Infrared Lock-in Non-destructive Testing Method Based on Adaptive Normalization Improvement Technology

Abstract

The present invention discloses an infrared lock-in non-destructive testing method based on an improved adaptive normalization technology, belonging to the technical field of infrared lock-in non-destructive testing methods. A problem point filtering algorithm is invented, which can find and filter out problem pixel points in the original infrared video to prevent the problem pixel points from causing noise interference to the detected phase result map. An adaptive normalization algorithm is invented. Using the adaptive normalization algorithm can effectively improve the contrast of the phase result map on the premise of maintaining the distribution characteristics of the original phase data structure, thereby improving the visual perception of the human eye. The effect is significantly improved compared with the traditional two normalization algorithms. A method for predicting the defect depth based on phase contrast is invented, realizing the ability to predict the depth of defects inside or on the back of the inspected plate.

Description

Technical Field

[0001] The present invention relates to an infrared lock-in non-destructive testing method, in particular to an infrared lock-in non-destructive testing method based on an adaptive normalization improvement technique, belonging to the technical field of infrared lock-in non-destructive testing methods. Background Technique

[0002] The traditional infrared lock-in thermography non-destructive testing algorithm has the following four problems:

[0003] 1. Lack of preprocessing of problem points in infrared videos: In the infrared lock-in thermography non-destructive testing technology, a heat source that changes in a sine wave is used to thermally stimulate the plate to be inspected, and at the same time, an infrared thermal imager is used to record the heating process of the plate to obtain the original infrared video. In a real scenario, due to software settings of the infrared thermal imager, damage to the photosensitive elements in the infrared thermal imager, and the existence of certain external interferences in the experimental environment, etc., the temperature change values of some pixel points in the original infrared video obtained do not conform to the law of rising in a sine wave. If these problem pixel points are not filtered out, it will cause great noise interference to the phase result map calculated by the traditional algorithm;

[0004] 2. The algorithm used for normalizing the original data into the phase result map has limitations: The traditional normalization algorithms include the direct normalization algorithm and the mean standard deviation normalization algorithm. Among them, the direct normalization algorithm can completely restore the structural distribution law of the original phase data, but the phase result map calculated by this algorithm lacks pertinence to defects, and the contrast between defects and the background in the phase result map is low, and the visual perception of defects by the naked eye is not obvious;

[0005] The standard deviation mean algorithm normalizes the original data based on the probability distribution characteristics of the data, which can better improve the contrast between defects and the background in the phase result map, enhance the visual perception of defects by the naked eye, and make it easier to observe the existence of defects. However, this algorithm has the defect of poor scene adaptability because there is a fixed adjustment parameter k in this algorithm. For infrared videos taken in different scenarios, the imaging quality of the phase result map obtained using different k values is also different. Therefore, the adaptive ability of this algorithm is poor, and in different experiments, the experimenter needs to repeatedly adjust the k value to improve the image quality of the phase result map;

[0006] 3. Only output the phase result map of the infrared video, and the buried depth of each defect cannot be obtained: The traditional infrared lock-in thermography non-destructive testing algorithm can only output the phase result map of the plate, and the detection personnel cannot obtain the depth distribution of each defect. Therefore, a method for detecting the depth of defects needs to be studied;

[0007] 4. It cannot automatically identify the type, location, and diameter of defects and requires manual analysis by professionals: Traditional infrared lock-in thermography non-destructive testing algorithms can only output the phase result map of the plate, and inspectors need to perform manual measurement and analysis on the defects.

[0008] Therefore, an infrared lock-in non-destructive testing method based on adaptive normalization improvement technology is designed to solve the above problems. Summary of the Invention

[0009] The main purpose of the present invention is to provide an infrared lock-in non-destructive testing method based on adaptive normalization improvement technology.

[0010] The object of the present invention can be achieved by adopting the following technical solutions:

[0011] An infrared lock-in non-destructive testing method based on adaptive normalization improvement technology includes the following steps:

[0012] Step 1: Calculate the SSIM structural similarity index of the phase result map to be evaluated according to the standard phase result map.

[0013]

[0014] Wherein:

[0015] x and y are the pixel values of two images;

[0016] μ x is the average value of image x;

[0017] μ y is the average value of image y;

[0018] is the variance of image x;

[0019] is the variance of image y;

[0020] C1 and C2 are two constants used to stabilize the denominator.

[0021] Step 2: Calculate the GMSD image high-frequency similarity index of the phase result map to be evaluated according to the standard phase result map.

