Water turbine non-stop state monitoring method and system, electronic equipment and product

Abstract

The invention belongs to the technical field of water turbine state monitoring, and aims to provide a water turbine non-stop state monitoring method and system, electronic equipment and a product. According to the method, local runner images at multiple different rotation angles are collected in the operation process of the water turbine, and technologies such as image feature extraction, image registration and splicing are combined to reconstruct a complete current runner image of the water turbine, so that full-coverage and high-precision monitoring of the health state of the surface of the runner of the water turbine is realized. Based on the method, on the basis of no sensor contact and high anti-interference capability, health state information such as surface damage and structural deformation of the turbine runner can be quickly identified, and the problems of poor environmental adaptability, low positioning precision, high maintenance cost and the like in the prior art are avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of water turbine state monitoring, and in particular relates to a method, system, electronic equipment and product for monitoring the state of a water turbine without shutting down. Background Art

[0002] Hydraulic turbines are typically installed in water conservancy projects such as rivers and reservoirs. Their operating environment is complex and affected by a variety of factors, including water flow, water pressure, and water quality. Furthermore, during operation, they are subject to significant mechanical stress and hydraulic shock, making them susceptible to various faults. Existing technologies typically perform comprehensive inspections of the turbine runner through regular shutdowns and overhauls. However, in the early stages of a fault, changes in the turbine's operating status are often subtle, and the fault signal is hidden in a large amount of background noise, making it difficult to detect directly. Therefore, utilizing advanced signal processing and feature extraction techniques to capture these weak fault signals from complex monitoring data is key to achieving early warning of turbine failures.

[0003] Non-stop status monitoring can effectively prevent potential failures, reduce maintenance costs, shorten downtime, and improve economic benefits. Currently, there are several methods for non-stop monitoring of the status of hydro turbines: 1) Laser monitoring technology, based on the characteristics of lasers, uses principles such as laser ranging and interference to determine the position, shape and other parameters of the turbine runner surface by measuring the time difference between laser emission and reflection and the change in interference fringes, thereby obtaining information such as the size and deformation of the runner; 2) Vibration monitoring technology, in which acceleration sensors are installed at key locations on the runner. When the runner vibrates, the mass block within the sensor generates a force proportional to the vibration acceleration. The force is converted into an electrical signal through a conversion element, thereby obtaining the vibration characteristics of the runner; 3) Stress and strain monitoring technology, based on the resistance strain effect of metal conductors, when a strain gauge is attached to the runner surface and deforms with the runner, its resistance value changes proportionally to the strain. The strain and stress of the runner are calculated by measuring the change in resistance value; 4) Acoustic monitoring technology, which utilizes the characteristics of ultrasonic waves that, when propagating in a medium, will be reflected, refracted, and scattered at the interface between different media. By analyzing the received ultrasonic signal, it is possible to determine whether there are defects inside the runner and the location and size of the defects.

[0004] However, in the process of using the existing technology, the inventors found that the existing technology has at least the following problems: The effective range of laser measurement is usually short. For large turbine runners, it may be necessary to move the measuring equipment multiple times or add measuring points to obtain complete information, which is complicated and time-consuming. Vibration monitoring technology can only obtain overall vibration information. For runners with complex structures, it is difficult to accurately determine the vibration source and the specific location of the fault. Stress and strain monitoring mainly measures the strain on the surface of the runner. The internal stress distribution cannot be directly obtained and needs to be inferred through complex calculations and assumptions. Ultrasonic waves may not be able to effectively detect defects of certain shapes and directions. For example, cracks parallel to the propagation direction of ultrasonic waves have low detection sensitivity. Based on this, the existing monitoring technology has the defect of limited monitoring defect types, making it difficult to achieve all-round monitoring of the turbine. Summary of the Invention

[0005] The present invention aims to solve the above technical problems at least to a certain extent, and provides a method, system, electronic equipment and product for monitoring the status of a hydraulic turbine without shutting down.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for monitoring the status of a hydraulic turbine without shutting down, comprising: Acquire multiple local runner images at different rotation angles during turbine operation; Perform feature extraction on each local wheel image to obtain a feature descriptor of each local wheel image; According to the feature descriptors of each local wheel image, feature matching is performed on all local wheel image combinations with adjacent acquisition times to obtain a matching point set; Obtaining, based on the matching point set, a perspective transformation matrix corresponding to a combination of all local wheel images acquired adjacently at the same time; According to the perspective transformation matrix corresponding to the combination of all local wheel images acquired at adjacent times, coordinate transformation processing is performed on each local wheel image to obtain multiple local wheel images after coordinate transformation in the same coordinate system; performing image stitching processing on the multiple local runner images after coordinate transformation to obtain a complete current runner image of the turbine; The health status information of the turbine is obtained according to the current runner image.

