Nondestructive testing system for substrate under coating based on terahertz spectrum and infrared thermal image

Abstract

The present application belongs to the technical field of industrial nondestructive testing, and relates to a coating-substrate nondestructive testing system based on terahertz spectroscopy and infrared thermal imaging, which comprises a detection data generation module, a thermophysical property inversion module, a dynamic feature extraction module, a feature fusion registration module, a defect classification and determination module, a regional scanning splicing module and a visual imaging module. The modules are sequentially executed to cooperatively detect, invert thermophysical parameters, extract terahertz dynamic features, register and fuse multiple source features, determine defects based on physical correlation, splice and fuse full-area scanning, and finally generate diagnostic images that can intuitively distinguish defect types, thereby solving the problems of information limitation of single detection method and difficulty in realizing defect classification under complex working conditions.

Description

Technical Field

[0001] This invention belongs to the technical field of industrial non-destructive testing, and relates to a non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography. Background Technology

[0002] Composite coating structures, such as thermal barrier coatings on aero-engine blades and anti-corrosion coatings on large equipment substrates, may develop various types of defects during service, including porosity, delamination, debonding, and cracks, at the coating body, the coating-substrate interface, and within the substrate. Accurate identification and classification of these multi-type, multi-scale defects are crucial for assessing structural integrity and preventing structural failure.

[0003] Currently, the industry commonly uses terahertz time-domain spectroscopy and pulsed thermography to independently inspect coating structures. Terahertz time-domain spectroscopy utilizes the penetrating power of terahertz waves through most non-metallic and non-polar coating materials. By analyzing the time-domain echo characteristics in the reflected signal, it can obtain information on coating thickness and internal interface reflection. However, its sensitivity to subtle changes in the physical properties of the coating material is usually limited. Pulsed thermography, on the other hand, applies transient thermal excitation to the coating surface. By analyzing the transient changes in the surface temperature field, it inverts the thermophysical parameters of the coating and substrate. It is more sensitive to the thermal conductivity of the material and the interfacial thermal contact, but its ability to detect defects in deeper areas below the coating surface is relatively weak.

[0004] The two methods described above have certain limitations in practical applications. While terahertz detection can detect interface reflection signals, when defects do not cause electromagnetic impedance mismatch, relying solely on the reflection information from a single interface often makes it difficult to distinguish the defect type corresponding to minute signal changes, such as coating pores and weak interface adhesion. The thermophysical parameters obtained by pulsed thermography are essentially equivalent overall parameters of the coating. Their spatial resolution and depth sensitivity are limited by the temporal and spatial scales of the thermal diffusion process. When the defect scale near the interface is small, the resulting surface temperature rise may be masked by the non-uniform noise of the material itself. Furthermore, using the two techniques independently and simply comparing the results usually involves superposition or comparison at the information level, which is insufficient at the information fusion level. This may lead to information redundancy or contradictions, and it is difficult to fully establish the intrinsic physical relationship between thermal and electromagnetic responses. Therefore, there is still room for improvement in the ability to classify defects under complex working conditions. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography.

[0006] A non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography includes:

[0007] The detection data generation module performs collaborative detection timing for the detection pixels of the composite coating structure under test, and generates a joint detection dataset including time-domain reference signal, thermal modulation state time-domain signal and surface temperature sequence.

[0008] The thermal property inversion module inverts and generates a surface thermal property parameter map characterizing the thermal physical properties of the detected pixel based on the surface temperature sequence in the joint detection dataset.

[0009] The dynamic feature extraction module calculates and extracts the dynamic change features of the terahertz signal caused by thermal excitation based on the time-domain reference signal and the thermally modulated time-domain signal in the joint detection dataset, and generates a terahertz dynamic differential feature map.

[0010] The feature fusion and registration module spatially registers and fuses the surface thermal property parameter map with the terahertz dynamic differential feature map to generate an aligned feature map that includes multi-dimensional feature parameters.

[0011] The defect classification and determination module executes physical association-based defect determination logic based on the combination of multi-dimensional feature parameters of each pixel in the aligned feature map, and generates a defect classification index for distinguishing defect types.

[0012] The area scanning and stitching module traverses the entire detection area of ​​the composite coating structure under test, and stitches together the defect classification index of all detected pixels to generate a defect distribution index map covering the entire detection area.

[0013] The visualization imaging module assigns preset visual identifiers and texture codes to different types of defect pixels based on the defect type markers in the defect distribution index map, generating a fused diagnostic image.

[0014] In a further embodiment of the present invention, the detection data generation module is specifically configured to perform the following operations:

[0015] The integrated scanning probe is driven to position itself at the detection pixel, triggering the terahertz time-domain spectroscopy module within it to acquire the reflection signal of that point in a non-thermal excitation state as a time-domain reference signal.

[0016] A pulsed thermal excitation source is synchronously activated to apply a thermal pulse, and an infrared thermal imager records the surface temperature at that point from the start of the thermal pulse to a certain period of time after the pulse ends, forming a surface temperature sequence.

[0017] Based on the coating thickness and thermal diffusivity of the composite coating structure under test, the characteristic time delay of heat diffusion to the coating-substrate interface is calculated. At this characteristic time delay, the reflected signal under the influence of the thermal pulse is collected again as the thermally modulated time domain signal.

[0018] In a further embodiment of the present invention, the thermal property inversion module is specifically configured to perform the following operations:

[0019] Extract cooling curve data after the thermal pulse ends from the surface temperature sequence;

[0020] The cooling curve data were fitted with the theoretical curve of the one-dimensional heat conduction model;

[0021] Based on the fitting results, the apparent thermal diffusivity and surface temperature rise of the detected pixel are derived, and a surface thermal property parameter map including the distribution of the apparent thermal diffusivity and surface temperature rise is generated.

[0022] In a further embodiment of the present invention, the dynamic feature extraction module is specifically configured to perform the following operations:

[0023] Time alignment processing is performed between the time-domain reference signal and the thermally modulated time-domain signal;

[0024] The two aligned signals are subtracted point by point to obtain the time-domain difference signal;

[0025] Identify the echo time window corresponding to the coating-substrate interface in the time-domain reference signal;

[0026] Within the echo time window of the time-domain differential signal, the differential peak amplitude of the signal is extracted, and the average phase difference between the time-domain reference signal and the thermally modulated time-domain signal within this window is calculated as the differential phase offset to generate a terahertz dynamic differential feature map.

[0027] In a further embodiment of the present invention, the feature fusion and registration module is specifically configured to perform the following operations:

[0028] The surface thermal property parameter map and the terahertz dynamic differential feature map are spatially registered at the pixel level to unify the coordinate system;

[0029] For each pixel, the apparent thermal diffusivity and surface temperature rise of the coating are read from its corresponding surface thermal property parameter map, and the differential peak amplitude and differential phase offset are read from its corresponding terahertz dynamic differential feature map.

[0030] The four parameters are combined according to pixel position to generate an aligned feature map for each pixel, which includes a four-dimensional feature vector.

[0031] In a further embodiment of the present invention, the defect classification and determination module is configured to perform the following operations:

[0032] Read the coating apparent thermal diffusivity, differential peak amplitude, and differential phase offset of the currently detected pixel;

[0033] If the apparent thermal diffusivity of the coating is lower than the preset lower limit of the normal threshold, and the differential peak amplitude and differential phase offset are both within the preset normal value range, then the pixel is determined to have a coating defect.

