth calibration defects to correct the defect depth prediction curve. Specifically, similar material parameters refer to thermal properties such as thermal conductivity, density, and specific heat capacity being within the same order of magnitude. For example, for similar polymer materials or composite materials, the differences in their thermal parameters are small, so the aforementioned mapping relationship (such as an exponential mapping relationship) can be used for prediction correction without repeating step five. By using the same exponential mapping relationship, a new prediction curve can be effectively constructed, enabling depth prediction on new materials. This method avoids repetitive and complex calculations and improves applicability and prediction efficiency for different materials. However, for materials with complex structures (such as multilayer composites or heterogeneous structures), due to significant differences in their thermal conductivity and diffusion characteristics, it is necessary to remodel in step five, establish a new mapping relationship, and then perform the correction in step six. Therefore, the processing of such materials is relatively more complex, but it still falls within the scope of this invention.

[0014] As another preferred embodiment, in step 4 of the present invention, the LSTM network is trained using simulation data; the LSTM network finds the value of "defect discovery time" based on the time domain sequence. In order to avoid human error, this value is specifically quantified. Furthermore, for samples of different materials, even if the thermal conduction and diffusion characteristics are different, the principle of the time delay value experienced by the output heating to manifestation is the same, and there is no need to retrain.

[0015] The Long Short-Term Memory (LSTM) neural network model is obtained through pre-training offline in step 3 of the data model. The data model, specifically the time-series sequence of the L1 norm ratio of the defect region relative to the reference region output in step 3, is used for regression training of an LSTM network with the manually labeled "time the defect was discovered," yielding a value for the "time the defect was discovered" without human intervention. It contains multiple layers of LSM neural units to learn the mapping relationship between thermal imaging sequences and defect depth. The model is trained under supervised supervision using a large amount of sample data with known defects of different depths, continuously adjusting the model parameters to enable it to predict the corresponding defect burial depth based on the temperature change over time. During actual inspection, the thermal imaging sequence of the workpiece 1 is input into the trained LSM neural network model. The model, combined with the mapping relationship curve between the "time the defect was first visible" obtained in step 6 and the defect depth, outputs the predicted defect depth.

[0016] The LSTM network is trained using simulation data. Through supervised learning, it utilizes a large amount of sample data with known defects of varying depths to train the model for predicting defect depths. During training, the model automatically optimizes its parameters to ensure high-accuracy prediction results.

[0017] The training process of LSTM networks relies on existing deep learning training methods. During training, a large amount of simulation data is used, and the model parameters are continuously optimized through supervised learning to ensure high-accuracy predictions. The training methods and procedures are standard skills for those skilled in the art.

[0018] As another preferred embodiment, the present invention controls and records the transmission of pulse signals by computer during data acquisition, and marks the specific time of pulse signal transmission in the time domain sequence.

[0019] As another preferred embodiment, for materials with high thermal conductivity, the power of the pulsed thermal excitation device 2 is increased or the time interval of infrared thermal image acquisition is shortened to capture a faster temperature decay process; for samples with low thermal conductivity, the acquisition time of the infrared camera 1 is extended to cover the complete cooling curve. For example, the excitation time is 2 ms, and the instantaneous energy released is approximately 20 J. For materials with high thermal conductivity, the recommended heating range is 20°C to 40°C, and the excitation time should always be kept at least two orders of magnitude less than the cooling time.

[0020] As another preferred approach, this invention improves the predictive versatility for samples made of different materials by establishing material-specific models, adding material parameters to a general model, or performing linear correction based on the typical thermal response deviation of the material after the general model outputs. Samples made of different materials may exhibit varying relationships between thermal characteristics and depth due to differences in thermal conductivity and diffusion properties. In specific implementations, material-specific models are established to improve predictive versatility. For example, for ABS plastic, carbon fiber composites, and PMMA plexiglass, a set of calibration blocks with known defect depths is prepared, and training data is obtained to adjust or fine-tune the step model, making its predictions more accurate for that material. Another strategy is to add material parameters to the general model: for example, using the material's thermal diffusivity as an input, or performing linear correction based on the material's typical thermal response deviation after the model outputs. Through these measures, the system of this invention can maintain good predictive performance when applied to different materials. In practical applications, the operator only needs to select the corresponding material settings, and the system can call the model or parameters optimized for that material to perform depth calculations, without needing to rebuild the detection system, demonstrating the flexibility and practical value of this method.

