An infrared image defect contour detection method for wind turbine blade composite materials

Abstract

The present invention discloses a method for detecting the defect contour of an infrared image of a composite material of a wind turbine blade, and relates to the field of nondestructive testing of composite materials. The method is to perform infrared flaw detection on the wind turbine blade, obtain the original thermal image, perform contrast enhancement processing using the grayscale bipolar method, and reduce noise using the dynamic filtering method; calculate the gradient amplitude and gradient direction, and perform non-maximum suppression on the gradient amplitude of the image to retain the true edge of the defect; use an adaptive double threshold method to connect edge points, and the adaptive double threshold method automatically solves the best high threshold. According to the infrared image feature research of the composite material, the low threshold is taken as six sevenths of the high threshold, so as to determine two thresholds; further morphological processing is performed on the defect contour obtained by the adaptive double threshold method, and the local gaps of the contour are connected. The method provided by the present invention has a good denoising effect, can effectively realize the defect edge detection of the composite material of the wind turbine blade, and the defect contour extraction is clear and accurate.

Description

Technical Field

[0001] The invention belongs to the field of infrared image processing, and in particular relates to an infrared image defect contour detection method for wind turbine blade composite materials. Background Art

[0002] With the development of the times, the earth's non-renewable resources such as coal and oil have been continuously developed and used. These resources are precious and limited. At the same time, with the gradual emergence of global warming, haze and other pollution phenomena, people have begun to develop and utilize clean resources. Among them, wind energy is a green resource that produces almost no greenhouse gases during the utilization process and is a renewable energy source. As an important component of wind turbines, wind turbine blades account for more than 20% of the cost of the entire machine. Their good design and reliable quality are the key to determining the performance of wind turbines. Wind turbine blades are usually a type of composite material product made of epoxy resin as the matrix and glass fiber cloth as the reinforcement, and vacuum casting. During the processing of the blades, defects such as pores, fiber breakage and inclusions will occur, and during the mold joining process with the beam, there will be glue shortage and debonding. The existence of defects will affect the performance of wind turbine blades during operation, shorten the life of the wind turbine, and even cause accidents, resulting in certain economic losses. Therefore, it is very important to detect defects in wind turbine blades.

[0003] Infrared nondestructive testing technology is widely used in material defect detection due to its advantages of non-contact detection, fast measurement speed, intuitive measurement results, and large detection area. When infrared detection technology is applied to the defect detection of wind turbine blade composite materials, due to the influence of the wind turbine blade structure, the infrared image is very non-uniform, the imaging effect is not ideal, and the defect edge is blurred and difficult to identify, which affects the detection effect of wind turbine blade composite material defects.

[0004] In order to improve the quality of infrared images, it is necessary to enhance the acquired infrared images, highlight the edge contours of defects, and improve the defect detection effect. Traditional infrared image processing methods have the problem of unclear image contrast when processing infrared images of wind turbine blade composite materials, and poor detection effect on wind turbine blade composite materials with low signal-to-noise ratio. In addition, several major threshold parameters that affect the detection results are usually set manually based on experience, and repeated tests are required to find the optimal parameters. The detection efficiency is low, and the extraction of the detected defect edge contour is incomplete. These problems lead to unsatisfactory defect edge detection effect and make it difficult to achieve high-precision detection of defect contours of wind turbine blade composite materials. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides an infrared image defect contour detection method for wind turbine blade composite materials, which is used for high-precision detection of defect edge contours of wind turbine blades, specifically:

[0006] Step 1: Use a pulsed infrared detection system to detect defective wind turbine blade composite material specimens, obtain the original thermal image, and use the grayscale bipolar method to perform contrast enhancement processing;

[0007] Step 2: Dynamically filter the contrast-enhanced image to reduce image noise and improve image quality;

[0008] Step 3: Use the Sobel operator to calculate the image gradient amplitude and gradient direction for the image after step 2;

[0009] Step 4: Use the non-maximum suppression method on the gradient amplitude and direction calculated in step 3 to retain the true edge of the defect;

[0010] Step 5: Adopt the adaptive double threshold method to connect the edge points in a certain area. The adaptive double threshold method automatically solves the optimal high threshold. According to the infrared image characteristics of composite materials, the low threshold is taken as six sevenths of the high threshold, so as to determine the two thresholds;

[0011] Step 6: Perform morphological closing operation on the image obtained in step 5 based on the optimal structural element, and connect the contours with local gaps;

