Nondestructive testing method based on multi-dimensional complementary ensemble empirical mode decomposition algorithm

Abstract

This invention discloses a non-destructive testing method based on a multidimensional complementary ensemble empirical mode decomposition (MEMD) algorithm. The method constructs a three-dimensional matrix of all thermal image data and transforms it into a two-dimensional matrix. MEMD is then performed on each row of the two-dimensional matrix, and the decomposed images are further processed column-wise. The resulting images are then combined using the MEMD algorithm to extract the defective image portion, yielding the clearest image. This invention utilizes both the temporal and spatial information of the thermal image sequence through row and column mode decomposition, comprehensively reflecting defect information. Furthermore, this invention has broad applicability, suitable for any thermal image sequence, including both long-pulse heating and eddy current pulse heating.

Description

Technical Field

[0001] This invention relates to the field of non-contact defect detection technology, and in particular to a non-destructive testing method based on a multidimensional complementary set empirical mode decomposition algorithm. Background Technology

[0002] Material defects are a major cause of many engineering accidents. Metallic and non-metallic materials are widely used in daily life, aerospace, military, and renewable energy fields. Composite materials are widely used due to their extremely high strength, stiffness, high temperature resistance, and corrosion resistance, while metallic materials are widely used due to their low density, water resistance, stain resistance, and corrosion resistance. However, different materials can experience fatigue damage, impact damage, and material aging, thus necessitating defect detection for various materials.

[0003] Non-destructive testing (NDT) technology detects defects in materials without damaging them or through contact. It primarily identifies material defects, obtains the real-time state of the object under test, and then assesses the material's reliability and ensures stable equipment operation. Currently, various NDT techniques are used to detect object defects, such as ultrasonic testing, mechatronics, eddy current testing, X-ray testing, and chemical analysis. These are some of the most commonly used NDT methods for defect detection.

[0004] In the 1960s, Green and Alzofon first formally introduced the basic principles and simple applications of infrared thermal imaging technology. In recent years, pulsed thermal imaging has been widely used in engineering practice, such as monitoring corrosion in metals and detecting internal cracks and defects in materials. Long-pulse thermal imaging has attracted widespread attention due to its advantages such as low cost, simple operation, and rapid scanning. Long-pulse thermal imaging can be used for the detection of both metals and non-metals, while eddy current pulse thermal imaging is only suitable for metallic materials. The rapid development of infrared thermal imagers has made it possible to directly detect the temperature of large areas, and technological advancements have improved the resolution, frame rate, and temperature sensitivity of thermal imagers. For infrared thermal imaging non-destructive testing technology, the presence of defects hinders heat diffusion under external heat source excitation, resulting in abnormal temperature distribution in the defect area. Therefore, infrared thermal imaging technology can detect the presence of defects in objects by acquiring and analyzing thermal image sequences of the object's surface. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a non-destructive testing method based on the multidimensional complementary set empirical mode decomposition algorithm.

[0006] This invention is achieved through the following technical solution:

[0007] A non-destructive testing method based on a multidimensional complementary set empirical mode decomposition algorithm specifically includes the following steps:

[0008] (1) Construct a three-dimensional matrix of all thermal image data and convert the three-dimensional matrix of thermal image data into a two-dimensional matrix;

[0009] (2) Perform complementary set empirical mode decomposition on each row of the two-dimensional matrix data, and perform complementary set empirical mode decomposition on each column of the decomposed image;

[0010] (3) Combine the decomposed images according to the multidimensional complementary integrated empirical mode decomposition algorithm to obtain the defective image part and obtain the clearest image.

[0011] Step (1) involves constructing a three-dimensional matrix of all thermal image data and converting the three-dimensional matrix of thermal image data into a two-dimensional matrix, as detailed below:

[0012] The Multidimensional Complementary Ensemble Empirical Mode Algorithm (MCEEMD) can be used to construct thermal signal images. Each frame of thermal image data is a two-dimensional matrix, and time is represented by the frame number. The data of each frame is represented by a two-dimensional matrix as shown in formula (1-1). All thermal image data constitute a three-dimensional matrix X. i ;

[0013]

[0014] The frame time series of thermal images is transformed from three-dimensional to two-dimensional. Each two-dimensional matrix is ​​encapsulated into a column, with each row representing the time series plot of each point and each column storing the spatial information of each image. The transformed result is shown in formula (1-2).