[0021] As another preferred embodiment, the pulsed thermal excitation device of the present invention uses a pulsed laser or a high-energy flash lamp, and the infrared imaging device uses an infrared thermal imager.

[0022] As another preferred embodiment, in steps 1 and 2 of this invention, the instantaneous heating time of the sample being tested (e.g., 2 ms) is more than two orders of magnitude shorter than the cooling time, thereby improving accuracy. Instantaneous heating means applying sufficient heat in a very short time.

[0023] As another preferred embodiment, in step 5 of the present invention, the typical defect depth is a standard defect of different depths, ranging from 0.5 mm to 2.2 mm. These depth standards are used to establish the relationship between the LSTM network and the defect depth.

[0024] Secondly, in step 2 of this invention, the image processing algorithm employs a grayscale mutation analysis algorithm, a shape recognition algorithm, or a threshold segmentation algorithm.

[0025] In addition, in step 3 of this invention, the L1 norm ratio of the defect region to the reference region is calculated, that is:

[0026]

[0027] Where T(t) represents the average temperature of the corresponding region at time t. The ratio sequence R(t) is input into the neural network as a feature vector reflecting the difference in thermal diffusion.

[0028] In step 1 of this invention, the excitation power of the pulsed thermal excitation device does not exhibit significant power attenuation during multiple heating cycles. Pulse excitation only requires one heating cycle. This does not refer to a specific region of a sample requiring multiple heating cycles, but rather that different defect regions must have sufficiently identical excitation heat sources when other conditions, such as the distance between the sample and the pulsed excitation device, are the same. For example, if a sample is large and needs to be divided into 100 regions for testing, this method requires that the heat sources for these 100 regions be sufficiently consistent. As long as the pulsed heat source meets the above requirements, the goal of this method can be achieved. Multiple heating cycles are only necessary because the heating area is not large; in most cases, a sample needs to be divided into multiple regions for testing, and each region only requires one pulse excitation cycle.

[0029] In step 2 of this invention, the specific frame rate selected has little impact on the final prediction result. For example, after pulse excitation, the outline of the defect is relatively clear after cooling for 10 seconds and 12 seconds. Whether the 10-second frame or the 12-second frame is selected, the extraction of the outline has little impact on the accuracy of depth prediction.

[0030] The beneficial effects of this invention.

[0031] The predictive modeling platform of this invention runs an LSTM deep learning model to analyze the extracted thermal signal ratio time series and output the prediction result of the first visible time of the defect; the depth correction module calculates the defect depth based on the predicted first visible time and the pre-stored time-depth mapping curve, and makes necessary corrections to the prediction model according to the thermal properties of the material.

[0032] This invention utilizes the norm ratio of the thermal signals from the defect area to the non-defect area within the heating range as a feature, improving the robustness of defect signal discrimination under different environments. It introduces an LSTM time series model to automatically determine the defect occurrence time, reducing human intervention and empirical judgment errors, and achieving intelligent defect detection and depth assessment. By converting time delay into depth through an empirical mapping curve, the method is intuitive and efficient, and can be easily adapted to different materials with minimal calibration. Furthermore, finite element simulation is used to perform parameter scanning analysis on defects of different depths, verifying and optimizing the time-depth mapping relationship, reducing the cost of calibrating a large number of samples. In addition, the method and system of this invention have good portability and flexibility; the algorithm framework can be implemented on various simulation and image processing platforms, not limited to a specific dedicated software environment, facilitating integration and deployment in existing detection equipment. In summary, this invention provides a practical and adaptable defect depth prediction scheme. This method is suitable for low-cost, rapid detection of low-depth defects in mass-produced items or large-area materials with identical material parameters, and has broad application prospects in areas such as delamination problems and the inspection of the outer shell of critical equipment. Attached Figure Description

[0033] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. The scope of protection of the present invention is not limited to the following description.

[0034] Figure 1 This is a schematic diagram of the device structure of the present invention. In the diagram, 1-a sample with defects, 2-infrared imaging equipment, 3-pulse excitation lamp, 4-pulse excitation lamp capacitor and control components, and 5-a computer that simultaneously acquires data from the infrared thermal imaging equipment and pulse time points.