[0007] In one possible design, feature extraction processing is performed on any local wheel image to obtain a feature descriptor of the local wheel image, including: Performing convolution operations on any of the local wheel images through Gaussian filters of different scales to generate a scale space; Calculating the Hessian matrix of each pixel in any local wheel image in the scale space; Obtaining a determinant response value of the Hessian matrix of each pixel point in any local wheel image, and screening candidate feature points from all pixels of any local wheel image based on the determinant response value of the Hessian matrix of each pixel point; The quasi-Newton method is used to perform sub-pixel optimization on the candidate feature points to obtain the final image feature points corresponding to the candidate feature points; Generate feature descriptors for image feature points.

[0008] In one possible design, the quasi-Newton method is used to perform sub-pixel optimization processing on any candidate feature point to obtain the image feature points corresponding to any candidate feature point, including: Get the approximate matrix of the Hessian matrix of any candidate feature point; By iteratively updating the position and scale of the candidate feature points, the updated feature points are obtained so as to maximize the determinant response value of the approximate matrix corresponding to the updated feature points and obtain the maximum value of the determinant response value of the current approximate matrix; The updated feature point corresponding to the maximum value of the determinant response value of the current approximation matrix is used as the image feature point corresponding to any one of the candidate feature points.

[0009] In a possible design, based on the feature descriptors of each local wheel image, feature matching is performed on all local wheel image combinations with adjacent acquisition times to obtain a matching point set, including: Calculate the Euclidean distance between each feature descriptor in any combination of local wheel images acquired at adjacent times, and use the image feature points corresponding to the feature descriptors with the closest Euclidean distance as the initial matching points of the current two local wheel images. The initial matching points of all combinations of local wheel images acquired at adjacent times constitute the initial matching point set. The RANSAC algorithm is used to remove mismatched points in the initial matching point set to obtain a final matching point set.

[0010] In one possible design, obtaining the health status information of the turbine according to the current runner image includes: Using the SURF algorithm to perform feature extraction processing on the current wheel image to obtain local feature information of the current wheel image; Inputting the current wheel image into a pre-trained feature extraction model to obtain global feature information of the current wheel image; Fusing the local feature information and the global feature information of the current wheel image to obtain fused feature information of the current wheel image; The fused feature information of the current runner image is input into a pre-trained health status assessment model to obtain the health status information of the turbine.

[0011] In one possible design, the health status assessment model adopts a convolutional neural network.

[0012] In one possible design, after obtaining the fused feature information of the current wheel image, the method further includes: Acquire a factory runner image of the turbine, and perform feature extraction processing on the factory runner image using a SURF algorithm to obtain local feature information of the factory runner image; Inputting the factory runner image into the feature extraction model to obtain global feature information of the factory runner image; fusing the local feature information and the global feature information of the factory runner image to obtain fused feature information of the factory runner image; Performing difference analysis on the fused feature information of the current wheel image and the fused feature information of the factory wheel image to obtain a feature difference between the current wheel image and the factory wheel image; Determining whether the feature difference is greater than a preset feature difference threshold; If so, the fused feature information of the current runner image is input into the health status assessment model to obtain the health status information of the turbine.

[0013] In a second aspect, the present invention provides a non-stop state monitoring system for a hydraulic turbine, comprising: An image acquisition module is used to obtain multiple local runner images at different rotation angles during the operation of the turbine; a feature extraction module, communicatively connected to the image acquisition module, for performing feature extraction processing on each local wheel image to obtain a feature descriptor of each local wheel image; a feature matching module, in communication with the feature extraction module, for performing feature matching on all combinations of local wheel images acquired adjacently in time based on feature descriptors of the local wheel images to obtain a matching point set; an image transformation module, communicatively connected to the feature matching module, for obtaining, based on the matching point set, a perspective transformation matrix corresponding to a combination of all local wheel images acquired adjacently at the same time; and for performing coordinate transformation processing on each local wheel image based on the perspective transformation matrix corresponding to the combination of all local wheel images acquired adjacently at the same time, to obtain a plurality of local wheel images after coordinate transformation in the same coordinate system; an image stitching module, communicating with the image transformation module, for stitching a plurality of local runner images after coordinate transformation to obtain a complete current runner image of the turbine; A health status detection module is in communication with the image stitching module and is used to obtain health status information of the turbine based on the current runner image.

[0014] In a third aspect, the present invention provides an electronic device, comprising: a memory for storing computer program instructions; and The processor is configured to execute the computer program instructions to thereby complete the operation of the method for monitoring the status of a hydraulic turbine without shutting down as described in any one of the above.

[0015] In a fourth aspect, the present invention provides a computer program product, comprising a computer program or instructions, which, when executed by a computer, implements a method for monitoring the status of a hydraulic turbine without shutting down as described in any one of the above.

[0016] The beneficial effects of the present invention are: The present invention discloses a method, system, electronic equipment and product for monitoring the status of a water turbine without stopping, which can achieve full coverage and high-precision monitoring of the surface health status of the water turbine runner. Specifically, during the implementation of the present invention, by collecting local runner images at multiple different rotation angles during the operation of the water turbine, and combining image feature extraction, image registration and splicing technologies, a complete current runner image of the water turbine is reconstructed, thereby achieving full coverage and high-precision monitoring of the surface health status of the water turbine runner. Based on the present invention, on the basis of no sensor contact and strong anti-interference ability, health status information such as surface damage and structural deformation of the water turbine runner can be quickly identified, avoiding the problems of poor environmental adaptability, low positioning accuracy and high maintenance cost existing in the prior art, and having good application value and promotion prospects.