[0012] Compared with the prior art, the present invention aims at the problems that the infrared image noise processing effect of wind turbine blade composite materials is poor, the efficiency of manual threshold search is low, the image enhancement effect is not obvious, and the defect contour extraction is not accurate. A method for detecting the defect contour of an infrared image of a wind turbine blade composite material is proposed. On the basis of the traditional infrared image processing method, according to the infrared image characteristics of the wind turbine blade composite material, the grayscale bipolar method is adopted to make the background area and the foreground area pixels form a contrast, and the image contrast is enhanced; the dynamic filtering method is adopted to effectively filter out environmental noise and improve the image quality; in addition, the adaptive dual threshold connection method is adopted to automatically give the best high and low thresholds of the infrared image, avoiding the uncertainty of the artificially given threshold and greatly improving the detection efficiency; based on the optimal structural element, the morphological closed operation is used to extract the defect contour completely, and the detection effect is improved. The infrared defect detection method provided by the present invention is effective and feasible, the defect contour is clear and accurate, the complexity and inconvenience caused by the manual selection of thresholds are avoided, the detection efficiency is greatly improved, and the high-precision detection of defects in wind turbine blade composite materials is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] By describing in detail the embodiments shown in conjunction with the accompanying drawings, the drawings described below are only some embodiments of the present invention, in which:

[0014] Figure 1 The overall flow chart of an infrared image defect contour detection method for wind turbine blade composite materials;

[0015] Figure 2This is the image contrast enhancement effect diagram;

[0016] Figure 3 This is a dynamic filtering effect diagram;

[0017] Figure 4 The effect diagram of the traditional Canny operator contour extraction and the effect diagram of the contour extraction based on the adaptive double threshold method;

[0018] Figure 5 Fill the hole with the effect diagram;

[0019] Figure 6 Calculate the graph for defect side length; DETAILED DESCRIPTION

[0020] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. The present invention proposes an infrared image defect contour detection method for wind turbine blade composite materials. Figure 1 This is the flow chart of the method. After obtaining the original thermal map, the grayscale bipolar method is used to enhance the image contrast, and the dynamic filtering method is used for denoising. The defect edges are extracted through gradient value and gradient direction calculation, non-maximum suppression, adaptive dual threshold connection and morphological closing operation, so as to achieve high-precision detection of defect contours of wind turbine blade composite materials.

[0021] The defective wind turbine blade laminate specimen prefabricated in this embodiment has a size of 220 mm in length, 110 mm in width, and 5 mm in thickness, and is divided into 5 layers, each layer is 1 mm thick. The defect sizes are set to (from left to right) 2 mm, 4 mm, 6 mm, 8 mm, 10 mm, and 15 mm, respectively; the defect depths are set to (from top to bottom) 1 mm, 1.5 mm, and 2 mm, respectively.

[0022] Step 1: Use a pulsed infrared detection system to detect defective wind turbine blade composite material specimens, obtain the original thermal image, and use the grayscale bipolar method to perform contrast enhancement processing;

[0023] Pulse infrared detection is used, with an excitation source of 2000W, a pulse width of 6s, and a pulse period of 20s. Heat is applied to the test piece for a total of 80s. The original infrared thermal image is obtained, and one frame of the infrared thermal image is selected for defect contour extraction.

[0024] The grayscale bipolar method is used to enhance the contrast of the acquired original thermal image. The original thermal image is first grayed out. In order to make the defective and healthy parts of the image have a larger contrast, the grayscale value of the image is statistically calculated, and the optimal grayscale distribution interval is automatically calculated. The grayscale distribution interval is mapped to [0,1] to achieve the effect of making the white part of the processed image whiter and the black part darker.

[0025] Image contrast enhancement effects such as Figure 2As shown, Figure 2 (a) is the original infrared image. Figure 2 (b) is the grayscale image of the color image. Figure 2 (c) is the image contrast enhancement effect diagram. Figure 2 By comparing (b) and 2(c), it can be seen that the defect information of the image without contrast enhancement is not obvious. After adding image contrast enhancement processing, the image contrast is significantly enhanced and the defect information is clearer.

[0026] Step 2: The contrast-enhanced image is subjected to dynamic filtering to reduce image noise and improve image quality;

[0027] Specifically, let f x,y is the gray value of point (x, y), D is the current working window, and the window size is 3*3 for the first time. max 、f min 、f med Window D x,y The maximum grayscale value, minimum grayscale value and median grayscale value, D max is the preset maximum window template. Determine whether the median point in the current area is a noise point. If f max >f med >f min , indicating that the median point is not a noise point, then it is necessary to determine whether the pixel value of (x, y) is a noise point. If it satisfies f max >f x,y >f min , the point is considered as a noise point, and the median point is used to replace the original gray value to filter out the noise. Otherwise, the gray value of the current pixel is retained. In the part of judging whether the median point in the current area is a noise point, if the median is identified as a noise point, the window D is automatically enlarged. x,y size.