[0022] Gradient magnitude similarity formula:

[0023]

[0024] Wherein: G x and G y are the gradient magnitudes of images x and y, respectively;

[0025] c is a constant used to stabilize the calculation.

[0026]

[0027] Wherein:

[0028] M is the total number of pixels of the image;

[0029] GMS i is the gradient magnitude similarity of the i-th pixel;

[0030] μ GMS is the average value of all GMS i values.

[0031] Step 3: Calculate the RMS contrast difference score of the phase result map to be evaluated;

[0032] Step 4: Perform a combined weighted scoring on the SSIM index and the GMSD index, and discard the phase result maps to be evaluated with a combined weighted score less than the ideal structural similarity score;

[0033] Step 5: Select the phase result map with the highest RMS contrast score among the undiscarded phase result maps to be evaluated as the final output phase result map of the algorithm.

[0034] Preferably, in Step 1, SSIM is used to measure the similarity between two images, and the formula is as follows:

[0035]

[0036] Wherein:

[0037] x and y are the pixel values of the two images;

[0038] μ x is the average value of image x;

[0039] μ y is the average value of image y;

[0040] is the variance of image x;

[0041] is the variance of image y;

[0042] C1 and C2 are two constants used to stabilize the denominator.

[0043] Preferably, in Step 4, GMSD is used to evaluate the similarity of image quality, and the formula is as follows:

[0044] Gradient magnitude similarity formula:

[0045]

[0046] Wherein: G x and Gy They are the gradient magnitudes of images x and y respectively;

[0047] c is a constant used to stabilize the calculation.

[0048] Preferably, the definition of GMSD;

[0049]

[0050] Where:

[0051] M is the total number of pixels in the image;

[0052] GMS i is the gradient magnitude similarity of the i-th pixel;

[0053] μ GMS is the average value of all GMS i values.

[0054] Preferably, in step three, RMS is used to measure the contrast of the image, and the formula is as follows:

[0055]

[0056] Where:

[0057] N is the total number of pixels in the image;

[0058] I i is the gray value of the -th pixel;

[0059] μ is the average gray value of the image.

[0060] Preferably, the application of the infrared lock-in non-destructive testing method based on the adaptive normalization improvement technology;

[0061] When it is necessary to detect the buried depth of defects in a certain fixed type of plate, the depth of the defects is predicted through phase contrast, and the specific method process is as follows:

[0062] S11: Fabricate a calibration plate with defects of different depths and different types, fix the excitation frequency and excitation time, perform infrared thermal wave lock-in excitation on the calibration plate, and record the infrared video during the excitation process;

[0063] S12: Analyze the infrared video to obtain the phase result map data of the calibration plate;

[0064] S13: Calculate the phase contrast of defects at different depths;

[0065] S14: Fit the mapping formula between the defect depth and the phase contrast of different defect types;

[0066] S15: Repeat the above S11 - S13 for the plate to be detected, calculate the phase contrast of the defect to be detected, and substitute the phase contrast into the mapping formula obtained in S14 to obtain the predicted value of the defect depth.

[0067] Preferably, the formula for the phase contrast in S13 is as follows:

[0068]

[0069] Where:

[0070] P background is the average value of the background original phase in the phase result map;

[0071] P defect is the average value of the original phase of the defect in the phase result map;

[0072] K is a constant fitting factor.

[0073] Preferably, the application of the infrared lock - in non - destructive testing method based on the adaptive normalization improvement technology;

[0074] Specific method flow:

[0075] S21: Customize plates with different types of internal defects. The defect types include: slit - type, circular, polygonal, and irregular. Set different light source excitation frequencies and light source powers to heat the plates with sine waves, and thus record the original infrared videos under various detection conditions;

[0076] S22: Perform algorithm processing on the original infrared videos to obtain phase result maps. Perform various processes on the phase result maps, such as rotation, scaling, shearing, noise processing, and affine transformation, to obtain a large amount of original data. Manually calibrate the original data, and divide the calibrated data to obtain the training set and test set of the model;

[0077] S23: Select a suitable neural network model to train the training set, thus obtaining the final neural network model. Use this model to identify the defect types, positions, and diameters of the phase result maps of the plates to be detected.

[0078] The beneficial technical effects of the present invention:

[0079] The infrared lock - in non - destructive testing method based on the adaptive normalization improvement technology provided by the present invention, (1) invented a problem - point filtering algorithm, which can find and filter out the problem pixel points in the original infrared videos, preventing the problem pixel points from causing noise interference to the phase result maps obtained by detection.