[0017] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of a method for monitoring the status of a hydraulic turbine without shutting down in an embodiment; Figure 2 3. It is a module block diagram of a method system for monitoring the status of a hydraulic turbine without shutting down in an embodiment; Figure 3 FIG. 4 is a block diagram of a module of an electronic device in an embodiment. DETAILED DESCRIPTION

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be briefly introduced below in conjunction with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0020] Example 1: This embodiment discloses a method for monitoring the status of a hydraulic turbine without shutting down. The method can be executed by, but is not limited to, a computer device or a virtual machine with certain computing resources, such as a personal computer, a smart phone, a personal digital assistant, or a wearable device, or by a virtual machine.

[0021] like Figure 1 As shown, a method for monitoring the status of a hydraulic turbine without shutting down may include, but is not limited to, the following steps: S1. Acquire multiple local runner images at different rotation angles during the operation of the turbine. It should be noted that this embodiment uses a camera combined with a stroboscope to capture local runner images, wherein the camera and stroboscope are disposed within the transparent top cover of the turbine and are positioned toward the runner. The stroboscope emits short pulses of light at a frequency that is the same as or an integer multiple of the turbine runner speed. The camera must have a frame rate synchronized with the stroboscope. When the stroboscope emits light, a trigger signal is synchronously sent to the camera to control it to capture images at the moment of the flash, ensuring that exposure is synchronized with the flash. The stroboscope setting enables a low-frame-rate camera to achieve high-precision imaging, significantly reducing the cost of image acquisition hardware while avoiding the problem of massive data caused by high-speed shooting.

[0022] Specifically, the original collected initial local wheel image is preprocessed as follows: Grayscale processing is performed on the initial local wheel image, and the color image is converted into a grayscale image to simplify subsequent calculations, thereby obtaining the grayscale processed local wheel image; The local wheel image after grayscale processing is subjected to noise reduction processing to obtain a noise-reduced local wheel image; it should be noted that due to factors such as electromagnetic interference and equipment noise in the monitoring environment, the collected image may contain noise. For this reason, in this embodiment, Gaussian filtering, mean filtering or median filtering are further used to remove noise interference in the local wheel image after grayscale processing, so as to improve image quality and provide a clear image basis for subsequent analysis.

[0023] Geometric correction is performed on the noise-reduced partial wheel image to obtain a preprocessed partial wheel image. It should be noted that due to factors such as shooting angle and lens distortion, the image may contain geometric distortion. Geometric correction of the image accurately represents the shape and position of the wheel, facilitating subsequent dimensional measurement and feature analysis.

[0024] S2. Perform feature extraction processing on each local wheel image to obtain a feature descriptor of each local wheel image.

[0025] In step S2, the SURF (Speeded-Up Robust Features) algorithm is used to perform feature descriptor matching. Specifically, feature extraction processing is performed on any local wheel image to obtain a feature descriptor of the local wheel image, including: S201. Perform convolution operations on any of the local wheel images using Gaussian filters of different scales to generate a scale space; wherein the scale space is: ; Where, The scale is s The function expression of the Gaussian filter is, , represents the convolution operator, represents any local wheel image, represents the pixel coordinates of any local wheel image in a plane rectangular coordinate system; In this embodiment, by performing convolution operations on any of the local wheel images through Gaussian filters of different scales, multi-scale feature point detection can be achieved, ensuring that features can be accurately extracted under different scaling conditions, thereby making the subsequently acquired image feature points scale-invariant.

[0026] S202. Calculate the Hessian matrix (Hessian matrix) of each pixel point in any local wheel image in the scale space; wherein, any local wheel image is s When the Hessian matrix is: ; Where, represents the Gaussian second-order differential The convolution result with any of the local runner images is, represents the Gaussian second-order differential The convolution result with any of the local runner images is, represents the Gaussian second-order differential The convolution result with any of the local runner images is, represents the Gaussian second-order differential convolution result with any of the local wheel images; S203. Obtain the determinant response value of the Hessian matrix of each pixel point in any local wheel image, and screen all pixels of any local wheel image based on the determinant response value of the Hessian matrix of each pixel point to obtain candidate feature points; wherein the determinant response value of the Hessian matrix of any pixel point is: ; Where, oh Represents the preset weight, generally set to 0.9, used to balance the approximation error.

[0027] Specifically, in the process of screening candidate feature points from all pixel points of any local wheel image, for any pixel point in any local wheel image, the larger the corresponding determinant response value, the more obvious the intensity of the image structure change, and the more likely the pixel point is a feature point such as a corner point or an edge intersection. In this embodiment, through non-maximum suppression, the determinant response value of any pixel point is compared with the determinant response values of 26 neighboring points of its adjacent scales (8 neighboring points of the same scale + 9 neighboring points of the upper and lower adjacent scales), and the neighboring points whose determinant response values are local maximum values are retained as candidate feature points. For example, if the determinant response value of any pixel point is the maximum value among them, it is set as a candidate feature point.