[0028] Figure 3 This is the dynamic filtering effect diagram. Figure 3 (a) is the original image after contrast enhancement in the previous step. Figure 3 (b) is the heat map after adding noise. Figure 3 (c) is the traditional Gaussian filtering effect. Figure 3 (d) is the dynamic filtering effect diagram. Figure 3 (c) and Figure 3 (d) It can be seen from the noise reduction effect that dynamic filtering has a good noise reduction effect and can effectively remove noise and protect the defect edges.

[0029] Step 3: Use the Sobel operator to calculate the image gradient amplitude and gradient direction for the image after step 2;

[0030] The gradient is a vector that contains magnitude and direction, indicating that the directional derivative of a function at that point takes the maximum value along that direction, that is, the function changes fastest along that direction (the direction of the gradient) at that point, and the rate of change is the largest (the modulus of the gradient), which is the gradient. Since the image is discrete, for an image, there are x-direction and y-direction, as follows Figure 3 *3 shows:

[0031] f(x-1, y-1) f(x-1,y) f(x-1, y+1) f(x, y-1) f(x, y) f(x, y+1) f(x+1,y-1) f(x+1,y) f(x+1, y+1)

[0032] Image F

[0033] Use the Sobel operator to perform convolution calculations on the templates in two directions with the original image F, and calculate the horizontal direction G x Or vertical direction G y Changes in pixel values.

[0034] Calculate the gradient value for each point of image F, and get the approximate gradient: G = |Gx 2 +Gy 2 |, calculate the gradient direction for each point of image F,

[0035] Step 4: Use the non-maximum suppression method on the gradient amplitude and direction calculated in step 3 to retain the true edge of the defect;

[0036] Specifically, based on the gradient value and gradient direction obtained in step 3, each pixel is detected in turn.

[0037] According to the gradient direction, the gradient amplitude is non-maximum suppressed. If a pixel is an edge point, the gradient value of this pixel in the gradient direction is the largest, otherwise its gray value is 0.

[0038] The gradient direction angles θ of 0 to 360° are merged into four directions: 0°, 45°, 90°, and 135°. For all edges, 180° = 0° and 225° = 45° are defined.

[0039] If θ(x, y) = 0°, then check pixels (x+1, y), (x, y), and (x-1, y);

[0040] If θ(x, y) = 45°, then check pixels (x+1, y+1), (x, y), and (x-1, y-1);

[0041] If θ(x, y) = 90°, then check pixels (x, y+1), (x, y) and (x, y-1);

[0042] If θ(x, y) = 135°, then check pixels (x+1, y-1), (x, y), and (x-1, y+1);

[0043] Compare the gradient values ​​of the three pixels to be tested. If the gradient value of the point (x, y) is greater than that of the other two points, the point (x, y) is considered to be an edge point and is marked. Otherwise, it is judged not to be an edge and is assigned 0. Repeat this method to traverse all the pixels in the image.

[0044] Step 5: Adopt the adaptive double threshold method to connect the edge points in a certain area. The adaptive double threshold method automatically solves the optimal high threshold. According to the infrared image characteristics of composite materials, the low threshold is taken as six sevenths of the high threshold, so as to determine the two thresholds;

[0045] Specifically, a threshold is used to divide all pixels in the image F into two parts, the foreground part S and the background part W form a class relationship, and the average grayscale of the foreground part is a, and the ratio of this part of pixels to all pixels in the image is N1. The average grayscale of the background part is b, and the ratio of this part of pixels to all pixels in the image is N2. The global grayscale mean of the image is calculated as mG = a*N1+b*N2; the inter-class variance σ = a(N1-mG) 2 +b(N2-mG) 2 , calculate the inter-class variance of all pixels from 0 to 255 according to the formula, and the maximum value is the high threshold T h , the low threshold is set to six sevenths of the high threshold.