[0035] Figure 2 This is a schematic diagram showing the ratio of the L1 norm of the defect region to the reference region of the example material at different depths.

[0036] Figure 3 This is a schematic diagram illustrating the fitting of the present invention to 300 sets of simulation datasets based on the predicted time values ​​of the LSTM network.

[0037] Figure 4 This invention uses a prediction curve to predict the depth of the thermal imaging detection result sequence at different defect depths in the output value of the LSTM network. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0039] like Figure 1 As shown, the defect depth prediction system of the present invention includes: an excitation device 3 (e.g., a pulsed laser or a high-energy flash lamp) for pulse heating, an infrared imaging device 2 for thermal signal acquisition, a computer 5 for synchronously recording the pulse excitation timing sequence and thermal image sequence, and a matching image processing and deep learning model analysis module. The pulse excitation device 3 provides stable and uniform heating power through a pulse excitation power supply and control module 4 (the control module 4 can be a power supply control switch for the flash lamp's battery. The computer 5 controls the battery to supply power to the flash lamp and simultaneously records when the flash begins, i.e., the moment the dataset begins heating), ensuring the consistency of thermal excitation in multiple samplings. The computer 5 accurately calibrates the excitation start time point through a synchronous control mechanism, ensuring the acquisition of a complete and time-aligned infrared thermal image sequence.

[0040] Step 1: Apply pulsed thermal excitation to the sample 1 under test, and the infrared imaging device 2 continuously acquires the surface temperature distribution of the sample at a fixed frame rate to form a thermal image sequence. The thermal excitation has a short duration and high energy, which can excite the thermal response process of the sample in a short time.

[0041] Step 2: At the end of the cooling phase of the thermal imaging sequence, when the temperature difference is significant, extract the contour of the defect region from the thermal image. This step can be done by manually selecting obvious defects or by automatically identifying them using image processing algorithms (such as grayscale change analysis, shape recognition, threshold segmentation, etc.). The extracted defect region and the adjacent defect-free reference region will serve as the basis for subsequent analysis.

[0042] Step 3: Extract the average temperature change curves of the defect area and the reference area throughout the entire thermal imaging sequence. Then calculate the L1 norm ratio of the temperature signals of the two areas in each frame, i.e.:

[0043]

[0044] Where T(t) represents the average temperature of the corresponding region at time t. The ratio sequence R(t) is input into the neural network as a feature vector reflecting the difference in thermal diffusion.

[0045] Step 4: Input the ratio sequence R(t) into the trained LSTM neural network model, perform time series analysis, and output the "first visible time point" t of the defect. d This moment corresponds to the location where the defect first becomes clearly visible on the thermal map, reflecting its dynamic response during thermal diffusion.

[0046] Step 5: For typical materials (such as ABS) and multiple defects of known depths (equivalent to anchor points in the mapping relationship between "defect discovery time" and "defect depth"), parametric modeling is performed using finite element simulation to obtain the first visible time t corresponding to different depths. d By using fitting techniques such as the least squares method, a relationship between depth d and t is established. d Mapping relationship between them (e.g.) Figure 2 (As shown). The relationship is usually exponential:

[0047]

[0048] Where a and b are both parameters; after fitting, a high-precision fitting curve (Ri) is obtained. 2 (It can reach over 0.998), providing a mathematical basis for predicting defect depth.

[0049] Step 6: For samples with defects of unknown depth, repeat steps 1-4 to obtain the first visible time t of the defect. d The depth is directly predicted using the mapping curve obtained from step 5. For example... Figure 3 As shown, for large-scale simulation datasets, this invention establishes a stable and high-precision fitting relationship between LSTM predictions and actual defect depths; Figure 4 The results demonstrate the predictive performance of this method on defects of different depths in experimental thermal imaging sequences, verifying the accuracy of the proposed method.

[0050] Material adaptation and correction mechanism: For homogeneous materials with similar parameters (such as similar polymers or composites), this invention only requires three or more sets of calibration samples at different depths, and the mapping curve can be adjusted in step 6 to achieve rapid adaptation. For heterogeneous or complex structural materials, step 5 can be repeated to perform new modeling. The LSTM network of this invention itself has strong cross-material adaptability and does not need to be retrained in most cases.