[0034] In a further embodiment of the present invention, the defect classification and determination module is also configured to perform the following operations:

[0035] Read the surface temperature rise value, differential peak amplitude, and differential phase offset of the currently detected pixel, and calculate the average surface temperature rise value of the surrounding pixels.

[0036] If the surface temperature rise is higher than the average surface temperature rise by a preset amount, and the absolute values ​​of the differential peak amplitude and the differential phase offset both exceed their respective preset abnormal thresholds, then the pixel is determined to have an interface bonding defect.

[0037] In a further embodiment of the present invention, the defect classification and determination module is also configured to perform the following operations:

[0038] Examine the temporal reference signal of the currently detected pixel and read all feature parameters in the aligned feature map.

[0039] If a structural reflection anomaly is identified within a time period later than the interface echo in the time-domain reference signal, but the coating apparent thermal diffusivity, surface temperature rise amplitude, differential peak amplitude, and differential phase offset in the alignment feature map do not trigger the anomaly judgment threshold, then the pixel is determined to have an internal defect in the substrate.

[0040] In a further embodiment of the present invention, the region scanning and stitching module is specifically configured to perform the following operations:

[0041] Based on the preset scanning path, the integrated scanning probe is driven to move to each detection pixel in row and column order; the defect classification index value of each point is obtained;

[0042] Based on the spatial coordinates of each pixel, all defect classification index values ​​are arranged into a two-dimensional matrix corresponding to the scanned area, forming a defect distribution index map.

[0043] In a further embodiment of the present invention, the visualization imaging module is specifically configured to perform the following operations:

[0044] A grayscale background is generated for defect-free pixels by mapping the apparent thermal diffusivity of their coating.

[0045] Pixels marked as defects in the coating body are rendered as the first preset visual identifier;

[0046] Pixels marked as interface adhesion defects are rendered as a second preset visual identifier, the brightness of which is positively correlated with the differential peak amplitude.

[0047] Pixels marked as internal defects of the matrix are rendered as third preset visual identifiers;

[0048] The rendered defective pixels are superimposed onto a grayscale background to generate a fused diagnostic image.

[0049] In summary, the present invention has the following beneficial technical effects:

[0050] 1. By designing a collaborative detection timing sequence, terahertz signals are acquired at characteristic time points when heat diffuses to the coating-substrate interface, so that the obtained "thermally modulated time-domain signal" contains modulation information of thermally induced changes in the physical state of the interface. The differential features extracted from this signal can reflect information such as thermal expansion and slight changes in dielectric constant at the interface, thus helping to distinguish weak bonding defects at the interface that are difficult to identify in traditional terahertz detection.

[0051] 2. By inverting the cooling stage data of the surface temperature sequence, the distribution information of the coating's apparent thermal diffusivity and surface temperature rise amplitude is obtained. The generated thermophysical parameter map can reflect the coating's own thermal properties and its internal uniformity. This parameter map can serve as an independent detection dimension to identify coating defects that are mainly manifested as abnormal thermal diffusivity, thus complementing the information from the terahertz dynamic differential features.

[0052] 3. By performing pixel-level registration and feature fusion between surface thermal property parameter maps and terahertz dynamic differential feature maps, an aligned feature map containing multi-dimensional physical parameters is constructed, and cross-validation is performed based on physical association rules. This method utilizes the differences in the influence patterns of different defect types on thermal and electromagnetic parameters, which helps to suppress false alarms caused by normal material inhomogeneities, improves the classification accuracy of coating bulk defects, interface bonding defects, and matrix internal defects, and enables the differentiation of the physical origin of defects. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This discloses a schematic diagram of the framework in the embodiments of this application.

[0055] Figure 2 This discloses a flowchart of an embodiment of this application. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] The following is in conjunction with the appendix Figures 1-2 A preferred description of the present invention is provided below.

[0058] See attached document Figures 1-2 This invention proposes a non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography, comprising the following modules:

[0059] The detection data generation module performs collaborative detection timing for the detection pixels of the tested composite coating structure, generating a joint detection dataset including a time-domain reference signal, a thermally modulated time-domain signal, and a surface temperature sequence. The thermal property inversion module, based on the surface temperature sequence in the joint detection dataset, inverts and generates a surface thermal property parameter map characterizing the thermophysical properties of the detection pixel. The dynamic feature extraction module, based on the time-domain reference signal and the thermally modulated time-domain signal in the joint detection dataset, calculates and extracts the dynamic change features of the terahertz signal induced by thermal excitation, generating a terahertz dynamic differential feature map. The feature fusion and registration module integrates the surface thermal property parameter map with the terahertz dynamic differential feature map. Spatial registration and feature fusion are performed to generate an aligned feature map including multi-dimensional feature parameters; the defect classification and judgment module executes a physically related defect judgment logic based on the combination of multi-dimensional feature parameters of each pixel in the aligned feature map to generate a defect classification index for distinguishing defect types; the region scanning and stitching module traverses the entire detection area of ​​the tested composite coating structure and stitches together the defect classification indices of all detected pixels to generate a defect distribution index map covering the entire detection area; the visualization imaging module assigns preset visual identifiers and texture codes to different types of defect pixels according to the defect type markers in the defect distribution index map to generate a fused diagnostic image.

[0060] In one embodiment of the present invention, the detection data generation module is specifically configured to perform the following operations:

[0061] The integrated scanning probe is driven to position itself at the detection pixel, triggering the terahertz time-domain spectroscopy module within it. The reflected signal of this point in the absence of thermal excitation is collected as a time-domain reference signal. Simultaneously, a pulsed thermal excitation source is activated to apply a thermal pulse, and an infrared thermal imager records the surface temperature of this point from the start of the thermal pulse to a certain time after its end, forming a surface temperature sequence. Based on the coating thickness and thermal diffusivity of the composite coating structure under test, the characteristic time delay of heat diffusion to the coating-substrate interface is calculated. At this characteristic time delay, the reflected signal under the influence of the thermal pulse is collected again as a thermally modulated time-domain signal.

[0062] Specifically, the collaborative detection of a single detection point includes: deploying an integrated scanning probe, which is a mechanical device that rigidly connects a pulsed thermal excitation source, an infrared thermal imager, and a terahertz time-domain spectroscopy module, ensuring that the detection axes of the three intersect at the same focal point. During detection, the terahertz time-domain spectroscopy module in the integrated scanning probe is positioned directly above the detection point. This terahertz time-domain spectroscopy module is based on the femtosecond laser pump detection principle and can generate and detect broadband terahertz pulses, acquiring the complete waveform of their electric field intensity over time. Timing and motion control are implemented through a programmable logic controller (PLC) or industrial computer, which sends start and stop commands to each hardware unit according to a preset logical sequence.

[0063] Next, a trigger command is sent to the terahertz time-domain spectroscopy module, which emits a femtosecond laser pulse to excite the photoconductive antenna to generate a broadband terahertz pulse. The pulse is incident perpendicularly on the surface of the sample under test. At this time, part of the energy is reflected on the upper surface of the coating, and the other part of the energy is transmitted into the interior of the coating. When the transmitted terahertz pulse encounters the interface between the coating and the substrate during propagation, it will generate a reflected echo due to the sudden change in the dielectric constant of the material.

[0064] The terahertz detection unit will receive all reflected signals, including those reflected from the upper surface and those reflected from the interface, and convert them into a voltage signal that varies with time using terahertz detection technology based on photoconductive antennas or electro-optic sampling. This signal is the time-domain reference signal, which is a discrete sequence of one-dimensional voltage amplitude with respect to time. Its time length covers the entire process from the emission of the terahertz pulse to the reception of all valid echoes. The sampling interval of the data points is determined by the delay line step accuracy of the terahertz time-domain spectroscopy system.

