Method for identifying morphology of micro-cracks in water turbine spiral case under weak light environment

Abstract

The application discloses a method for identifying the morphology of a small crack of a water turbine volute in a weak light environment, and is specifically implemented according to the following steps: video recording volute movement; selecting a target image, partitioning the measured volute structure, screening pixels in different regions as feature points, and calculating an optical flow; constructing a sparse optical flow matrix of different regions, calculating the sparse optical flow information entropy of different regions, and positioning the crack position according to the size of the sparse optical flow information entropy; dividing the region with the crack into subdomains, calculating the dense optical flow of the pixels in different subdomains; constructing a dense optical flow information matrix of different subdomains, calculating the size of the dense optical flow information entropy of different subdomains, and adopting an adaptive dense optical flow information entropy threshold judgment method to identify the morphology of the volute crack. The application solves the problems that the existing volute crack identification method cannot identify the morphology of the crack, the randomness of the identification result is high, and the identification cost is high.

Description

Technical Field

[0001] This invention belongs to the field of structural health monitoring technology, specifically relating to a method for identifying the morphology of microcracks in the spiral casing of a water turbine under low light conditions. Background Technology

[0002] As a crucial component of a hydraulic turbine, the spiral casing plays a vital role in altering water flow direction, increasing water velocity, enhancing turbine efficiency, and preventing water erosion of the turbine blades. Cracks in the spiral casing significantly increase the burden on the drainage pumps in the sump, and in severe cases, can lead to flooding of the power plant. Therefore, crack detection is one of the most important tasks for maintenance personnel in the current inspection and maintenance of hydraulic turbine spiral casings.

[0003] Existing methods for identifying cracks in spiral casings are mainly divided into three types: visual methods, model-based methods, and motion-based methods. Visual methods rely heavily on engineer inspections, which are labor-intensive, have high accuracy depending on the engineer's experience, and exhibit significant randomness in crack identification results. Furthermore, these methods depend on texture information such as color in crack images, making it difficult to distinguish targets with similar colors to cracks, such as rust. Model-based methods primarily utilize deep learning to identify cracks. These methods train neural network models on images containing cracks and then use the trained model to identify cracks. However, deep learning-based methods require a large number of high-quality image samples for training, and their detection accuracy is relatively low in low-light environments, posing a challenge to their application in practical engineering. Additionally, these methods can only identify static cracks and cannot track the dynamic evolution of cracks in real time.

[0004] Compared to crack identification algorithms based on color and images, motion-based crack identification primarily identifies cracks by observing changes in structural motion characteristics before and after crack initiation. For example, it can locate cracks by calculating structural mode shapes and utilizing their singularities. However, the initiation of volute cracks often begins as microcracks, which typically affect higher-order mode shapes. This necessitates motion-based crack identification methods requiring high-speed cameras or numerous high-sensitivity sensors, making mode shape testing expensive and hindering widespread practical application. Furthermore, the existing three methods can only identify crack location, not crack morphology, making it difficult to provide further support for subsequent volute structure repair and lifespan prediction. Summary of the Invention

[0005] The purpose of this invention is to provide a method for identifying the morphology of microcracks in the spiral casing of a water turbine under low light conditions, which solves the problems of existing spiral casing crack identification methods being unable to identify crack morphology, having high randomness in identification results, and high identification costs.

[0006] The technical solution adopted in this invention is a method for identifying the morphology of microcracks in the spiral casing of a water turbine under low light conditions, which is implemented according to the following steps:

[0007] Step 1: Record the movement of the volute via video;

[0008] Step 2: Select the target image, divide the volute structure under test into partitions on the target image, select pixels in different regions as feature points, and calculate the optical flow of feature points in different partitioned regions.

[0009] Step 3: Construct sparse optical flow matrices for different regions, calculate the sparse optical flow information entropy for different regions using the sparse optical flow matrices, and locate the crack position based on the magnitude of the sparse optical flow information entropy.

[0010] Step 4: Divide the cracked region into subdomains and use the image phase optical flow-based subdomain pixel dense optical flow calculation method to calculate the dense optical flow of pixels in different subdomains.