[0046] Figure 4 This is a comparison between the contour extraction effect of the traditional Canny operator and the contour extraction effect based on the adaptive dual threshold method. Figure 4 (a) is the effect diagram of the traditional Canny operator. Figure 4 (b) is the edge extraction diagram of the adaptive dual threshold algorithm. Figure 4 As can be seen from (a), the traditional Canny operator does not retain enough edge details in the defect contour extraction image, resulting in discontinuity of the target contour and loss of edge detail information, which is very different from the actual defect morphology. The algorithm proposed in this embodiment retains the edge details well, highlights the defect contour, and restores the defect morphology well. Figure 4 In addition, compared with the traditional Canny operator contour extraction which requires manual selection of thresholds, the adaptive dual threshold method automatically selects the best high and low thresholds, which can effectively avoid the complexity and inconvenience of manual operation and greatly improve the detection efficiency.

[0047] Step 6: Perform morphological closing operation on the image obtained in step 5 based on the optimal structural element, and connect the contours with local gaps;

[0048] Specifically, due to the discontinuity of the image grayscale value, there are still disconnected parts at the contour, so the morphological closing operation method is used to first dilate the image and then erode the image to connect the disconnected contour lines. First, according to the infrared image characteristics of the composite material of the wind turbine blade, a convolution kernel X with a structural element of 5*5 is created:

[0049]

[0050] The image and the structural element obtained in step 5 are first dilated and then eroded. The convolution kernel traverses each pixel of the image, connects the disconnected contour lines at the defect, and smoothes its boundaries.

[0051] In step 6, the contour image 4(b) obtained in the previous step is subjected to morphological closing operation, and the contour lines that have not been connected after the previous step are connected again, and the following is obtained: Figure 5 (a) As can be seen from the figure, through the morphological closing operation, the edges of the defect contour have been connected to form a complete defect contour morphology. Figure 5 (b) shows the quantitative calculation of defects in the image after hole filling. Figure 6 As shown, Figure 6 (a) is the defect number statistics chart, Figure 6 (b) is the calculation diagram of the side length of each defect, and 6(c) is the comparison diagram between the actual size and the measured size. Figure 6 (c) It can be seen that the measured size of the defect is consistent with the actual size, and the defect contour extraction accuracy is high.

[0052] Finally, it is necessary to explain here that the above-mentioned specific embodiments of the present invention are only used to illustrate or explain the principles of the present invention, and do not constitute a limitation of the present invention. Some non-essential improvements and adjustments made by technicians in this field based on the above-mentioned contents of the present invention belong to the protection scope of the present invention.

Claims

1. A method for detecting infrared image defect contours of composite materials for wind turbine blades, characterized in that: The method comprises the following steps: Step 1: Use a pulsed infrared detection system to detect defective wind turbine blade composite material specimens, obtain the original thermal image, and use the grayscale bipolar method to perform contrast enhancement processing; Step 2: Dynamically filter the contrast-enhanced image to reduce image noise and improve image quality; Step 3: Use the Sobel operator to calculate the image gradient amplitude and gradient direction for the image after step 2; Step 4: Use the non-maximum suppression method on the gradient amplitude and direction calculated in step 3 to retain the true edge of the defect; Step 5: Adopt the adaptive double threshold method to connect the edge points in a certain area. The adaptive double threshold method automatically solves the optimal high threshold. According to the infrared image characteristics of composite materials, the low threshold is taken as six sevenths of the high threshold, so as to determine the two thresholds. The specific steps are as follows: Step 5.1: Divide all pixels in the image F into two parts, the foreground part S and the background part W; Step 5.2: The average grayscale of the foreground part is a, and the ratio of this part of pixels to all pixels of the image is N1; Step 5.3: The average grayscale of the background part is b, and the ratio of this part of pixels to all pixels of the image is N2; Step 5.4: The global grayscale mean of the image is mG, and the inter-class variance is denoted as σ; a*N1+b*N2=mG The expression of between-class variance is: σ=a(N1-mG) 2 +b(N2-mG) 2 According to the formula, the intra-class variance of all pixels between 0 and 255 is calculated, and the maximum value is the high threshold T h ; Step 5.5: Low Threshold T l Set to six sevenths of the high threshold. If the gradient value of an edge pixel is higher than the high threshold, it is marked as a strong edge pixel, and the pixel at this position is 255 and set to 1. If the gradient value of an edge pixel is less than the high threshold and greater than the low threshold, it is marked as a weak edge pixel, and the domain search range is expanded to determine the state of the weak edge. If the gradient value of an edge pixel is less than the low threshold, it is suppressed, and the pixel value at this position is set to 0. Step 6: Based on the optimal structural element, perform morphological closing operation on the image obtained in step 5 to connect the contours with local gaps. The specific steps are as follows: Step 6.1: Create a convolution kernel X with a structure element of 5*5: Step 6.2: Perform a morphological closing operation on the image obtained in step 5 and the structural element. Image closing operation is a process in which the image undergoes dilation and erosion in sequence. The image is first dilated and then eroded. The closing operation of F on X is represented by F·X, which is defined as: Step 6.3: The convolution kernel traverses every pixel of the image, connects the disconnected contours at the defect, and smoothes its boundaries.