[0065] Subsequently, the pulsed thermal excitation source and the infrared thermal imager are started simultaneously. The pulsed thermal excitation source releases a rectangular thermal pulse with an energy density sufficient to cause a detectable temperature rise on the surface under test within a time width of milliseconds, and uniformly irradiates the surface of the detection point. At the same time, the infrared thermal imager begins to acquire thermal images at a fixed rate not lower than that required by the sampling theorem, such as more than 100 frames per second. After the thermal pulse ends, heat diffuses from the coating surface inwards and towards the substrate. The infrared thermal imager continuously records this process until it obtains a curve showing the surface temperature change over time, which includes the complete cooling process—that is, the surface temperature sequence. It should be noted that the surface temperature sequence is a one-dimensional array output by the infrared thermal imager, recording the temperature change of individual pixels over time after the thermal pulse. Its duration and sampling rate are set by the frame rate and recording duration of the infrared thermal imager. A higher frame rate helps to satisfy the sampling theorem to accurately capture the details of transient temperature changes caused by the thermal pulse. A pulsed thermal excitation source is a device that can release a high-energy-density light spot within milliseconds to achieve rapid heating of the material surface. An infrared thermal imager is a device that uses an infrared detector array to convert the infrared radiation emitted by the surface of an object into a temperature distribution image.

[0066] Subsequently, based on the pre-calculated characteristic time delay, at the point after the thermal pulse ends, the control unit triggers the terahertz time-domain spectroscopy module again. This module repeats the process of transmitting and receiving terahertz pulses, collects the reflected signal at this moment, and obtains the thermally modulated time-domain signal. The thermally modulated time-domain signal is a one-dimensional voltage amplitude sequence with the same time length and sampling format as the time-domain reference signal, and its acquisition time is strictly controlled by the characteristic time delay.

[0067] Finally, the time-domain reference signal, thermally modulated time-domain signal, and surface temperature sequence obtained in the above steps are packaged according to a unified time reference to form a data file containing three data types and indicating the spatial coordinates of the detection points, namely the joint detection dataset. The joint detection dataset is a structured electronic information file, the contents of which include at least three data arrays: time-domain reference signal, thermally modulated time-domain signal, and surface temperature sequence, and a unified timestamp index is established for the three.

[0068] To ensure that heat diffuses precisely to the interface between the coating and the substrate after the thermal pulse ends, its characteristic time delay needs to be calculated. The characteristic time delay is equal to the square of the coating thickness divided by four times the thermal diffusivity of the coating material, expressed as:

[0069]

[0070] Among them, coating thickness This refers to the vertical distance from the top surface of the coating to the interface between the coating and the substrate. Its value is obtained from known design drawings or measured in non-critical areas of the sample using a micrometer or eddy current thickness gauge. For common industrial anti-corrosion coatings or thermal barrier coatings, The typical range is between 50 micrometers and 500 micrometers. For example, for a thermal barrier coating on an aero-engine blade, its nominal design thickness is... The thermal diffusivity of the coating material is 200 micrometers. This is a physical parameter characterizing the thermal conductivity of a coating material. Its value is based on typical values ​​for the coating system published in the material handbook or obtained by calibration on standard samples using an independent flash thermal conductivity meter. For example, for epoxy resin-based coatings, the typical value ranges from 0.1 to 1.0 mm² / s. Characteristic time delay It is the time point calculated based on the first two parameters. Its physical meaning is the time required for heat to diffuse to a characteristic depth after an instantaneous heat source is applied to the surface. At this moment, the temperature rise response at the interface is the most obvious, which is used to trigger the second terahertz acquisition.

[0071] For example, the test was conducted on a sample with a nominal coating thickness of 200 micrometers and a material thermal diffusivity α of 0.8 square millimeters per second. The coating thickness was set. The value is 200 micrometers, or 0.0002 meters. This sets the thermal diffusivity of the coating material. It is 0.8 square millimeters per second, or 0.8 multiplied by 10. -6 Square meters per second. Substituting the above values ​​into the formula, the characteristic time delay is calculated. Approximately 0.0002 squared divided by 4 multiplied by 0.8 multiplied by 10 -6 The control unit then triggers terahertz acquisition precisely at 15.6 milliseconds after the thermal pulse ends, obtaining the thermally modulated time-domain signal. The infrared thermal imager records the surface temperature sequence at a frame rate of 200 Hz, acquiring a total of 61 frames of temperature data within a total duration of 300 milliseconds. The final encapsulated joint detection dataset includes a time-domain reference signal for this specific detection point, a thermally modulated time-domain signal acquired with a 15.6 millisecond delay, and a surface temperature sequence array of length 61.

[0072] In one embodiment of the present invention, the thermal property inversion module is specifically configured to perform the following operations:

[0073] Extract the cooling curve data after the thermal pulse ends from the surface temperature sequence; fit the cooling curve data with the theoretical curve of the one-dimensional heat conduction model; based on the fitting results, deduce the coating apparent thermal diffusivity and surface temperature rise of the detected pixel, and generate a surface thermal property parameter map including the distribution of coating apparent thermal diffusivity and surface temperature rise.

[0074] Based on the obtained joint detection dataset, the surface temperature sequence data included therein is processed to obtain two key parameters characterizing the thermophysical properties of that point. Specifically, a data segment from the end of the heat pulse application to the end of the recording is extracted from the surface temperature sequence. This data segment records the process of the surface temperature naturally cooling down from the peak, forming a curve corresponding to time and temperature rise, i.e., the cooling curve data. Then, this measured cooling curve data is mathematically fitted with a theoretical model based on the one-dimensional heat conduction assumption.

[0075] This theoretical model is based on the physical assumption that heat is mainly conducted one-dimensionally along the vertical direction within the coating. The specific mathematical expression for its theoretical curve is: surface temperature rise during the cooling process. It's time for the temperature to drop. The function is expressed as:

[0076]

[0077] The undetermined parameters in the theoretical model, namely the surface temperature rise amplitude, are iteratively adjusted using the least squares fitting algorithm. and the apparent thermal diffusivity of the coating This minimizes the overall error between the theoretically calculated curve and the measured cooling curve data points. Finally, the successfully fitted curve is inverted... Value and The values ​​are used as quantization descriptors for the detected pixel in two different feature dimensions, and are combined to form the surface thermal property parameter map corresponding to that pixel. This map is essentially a pseudo-image of two superimposed data layers, where the first data layer stores numerical representations of... The spatial distribution of the data is such that the second data layer stores numerical representations. Spatial distribution.

[0078] in, The independent variable is time. The temperature rise function, The timing starts from the moment the thermal pulse ends. It is the first parameter that needs to be inverted through fitting, representing the surface temperature rise amplitude, that is, the temperature rise amplitude of the cooling curve at the zero point of time (after the ideal instantaneous pulse). The second parameter that needs to be obtained through fitting and inversion is the apparent thermal diffusivity of the coating. This comprehensively reflects the material's inherent thermal diffusivity as well as the impact of internal structural defects on heat transfer. For a uniform and intact coating... It should be close to the nominal value of the material. When defects exist It will decrease. The known coating thickness is calculated using the same set value as in the previous step when calculating the characteristic time delay, derived from design drawings or offline measurements. In the actual fitting operation, the measured cooling curve data points are used... Input the data into the computer and call optimization algorithms, such as the Levonberg-Marquardt algorithm, to continuously adjust the settings. and The estimated value makes the value calculated by the theoretical model differ from the measured value. Minimize the sum of squared residuals to determine the optimal solution for these two parameters.