[0015] M p =[X1,X2,…,X p (1-2)

[0016] Save the thermal image data using the method described above to obtain a set of two-dimensional data.

[0017] Step (2) involves performing Complementary Set Empirical Mode Decomposition (CEEMD) on each row of the two-dimensional matrix data, and then performing CEEMD on the decomposed image column by column, as follows:

[0018] Assuming the two-dimensional data is m×n, each row of the data is decomposed according to the form of complementary set empirical mode decomposition, resulting in i modal components (IMFs) and a residual value. The modal components of different rows are taken out and recombined. The first i IMFs form i new images, and the remaining residuals form a new image. The selection of the number i of modal components (IMFs) follows the empirical formula (1-3).

[0019]

[0020] After row-wise decomposition, (i+1) images are obtained. To include both temporal and spatial information in the data, complementary set empirical mode decomposition (CEM) is performed on the (i+1) images column-wise. Similarly, each image is further decomposed into (i+1) images, and CEM is performed on the columns-wise, resulting in a total of (i+1) images. 2 Zhang image.

[0021] Step (3) involves combining the decomposed images using a multidimensional complementary ensemble empirical mode decomposition algorithm to obtain the defective image portion, resulting in the clearest image, as detailed below:

[0022] A single image is insufficient to reflect the information of the original image. Therefore, an image combination strategy is adopted to combine the decomposed (i+1) images. 2 The images are combined according to the multidimensional complementary integrated empirical mode decomposition algorithm. The images decomposed by column are arranged in the first row, and the remaining images decomposed in the first stage are arranged in order. The specific combination form is according to equation (1-4).

[0023]

[0024] Where C i Image represents the image of the i-th component. i,k Image represents a column of images. j,i Represents a row image, where Image i,k In the image, i and k represent the row and column, respectively. j,i In this context, j and i represent the row and column of the image, K is the number of modal components expanded by column, and J is the number of modal components expanded by row. K and J are equal.

[0025] In complementary set empirical mode decomposition, the first few components are high-frequency signals, and the last few components are low-frequency signals. Correspondingly, in infrared thermal images, the first few components are high-frequency heating noise, and the last few components are low-frequency background noise. Therefore, the image after noise removal and the image after removing the non-uniform background are represented as follows:

[0026]

[0027] Where R is the reconstructed image, a is the initial signal image, and b is the final signal image; in summary, it is first expanded by row and then by column to obtain (i+1). 2 The images are synthesized to obtain C. i The defective image portion is extracted from it to obtain the clearest image.

[0028] A signal-to-noise ratio (SNR) > 2.5 was used as the standard to identify defects and select components for reconstruction, where G... def and G inThese are the average gray values ​​of the non-defect area and the defect area, respectively, σ in The standard deviation of the non-defect region is given by the following formula:

[0029]

[0030] The advantages of this invention are:

[0031] 1. This invention utilizes the temporal and spatial information of thermal image sequences through row-level and column-level mode decomposition, enabling a comprehensive reflection of defect information;

[0032] 2. This invention uses a combination strategy to divide noise into high-frequency noise and low-frequency noise, and removes the modes corresponding to the noise during mode reconstruction, thereby achieving noise reduction and improving the defect detection rate.

[0033] 3. This invention has wide applicability and can be applied to any thermal image sequence, whether it is long pulse heating or eddy current pulse heating.