[0011] Step 5: Construct dense optical flow information matrices for different subdomains, calculate the magnitude of dense optical flow information entropy for different subdomains, and propose an adaptive dense optical flow information entropy threshold judgment method to identify the morphology of volute cracks.

[0012] The invention is further characterized by:

[0013] Step 2 is as follows:

[0014] Step 2.1: Select the initial frame of the volute running video as the target image, and divide the structure under test into regions on the target image, with each region being 100×100 pixels in size; use the proposed K-factor to filter the feature points in different regions through the feature point filtering method in different regions to obtain the pixel coordinates of the feature points in different regions of the initial frame target image.

[0015] Feature points within different regions are filtered out. The feature points within the filtered regions are:

[0016]

[0017] In the formula, λ1 and λ2 are the pixel brightness matrices A and B, respectively. T Eigenvalues ​​of A;

[0018] Step 2.2: Based on the pixel coordinates of the feature points in the target image of the initial frame obtained in Step 2.1, the sparse optical flow algorithm is used to calculate the optical flow of the feature points in different sub-regions;

[0019] The optical flow equation based on image grayscale is described as follows:

[0020] I x u+Iy v+I t =0 (2)

[0021] In the formula, I x and I y These are the gradients of the feature point in the horizontal and vertical directions, respectively, I. t Let be the derivative of the feature point brightness with respect to time t, and u and v be the optical flow of the feature point in the horizontal and vertical directions, respectively.

[0022] In step 2.2, the optical flow equation is an overdetermined static equation, which cannot be solved using the brightness of a single feature point. Based on the assumption of uniform pixel motion within the region, a feature point optical flow calculation method based on least squares is adopted. A region is selected around the feature point, namely a feature point p and k pixels within its neighborhood window. Since the pixels in the window have the same motion, there are a total of k equations that satisfy the optical flow equation, namely:

[0023]

[0024] Equation (3) is defined as follows:

[0025]

[0026] Equation (4) is further described as follows:

[0027]

[0028] Step 2.3: Since there are k pixels, there are k equations corresponding to equation (5), with two unknowns u and v. The least squares method is used to solve equation (5), and the optical flow of the feature point in the horizontal and vertical directions is calculated as follows:

[0029]

[0030] Step 3 specifically involves:

[0031] Step 3.1, the sparse optical flow matrix of different regions is described as follows:

[0032]

[0033] In the formula, m is the number of frames in the captured video, and d k (t) is the optical flow calculation result of feature point α at time t, i.e.;

[0034]

[0035] In the formula, u t and v t These represent the optical flow of the feature point at time t in the horizontal and vertical directions, respectively.

[0036] Step 3.2: Perform singular value decomposition on the optical flow information matrix, i.e.:

[0037]

[0038] In the formula, Σ is a diagonal matrix, R and S are the left and right singular value vectors of the sparse optical flow information matrix, respectively, and σ i This represents the i-th singular value of the sparse optical flow information matrix;

[0039] Step 3.3: Using the singular value decomposition results of the optical flow information, calculate the singular value distribution probability, i.e.:

[0040]

[0041] Step 3.4: According to the definition of information entropy, region Ω i The sparse optical flow information entropy is described as follows:

[0042]

[0043] The sparse optical flow information entropy in different regions is calculated, and the region with cracks is determined by the magnitude of the sparse optical flow information entropy in different regions, and the location of the cracks is located.

[0044] Step 4 specifically involves:

[0045] Step 4.1: Divide the cracked area into subdomains of 10×10 pixels. The optical flow equation based on the image phase is:

[0046]

[0047] In the formula, φ θ The phase of the image is represented in the θ direction, and u and v represent the optical flow in the horizontal and vertical directions, respectively.

[0048] Step 4.2: Based on step 2, the optical flow equation based on image phase is used to calculate the optical flow information of all pixels in the subdomain. During the calculation, when the θ direction is selected as horizontal and vertical, i.e. θ = 0° and θ = 90°, the following conditions are met.

[0049]

[0050] The optical flow of any pixel in the subdomain in the horizontal and vertical directions is calculated as follows:

[0051]

[0052]

[0053] The optical flow of any pixel in the image subdomain is calculated by Equation (14), thus obtaining the dense optical flow of all pixels in any subdomain.