[0079] It should be noted that the apparent thermal diffusivity of the coating is a physical parameter obtained through this inversion step, reflecting the equivalent thermal diffusivity of the coating at a specific test point. Its value may deviate from the nominal value due to material inhomogeneity or defects, and is expressed in square meters per second. The surface temperature rise amplitude is a physical parameter obtained through this inversion step, characterizing the highest temperature rise amplitude that a thermal pulse can reach on the surface at this test point. Its value is affected by factors such as incident heat flux density, coating thermal reflectivity, and heat capacity, and is expressed in Kelvin or degrees Celsius. The surface thermal property parameter map is a matrix-style data set including two independent data channels. Each pixel position includes two floating-point values, corresponding to... and The spatial resolution of the image is consistent with the scanning step size of the detection points; the cooling curve data is extracted from the surface temperature sequence, and the data of the heat pulse application stage is removed, retaining only the one-dimensional array of the time-temperature rise correspondence of the natural cooling stage; the one-dimensional heat conduction model is a mathematical model established based on the simplified physical assumption that heat is mainly conducted along the direction perpendicular to the coating surface, and is used to describe the decay law of surface temperature rise over time; the least squares fitting is a standard mathematical optimization method to estimate the model parameters by minimizing the sum of squares of the differences between theoretical predictions and actual observations.

[0080] For example, for a specific detection point in a jointly detected dataset, the cooling curve data after the thermal pulse ends is extracted from its surface temperature sequence, with the time range from the 2nd millisecond (pulse end) to the 302nd millisecond, resulting in 60 data points. The coating thickness L = 0.0002 meters is used as a known input. A fitting program is then called to process the 60 data points. With theoretical models A fitting process is performed. Through algorithm iteration, the apparent thermal diffusivity of the coating at that point is finally determined. =7.5×10 -7 The surface temperature rise is measured in square meters per second, or 0.75 square millimeters per second. =8.5 Kelvin. These two values ​​are then recorded as feature values ​​for that pixel in the surface thermal property parameter map. This example verifies the effectiveness of retrieving the apparent thermal diffusivity and surface temperature rise of the coating by fitting a cooling curve, and demonstrates the complete operational chain from joint detection dataset to generating a surface thermal property parameter map.

[0081] In one embodiment of the present invention, the dynamic feature extraction module is specifically configured to perform the following operations:

[0082] The time-domain reference signal and the thermally modulated time-domain signal are time-aligned; the two aligned signals are subtracted point by point to obtain the time-domain differential signal; the echo time window corresponding to the coating-substrate interface is identified in the time-domain reference signal; and the differential peak amplitude of the signal is extracted within the echo time window of the time-domain differential signal. And calculate the average phase difference between the time-domain reference signal and the thermally modulated time-domain signal within this window. As a differential phase offset, a terahertz dynamic differential feature map is generated.

[0083] Based on the obtained joint detection dataset, the included time-domain reference signal and thermally modulated time-domain signal are processed to extract the dynamic change features of the signal induced by thermal excitation. Specifically, two time-domain signals from the same detection pixel are time-aligned. Time alignment is achieved through a cross-correlation algorithm in digital signal processing to synchronize the signals. Its purpose is to eliminate the random time delay between two independent acquisitions and ensure the effectiveness of subsequent differential operations.

[0084] During operation, a time-domain reference signal is used as a benchmark. By calculating the cross-correlation coefficient between the reference signal and the thermally modulated time-domain signal and finding the peak value, the relative offset between the two signals on the time axis is determined. Then, by cyclically shifting the thermally modulated time-domain signal, the time base deviation caused by minor trigger jitter that may exist between the two acquisitions is eliminated, ensuring that the two signals are synchronized on the time axis. Next, point-by-point subtraction is performed. Point-by-point subtraction is a basic operation of algebraic subtraction of corresponding elements of two discrete sequences of equal length and time alignment.

[0085] After the two signals are aligned in time, at each sampling point at the same time, the voltage amplitude of the time-domain reference signal at the corresponding point is subtracted from the voltage amplitude of the thermally modulated time-domain signal to obtain a new time-domain signal sequence, namely the time-domain differential signal. The time-domain differential signal is a one-dimensional discrete sequence obtained by subtracting point by point. Its amplitude characterizes the change in the terahertz reflection signal caused by thermal excitation. This signal intuitively reflects the dynamic change of the terahertz reflection signal before and after the thermal pulse.

[0086] Subsequently, based on the known propagation speed of terahertz waves in the coating material and the coating thickness parameters, the theoretical two-way travel time of the electromagnetic wave from the coating surface to the coating-substrate interface and back is calculated in the original time-domain reference signal. Using this theoretical time as the center, a set time tolerance window is extended before and after it. The signal segment within this time window is identified and extracted. This signal segment is determined to be the area where the main echo generated by the reflection from the coating-substrate interface is located, i.e., the echo time window. It is a specific time interval calculated based on the known coating thickness and the average refractive index of terahertz waves in the coating. It is used to locate the signal segment that includes interface reflection information. The setting is based on the time-flight principle of electromagnetic wave propagation.

[0087] Within the identified echo time window, feature extraction is performed on the time-domain differential signal: On one hand, the absolute maximum value of the time-domain differential signal amplitude is directly searched within the window and recorded as the peak amplitude of the differential signal. This value directly measures the intensity of the amplitude change of the interface reflection signal caused by thermal excitation. For areas without defects or with minimal thermal modulation influence, the differential peak amplitude is close to 0. Its numerical range is determined by system noise and signal variation amplitude, with an exemplary range between -0.5 mV and +0.5 mV. On the other hand, the instantaneous phases of the time-domain reference signal and the thermally modulated time-domain signal within the window are calculated using Hilbert transform. The instantaneous phase values ​​of the two signals at the same time point are then subtracted, and finally, the average value of the phase difference within the entire window is calculated and recorded as the phase offset of the differential signal. The phase offset is calculated by comparing the average difference in phase of the interface echo signal before and after thermal excitation within the echo time window, reflecting the degree of disturbance of the signal phase by thermal modulation.

[0088] Finally, the peak amplitude value and phase offset value extracted from the pixel are used as two independent feature parameters and integrated to generate the terahertz dynamic differential feature map corresponding to the point. This map is also a data structure that includes two data layers, namely a matrix set including two independent feature parameter data layers of peak amplitude and phase offset.

[0089] For example, following the steps above, for the same detection point, both the time-domain reference signal and the thermally modulated time-domain signal acquired with a 15.6 ms delay are sequences of length N. First, cross-correlation is performed on the two signals, revealing a 0.3-sampling-point delay between the thermally modulated signal and the reference signal. Interpolation and translation are then applied to align the two signals. Next, the two aligned signals are subtracted point-by-point to obtain the time-domain difference signal. Given a coating thickness of 200 micrometers and a refractive index of approximately 2.0 for terahertz waves, the theoretical two-way travel time of the interface echo is calculated to be approximately 2.67 picoseconds. Using this as the center, a signal segment within a time window from 2.5 picoseconds to 2.9 picoseconds is extracted. Within this window, the maximum absolute amplitude of 0.15 millivolts is found in the time-domain difference signal and recorded as the peak amplitude. Simultaneously, the average instantaneous phase difference between the time-domain reference signal and the thermally modulated time-domain signal within this window is calculated, yielding a phase offset of -0.12 radians. Finally, the peak amplitude of 0.15 mV and the phase offset of -0.12 radians are used as the feature values ​​of this pixel and stored in the terahertz dynamic differential feature map. This example directly verifies the effectiveness and feasibility of a series of technical feature operations, from alignment and differentiation to extracting peak amplitude and phase offset within a specific echo time window.