[0028] S204. Perform sub-pixel optimization on the candidate feature points using the quasi-Newton method to obtain the final image feature points corresponding to the candidate feature points; In this embodiment, the quasi-Newton method is used to perform sub-pixel optimization processing on any candidate feature point to obtain image feature points corresponding to any candidate feature point, including: S2041. Obtain an approximate matrix of the Hessian matrix of any candidate feature point; S2042. By iteratively updating the position and scale of the candidate feature points, an updated feature point is obtained to maximize the determinant response value of the approximate matrix corresponding to the updated feature point, thereby obtaining the maximum value of the determinant response value of the current approximate matrix; S2043. Use the updated feature point corresponding to the maximum value of the determinant response value of the current approximation matrix as the image feature point corresponding to any one of the candidate feature points.

[0029] It should be noted that in the feature point positioning stage, the existing SURF algorithm uses the Newton method to iterate the candidate feature points to obtain the image feature points, which can determine the precise position and optimal scale of the image feature points and achieve sub-pixel precision positioning of the image feature points. The biggest advantage of the Newton method is that the iterative algorithm has a second-order convergence speed. However, the non-positive Hessian matrix cannot guarantee that the algorithm iteration direction is the objective function. f ( x , y ,σ)=|det(H( x , y ,σ))|at the feature point position and scale ( x , y ,σ). In addition, when using the Newton method, each iteration requires the calculation of the Hessian matrix. For multivariable problems, the huge amount of calculation will consume a lot of time and computer resources, and the Hessian matrix may become singular during iteration, which will make it difficult for the algorithm to converge. To this end, this embodiment adopts the quasi-Newton method to obtain image feature points and uses an approximate matrix to replace the Hessian matrix. This not only retains the advantage of the second-order convergence speed of the Newton method, but also solves the defect that the Hessian matrix needs to be calculated in each iteration.

[0030] Specifically, in this embodiment, the x k Indicates the k The position and scale of the feature points at the iteration, that is, x k =( u k , v k , s k ),( u k , v k ,) is the first k The two-dimensional coordinates of the feature points in the UV coordinate system at the iteration (sub-pixel accuracy), s k For the k The scale of the feature points at the iteration (reflecting the detection scale of the feature points) is used to initialize the approximate matrix of the Hessian matrix of the candidate feature points: B 0= I (Unit matrix), assuming that the initial scale is 1 and the coordinates are the coarse positioning results of the candidate feature points, the iterative derivation process of the approximate matrix during the process of iteratively updating the position and scale of the candidate feature points is as follows: As the objective function f ( x’) is twice continuously differentiable on its domain, and the objective function is obtained f ( x’ ) k +1 The position and scale of the feature points at iteration x k+1 The quadratic approximation model at is formula (1).

[0031] (1) In formula (1), f ( x k+1 ) indicates the k The objective function at +1 iteration, g k+1 Represents the objective function f ( x’ ) k +1 The position and scale of the feature points at iteration x k+1 The gradient at x’ Indicates the current position and scale of the feature point, T represents the transpose symbol, G k+1 Represents the objective function f ( x’ ) k +1 The position and scale of the feature points at iteration x k+1 The Hessian matrix at .

[0032] Derivative of formula (1): (2) Let the gradient change , , feature point displacement , then: (3) Use a matrix that approximates the Hessian matrix B k+1 replace G k+1 We get formula (4), which is usually called the quasi-Newton condition: (4) generally B k It has the following three characteristics.

[0033] (1) In a sense, B k ≈ G k, so that the direction generated by the algorithm is similar to Newton's method, but ensures that the algorithm has a faster convergence speed.

[0034] (2) For the number of iterations k =1, 2, 3..., only approximate matrices B k Is positive, to ensure that the descent direction obtained by iteration is consistent with the objective function f ( x )exist x k The descending direction is consistent.

[0035] (3) Approximate Matrix B k Usually a rank 1 or rank 2 matrix is used for correction, and the selection can be made after comparing the correction effects.

[0036] (5) In formula (5), E k is a rank 1 or rank 2 matrix. If E k Take a rank 1 matrix, then ,in, a represents a preset positive real number, u k represents the rank-1 correction matrix, .

[0037] By quasi-Newton conditions , that is, formula (6): (6) Formula (6) shows that u k Collinear with vector , that is, there is a constant β Make , that is, formula (7): (7) In formula (7), B s Indicates the s The approximate matrix of the Hessian matrix at the iteration, k s Indicates the s A constant term related to the number of iterations.

[0038] Then we have formula (8).

[0039] (8) Therefore, if Desirable , that is, formula (9): , (9) So we get the approximate matrix B k The formula for rank 1 correction is as follows (10): (10) Formula (10) is the approximate matrix B k The iterative derivation is used to replace the complex Hessian matrix to reduce the computational cost. In addition, in the noise area of the wheel image, the real Hessian matrix may have negative eigenvalues due to local interference, and the approximate matrix B k The correction is used to maintain positive definiteness and ensure stable convergence of the iteration.