[0054] Step 5 specifically involves:

[0055] Step 5.1: Based on the dense optical flow in different subdomains, construct the dense optical flow information matrix for different subdomains and calculate the magnitude of the dense optical flow information entropy in different subdomains;

[0056] Step 5.2: Identify the crack morphology of the volute shell based on the crack morphology identification strategy of adaptive threshold screening.

[0057] The dense optical flow information matrix in step 5.1 is described as follows:

[0058]

[0059] In the formula, m is the number of frames in the captured video, and h g (t) is the dense optical flow calculation result of subdomain pixel g at time t;

[0060] Based on the dense optical flow information entropy matrix, the optical flow information entropy calculation method in step 3 is repeated to calculate the dense optical flow information entropy in different subdomains.

[0061] Step 5.2 The specific process is as follows:

[0062] Step 5.2.1: Calculate the dense optical flow information entropy and count the number of intervals. Utilize the difference in optical flow information entropy between the cracked region and the intact region, and automatically determine the threshold based on the distribution of optical flow information entropy in different intervals.

[0063]

[0064] In the formula, M is the calculated result of dense optical flow information entropy in different intervals, and Q is the number of subdomains;

[0065] Step 5.2.2: Identify the morphology of the volute crack by using the determined threshold.

[0066] The beneficial effects of this invention are as follows: The method for identifying the morphology of microcracks in the turbine casing under low light conditions achieves accurate identification of the location and morphology of the casing crack through sparse optical flow information entropy and dense optical flow information entropy. Based on the difference in the motion characteristics of the structure before and after the initiation of cracks in the casing, and considering the advantage of optical flow algorithms in structural motion measurement that do not require marking, a sparse optical flow information entropy algorithm for casing crack location and a dense optical flow information entropy algorithm for crack morphology identification are proposed. This solves the problems of low accuracy, high location cost, and inability to identify crack morphology in existing casing crack location methods, and achieves accurate identification of the location and morphology of casing cracks under unmarked conditions. Attached Figure Description

[0067] Figure 1 This is a flowchart of the method for identifying the morphology of microcracks in the spiral casing of a water turbine under low light conditions, according to the present invention.

[0068] Figure 2(a) is a schematic diagram of the crack loading platform in Embodiment 3 of the present invention;

[0069] Figure 2(b) is a schematic diagram of the specimen structure of Embodiment 3 of the present invention;

[0070] Figure 3 This is the result of crack region division in Embodiment 3 of the present invention;

[0071] Figure 4(a) is a schematic diagram of any crack region selected in Embodiment 3 of the present invention;

[0072] Figure 4(b) is a schematic diagram of the regional feature point detection results in Embodiment 3 of the present invention;

[0073] Figure 5 This is a graph showing the calculation results of sparse optical flow information entropy in different regions in Embodiment 3 of the present invention;

[0074] Figure 6(a) is a magnified 10 times of the region with large sparse optical flow information entropy in Embodiment 3 of the present invention;

[0075] Figure 6(b) is a magnified 100 times view of the region with large sparse optical flow information entropy in Embodiment 3 of the present invention;

[0076] Figure 7 This is a diagram showing the subdomain division result of the cracked region in Embodiment 3 of the present invention;

[0077] Figure 8 This is a graph showing the calculation results of dense optical flow information entropy in different subdomains in Embodiment 3 of the present invention;

[0078] Figure 9 This is a schematic diagram of the threshold distribution results in Embodiment 3 of the present invention;

[0079] Figure 10 This is the crack morphology identification result based on dense optical flow information entropy in Embodiment 3 of the present invention. Detailed Implementation

[0080] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0081] Example 1

[0082] The present invention relates to a method for identifying the morphology of microcracks in the spiral casing of a water turbine under low-light conditions, such as... Figure 1 As shown, this embodiment is implemented according to the following steps:

[0083] Step 1: Record the movement of the volute on video;

[0084] Step 2: Select the target image, divide the volute structure under test into partitions on the target image, select pixels in different regions as feature points, and calculate the optical flow of feature points in different partitioned regions.

[0085] Step 3: Construct sparse optical flow matrices for different regions, calculate the sparse optical flow information entropy for different regions using the sparse optical flow matrices, and locate the crack position based on the magnitude of the sparse optical flow information entropy.