[0090] In one embodiment of the present invention, the feature fusion registration module is specifically configured to perform the following operations:

[0091] The surface thermal property parameter map and the terahertz dynamic differential feature map are spatially registered at the pixel level to unify the coordinate system. For each pixel, the apparent thermal diffusivity and surface temperature rise of the coating are read from its corresponding surface thermal property parameter map, and the differential peak amplitude and differential phase offset are read from its corresponding terahertz dynamic differential feature map. The four parameters are combined according to the pixel position to generate an aligned feature map for each pixel, including a four-dimensional feature vector.

[0092] Specifically, based on the aforementioned surface thermal property parameter map and terahertz dynamic differential feature map, spatial registration and feature fusion operations are performed. Spatial registration is a process of geometrically transforming two image data sets describing the same scene acquired at different times or by different sensors to ensure that their pixels correspond one-to-one in space. First, the coordinate system and physical size of the two feature maps are unified. Since both maps originate from data collected point-by-point from the same detection area using the same scanning step size, the physical coordinates of their corresponding pixels on the sample surface have a one-to-one correspondence. During the spatial registration and feature fusion operation, based on the mechanical coordinates of each detected pixel recorded by the scanning control unit, the same two-dimensional spatial index matrix is ​​established for each pixel in the surface thermal property parameter map and each pixel in the terahertz dynamic differential feature map, thereby ensuring that the two maps are spatially aligned.

[0093] If slight positional deviations exist due to factors such as non-rigid deformation, a fine-tuning algorithm based on maximizing mutual information is further employed. Based on this spatial alignment, the feature parameters carried by pixels corresponding to the same physical location in the two images are fused. Specifically, for any pixel, two parameters—apparent thermal diffusivity and surface temperature rise—are read from its corresponding surface thermal property parameter image, and two parameters—differential peak amplitude and differential phase shift—are read from its corresponding terahertz dynamic differential feature image. These four values ​​are used as a four-dimensional feature vector and assigned to the pixel. This process is repeated for all pixels, ultimately generating a new dataset where each pixel includes a four-dimensional feature vector, i.e., the alignment feature map. This map fully preserves the spatial structure of the original image, with each pixel being a feature unit containing four channels of data. The alignment feature map is a three-dimensional array, where the first two dimensions represent the spatial position rows and columns of the image, and the third dimension is a channel with a depth of 4, storing the coating apparent thermal diffusivity, surface temperature rise, differential peak amplitude, and differential phase shift in sequence.

[0094] For example, following the previous steps, assume the detection point is located at the 10th row and 20th column pixel position in the scanning area. From the generated surface thermal property parameter map, read the coating apparent thermal diffusivity value of the pixel at the 10th row and 20th column as 0.75 mm² / s and the surface temperature rise value as 8.5 Kelvin. From the generated terahertz dynamic differential feature map, read the differential peak amplitude value of the pixel at the same position as 0.15 mV and the differential phase offset value as -0.12 radians. Combine these four values ​​in a fixed order to form the feature vector of this pixel: 0.75, 8.5, 0.15, -0.12. After performing this operation on all pixels, the four-dimensional feature vector is stored at the 10th row and 20th column position in the generated aligned feature map.

[0095] In one embodiment of the present invention, the defect classification and determination module is configured to perform the following operations:

[0096] Read the coating apparent thermal diffusivity, differential peak amplitude, and differential phase offset of the currently detected pixel; if the coating apparent thermal diffusivity is lower than the preset normal threshold lower limit, and the differential peak amplitude and differential phase offset are both within the preset normal value range, then it is determined that the pixel has a coating defect.

[0097] Read the surface temperature rise amplitude, differential peak amplitude, and differential phase offset of the currently detected pixel, and calculate the average surface temperature rise of the surrounding pixels. If the surface temperature rise amplitude is higher than the average surface temperature rise by a preset amplitude, and the absolute values ​​of the differential peak amplitude and differential phase offset both exceed their respective preset abnormal thresholds, then the pixel is determined to have an interface bonding defect.

[0098] Check the temporal reference signal of the currently detected pixel and read all feature parameters in the alignment feature map. If a structural reflection anomaly is identified in the time period after the interface echo, but the coating apparent thermal diffusivity, surface temperature rise amplitude, differential peak amplitude, and differential phase offset in the alignment feature map do not trigger the anomaly judgment threshold, then the pixel is determined to have an internal defect in the matrix.

[0099] Furthermore, the identification of structural reflection anomalies can be achieved through one or more quantifiable signal analysis methods, which are configured and executed by the defect classification and determination module:

[0100] Threshold comparison method:

[0101] In the time-domain reference signal, a time interval earlier than any expected echo (typically before the echo reflected from the upper surface) is selected as the noise reference segment. The root mean square value of the signal amplitude within this noise reference segment is calculated and denoted as . Based on the known sound velocity or pre-calibrated wave velocity of the substrate material being tested, the time window in which the reflected echo from a specific depth within the substrate should appear is estimated, and this window is defined as the signal segment to be analyzed. Within the signal segment to be analyzed, local maxima of the signal are sought. Calculate the ratio of the local maximum to the root mean square value of the background noise. If the ratio R exceeds a preset anomaly threshold multiple N, a structural reflection anomaly is determined to exist. The anomaly threshold multiple N typically ranges from 3 to 5, meaning the signal-to-noise ratio is significantly higher than the background. The specific value can be calibrated based on the sensitivity of the detection system and the requirements for the false alarm rate. For example, N can be set to 3.5, meaning that when R > 3.5, an abnormal reflection is considered to have occurred.

[0102] Peak detection method:

[0103] This method is applicable to situations with additional discrete reflective interfaces. The envelope of the time-domain reference signal is extracted, for example, by obtaining the analytical envelope of the signal through Hilbert transform. Local maxima points on the envelope are searched within a time window later than the coating-substrate interface echo time window. If a local maximum value exceeding a preset amplitude threshold (e.g., 0.1 × 10⁻⁶) is detected within a time interval where no major reflective interface is expected, the method is successful. ,in If the amplitude of the main peak of the interface echo is given, and the time interval between this maximum point and the known interface echo peak matches the depth location of a possible defect in the matrix, then a structural reflection anomaly is determined to exist, which may correspond to a new reflection interface inside the matrix.

[0104] Cross-correlation analysis:

[0105] This method is applicable to situations involving signal waveform distortion. It obtains the time-domain reference signal from a known, defect-free reference region or through theoretical model simulation, and extracts a typical signal segment starting from the interface echo as a reference template. For the temporal reference signal of the currently detected pixel, the signal segment to be tested is extracted within the same time period. .calculate and The normalized cross-correlation coefficient is calculated. If the cross-correlation coefficient is lower than a preset similarity threshold, such as 0.85, it indicates that the current signal waveform differs from the defect-free reference signal, and a structural reflection anomaly is determined to exist. This difference may be caused by scattering, diffraction, or complex reflection within the matrix.

[0106] Those skilled in the art should understand that the above methods are examples intended to provide clear and feasible judgment criteria. In actual systems, multiple methods can be combined for comprehensive judgment to improve the robustness and accuracy of structural reflection anomaly identification. The preset thresholds, such as N, amplitude threshold, and similarity threshold, can be determined through statistical analysis and performance optimization on a representative sample set (containing samples with known defects and samples without defects in the matrix).