[0080] (2) An adaptive normalization algorithm was invented. Using the adaptive normalization algorithm can effectively improve the contrast of the phase result map while maintaining the distribution characteristics of the original phase data structure, thereby enhancing the visual perception by the naked eye. The effect is significantly improved compared with the traditional two normalization algorithms.

[0081] (3) A method for predicting the defect depth based on phase contrast was invented, realizing the ability to predict the depth of defects inside or on the back of the inspected plate.

[0082] (4) The use of a neural network realized the automatic recognition of the type, position, and diameter of defects in the phase result map, and achieved the ability to automatically generate defect detection reports. Description of the Drawings

[0083] Figure 1 It is a flowchart of a preferred embodiment of the infrared lock-in nondestructive testing method based on the adaptive normalization improvement technology according to the present invention. Detailed Embodiment

[0084] To make the technical solutions of the present invention clearer and more definite to those skilled in the art, the present invention will be further described in detail below in conjunction with the embodiments and the drawings. However, the embodiments of the present invention are not limited thereto.

[0085] In infrared lock-in thermography nondestructive testing, after heating the plate with a sinusoidal thermal excitation, the normal original infrared video obtained should conform to the rule that the amplitudes of all pixel points show a sinusoidal rise, and the overall amplitude average value of the time-domain waveforms of all pixel points should be within a reasonable fluctuation range. Therefore, based on these two characteristics, we developed an algorithm for finding problem pixel points, which can quickly find the problem pixel points. After finding the problem pixel points, the values of the problem pixel points are replaced with the average value of all image pixel points, thereby achieving the goal of filtering the problem pixel points.

[0086] The principle of the algorithm for finding problem pixel points is as follows:

[0087] Step 1: First, calculate the time-domain waveform of the average value of all pixel points, use this time-domain waveform as the standard reference waveform, and calculate the standard deviation, average value, and data range of the standard reference waveform.

[0088] Step 2: Then, calculate the standard deviation, average value, and data range of the time-domain waveform of the pixel points to be analyzed.

[0089] Step 3: Finally, analyze and compare the three major parameters of the pixel points to be analyzed with the three major parameters of the standard reference waveform to determine whether the pixel points to be analyzed are problem pixel points.

[0090] The formula for the standard deviation of the amplitude of the time-domain waveform of a single pixel point in the infrared video is:

[0091]

[0092] Among them, σ is the standard deviation, μ is the average amplitude, N is the total number of recording points, and Ii is the amplitude of the i-th point.

[0093] The formula for the average value of the amplitude of the time-domain waveform of a single pixel point in the infrared video is:

[0094]

[0095] Among them, μ is the average amplitude, N is the total number of recording points, and Ii is the amplitude of the i-th point. The formula for the data range of the amplitude of the time-domain waveform of a single pixel point in the infrared video is:

[0096] R = Imax - Imin

[0097] Imax is the maximum amplitude in the time-domain waveform of this pixel point, and Imin is the minimum amplitude in the time-domain waveform of this pixel point.

[0098] As Figure 1 shown, the infrared lock-in non-destructive testing method based on the improved adaptive normalization technology provided by this embodiment includes the following steps:

[0099] Step 1: Calculate the SSIM structural similarity index of the phase result map to be evaluated according to the standard phase result map;

[0100]

[0101] Among them:

[0102] x and y are the pixel values of two images;

[0103] μ x is the average value of image x;

[0104] μ y is the average value of image y;

[0105] is the variance of image x;

[0106] is the variance of image y;

[0107] C1 and c2 are two constants used to stabilize the denominator.

[0108] Step 2: Calculate the GMSD image high-frequency similarity index of the phase result map to be evaluated according to the standard phase result map;

[0109] Gradient amplitude similarity formula:

[0110]

[0111] where: G x and G y are the gradient magnitudes of images x and y respectively;

[0112] c is a constant used to stabilize the calculation.

[0113] In this embodiment, the definition of GMSD

[0114]

[0115] where:

[0116] M is the total number of pixels in the image;

[0117] GMS i is the gradient magnitude similarity of the i-th pixel;

[0118] μ GMS is the average value of all GMS i values.

[0119] Step 3: Calculate the RMS contrast difference score of the phase result map to be evaluated;

[0120] Step 4: Perform a joint weighted scoring on the SSIM index and the GMSD index, and discard the phase result maps to be evaluated whose joint weighted scores are less than the ideal structural similarity score;

[0121] Step 5: Select the phase result map with the highest RMS contrast score among the undiscarded phase result maps to be evaluated as the final output phase result map of the algorithm.