[0086] Step 4: Divide the cracked region into subdomains and use the image phase optical flow-based subdomain pixel dense optical flow calculation method to calculate the dense optical flow of pixels in different subdomains.

[0087] Step 5: Construct dense optical flow information matrices for different subdomains, calculate the magnitude of dense optical flow information entropy for different subdomains, and use an adaptive dense optical flow information entropy threshold judgment method to identify the morphology of the volute crack.

[0088] Example 2

[0089] The present invention relates to a method for identifying the morphology of microcracks in the spiral casing of a water turbine under low-light conditions, such as... Figure 1 As shown, this embodiment is implemented according to the following steps:

[0090] Step 1: Record the movement of the volute using a camera;

[0091] Step 2: Select the target image, divide the volute structure under test into partitions on the target image, select pixels in different regions as feature points, and calculate the optical flow of feature points in different partitioned regions.

[0092] Step 2.1: Select the initial frame of the volute running video as the target image, and divide the structure under test into regions on the target image, with each region being 100×100 pixels in size; use the proposed K-factor to filter the feature points in different regions through the feature point filtering method in different regions to obtain the pixel coordinates of the feature points in different regions of the initial frame target image.

[0093] Feature points within different regions are filtered out. The feature points within the filtered regions are:

[0094]

[0095] In the formula, λ1 and λ2 are the pixel brightness matrices A and B, respectively. T Eigenvalues ​​of A;

[0096] Step 2.2: Based on the pixel coordinates of the feature points in the target image of the initial frame obtained in Step 2.1, the sparse optical flow algorithm is used to calculate the optical flow of the feature points in different sub-regions;

[0097] The optical flow equation based on image grayscale is described as follows:

[0098] I x u+I y v+I t =0 (2)

[0099] In the formula, I x and I y These are the gradients of the feature point in the horizontal and vertical directions, respectively, I. t Let be the derivative of the feature point brightness with respect to time t, and u and v be the optical flow of the feature point in the horizontal and vertical directions, respectively.

[0100] In step 2.2, the optical flow equation is an overdetermined static equation, which cannot be solved using the brightness of a single feature point. Based on the assumption of uniform pixel motion within the region, a feature point optical flow calculation method based on least squares is adopted. A region is selected around the feature point, namely a feature point p and k pixels within its neighborhood window. Since the pixels in the window have the same motion, there are a total of k equations that satisfy the optical flow equation, namely:

[0101]

[0102] Equation (3) is defined as follows:

[0103]

[0104] Equation (4) is further described as follows:

[0105]

[0106] Step 2.3: Since there are k pixels, there are k equations corresponding to equation (5), with two unknowns u and v. The least squares method is used to solve equation (5), and the optical flow of the feature point in the horizontal and vertical directions is calculated as follows:

[0107]

[0108] Step 3: Based on the optical flow results of feature points in different regions, construct sparse optical flow matrices for different regions, calculate the sparse optical flow information entropy for different regions using the sparse optical flow matrix, and locate the crack position based on the magnitude of the sparse optical flow information entropy.

[0109] Step 3.1, the sparse optical flow matrix of different regions is described as follows:

[0110]

[0111] In the formula, m is the number of frames in the captured video, and d k (t) is the optical flow calculation result of feature point α at time t, i.e.;

[0112]

[0113] In the formula, u t and v t Let be the optical flow of the feature point at time t in the horizontal and vertical directions, respectively.

[0114] Step 3.2: Perform singular value decomposition on the optical flow information matrix, i.e.:

[0115]

[0116] In the formula, Σ is a diagonal matrix, R and S are the left and right singular value vectors of the sparse optical flow information matrix, respectively, and σ i It represents the i-th singular value of the sparse optical flow information matrix.

[0117] Step 3.3: Using the singular value decomposition results of the optical flow information, calculate the singular value distribution probability, i.e.:

[0118]

[0119] Step 3.4: According to the definition of information entropy, region Ω i The sparse optical flow information entropy is described as follows:

[0120]

[0121] The sparse optical flow information entropy in different regions is calculated, and the region with cracks is determined by the magnitude of the sparse optical flow information entropy in different regions, and the location of the cracks is located.