[0107] Based on the combination of multi-dimensional feature parameters included in each pixel of the aligned feature map, where the multi-dimensional feature parameters refer to the four specific physical or signal feature quantities integrated into each pixel, a progressive logical judgment operation based on physical association is executed sequentially to generate a pixel-level defect classification index. It should be noted that the defect judgment logic based on physical association refers to the "if-then" rule set established by associating different types of defects with different feature parameter anomaly patterns based on the physics of heat conduction and the principle of terahertz wave reflection. The defect classification index is a two-dimensional integer array, in which the integer code stored at each position corresponds one-to-one with a specific defect type or defect-free state.

[0108] Specifically, from the alignment feature map, the pixel to be judged is selected, and its four-dimensional feature parameters are read: the coating apparent thermal diffusivity, surface temperature rise amplitude, differential peak amplitude, and differential phase offset. Then, sequential judgment is performed according to preset rules:

[0109] The first judgment rule targets coating defects. The logic is as follows: if the apparent thermal diffusivity value of the coating read by the pixel is lower than the preset normal threshold, and the differential peak amplitude value read by the pixel is within its normal range, and the differential phase offset value read by the pixel is also within its normal range, then a specific integer code representing a coating defect is marked in the defect classification index of the pixel.

[0110] The second judgment rule targets interface adhesion defects. The logic is as follows: if the surface temperature rise value read by the pixel is higher than the average temperature rise value of the pixels in its surrounding area, and the differential peak amplitude value read by the pixel also exceeds the preset abnormal threshold for this parameter, and the differential phase offset value read by the pixel also exceeds the preset abnormal threshold for this parameter, then a specific integer code representing the interface defect is marked in the defect classification index of the pixel.

[0111] The third judgment rule targets defects within the matrix. The logic is as follows: additionally access the original time-domain reference signal data corresponding to the pixel. If a structural anomaly echo signal corresponding to a deeper layer within the matrix is ​​identified, but the feature parameter values ​​of all four dimensions read from the aligned feature map for the pixel do not trigger any anomaly markers after the judgment of the aforementioned two rules, then a specific integer code representing a matrix defect is marked in the defect classification index of the pixel.

[0112] If a pixel does not trigger any of the above judgment rules, it is marked as defect-free in its defect classification index. The process of reading parameters and making logical judgments is repeated for each pixel in the aligned feature map, ultimately generating a two-dimensional matrix of the same size as the original image, where the value of each element is the defect classification index code for that pixel.

[0113] Among them, coating-body defects refer to defects that mainly exist within the coating material, such as pores, inclusions, or poor curing, and do not affect the interface between the coating and the substrate. Interface adhesion defects refer to defects such as debonding and weak adhesion that occur at the interface between the coating and the substrate. Substrate-internal defects refer to deep defects existing within the substrate material, such as delamination and cracks.

[0114] The defect classification and judgment module is configured to execute judgment logic based on physical correlation. Its core lies in setting reasonable judgment thresholds or normal ranges for each feature parameter. The determination of the above thresholds and ranges can be based on any one or more of the following methods:

[0115] Historical data statistics: Collect a large amount of sample data in known states, including defect-free and defect-free samples.

[0116] Traps, statistical analysis of various characteristic parameters ( , , , The distribution of the defective samples is determined by taking the mean of the normal samples ± N times the standard deviation as the normal range, or taking the boundary between the distributions of defective samples and normal samples as the abnormal threshold.

[0117] Physical model simulation: Establish a physical model containing different defects, and calculate the parameters through simulation.

[0118] Based on the theoretical trend of change, a threshold is set.

[0119] Expert experience calibration: This is performed by domain experts based on the correspondence between typical defect maps and parameter values.

[0120] Initially set thresholds, then fine-tune them using actual testing cases.

[0121] The specific judgment logic example is as follows:

[0122] Defects in the coating itself: If and and If so, it is determined to be a defect in the coating itself. The nominal value can be set as the lower limit of the apparent thermal diffusivity of the coating. 70%-85%, for example, nominal If it is 0.8 square millimeters per second, then It is 0.65 square millimeters per second. -0.05 millivolts +0.05 millivolts It is -0.05 radians. It is +0.05 radians.

[0123] Interface bonding defect judgment: If and and If it is, then it is determined to be an interface bonding defect, in which case, The 8 neighboring pixels of this point The average value, To indicate the extent of the temperature rise, it can be set to... 15%-30% or a fixed value such as 1.5 Kelvin, The threshold for abnormal differential peak amplitude can be set to 0.1 millivolts. The threshold for differential phase offset anomalies can be set to 0.1 radians.

[0124] Matrix internal defect determination: If structural reflection anomalies (such as the appearance of additional echo peaks or signal envelope distortion) are identified in the time-domain reference signal at a specific time period later than the interface echo time window (e.g., corresponding to a certain depth inside the matrix), and all based on , , , If none of the judgment conditions are met, it is judged to be an internal defect of the matrix. Structural reflection anomalies can be judged by setting a threshold that the signal amplitude exceeds the background noise level by a certain multiple, such as 3-5 times, or by using a pattern recognition algorithm.

[0125] Threshold optimization example: The system can pre-embed a threshold learning mode. In this mode, the operator inputs a batch of sample data of known defect types and their ground truth labels. The system automatically calculates the cluster centers and dispersion of the corresponding parameter values ​​for each type of defect, and uses linear discriminant analysis or support vector machine methods to find the optimal classification boundary, thereby automatically optimizing or suggesting each judgment threshold.

[0126] For example, for this detection point pixel, its feature vector is 0.75 mm² / s, 8.5 Kelvin, 0.15 mV, and -0.12 radians. First, the apparent thermal diffusivity value of the coating is read as 0.75. The preset lower limit of the normal threshold is known to be 0.8 mm² / s; 0.75 is below this threshold, satisfying the first condition. Next, the differential peak amplitude is read as 0.15 mV, whose normal range is -0.05 to +0.05 mV. 0.15 exceeds this range and does not satisfy the condition of "within the normal range." Therefore, the coating body defect rule judgment is invalid. Next, the interface defect rule judgment is executed. The surface temperature rise amplitude is read as 8.5 Kelvin, and the average temperature rise of its eight neighboring pixels is calculated to be approximately 7.0 Kelvin. 8.5 is above this average, satisfying the first condition. The differential peak amplitude is read as 0.15 mV, whose preset abnormal threshold is 0.1 mV. 0.15 exceeds this threshold, satisfying the second condition. The differential phase offset is read as -0.12 radians. The preset anomaly threshold is an absolute value of 0.1 radians. If the absolute value of -0.12 exceeds this threshold, the third condition is met. If all three conditions are met, the corresponding position in the defect classification index matrix of that pixel is marked with the code representing the interface defect, such as the number 2.

[0127] In one embodiment of the present invention, the region scanning and stitching module is specifically configured to perform the following operations:

[0128] Based on the preset scanning path, the integrated scanning probe is driven to move to each detection pixel in row and column order; the defect classification index value of each point is obtained; according to the spatial coordinates of each pixel, all defect classification index values ​​are arranged into a two-dimensional matrix corresponding to the scanning area to form a defect distribution index map.