[0034] 4. Compared with the traditional principal component analysis algorithm, this invention retains all thermal image information, which can more comprehensively represent the image, improve the signal-to-noise ratio, and discover more defects. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the detection device of the present invention;

[0036] Figure 2 Dimensions of the carbon fiber sheets used;

[0037] Figure 3 A schematic diagram of CEEMD decomposition and recombination of modal components in two-dimensional data;

[0038] Figure 4 This is a schematic diagram of the MCEEMD combination strategy;

[0039] Figure 5 To decompose the eight components of the combination using MCEEMD;

[0040] Figure 6 Comparison images of the original images with different defect diameters and the normalized grayscale values ​​at the defect points after processing with MCEEMD. Figure 6 (a)(c)(e) are normalized grayscale images of defect points in the original image, and (b)(d)(f) are the images after processing with MCEEMD.

[0041] Figure 7 The image shows a comparison of the results processed by different methods: (a) original image, (b) image processed by principal component analysis, (c) image processed by thermal signal reconstruction algorithm, and (d) image processed by MCEEMD method in this invention.

[0042] Figure 8 This is a flowchart of CEEMD. Detailed Implementation

[0043] like Figure 1 As shown, two halogen lamps were used to heat the sample plate in long-pulse thermal imaging. The selected halogen lamps were Philips QVF137 lamps, with a rated voltage of 220V and a rated power of 2000W. The infrared thermal imager used was a FLIR A40MWIR model, with a focal length range of 15-50cm, a resolution of 320×240, a temperature sensitivity of 80mK, and a detection band (7.5-13μm) in the mid-infrared band. Long pulses were used to heat the carbon fiber plate. The carbon fiber plate measures 150mm×100mm×3mm, with a length of 150mm, a width of 100mm, and a thickness of 3mm. It was divided into three groups of defect points with different diameters, each group having a defect depth between 0.5mm and 2.0mm. Specific defect parameters are as follows... Figure 2 The heating time was set to 8 seconds, and the frame rate of the infrared thermal imager was set to 50 Hz. Thermal image data was recorded from the heating stage to the cooling stage, with a total of 13 seconds of thermal image data recorded for the heating stage (8 seconds) and the cooling stage (5 seconds).

[0044] Selecting data from the cooling phase, before data processing, logarithmic smoothing is performed on the data using equation (1-6) to obtain better results. The fitted data is organized into a matrix form as shown in (1-2), and CEEMD decomposition is performed on it row by row. After row decomposition, the modal components are recombined to obtain a new image. The number of modal components is selected as 7, which decomposes it into 7 individual modal components and one residual value. The residual value forms a modal component, for a total of 8 modal components. The process of recombining the modal components row by row is as follows: Figure 3 As shown.

[0045] ln(T(t))=a0+a1ln(t)+a2[ln(t)] 2 +…+a n [ln(t)] n (1-6)

[0046] The eight generated IMFs are decomposed column-wise. Each IMF will yield seven new IMFs and one residual value. The modal components and residual values ​​are then constructed in the same manner as above, as follows: Figure 4 As shown, each IMF will receive eight new images, for a total of sixty-four images. These images will be combined according to equation (1-5) to recombine the sixty-four images into eight new images, as shown below. Figure 5 , Figure 5 A combined image of the MCEEMD strategy.

[0047] Analyzing C1-C8, the first two images mainly contain high-frequency components, representing high-frequency heating noise, while the latter two images contain low-frequency components, mainly representing low-frequency background noise. The high-frequency and low-frequency noises are removed, retaining only the components reflecting defect information. C3-C6 are then overlaid to obtain the desired image representing the defect, as shown below. Figure 7 As shown in (d), the MCEEMD reconstructed image shows that all ten defect points can be detected.

[0048] Compare the original image with the reconstructed image. Figure 6 This is a comparison image of three groups of defect points with different diameters in the horizontal direction. The horizontal axis represents the width, and the vertical axis represents the normalized grayscale value. Figure 6 (a)(c)(e) are the normalized grayscale values ​​of the defect points in the original image, and (b)(d)(f) are the results after processing with MCEEMD. The dashed lines are the straight lines for comparing defect-free areas. The processed image can detect all four defect points with a diameter of 10mm, and the four defect points with a diameter of 5mm are also very clear. The three defect points with a diameter of 2.5mm can be clearly detected. Compared with the original image, the processed image has a more obvious defect contrast.