[0122] Step 4: Divide the cracked region into subdomains. Based on the optical flow equation in Step 2, use the image phase optical flow-based subdomain pixel dense optical flow calculation method to calculate the dense optical flow of pixels in different subdomains.

[0123] Step 4.1: Divide the cracked area into subdomains of 10×10 pixels. The optical flow equation based on the image phase is:

[0124]

[0125] In the formula, φ θ The phase of the image is represented in the θ direction, and u and v represent the optical flow in the horizontal and vertical directions, respectively.

[0126] Step 4.2: Based on step 2, the optical flow equation based on image phase is used to calculate the optical flow information of all pixels in the subdomain. During the calculation, when the θ direction is selected as horizontal and vertical, i.e. θ = 0° and θ = 90°, the following conditions are met.

[0127]

[0128] The optical flow of any pixel in the subdomain in the horizontal and vertical directions is calculated as follows:

[0129]

[0130] The optical flow of any pixel in the image subdomain is calculated by Equation (14), thus obtaining the dense optical flow of all pixels in any subdomain.

[0131] Step 5: Based on the dense optical flow of different subdomains, construct the dense optical flow information matrix of different subdomains, calculate the magnitude of the dense optical flow information entropy of different subdomains, and propose an adaptive dense optical flow information entropy threshold judgment method to identify the morphology of the volute crack.

[0132] Step 5.1: Based on the dense optical flow in different subdomains, construct the dense optical flow information matrix for each subdomain, and calculate the magnitude of the dense optical flow information entropy in each subdomain. The dense optical flow information matrix is ​​described as follows:

[0133]

[0134] In the formula, m is the number of frames in the captured video, and h g (t) is the dense optical flow calculation result of subdomain pixel g at time t;

[0135] Based on the dense optical flow information entropy matrix, the optical flow information entropy calculation method in step 3 is repeated to calculate the dense optical flow information entropy in different subdomains.

[0136] Step 5.2: Identify the crack morphology of the volute shell based on the crack morphology identification strategy of adaptive threshold screening.

[0137] Step 5.2 The specific process is as follows:

[0138] Step 5.2.1: Calculate the dense optical flow information entropy and count the number of intervals. Utilize the difference in optical flow information entropy between the cracked region and the intact region, and automatically determine the threshold based on the distribution of optical flow information entropy in different intervals.

[0139]

[0140] In the formula, M represents the calculated dense optical flow entropy in different intervals, and Q represents the number of subdomains.

[0141] Step 5.2.2: Identify the morphology of the volute crack by using the determined threshold.

[0142] Example 3

[0143] The crack loading platform is shown in Figure 2(a). Under simulated low-light conditions, speckle processing was applied to the surface of the metal specimen, as shown in Figure 2(b). The crack region division results are as follows: Figure 3As shown, due to the influence of speckle on the surface of the metal structure, it is impossible to identify the location and morphology of the crack.

[0144] The method for identifying the morphology of microcracks in the spiral casing of a water turbine under low-light conditions, as described in this embodiment, is implemented according to the following steps:

[0145] Step 1: Record the movement of the volute using a camera;

[0146] Step 2: Select the target image and divide the test volute structure into partitions on the target image. Select any crack region as shown in Figure 4(a). Filter the pixels in different regions as feature points and calculate the optical flow of the feature points in different partitioned regions.

[0147] Step 2.1: Select the initial frame of the volute running video as the target image, and divide the structure under test into regions on the target image, with each region being 100×100 pixels in size; use the proposed K-factor to filter the feature points in different regions through the feature point filtering method in different regions to obtain the pixel coordinates of the feature points in different regions of the initial frame target image.

[0148] Feature points within different regions are filtered out. The feature points within the filtered regions are:

[0149]

[0150] In the formula, λ1 and λ2 are the pixel brightness matrices A and B, respectively. T The feature values ​​of A; the feature point detection results are shown in Figure 4(b). The proposed feature point screening method can effectively extract structural feature points.

[0151] Step 2.2: Based on the pixel coordinates of the feature points in the target image of the initial frame obtained in Step 2.1, the sparse optical flow algorithm is used to calculate the optical flow of the feature points in different sub-regions;

[0152] The optical flow equation based on image grayscale is described as follows:

[0153] I x u+I y v+I t =0 (2)

[0154] In the formula, I x and I y These are the gradients of the feature point in the horizontal and vertical directions, respectively, I. t Let be the derivative of the feature point brightness with respect to time t, and u and v be the optical flow of the feature point in the horizontal and vertical directions, respectively.