[0040] S205. Generate feature descriptors of image feature points. Specifically, in the process of generating feature descriptors, first take the image feature point as the center and calculate its radius of 6 s The Harr wavelet features of the pixels in the neighborhood of are calculated in the horizontal and vertical directions, and then Gaussian weight coefficients are assigned to these response values. Then, the responses within the 60° range are accumulated to form a new vector. Finally, the entire circular area is traversed and the longest vector direction is selected as the main direction of the image feature point. Subsequently, with the image feature point as the center, the coordinate axis is rotated to the main direction, and a vector with a side length of 20 is selected according to the main direction. s The square area of the window is divided into 4×4 sub-areas, and the 5 s ×5 s The haar wavelet features in the horizontal and vertical directions of the sampling point are then added together to form the feature vector of each sub-region. For each image feature point, a total of 64-dimensional feature vectors can be generated. The 64-dimensional feature vector is the feature descriptor of the image feature point.

[0041] It should be noted that wheel photography presents issues such as scale, rotation, illumination variations, and noise interference. The SURF algorithm, however, extracts feature descriptors quickly and can rapidly process large numbers of wheel images. Its robustness ensures stable feature point extraction and generation of stable feature descriptors under complex conditions, facilitating subsequent accurate feature matching. In this embodiment, during the screening of image feature points, the quasi-Newton method is employed to obtain image feature points, and an approximate matrix is used to replace the Hessian matrix. This method retains the advantage of the Newton method's second-order convergence speed while addressing the drawback of requiring the calculation of the Hessian matrix for each iteration.

[0042] S3. Based on the feature descriptors of each local wheel image, feature matching is performed on all combinations of local wheel images that are acquired adjacently at the same time to obtain a set of matching points.

[0043] In step S3, based on the feature descriptors of each local wheel image, feature matching is performed on all local wheel image combinations with adjacent acquisition times to obtain a matching point set, including: S301. Calculate the Euclidean distance between each feature descriptor in any combination of adjacent local wheel images acquired at any time, and use the image feature points corresponding to the feature descriptors with the closest Euclidean distance as the initial matching points of the current two local wheel images. The initial matching points of all combinations of adjacent local wheel images acquired at any time constitute the initial matching point set. S302. To improve matching accuracy, the RANSAC algorithm is used to remove incorrect matching points from the initial matching point set to obtain a final matching point set. Specifically, a set of initial matching points corresponding to any two adjacent partial wheel images is randomly selected and assumed to be a correct match. A perspective transformation model is then calculated between these initial matching points. This perspective transformation model is then used to verify the initial matching points of the remaining sets. Initial matching points that meet the calculation rules of the perspective transformation model are considered inliers, while those that do not meet the rules are considered outliers. After multiple iterations, an accurate set of matching points is obtained.

[0044] In this embodiment, the Euclidean distance between the descriptors of each feature point in two adjacent local wheel images is calculated, and the most similar descriptor pair is found to establish a matching relationship between the two adjacent local wheel images, so as to achieve accurate matching between the two adjacent local wheel images, which can provide a reliable basis for subsequent image stitching and health status detection.

[0045] S4. Based on the matching point set, obtain the perspective transformation matrix corresponding to all combinations of local wheel images acquired adjacently at the time of acquisition. Specifically, for any combination of local wheel images acquired adjacently at the time of acquisition, a least squares method is used to solve the perspective transformation matrix between the two local wheel images in the current local wheel image combination based on the coordinate information of the matching points between the two. This perspective transformation matrix can determine the transformation parameters such as rotation, scaling, and translation between the two local wheel images, so that the matching points of the two local wheel images overlap as much as possible after the image transformation.

[0046] S5. Based on the perspective transformation matrix corresponding to the combination of all local wheel images acquired adjacently at the same time, coordinate transformation processing is performed on each local wheel image to obtain multiple local wheel images after coordinate transformation in the same coordinate system. Specifically, in this embodiment, the cv2.warpPerspective function in the Python OpenCV library is used to transform all local wheel images into the same coordinate system for subsequent image stitching.

[0047] S6. Performing image stitching on the multiple coordinate-transformed local runner images to obtain a complete current runner image of the turbine. It should be noted that during the image stitching process, for pixels in the overlapping regions, a weighted average is calculated based on their distance to the image center or other weighting function, and this is used as the pixel value of the pixel in the fused current runner image.

[0048] S7. Obtain health status information of the turbine according to the current runner image.

[0049] In step S7, the health status information of the turbine is obtained according to the current runner image, including: S701. Perform feature extraction on the current wheel image using the SURF algorithm to obtain local feature information of the current wheel image; S702. Input the current wheel image into a pre-trained feature extraction model to obtain the global feature information of the current wheel image; in this embodiment, the feature points of the current wheel image can also be pre-screened in advance to retain feature points with higher response strength, and then the screened feature points are input into the feature extraction model. These strong response feature points can better represent the key features of the image, which helps to reduce the computational complexity of the subsequent feature extraction model.