[0049] Principal component analysis and thermal signal reconstruction algorithms were used to process long pulse data, and the results were compared with the MCEEMD method. Figure 7 (b) is principal component analysis, which can detect eight defect points. Figure 7 (c) is a thermal signal reconstruction algorithm that can detect eight defect points. In terms of the number of defects identified, MCEEMD has a better effect.

[0050] Example:

[0051] This embodiment selects defective carbon fiber plate material as the research object to verify the multidimensional complementary set empirical mode decomposition algorithm proposed in this invention.

[0052] Long-pulse thermal imaging was used, employing two 2000W halogen lamps to heat the sample plate. An A40MWIR infrared thermal imager was selected, with a frame rate of 50Hz. The counterfeit plate was a defective carbon fiber plate, measuring 150mm × 100mm × 3mm. The material contained non-porous blind holes as defects, divided into three groups of defects with different diameters. The depth of each group ranged from 0.5mm to 2.0mm. Specific defect parameters are as follows... Figure 2 The heating time was set to 8 seconds and the cooling time to 5 seconds. Thermal image data from the heating stage to the cooling stage was recorded in three-dimensional data in frame time format.

[0053] The frame time series of thermal images is transformed from three-dimensional to two-dimensional. Each two-dimensional matrix is ​​encapsulated into a column, with each row representing the time series graph of each point and each column storing the two-dimensional information of an image. After the transformation, a two-dimensional matrix M is obtained, with a data volume of m×n.

[0054] Before data processing, data smoothing is performed in the logarithmic domain using equation (1-6). In the logarithmic domain, the defect region is non-linear, while the non-defect region is a straight line. The difference between the two is more obvious, thus obtaining better defect effects.

[0055] Perform complementary set empirical mode decomposition (CEEMD) on the rows of the two-dimensional matrix M. The CEEMD process is as follows: Figure 8 Each row of data can be decomposed into 7 IMFs and one residual. The residual is also treated as an IMF, and then... Figure 3 The combination shown is Image1-Image8.

[0056] Image1 through Image8 are decomposed column-wise, with each image consisting of 7 IMFs and one residual. The residual is also treated as an IMF. Following the same combination method as in the previous step, a total of 64 image data points are generated from the 8 images. (1,1) -Image (8,8) .

[0057] The 64 decomposed image data obtained are synthesized according to the combination strategy in (1-4) to obtain 8 synthesized image data C1-C8, which contain noise signals and data signals.

[0058] Noise removal was performed on the combined C1-C8 signals, using the signal-to-noise ratio (SNR) as the distinguishing metric. Signals with an SNR < 2.5 were considered noise signals. The data was divided into high-frequency heating noise, low-frequency background noise, and defect signals sandwiched between the two frequencies. Defect signals were extracted according to the combination strategy (1-5), where C1 and C2 represent high-frequency heating noise, and C7 and C8 represent low-frequency background noise. The combined image after noise removal is the defect data image, as shown below. Figure 7 (d)

[0059] To verify the effectiveness of this invention, the grayscale curve of the original image will be compared with the method described herein. Figure 6 This is a comparison image of three groups of defect points with different diameters in the horizontal direction. The horizontal axis represents the width, and the vertical axis represents the normalized grayscale value. Figure 6 (a)(c)(e) are the normalized grayscale values ​​of the defect points in the original image, (b)(d)(f) are the results after processing with MCEEMD, and the red dashed line is the straight line for comparing the defect-free area. The original image has 7 defect points from the grayscale curve, and 11 defect points can be obtained after processing with the method in this paper.

[0060] To verify the superiority of this invention in defect point extraction, it was compared with traditional principal component analysis and thermal signal reconstruction methods. The number of defects extracted was characterized by the signal-to-noise ratio (SNR>2.5). Table 1 compares the number of defects identified by different image processing algorithms in carbon fiber plates. The algorithm proposed in this invention can detect 10 out of 12 defects in carbon fiber plates, while traditional principal component analysis and thermal signal reconstruction algorithms can only detect 8 defects. Compared with traditional methods, the method proposed in this invention improves the defect point detection rate by 16.7% and has a better defect feature extraction effect.