[0155] In step 2.2, the optical flow equation is an overdetermined static equation, which cannot be solved using the brightness of a single feature point. Based on the assumption of uniform pixel motion within the region, a feature point optical flow calculation method based on least squares is adopted. A region is selected around the feature point, namely a feature point p and k pixels within its neighborhood window. Since the pixels in the window have the same motion, there are a total of k equations that satisfy the optical flow equation, namely:

[0156]

[0157] Equation (3) is defined as follows:

[0158]

[0159] Equation (4) is further described as follows:

[0160]

[0161] Step 2.3: Since there are k pixels, there are k equations corresponding to equation (5), with two unknowns u and v. The least squares method is used to solve equation (5), and the optical flow of the feature point in the horizontal and vertical directions is calculated as follows:

[0162]

[0163] Step 3: Based on the optical flow results of feature points in different regions, construct sparse optical flow matrices for different regions, calculate the sparse optical flow information entropy for different regions using the sparse optical flow matrix, and locate the crack position based on the magnitude of the sparse optical flow information entropy.

[0164] Step 3.1, the sparse optical flow matrix of different regions is described as follows:

[0165]

[0166] In the formula, m is the number of frames in the captured video, and d k (t) is the optical flow calculation result of feature point α at time t, i.e.;

[0167]

[0168] In the formula, u t and v t Let be the optical flow of the feature point at time t in the horizontal and vertical directions, respectively.

[0169] Step 3.2: Perform singular value decomposition on the optical flow information matrix, i.e.:

[0170]

[0171] In the formula, Σ is a diagonal matrix, R and S are the left and right singular value vectors of the sparse optical flow information matrix, respectively, and σ iIt represents the i-th singular value of the sparse optical flow information matrix.

[0172] Step 3.3: Using the singular value decomposition results of the optical flow information, calculate the singular value distribution probability, i.e.:

[0173]

[0174] Step 3.4: According to the definition of information entropy, region Ω i The sparse optical flow information entropy is described as follows:

[0175]

[0176] The sparse optical flow information entropy in different regions is calculated, and the region with cracks is determined by the magnitude of the sparse optical flow information entropy in different regions, and the location of the cracks is located.

[0177] These are the calculation results of sparse optical flow information entropy in different regions and the crack location identification results in Embodiment 3 of the present invention, such as... Figure 5 As shown, compared to other regions, regions 6 and 8 have higher sparse optical flow information entropy, indicating the highest probability of cracks. Enlarged images of regions 6 and 8 are shown in Figures 6(a) and 6(b). The results demonstrate the presence of cracks in regions 6 and 8, proving the effectiveness of the proposed crack identification method based on sparse optical flow.

[0178] Step 4: Divide the cracked region into subdomains. Based on the optical flow equation in Step 2, use the image phase optical flow-based subdomain pixel dense optical flow calculation method to calculate the dense optical flow of pixels in different subdomains.

[0179] Step 4.1: Further subdivide the cracked regions 6 and 8 into subdomains. The subdivision results are as follows: Figure 7 As shown, each subdomain is 10×10 pixels in size, and a total of 200 subdomains are divided. The optical flow equation based on the image phase is:

[0180]

[0181] In the formula, φ θ The phase of the image is represented in the θ direction, and u and v represent the optical flow in the horizontal and vertical directions, respectively.

[0182] Step 4.2: Based on step 2, the optical flow equation based on image phase is used to calculate the optical flow information of all pixels in the subdomain. During the calculation, when the θ direction is selected as horizontal and vertical, i.e. θ = 0° and θ = 90°, the following conditions are met.

[0183]

[0184] The optical flow of any pixel in the subdomain in the horizontal and vertical directions is calculated as follows:

[0185]

[0186] The optical flow of any pixel in the image subdomain is calculated by Equation (14), thus obtaining the dense optical flow of all pixels in any subdomain.