[0129] Specifically, the integrated scanning probe is driven to reset to the starting point of the detection area. According to the preset scanning path plan, the detection area is divided into a pixel array arranged in multiple rows and columns. The number of rows and columns, as well as the row spacing and column spacing (i.e., the scanning step size), are all predefined according to the detection resolution requirements. Subsequently, the control system starts a nested loop program, where the outer loop controls the probe to move along the row direction, and the inner loop controls the probe to move along the column direction within each row. For each target pixel that is moved to, i.e., the current detection pixel, the aforementioned complete operation process is automatically called and executed sequentially, namely: performing collaborative detection to generate a joint detection dataset for that point; inverting thermal properties to generate a surface thermal property parameter map; extracting dynamic features of the signal to generate a terahertz dynamic differential feature map; fusing to generate an aligned feature map; and finally executing defect judgment logic to generate a defect classification index value for that single pixel.

[0130] For each pixel determined, its spatial coordinates and corresponding defect classification index value are recorded as a pair and temporarily stored in computer memory or written to a data file in real time. Once the inner loop has completed scanning all columns of the current row, the probe moves to the starting column of the next row and repeats the process. This process continues row by row and point by point until the outer loop has completed scanning all rows, meaning the entire inspection area of ​​the composite coating structure under test has been traversed.

[0131] After all defect classification index data for all pixels has been collected, the values ​​are rearranged into a two-dimensional matrix based on the spatial coordinate information of each pixel record pair. The row and column indices of this matrix strictly correspond to the physical row and column order during detection. The element value at each position in the matrix is ​​the defect classification index code of the pixel at that coordinate. This complete two-dimensional matrix constitutes a complete defect distribution index map covering the entire detection area. The complete defect distribution index map is a two-dimensional integer array, the size of which is equal to the total number of rows and columns of the detection area. Each element in the array represents the defect type code of the pixel at the corresponding spatial location. This map provides a complete spatial distribution of defect types and locations throughout the entire tested area.

[0132] It should be noted that the detection area refers to the designated surface area on the composite coating structure to be fully inspected. Its size, shape, and internal pixel grid division are pre-set according to the requirements of the inspection task. For example, a 100 mm by 100 mm square area is divided into a pixel grid of 100 rows and 100 columns. Repeated execution row by row and point by point refers to the automated and sequential implementation of a fixed inspection and analysis process on each predetermined location point in the detection area grid through a program loop structure.

[0133] For example, in the instance of detecting a single specific pixel in the aforementioned steps, assume that the point is located in the second row and second column of a detection area planned as 3 rows and 3 columns, with a scanning step size of 1 mm. Scanning begins from the first row and first column at the top left corner of the area. When the probe moves to the second row and second column, based on the previous example, the defect classification index value of this point is determined to be code 2, representing an interface defect. This process continues to traverse all other 8 points, obtaining their respective index values. For example, a value of 0 for the point in the first row and first column represents no defect, while a value of 1 for the point in the first row and second column represents a coating defect, etc. After traversing all 9 points, the index values ​​of all points are combined into a 3 row and 3 column matrix according to their row and column positions. For example, the first row is 0, 1, 0; the second row is 1, 2, 0; and the third row is 0, 0, 0. This matrix is ​​the complete defect distribution index map of the 9-pixel detection area.

[0134] In one embodiment of the present invention, the visualization imaging module is specifically configured to perform the following operations:

[0135] A grayscale background is generated for defect-free pixels based on their apparent thermal diffusivity values. Pixels marked as defects in the coating body are rendered as a first preset visual identifier, such as a blue grid texture. Pixels marked as interface bonding defects are rendered as a second preset visual identifier, such as a red area, the brightness of which is positively correlated with the differential peak amplitude. Pixels marked as defects inside the substrate are rendered as a third preset visual identifier, such as a yellow semi-transparent filled area. The rendered defect pixels are then superimposed on the grayscale background to generate a fused diagnostic image.

[0136] Specifically, the complete defect distribution index map is read. This map is a two-dimensional integer array, where each element represents the defect type code of the corresponding pixel. Then, according to the preset mapping rules, specific color, transparency, and texture attributes are assigned to each pixel, thereby initializing a multi-channel image canvas with the same spatial size in memory.

[0137] Rendering operations are performed according to defect type:

[0138] For pixels whose defect codes are marked as coating defects, their color is set to, for example, pure blue. At the same time, a grid pattern centered on the pixel is superimposed in the local neighborhood of the pixel, thereby visually presenting the effect of a blue grid texture.

[0139] For pixels marked as interface defects by defect encoding, their color is set to, for example, pure red. Simultaneously, a brightness coefficient is calculated based on the differential peak amplitude value stored in the alignment feature map for that pixel. This brightness coefficient is linearly proportional to the differential peak amplitude value; that is, the larger the differential peak amplitude, the higher the brightness coefficient. Multiplying the original intensity of the red component by this brightness coefficient ensures that the final display brightness of the red pixel area is positively correlated with the differential peak amplitude. This means that there is a monotonically increasing functional relationship between the display brightness value of the interface defect pixel and its differential peak amplitude value. The brightness coefficient can be calculated as follows: the brightness coefficient equals the differential peak amplitude divided by a reference amplitude value.

[0140] For pixels whose defect encoding is marked as matrix defects, their color is set to, for example, pure yellow, and a specific transparency value is assigned to them, making them appear as a yellow semi-transparent filled area.

[0141] For pixels marked as defect-free, a fixed color is not directly assigned. Instead, background grayscale is calculated. Generating background grayscale involves linearly normalizing and mapping the continuously varying apparent thermal diffusivity parameter of the coating to 256 grayscale levels. The apparent thermal diffusivity value of the coating stored in the aligned feature map for that pixel is read and linearly mapped to an integer range of 0 to 255. The resulting integer value serves as the brightness value of that pixel in the grayscale channel, thus generating a continuous grayscale background image. The grayscale depth reflects the spatial distribution of the coating's thermal diffusivity. Specifically, for defect-free pixels... The value is mapped to a grayscale value of 0-255 through linear normalization. The mapping formula is: ,in and For the entire detection area The minimum and maximum values, or the upper and lower limits of a pre-defined typical range.

[0142] Set reference range For example, 0.2 millivolts, luminance coefficient Multiply the original intensity of the red channel by An upper limit can be set, such as 2.0. For example, at a certain point... If it is 0.15 millivolts, then The value is 0.75, and the red display brightness at this point is 75% of the normal value.

[0143] Finally, all rendered foreground pixels with different colors, textures, and transparency are superimposed onto the grayscale background image according to their spatial coordinates to synthesize the final image, namely the fused diagnostic image. During the synthesis process, the semi-transparent yellow areas allow the background grayscale information underneath to show through, while the bright red and gridded blue areas cover the background. It should be noted that the fused diagnostic image is a color bitmap that visualizes different defect types encoded with specific colors and textures and fuses multi-dimensional physical parameter information in the form of background grayscale to provide intuitive detection results.

[0144] For example, suppose the complete defect distribution index map is a 3x3 matrix, where the pixel value in the 2nd row and 2nd column is 2, representing an interface defect; the pixel value in the 1st row and 2nd column is 1, representing a coating defect; and the remaining pixels are 0, representing no defects. First, a background grayscale is generated for the defect-free pixels. For example, the apparent thermal diffusivity of the coating at the 1st row and 1st column pixel is read as 0.8 mm² / s. Assuming a mapping range of 0.5 to 1.0 mm² / s corresponding to 0 to 255, a grayscale value of 204 is obtained through linear mapping. Similarly, grayscale values ​​are calculated for all defect-free points to form a grayscale background. For the coating defect pixel in the 1st row and 2nd column, it is rendered as blue and a mesh texture is overlaid. For the interface defect pixel in the 2nd row and 2nd column, assuming its differential peak amplitude is 0.15 mV and the reference amplitude is set to 0.1 mV, the brightness coefficient is 1.5. The red channel intensity is multiplied by 1.5 to obtain a bright red pixel. Finally, all the rendered foreground pixels are overlaid onto the grayscale background to synthesize a fused diagnostic image.