[0050] S703. Fusing the local feature information and the global feature information of the current wheel image to obtain fused feature information of the current wheel image; It should be noted that the SURF algorithm excels at capturing local details (such as the characteristic points of cracks on blade edges and wear pits) and is insensitive to noise, but it cannot characterize global structures (such as overall blade distortion) and fails in smooth curved areas (without significant characteristic points). The use of a feature extraction model can automatically learn global features (such as the curvature distribution of the blade surface and the wear pattern formed by water impact), but is less sensitive to local subtle defects (such as early microcracks) than traditional algorithms. In this embodiment, by fusing the local feature information obtained based on the SURF algorithm with the global feature information obtained through the feature extraction model, the local and global features can be complemented, filling the blind spots of a single algorithm. In the complex curved surface scenario of the runner, the fused features can increase the recall rate of defect detection from 75% to 92%, avoiding missed detections due to a single feature.

[0051] In step S703, after obtaining the fused feature information of the current wheel image, the method further includes: a1 obtain the turbine runner image, and use the SURF algorithm to extract the image of the runner to obtain local feature information of the runner image; a2. The factory runner image is input into the feature extraction model to obtain global feature information of the factory runner image; a3. The local feature information and global feature information of the factory runner image are fused to obtain the fusion feature information of the factory runner image; a4. Perform difference analysis on the fusion feature information of the current runner image and the fusion feature information of the factory runner image to obtain the characteristic difference between the current runner image and the factory runner image; a5. Determine whether the feature difference is greater than a preset feature difference threshold; a6. If yes, input the fused feature information of the current runner image into the health status assessment model to obtain the health status information of the turbine.

[0052] It should be noted that before performing the health status assessment of the turbine, the feature differences between the current runner image and the factory runner image are compared in advance, and when the feature difference is large, the health status assessment is performed using the health status assessment model. This can solve the limitation of using a single health status assessment model for health status assessment.

[0053] S704. Input the fused feature information of the current runner image into a pre-trained health status assessment model to obtain the health status information of the turbine.

[0054] In this embodiment, the health status assessment model adopts Convolutional Neural Networks (CNN).

[0055] It should be noted that convolutional neural networks are deep learning networks specifically designed for processing grid-structured data, such as images. They primarily consist of convolutional layers, pooling layers, and fully connected layers. Convolutional layers perform convolution operations by sliding multiple convolution kernels with different parameters over the image. Each kernel acts as a feature detector, extracting features from a specific local region of the image. Different kernels can extract different features, such as edges and textures. For example, small kernels can extract low-level features like edges and corners, while large kernels can extract more complex high-level features like shapes and textures. During the convolution process, the kernel is multiplied by the pixel values of the corresponding image region and the sum is calculated to produce the convolution result. This process also introduces a bias value and increases model nonlinearity through activation functions (such as ReLU), enabling the model to learn more complex features. Pooling layers typically follow convolutional layers to downsample the feature maps output by the convolutional layers. Common pooling methods include max pooling and average pooling. Max pooling selects the maximum value within the pooling window as its output, while average pooling calculates the average value within the window as its output. This reduces the amount of data and computation, lowering the risk of overfitting while preserving key features. The fully connected layer connects the output of the pooling layer, integrating all features and calculating the final output using a weight matrix for tasks such as classification and regression. During training, the backpropagation algorithm calculates the gradient of the loss function with respect to each layer's parameters, continuously updating the weights to bring the model's predictions closer to the true values.

[0056] In this embodiment, the feature extraction model also utilizes a convolutional neural network. In wheel image analysis, the convolutional neural network's powerful feature learning capabilities can automatically learn complex features under different health conditions from a large number of wheel images. For wheel images taken from different shooting angles and lighting conditions, the convolutional neural network's convolution operation can effectively extract global features from the image. Combined with the translational invariance of the pooling operation, the model is robust to geometric transformations of the image and can stably analyze wheel images in complex environments. Furthermore, the global features extracted by the convolutional neural network complement the local features extracted by the SURF algorithm. Fusion of the two can provide a more comprehensive feature representation for differential analysis of wheel images, thereby improving the accuracy of subsequent wheel health assessments.

[0057] In the wheel health status assessment task in this embodiment, by training on a large number of labeled wheel images in different states, such as normal and worn, the convolutional neural network can capture subtle feature changes in the image that are difficult to detect with the naked eye, thereby accurately judging the health status of the wheel.

[0058] It's also important to note that the mean square error (MSE) loss function commonly used in convolutional neural networks for image analysis is based on the least squares method. Using the MSE loss function guides the model to learn the mapping between image features and true labels, minimizing the model's prediction error on the training data. However, the least squares method is primarily applicable to linear models or models that can be converted to linear models. Its expressive power is limited for complex nonlinear models and models with high-order interactions. Furthermore, as model complexity increases, such as with an increase in the number of independent variables or the introduction of higher-order terms, the computational complexity of the least squares method increases significantly, and overfitting may occur. Furthermore, multicollinearity can make the regression coefficients estimated by the least squares method unstable and their variance increased, making it difficult to accurately determine the individual impact of each independent variable on the dependent variable. Furthermore, the model becomes highly sensitive to small changes in the data, reducing its predictive stability.