[0122] (1) Invented a problem point filtering algorithm, which can find and filter out the problem pixel points in the original infrared video to prevent the problem pixel points from causing noise interference to the detected phase result map.

[0123] (2) Invented an adaptive normalization algorithm. Using the adaptive normalization algorithm can effectively improve the contrast of the phase result map while maintaining the distribution characteristics of the original phase data structure, thereby improving the visual perception by the naked eye, and the effect is significantly improved compared with the traditional two normalization algorithms.

[0124] (3) Invented a method for predicting the defect depth based on phase contrast, realizing the ability to predict the defect depth inside or on the back of the inspected plate.

[0125] (4) Used a neural network to realize the automatic recognition of the types, positions and diameters of defects in the phase result map, and realized the ability to automatically generate defect detection reports.

[0126] In this embodiment, in Step 1, SSIM is used to measure the similarity between two images, and the formula is as follows:

[0127]

[0128] Wherein:

[0129] x and y are the pixel values of two images;

[0130] μ x is the average value of image x;

[0131] μ y is the average value of image y;

[0132] is the x variance of the image;

[0133] is the y variance of the image;

[0134] C1 and C2 are two constants used to stabilize the denominator.

[0135] In this embodiment, in step four, GMSD is used to evaluate the similarity of image quality, and the formula is as follows:

[0136] Gradient magnitude similarity formula:

[0137]

[0138] Wherein: G x and G y are the gradient magnitudes of images x and y respectively;

[0139] c is a constant used to stabilize the calculation.

[0140] In this embodiment, the definition of GMSD

[0141]

[0142] Wherein:

[0143] M is the total number of pixels of the image;

[0144] GMS i is the gradient magnitude similarity of the i-th pixel;

[0145] μ GMS is the average value of all GMS i ;

[0146] In this embodiment, in step three, RMS is used to measure the contrast of the image, and the formula is as follows:

[0147]

[0148] Wherein:

[0149] N is the total number of pixels of the image;

[0150] I i is the gray value of the \(i\)-th pixel;

[0151] \(\mu\) is the average gray value of the image.

[0152] In this embodiment, the application of the infrared lock-in non-destructive testing method based on the adaptive normalization improvement technology;

[0153] When it is necessary to detect the buried depth of defects in a certain fixed type of plate, the depth of the defects is predicted through phase contrast. The specific method process is as follows:

[0154] S11: Fabricate a calibration plate containing defects of different depths and different types, fix the excitation frequency and excitation time, perform infrared thermal wave lock-in excitation on the calibration plate, and record the infrared video during the excitation process;

[0155] S12: Analyze the infrared video to obtain the phase result map data of the calibration plate;

[0156] S13: Calculate the phase contrast of the defects at different depths;

[0157] S14: Fit the mapping formula between the defect depth and the phase contrast for different defect types;

[0158] S15: Repeat the above S11 - S13 for the plate to be detected, calculate the phase contrast of the defect to be detected, and substitute the phase contrast into the mapping formula obtained in S14 to obtain the predicted value of the defect depth.

[0159] In this embodiment, the formula for the phase contrast in S13 is as follows:

[0160]

[0161] Where:

[0162] P background is the average background original phase in the phase result map;

[0163] P defect is the average original phase of the defect in the phase result map;

[0164] K is a constant fitting factor.

[0165] In this embodiment, the application of the infrared lock-in non-destructive testing method based on the adaptive normalization improvement technology;

[0166] Specific method process:

[0167] S21: Customize plates with different types of internal defects. The defect types include: slit type, circular, polygonal, and irregular. Set different light source excitation frequencies and light source powers to perform sinusoidal heating on the plates, and thus record the original infrared videos under various detection conditions;

[0168] S22: Perform algorithm processing on the original infrared videos to obtain a phase result map. Perform various processes on the phase result map, such as rotation, scaling, shearing, noise processing, and affine transformation, to obtain a large amount of original data. Manually calibrate the original data, and divide the calibrated data to obtain the training set and test set of the model;

[0169] S23: Select a suitable neural network model to train the training set, thereby obtaining the final neural network model. Use this model to identify the defect type, position, and diameter of the phase result map of the plate to be tested.