[0145] Those skilled in the art should understand that the specific values, ranges, and thresholds given in the specification are merely examples and not intended to limit the scope of protection of this invention. In practical applications, the parameters (such as...) , ), threshold (e.g.) , ) and mapping coefficients (such as It can be adaptively adjusted or recalibrated according to the specific material, coating system, process conditions and detection accuracy requirements of the object being tested.

[0146] The adjustment or calibration process, such as by collecting representative sample data and applying the aforementioned statistical, modeling, or learning methods, is an operation that can be performed by those skilled in the art after reading the disclosure of this invention.

[0147] It should be noted that the formulas described above, through the principle of dimensional consistency and mathematical standardization methods (such as normalization, dimensionless parameter conversion, or unit system unification), can translate physical quantities with different properties into unitless standard values ​​or superimposed parameters of the same dimension. This eliminates the interference of different dimensions on the computational logic, allowing the formulas to retain the original data distribution characteristics while possessing mathematical rationality and adaptability to objective laws. The descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the invention.

[0148] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0149] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography, characterized in that, include: The detection data generation module performs collaborative detection timing for the detection pixels of the composite coating structure under test, and generates a joint detection dataset including time-domain reference signal, thermal modulation state time-domain signal and surface temperature sequence. The thermal property inversion module inverts and generates a surface thermal property parameter map characterizing the thermal physical properties of the detected pixel based on the surface temperature sequence in the joint detection dataset. The dynamic feature extraction module calculates and extracts the dynamic change features of the terahertz signal caused by thermal excitation based on the time-domain reference signal and the thermally modulated time-domain signal in the joint detection dataset, and generates a terahertz dynamic differential feature map. The feature fusion and registration module spatially registers and fuses the surface thermal property parameter map with the terahertz dynamic differential feature map to generate an aligned feature map that includes multi-dimensional feature parameters. The defect classification and determination module executes physical association-based defect determination logic based on the combination of multi-dimensional feature parameters of each pixel in the aligned feature map, and generates a defect classification index for distinguishing defect types. The area scanning and stitching module traverses the entire detection area of ​​the composite coating structure under test, and stitches together the defect classification index of all detected pixels to generate a defect distribution index map covering the entire detection area. The visualization imaging module assigns preset visual identifiers and texture codes to different types of defect pixels based on the defect type markers in the defect distribution index map, generating a fused diagnostic image.

2. The non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography according to claim 1, characterized in that, The specific configuration of the probe data generation module is used to perform the following operations: The integrated scanning probe is driven to position itself at the detection pixel, triggering the terahertz time-domain spectroscopy module within it to acquire the reflection signal of that point in a thermally unexcited state as a time-domain reference signal. A pulsed thermal excitation source is synchronously activated to apply a thermal pulse, and an infrared thermal imager records the surface temperature at that point from the start of the thermal pulse to a certain period of time after the pulse ends, forming a surface temperature sequence. Based on the coating thickness and thermal diffusivity of the composite coating structure under test, the characteristic time delay of heat diffusion to the coating-substrate interface is calculated. At this characteristic time delay, the reflected signal under the influence of the thermal pulse is collected again as the thermally modulated time domain signal.

3. The non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography according to claim 1, characterized in that, The thermal property inversion module is specifically configured to perform the following operations: Extract cooling curve data after the thermal pulse ends from the surface temperature sequence; The cooling curve data were fitted with the theoretical curve of the one-dimensional heat conduction model; Based on the fitting results, the apparent thermal diffusivity and surface temperature rise of the detected pixel are derived, and a surface thermal property parameter map including the distribution of the apparent thermal diffusivity and surface temperature rise is generated.

4. The non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography according to claim 1, characterized in that, The dynamic feature extraction module is specifically configured to perform the following operations: Time alignment processing is performed between the time-domain reference signal and the thermally modulated time-domain signal; The two aligned signals are subtracted point by point to obtain the time-domain difference signal; Identify the echo time window corresponding to the coating-substrate interface in the time-domain reference signal; Within the echo time window of the time-domain differential signal, the differential peak amplitude of the signal is extracted, and the average phase difference between the time-domain reference signal and the thermally modulated time-domain signal within this window is calculated as the differential phase offset to generate a terahertz dynamic differential feature map.

5. The non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography according to claim 1, characterized in that, The feature fusion and registration module is specifically configured to perform the following operations: The surface thermal property parameter map and the terahertz dynamic differential feature map are spatially registered at the pixel level to unify the coordinate system; For each pixel, the apparent thermal diffusivity and surface temperature rise of the coating are read from its corresponding surface thermal property parameter map, and the differential peak amplitude and differential phase offset are read from its corresponding terahertz dynamic differential feature map. The four parameters are combined according to pixel position to generate an aligned feature map for each pixel, which includes a four-dimensional feature vector.

6. The non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography according to claim 1, characterized in that, The defect classification and determination module is configured to perform the following operations: Read the coating apparent thermal diffusivity, differential peak amplitude, and differential phase offset of the currently detected pixel; If the apparent thermal diffusivity of the coating is lower than the preset lower limit of the normal threshold, and the differential peak amplitude and differential phase offset are both within the preset normal value range, then the pixel is determined to have a coating defect.

7. The non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography according to claim 1, characterized in that, The defect classification and determination module is also configured to perform the following operations: Read the surface temperature rise value, differential peak amplitude, and differential phase offset of the currently detected pixel, and calculate the average surface temperature rise value of the surrounding pixels. If the surface temperature rise is higher than the average surface temperature rise by a preset amount, and the absolute values ​​of the differential peak amplitude and the differential phase offset both exceed their respective preset abnormal thresholds, then the pixel is determined to have an interface bonding defect.

8. The non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography according to claim 1, characterized in that, The defect classification and determination module is also configured to perform the following operations: Examine the temporal reference signal of the currently detected pixel and read all feature parameters in the aligned feature map. If a structural reflection anomaly is identified within a time period later than the interface echo in the time-domain reference signal, but the coating apparent thermal diffusivity, surface temperature rise amplitude, differential peak amplitude, and differential phase offset in the alignment feature map do not trigger the anomaly judgment threshold, then the pixel is determined to have an internal defect in the substrate.

9. The non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography according to claim 1, characterized in that, The specific configuration of the region scanning and stitching module is used to perform the following operations: Based on the preset scanning path, the integrated scanning probe is driven to move to each detection pixel in row and column order; the defect classification index value of each point is obtained; Based on the spatial coordinates of each pixel, all defect classification index values ​​are arranged into a two-dimensional matrix corresponding to the scanned area, forming a defect distribution index map.

10. The non-destructive testing system for substrates under coatings based on terahertz spectroscopy and infrared thermography according to claim 1, characterized in that, The visualization imaging module is specifically configured to perform the following operations: A grayscale background is generated for defect-free pixels by mapping the apparent thermal diffusivity of their coating. Pixels marked as defects in the coating body are rendered as the first preset visual identifier; Pixels marked as interface adhesion defects are rendered as a second preset visual identifier, the brightness of which is positively correlated with the differential peak amplitude. Pixels marked as internal defects of the matrix are rendered as third preset visual identifiers; The rendered defective pixels are superimposed onto a grayscale background to generate a fused diagnostic image.

Images