[0059] For a well-posed problem, the matrix of the input convolutional neural network X Is full column rank, the matrix X The result obtained after inputting the convolutional neural network can be expressed as: (11) In formula (11), the matrix X The dimension is , N is the sample size, D is the feature dimension of a single sample, i Represents the weight vector, whose dimension is , the resulting vector y The dimension of ' is , which can represent health status information such as the health status label of the wheel and the quantitative value of the wear level.

[0060] If the loss function of the convolutional neural network is defined as the square of the residual, the loss function minimized by the least squares method can be expressed as formula (12).

[0061] (12) Formula (12) can be solved using the gradient descent method, or directly using Formula (13).

[0062] (13) The matrix X column is not full rank, then , Close to singular solution The error will be very large. To solve this problem, a regularized compensation term can be added to the loss function, which is formula (14).

[0063] (14) Among them, the regularization term is defined as ,in a is a preset positive real number, I is the identity matrix, so: (15) In formula (15), i ( a ) represents the optimal solution of the loss function after adding the regularization term.

[0064] Mathematically deriving the above formula (15) we have: (16) (17) Formula (17) shows that yes Compression towards the origin. So there is Make the mean square error , so under the mean square error criterion, this method is better than linear regression. From a mathematical logic point of view, this method abandons the unbiasedness of the least squares method. Although it loses some information at the cost of more reliable regression coefficients and more realistic regression estimates, it has better training effects on pathological data (expressed as pathological matrices in the algorithm, where a small change in the value of an element in the matrix can cause a large calculation error. This type of matrix is called a "pathological matrix." For example, in the Gaussian elimination method, if the values of the elements on the matrix principal are very small, then this type of matrix will show pathological characteristics when performing Gaussian elimination calculations). The above method can improve the stability of the model's prediction solution, that is, through the regularization parameter a Shrinking the solution toward the origin can reduce the impact of multicollinearity between features, reduce the model's sensitivity to noise, and avoid overfitting.

[0065] It should be noted that during the training of the health status assessment model, a large amount of wheel image data with different health status labels (such as normal, light wear, moderate wear, and severe wear) is first collected and carefully annotated. The annotated data is then divided into training, validation, and test sets. The convolutional neural network is trained using the training data. By adjusting the network parameters, the network accurately learns the characteristic patterns of wheel images in different health states. During the training process, a loss metric such as the cross-entropy loss function is used, and the backpropagation algorithm is used to continuously update the network weights to improve the accuracy of the health status assessment model. Finally, the model's prediction results are validated using the test set data, and metrics such as precision and recall are calculated to evaluate the model's performance. If the model performance is unsatisfactory, the convolutional neural network structure can be adjusted, such as by adding or removing convolutional layers or adjusting the convolution kernel size. Alternatively, the model can be retrained using more training data to optimize performance and obtain more accurate health assessment results.

[0066] In this embodiment, the health status information of the turbine is obtained by inputting the fused feature information of the current impeller image into the health status assessment model. The health status information can be set as a health status label. For example, if it is "moderate wear", it means that the current impeller is in a health state of moderate wear. As another form of health status information, the probability of each health status category can also be output through a softmax function to obtain a health score to more intuitively reflect the health of the impeller.

[0067] This embodiment can achieve full coverage and high-precision monitoring of the surface health status of the turbine runner. Specifically, during the implementation of this embodiment, by collecting local runner images at multiple different rotation angles during the operation of the turbine, and combining image feature extraction, image registration and stitching technologies, a complete current runner image of the turbine is reconstructed, thereby achieving full coverage and high-precision monitoring of the surface health status of the turbine runner. Based on this embodiment, on the basis of no sensor contact and strong anti-interference ability, health status information such as surface damage and structural deformation of the turbine runner can be quickly identified, avoiding the problems of poor environmental adaptability, low positioning accuracy and high maintenance costs existing in the existing technology, and has good application value and promotion prospects.

[0068] Example 2: This embodiment discloses a non-stop state monitoring system for a water turbine for implementing the non-stop state monitoring method for a water turbine in embodiment 1; Figure 2 As shown, the non-stop turbine status monitoring system includes: An image acquisition module is used to obtain multiple local runner images at different rotation angles during the operation of the turbine; a feature extraction module, communicatively connected to the image acquisition module, for performing feature extraction processing on each local wheel image to obtain a feature descriptor of each local wheel image; a feature matching module, in communication with the feature extraction module, for performing feature matching on all combinations of local wheel images acquired adjacently in time based on feature descriptors of the local wheel images to obtain a matching point set; an image transformation module, communicatively connected to the feature matching module, for obtaining, based on the matching point set, a perspective transformation matrix corresponding to a combination of all local wheel images acquired adjacently at the same time; and for performing coordinate transformation processing on each local wheel image based on the perspective transformation matrix corresponding to the combination of all local wheel images acquired adjacently at the same time, to obtain a plurality of local wheel images after coordinate transformation in the same coordinate system; an image stitching module, communicating with the image transformation module, for stitching a plurality of local runner images after coordinate transformation to obtain a complete current runner image of the turbine; A health status detection module is in communication with the image stitching module and is used to obtain health status information of the turbine based on the current runner image.