[0170] The above are only further embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the scope disclosed by the present invention, according to the technical solution and its concept of the present invention, makes equivalent substitutions or changes, all belong to the protection scope of the present invention.

Claims

1. An infrared lock-in non-destructive testing method based on an adaptive normalization improvement technique, characterized in that: It includes the following steps: Step 1: Calculate the SSIM structural similarity index of the phase result map to be evaluated according to the standard phase result map; ； Where: and are the pixel values of two images; is the average value of the image ; is the image average value; is the variance of the image; is the variance of the image; and are two constants used to stabilize the denominator; Step 2: Calculate the GMSD image high-frequency similarity index of the phase result map to be evaluated according to the standard phase result map; Gradient magnitude similarity formula: ； Wherein: and are respectively the gradient magnitudes of the images and ; is a constant used for stable calculation; ； Where: is the total number of pixels of the image; is the gradient magnitude similarity of the nth pixel; is the average of all ; Step 3: Calculate the RMS contrast difference score of the phase result map to be evaluated; ； Where: is the total number of pixels of the image; is the gray value of the pixel; Step 4: Perform a combined weighted scoring on the SSIM index and the GMSD index, and discard the phase result maps to be evaluated whose combined weighted score is less than the ideal structural similarity score; Step 5: Select the phase result map with the highest RMS contrast score among the undiscarded phase result maps to be evaluated as the final output phase result map of the algorithm; Application of the infrared lock-in non-destructive testing method based on the adaptive normalization improvement technology; When it is necessary to detect the buried depth of defects in a certain fixed type of plate, the depth of the defects is predicted through phase contrast. The specific method process is as follows: S11: Fabricate a calibration plate containing defects of different depths and different types, fix the excitation frequency and excitation time, perform infrared thermal wave lock-in excitation on the calibration plate, and record the infrared video during the excitation process; S12: Analyze the infrared video to obtain the phase result map data of the calibration plate; S13: Calculate the phase contrast of defects at different depths; S14: Fit the mapping formula between the defect depth and the phase contrast of different defect types; S15: Repeat the above S11 - S13 for the plate to be detected, calculate the phase contrast of the defect to be measured, and substitute the phase contrast into the mapping formula obtained in S14 to obtain the predicted value of the defect depth; Application of the infrared lock-in non-destructive testing method based on the adaptive normalization improvement technology; Specific method process: S21: Customize a plate containing different types of internal defects. The defect types include: slit type, circular, polygonal, and irregular. Set different light source excitation frequencies and light source powers to heat the plate with a sine wave, and thus record the original infrared videos under various detection conditions; S22: Perform algorithm processing on the original infrared video to obtain a phase result map. Perform various processes such as rotation, scaling, shearing, noise processing, and affine transformation on the phase result map to obtain a large amount of original data. Manually calibrate the original data, and divide the calibrated data to obtain the training set and test set of the model; S23: Select a suitable neural network model to train the training set, so as to obtain the final neural network model, and use this model to identify the defect type, position, and diameter of the phase result map of the plate to be detected.

2. The infrared lock-in non-destructive testing method based on the improved adaptive normalization technology according to claim 1, characterized in that: In Step 1, SSIM is used to measure the similarity between two images. The formula is as follows: ； Where: and are the pixel values of two images; is the image average value; is the image average value; is the variance of the image; is the variance of the image; and are two constants used to stabilize the denominator.

3. The infrared lock-in non-destructive testing method based on the improved adaptive normalization technology according to claim 1, characterized in that: In Step 4, GMSD is used to evaluate the similarity of image quality. The formula is as follows: Gradient magnitude similarity formula: ； Wherein: and are respectively the gradient magnitudes of the images and ; is a constant used for stable calculation.

4. The infrared lock-in non-destructive testing method based on the improved adaptive normalization technology according to claim 3, wherein: Definition of GMSD ； Where: is the total number of pixels of the image; is the gradient magnitude similarity of the nth pixel; is the average of all values.

5. The infrared lock-in non-destructive testing method based on the improved adaptive normalization technology according to claim 4, characterized in that: In Step 3, RMS is used to measure the contrast of an image. The formula is as follows: ； Where: is the total number of pixels of the image; is the gray value of the [n]th pixel; is the average gray value of the image.

6. The infrared lock-in non-destructive testing method based on the improved adaptive normalization technology according to claim 5, characterized in that: The formula for the phase contrast in S13 is as follows: ； Where: is the average value of the background original phase in the phase result diagram; is the original phase average of the defects in the phase result diagram; is a constant fitting factor.

Images