[0187] Step 5: Based on the dense optical flow of different subdomains, construct the dense optical flow information matrix of different subdomains, calculate the magnitude of the dense optical flow information entropy of different subdomains, and propose an adaptive dense optical flow information entropy threshold judgment method to identify the morphology of the volute crack.

[0188] Step 5.1: Based on the dense optical flow in different subdomains, construct the dense optical flow information matrix for each subdomain, and calculate the magnitude of the dense optical flow information entropy in each subdomain. The dense optical flow information matrix is ​​described as follows:

[0189]

[0190] In the formula, m is the number of frames in the captured video, and h g (t) is the dense optical flow calculation result of subdomain pixel g at time t;

[0191] Based on the dense optical flow information entropy matrix, the optical flow information entropy calculation method in step 3 is repeated to calculate the dense optical flow information entropy in different subdomains.

[0192] Step 5.2: Identify the crack morphology of the volute shell based on the crack morphology identification strategy of adaptive threshold screening.

[0193] Step 5.2 The specific process is as follows:

[0194] Step 5.2.1: Calculate the dense optical flow information entropy and count the number of intervals. Utilize the difference in optical flow information entropy between the cracked region and the intact region, and automatically determine the threshold based on the distribution of optical flow information entropy in different intervals.

[0195]

[0196] In the formula, M represents the calculated dense optical flow entropy in different intervals, and Q represents the number of subdomains.

[0197] Step 5.2.2: Identify the morphology of the volute crack by using the determined threshold.

[0198] The phase optical flow algorithm of Embodiment 3 of this invention is used to calculate the optical flow of all pixels in the subdomain, and the dense optical flow information entropy of each subdomain is further obtained. The threshold is then calculated using the threshold discrimination method proposed in this invention. Figure 8As shown, the threshold distribution ranges from 0.17 to 0.18. The calculated dense optical flow entropy is then filtered using the threshold distribution results, as follows: Figure 9 As shown, regions exceeding a threshold are considered cracked areas. Sub-regions exceeding the threshold are then labeled, such as... Figure 10 As shown, the results indicate that subdomains above the threshold can completely identify the morphology of cracks, proving the effectiveness of the method for identifying the morphology of microcracks in turbine volutes under low light conditions. The identification results are highly accurate and reliable.

Claims

1. A method for identifying the morphology of microcracks in the spiral casing of a water turbine under low-light conditions, characterized in that, The specific steps are as follows: Step 1: Record the movement of the volute via video; Step 2: Select the target image, divide the volute structure under test into partitions on the target image, select pixels in different regions as feature points, and calculate the optical flow of feature points in different partitioned regions. Step 3: Construct sparse optical flow matrices for different regions, calculate the sparse optical flow information entropy for different regions using the sparse optical flow matrices, and locate the crack position based on the magnitude of the sparse optical flow information entropy. Step 4: Divide the cracked region into subdomains and use the image phase optical flow-based subdomain pixel dense optical flow calculation method to calculate the dense optical flow of pixels in different subdomains. Step 5: Construct dense optical flow information matrices for different subdomains, calculate the magnitude of dense optical flow information entropy for different subdomains, and propose an adaptive dense optical flow information entropy threshold judgment method to identify the morphology of volute cracks. Step 5 specifically involves: Step 5.1: Based on the dense optical flow in different subdomains, construct the dense optical flow information matrix for different subdomains and calculate the magnitude of the dense optical flow information entropy in different subdomains; Step 5.2: Identify the crack morphology of the volute shell based on the crack morphology identification strategy of adaptive threshold screening; The dense optical flow information matrix in step 5.1 is described as follows: （15） In the formula, m The number of frames in the captured video. yes t Time subdomain pixels g Results of dense optical flow calculations; Based on the dense optical flow information entropy matrix, repeat the optical flow information entropy calculation method in step 3 to calculate the dense optical flow information entropy in different subdomains; Step 5.2 The specific process is as follows: Step 5.2.1: Calculate the dense optical flow information entropy and count the number of intervals. Utilize the difference in optical flow information entropy between the cracked region and the intact region, and automatically determine the threshold based on the distribution of optical flow information entropy in different intervals. （16） In the formula, M is the calculated result of dense optical flow information entropy in different intervals, and Q is the number of subdomains; Step 5.2.2: Identify the morphology of the volute crack by using the determined threshold.

Images