[0051] System integration features: The entire method of this invention can be embedded into an industrial infrared inspection system. The equipment consists of a heat source, a thermal imager, a data synchronization and control module, image processing software, and a defect prediction platform. All modules can be integrated in the same software interface, making it suitable for automated inspection or batch evaluation.

[0052] This invention extracts the ratio of the relative norm at the defective location to that at the non-defective location, uses this ratio as a training feature, thereby truly quantifying the "time it takes for a defect to be discovered," and also uses simulation to find the mapping relationship between the "time it takes for a defect to be discovered" and the "defect depth."

[0053] This invention rationally quantifies the "time of defect identification" and uses batch simulation to reasonably map this indicator to the depth of the defect; it also has correction strategies for materials with different properties.

[0054] Example:

[0055] A defect depth prediction method based on infrared thermography and LSTM technology was validated in a thermal imaging inspection experimental setup. The sample used was an ABS thermoplastic sample manufactured by SLA 3D printing, with a total thickness of 3.0 mm. It contained multiple artificial defects (cylindrical cavities) with depths ranging from 0.5 mm to 2.2 mm, spaced 0.1 mm apart. Each defect had a diameter of approximately 3 mm, and its surface was coated with a matte finish to enhance infrared absorption.

[0056] Step 1: Dual flash lamp pulse excitation (6.4 kJ output energy per lamp, pulse width approximately 2 ms) was used to instantaneously heat the sample surface; an infrared thermal imager (1.5–5.4 μm band, 640 × 512 pixels resolution, 50 Hz frame rate) was used to acquire 30 seconds of cooling process data, recording a total of 1500 frames of infrared thermal images. All thermal image data and pulse start signals were controlled and recorded by a connected computer.

[0057] Step 2: After acquisition, the 1200th frame of the thermal image is selected, as the defect outline in the temperature field of this frame is the clearest and most obvious, and it is used as a reference image for defect area identification. The researchers used the Hough circle algorithm to select each defect area and manually selected reference areas in adjacent defect-free areas, and extracted the average temperature sequence of the two areas in the entire time domain.

[0058] Step 3: Calculate the L1 norm for the temperature sequences of the defect region and the reference region respectively, and form a ratio sequence reflecting the difference in thermal diffusion based on the time point ratios. Input the ratio sequence into the trained LSTM neural network model, and the model outputs the predicted "first visible time" for each set of data. This value is the moment when the defect first appears significantly in the thermal image, in seconds.

[0059] Step 4: Using finite element method (FEM) simulation, a corresponding ABS material model was constructed in COMSOL Multiphysics software to simulate the heating response behavior of defects at different depths, collecting a total of 300 sets of simulation data. A mapping relationship was established between the first visible time predicted by the LSTM network and the known depth, and the R-value under various mapping conditions was evaluated. 2 Values, select the exponential mapping relationship and fit:

[0060]

[0061] Where d is the defect depth, t is the time point of first visibility, and the fitted parameters are a = 0.21, b = 0.65, and the fitting coefficient R0 is 0.65. 2 Up to 0.998.

[0062] Step 5: Input the predicted time of the experimental sample into the mapping curve to obtain the predicted depth value of each defect. Compare the predicted value with the actual defect depth of the sample. For the case where the depth of 15 other defects was predicted using 3 known defects, at a 95% confidence level, the depth ranges corresponding to each time point are: 1 second, depth range is [0.39, 0.45]; 2 seconds, depth range is [0.67, 0.76]; 4 seconds, depth range is [1.08, 1.23]; 6 seconds, depth range is [1.38, 1.58]; 8 seconds, depth range is [1.62, 1.84]. The accuracy within the corresponding depth range is controlled within ±0.11 mm, verifying the accuracy of this method.

[0063] It is understood that the above specific description of the present invention is only for illustrating the present invention and is not limited to the technical solutions described in the embodiments of the present invention. Those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention to achieve the same technical effect; as long as the use needs are met, they are all within the protection scope of the present invention.