2. The infrared image defect contour detection method for wind turbine blade composite materials according to claim 1 is characterized in that: In step 1, a pulsed infrared detection system is used to detect defects in a wind turbine blade composite material specimen, and a grayscale bipolar method is used to perform contrast enhancement processing after obtaining the original thermal image. The specific steps are as follows: Step 1.1: Use a pulsed infrared detection system to detect the defective wind turbine blade specimen and obtain the original infrared thermal image; Step 1.2: Grayscale the original infrared thermal image, count the grayscale values ​​of the image, calculate the optimal grayscale distribution range of the image based on the grayscale value distribution of the image, and linearly map the grayscale distribution range to [0,1]. The grayscale value of the image is correspondingly controlled between 0 and 1, making the white part of the image whiter and the black part darker, achieving a contrast enhancement effect.

3. The infrared image defect contour detection method for wind turbine blade composite materials according to claim 1 is characterized in that: In step 2, dynamic filtering is performed after contrast enhancement, and the specific steps are as follows: Step 2.1: Let f x,y is the grayscale of point (x, y), D is the current working window, and f max 、f min 、f med Window D x,y The maximum grayscale value, minimum grayscale value and median grayscale value, D max The maximum window template allowed by default; Step 2.2: Determine f max 、f min 、f med If f max >f med >f min , jump to step 2.3, otherwise, increase the window D x,y Size, repeat step 2.2, set the window size to an odd number, and increase the size by two each time; Step 2.3: Determine f max 、f min 、f x,y If f max >f x,y >f min , then the output f x,y Otherwise, output f med .

4. The infrared image defect contour detection method for wind turbine blade composite materials according to claim 1 is characterized in that: In step 3, the gradient amplitude and gradient direction are calculated using the first-order differential operator Sobel operator. The specific steps are as follows: Step 3.1: Use the Sobel operator to calculate the first-order derivative of the pixel grayscale of image F; Step 3.2: Calculate the approximate value G of the horizontal partial derivative of image F x and the approximate value of the vertical partial derivative G y , choose a 3*3 kernel; Convolve the Sobel operator with the original image F to calculate the changes in pixel values ​​in the horizontal and vertical directions; Horizontal direction: G x =F*Sobel horizontal operator template =[f(x-1,y+1)+2f(x,y+1)+f(x+1,y+1)]-[f(x-1,y-1)+2f(x,y-1)+f(x+1,y-1)] Vertical direction: G y =F*Sobel vertical operator template =[f(x-1,y-1)+2f(x-1,y)+f(x-1,y+1)]-[f(x+1,y-1)+2f(x+1,y)+f(x+1,y+1)] Step 3.3: Calculate the gradient value for each point of image F, and get the approximate gradient as: Step 3.4: Calculate the gradient direction for each point of image F, 5. The infrared image defect contour detection method for wind turbine blade composite materials according to claim 1 is characterized in that: In step 4, the non-maximum suppression method is used for the gradient amplitude and direction calculated in step 3 to retain the true edge of the defect; the specific steps are as follows: Step 4.1: Combine the gradient value and gradient direction obtained in step 3 and detect each pixel point in turn; Step 4.2: According to the gradient direction, calculate the non-maximum value of the gradient amplitude. If a pixel is an edge point, the gradient value of this pixel in the gradient direction is the largest, otherwise its gray value is 0. The calculation formula is as follows: The gradient direction angle θ from 0 to 360° is merged into four directions: 0°, 45°, 90°, and 135°. For all edges, 180° = 0° and 225° = 45° are defined. If θ(x, y) = 0°, then check pixels (x+1, y), (x, y) and (x-1, y); If θ(x, y) = 45°, then check pixels (x+1, y+1), (x, y), and (x-1, y-1); If θ(x, y) = 90°, then check pixels (x, y+1), (x, y) and (x, y-1); If θ(x, y) = 135°, then check pixels (x+1, y-1), (x, y), and (x-1, y+1); Step 4.3: Compare the gradient values ​​of the three detected pixels. If the gradient value of the point (x, y) is greater than that of the other two points in this gradient direction, the point (x, y) is considered to be an edge point and is marked. Otherwise, it is judged not to be an edge point and is assigned 0. Step 4.4: Repeat step 4.3 to traverse all pixels in the image.

Images