[0061] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0062] Table 1 Comparison of the number of defects identified in carbon fiber plates by different image processing algorithms

[0063]

Claims

1. A non-destructive testing method based on a multidimensional complementary set empirical mode decomposition algorithm, characterized in that: Specifically, the steps include the following: (1) Construct a three-dimensional matrix of all thermal image data and convert the three-dimensional matrix of thermal image data into a two-dimensional matrix; (2) Perform complementary set empirical mode decomposition on each row of the two-dimensional matrix data, and perform complementary set empirical mode decomposition on each column of the decomposed image; (3) Combine the decomposed images using the multidimensional complementary set empirical mode decomposition algorithm to obtain the defective image portion and obtain the clearest image; the specific steps of step (3) are as follows: The (i+1)² decomposed images are combined using the multidimensional complementary set empirical mode decomposition algorithm. The images decomposed by column are arranged in the first row, and the remaining images decomposed in the first stage are arranged sequentially below. The specific combination form is according to equation (1-4). (1-4) in Representing the Image of one component, Representative column image, Represents a row image, where middle 'k' represents the row and column of the image. j and K represents the row and column of the image, K is the number of modal components expanded by column, and J is the number of modal components expanded by row. K and J are equal. In complementary set empirical mode decomposition, the first few components are high-frequency signals, and the last few components are low-frequency signals. Correspondingly, in infrared thermal images, the first few components are high-frequency heating noise, and the last few components are low-frequency background noise. Therefore, the image after noise removal and the image after removing the non-uniform background are represented as follows: (1-5) Where R is the reconstructed image, A is the initial signal image, and B is the final signal image; A signal-to-noise ratio (SNR) greater than 2.5 was used as the standard for defect identification, and components were selected for reconstruction. and These represent the average grayscale values ​​of the non-defect area and the average grayscale value of the defect area, respectively. The standard deviation of the non-defect region is given by the following formula: （1-6）。 2. The non-destructive testing method based on the multidimensional complementary set empirical mode decomposition algorithm according to claim 1, characterized in that: Step (1) involves constructing a three-dimensional matrix of all thermal image data and converting the three-dimensional matrix of thermal image data into a two-dimensional matrix, as detailed below: Each frame of thermal image data is a two-dimensional matrix, with time represented by the frame number. The data of each frame is represented by a two-dimensional matrix as shown in formula (1-1). All thermal image data constitute a three-dimensional matrix X. i ; (1-1) The frame time series of thermal images is transformed from three-dimensional to two-dimensional. Each two-dimensional matrix is ​​encapsulated into a column, with each row representing the time series plot of each point and each column storing the spatial information of each image. The transformed result is shown in formula (1-2). (1-2) Save the thermal image data using the method described above to obtain a set of two-dimensional data.

3. The non-destructive testing method based on the multidimensional complementary set empirical mode decomposition algorithm according to claim 2, characterized in that: Step (2) involves performing complementary set empirical mode decomposition on each row of the two-dimensional matrix data, and then performing complementary set empirical mode decomposition on the decomposed image column by column, as follows: Assuming the two-dimensional data is m×n, each row of the data is decomposed according to the form of complementary set empirical mode decomposition, resulting in i modal components (IMFs) and a residual value. The modal components of different rows are taken out and recombined. The first i IMFs form i new images, and the remaining residuals form a new image. The selection of the number i of modal components (IMFs) follows the empirical formula (1-3). (1-3) After row decomposition, (i+1) images are obtained. In order to include the temporal and spatial information in the data, the generated (i+1) images are subjected to complementary set empirical mode decomposition by column. Similarly, each image is decomposed into (i+1) images, and complementary set empirical mode decomposition is performed by column, resulting in a total of (i+1)² images.

Images