Claims

1. A method for predicting defect depth based on pulsed infrared thermographic sequences, characterized in that... Includes the following steps: Step 1: Instantly heat the sample under test using a pulsed thermal excitation device, and non-contactly acquire the infrared thermal image sequence of the sample surface using an infrared imaging device; Step 2: During the cooling process, select a frame with a strong thermal contrast between the defect and the background when the defect outline is clear during the cooling process after pulse excitation. Use image processing algorithms to identify the outline of the defect area in the thermal image and determine the corresponding defect-free reference area. Step 3: Extract the temperature-time signal curves of the defective region and the non-defective reference region over the entire time range; for the two sets of time-domain temperature signals obtained, calculate the L1 norm ratio of the defective region to the non-defective region within the heating range under different defect depth conditions, that is, the ratio of the sum of the absolute values ​​of the average temperature signals of the defective region and the non-defective region at each time point, thereby forming a ratio time series reflecting the difference in thermal response between the two over time. Step 4: Input the above ratio time series into the pre-trained LSTM network model to perform dynamic feature analysis on the defect and predict the first visible moment of the defect; the first visible moment is defined as the time point when the defect sign is first clearly detectable in the thermal image sequence, reflecting the time delay experienced by the defect from being heated to becoming visible; Step 5: Perform extensive 3D finite element simulation analysis on typical defect depths, and use a parametric scanning method for defect depth to model the curve of "first visible moment of defect" as output by the LSTM network model described in Step 4 with respect to defect depth; and calculate the result determination coefficient R. 2 Find the depth of the defect d With the first visible moment t d The mapping relationship between two variables; Step 6: Using the mapping relationship in Step 5 and the known depth of the defect, establish and calibrate the mapping relationship curve between the "first visible time of the defect" and the defect depth of the current material; through pulse excitation, collect data on the defect of unknown depth, input the ratio time series into the "first visible time of the defect" output by the LSTM network model, and input the mapping relationship curve between the "first visible time of the defect" and the defect depth to quickly and accurately predict the defect depth.

2. The defect depth prediction method based on pulsed infrared thermographic sequences according to claim 1, characterized in that... For different isotropic and uniform materials with thermal conductivity, density, and specific heat capacity within the same order of magnitude, step 6 is performed using three or more sets of known-depth calibration defects to correct the defect depth prediction curve without repeating step 5.

3. The defect depth prediction method based on pulsed infrared thermographic sequences according to claim 1, characterized in that... In step 4, the LSTM network model is trained using simulation data. The LSTM network model finds the value of the "first visible moment of the defect" based on the time domain sequence, and performs a specific quantification on this value. Furthermore, for samples of different materials, even if the thermal conduction and diffusion characteristics are different, the principle of the time delay value experienced by the LSTM network model from heating to manifestation is the same, and there is no need to retrain.

4. The defect depth prediction method based on pulsed infrared thermographic sequences according to claim 1, characterized in that... The LSTM network model is obtained by pre-training offline in step 3 of the data model. The data model is the time series of the L1 norm ratio of the defective region to the non-defective reference region output by step 3. The LSTM network model is then used for regression training with the artificial label "defect first visible time" to obtain the value of "defect first visible time" without human intervention.

5. The defect depth prediction method based on pulsed infrared thermographic sequences according to claim 1, characterized in that... For materials with high thermal conductivity, increase the power of the pulsed thermal excitation device or shorten the time interval of infrared thermal image acquisition to capture a faster temperature decay process; for samples with low thermal conductivity, extend the acquisition time of the infrared camera to cover the complete cooling curve.

6. The defect depth prediction method based on pulsed infrared thermographic sequences according to claim 1, characterized in that... For samples of different materials, the general applicability of predictions can be improved by establishing material-specific models, adding material parameters to general models, or performing linear correction based on the typical thermal response deviation of the material after the output of the general model.

7. The defect depth prediction method based on pulsed infrared thermographic sequences according to claim 1, characterized in that... In steps 1 and 2, the instantaneous heating time of the sample being tested is more than two orders of magnitude shorter than the cooling time.

8. The defect depth prediction method based on pulsed infrared thermographic sequences according to claim 1, characterized in that... In step 5, the typical defect depth refers to standard defects of different depths, ranging from 0.5 mm to 2.2 mm. These depth standards are used to establish the relationship between the LSTM network model and the defect depth.

9. The defect depth prediction method based on pulsed infrared thermographic sequences according to claim 1, characterized in that... In step 2, the image processing algorithm employs a grayscale mutation analysis algorithm, a shape recognition algorithm, or a threshold segmentation algorithm.

Images