[0069] It should be noted that the working process, working details and technical effects of the turbine non-stop status monitoring system provided in this embodiment 2 can be found in embodiment 1 and will not be described in detail here.

[0070] Example 3: Based on the embodiment 1 or 2, this embodiment discloses an electronic device, which may be a smart phone, a tablet computer, a laptop computer or a desktop computer. The electronic device may be called a user terminal, a portable terminal, a desktop terminal, etc. Figure 3 As shown, the electronic equipment includes: a memory for storing computer program instructions; and A processor is used to execute the computer program instructions to complete the operation of a method for monitoring the status of a turbine without shutting down as described in any one of the embodiments 1.

[0071] Specifically, the processor 301 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 301 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 301 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 301 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen.

[0072] Memory 302 may include one or more computer-readable storage media, which may be non-transitory. Memory 302 may also include high-speed random access memory and non-volatile memory, such as one or more magnetic disk storage devices or flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in memory 302 is used to store at least one instruction, which is executed by processor 301 to implement the method for non-stop status monitoring of a hydraulic turbine provided in Example 1 of this application.

[0073] In some embodiments, the terminal may optionally include a communication interface 303 and at least one peripheral device. The processor 301, memory 302, and communication interface 303 may be connected via a bus or signal lines. Each peripheral device may be connected to the communication interface 303 via a bus, signal lines, or circuit boards. Specifically, the peripheral device may include at least one of a radio frequency circuit 304, a display screen 305, and a power supply 306.

[0074] The communication interface 303 can be used to connect at least one I / O (Input / Output)-related peripheral device to the processor 301 and the memory 302. In some embodiments, the processor 301, the memory 302, and the communication interface 303 are integrated on the same chip or circuit board. In other embodiments, any one or two of the processor 301, the memory 302, and the communication interface 303 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0075] The radio frequency circuit 304 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 304 communicates with a communication network and other communication devices via electromagnetic signals.

[0076] The display screen 305 is used to display a UI (User Interface). The UI may include any combination of graphics, text, icons, and videos.

[0077] The power supply 306 is used to supply power to various components in the electronic device.

[0078] Example 4: Based on any one of Embodiments 1 to 3, this embodiment discloses a computer program product, including a computer program or instructions. When executed by a computer, the computer program or instructions implement the method for monitoring the status of a hydraulic turbine without shutting down as described in any one of Embodiments 1. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0079] Obviously, those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computing device. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0080] Finally, it should be noted that the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art will appreciate that modifications may be made to the technical solutions described in the above embodiments, or that some of the technical features may be replaced with equivalents. Such modifications or replacements do not deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for monitoring the status of a hydraulic turbine without shutting down, characterized in that: include: Acquire multiple local runner images at different rotation angles during turbine operation; Perform feature extraction on each local wheel image to obtain a feature descriptor of each local wheel image; According to the feature descriptors of each local wheel image, feature matching is performed on all local wheel image combinations with adjacent acquisition times to obtain a matching point set; Obtaining, based on the matching point set, a perspective transformation matrix corresponding to a combination of all local wheel images acquired adjacently at the same time; According to the perspective transformation matrix corresponding to the combination of all local wheel images acquired at adjacent times, coordinate transformation processing is performed on each local wheel image to obtain multiple local wheel images after coordinate transformation in the same coordinate system; performing image stitching processing on the multiple local runner images after coordinate transformation to obtain a complete current runner image of the turbine; The health status information of the turbine is obtained according to the current runner image.

2. A method for monitoring the status of a hydraulic turbine without stopping according to claim 1, characterized in that: Performing feature extraction processing on any local wheel image to obtain a feature descriptor of the any local wheel image includes: Performing convolution operations on any of the local wheel images through Gaussian filters of different scales to generate a scale space; Calculating the Hessian matrix of each pixel in any local wheel image in the scale space; Obtaining a determinant response value of the Hessian matrix of each pixel point in any local wheel image, and screening candidate feature points from all pixels of any local wheel image based on the determinant response value of the Hessian matrix of each pixel point; The quasi-Newton method is used to perform sub-pixel optimization on the candidate feature points to obtain the final image feature points corresponding to the candidate feature points; Generate feature descriptors for image feature points.

3. A method for monitoring the status of a hydraulic turbine without stopping according to claim 2, characterized in that: The quasi-Newton method is used to perform sub-pixel optimization processing on any candidate feature point to obtain the image feature points corresponding to any candidate feature point, including: Get the approximate matrix of the Hessian matrix of any candidate feature point; By iteratively updating the position and scale of the candidate feature points, the updated feature points are obtained so as to maximize the determinant response value of the approximate matrix corresponding to the updated feature points and obtain the maximum value of the determinant response value of the current approximate matrix; The updated feature point corresponding to the maximum value of the determinant response value of the current approximation matrix is used as the image feature point corresponding to any one of the candidate feature points.

4. The method for monitoring the status of a hydraulic turbine without shutting down according to claim 1, characterized in that: Based on the feature descriptors of each local wheel image, feature matching is performed on all local wheel image combinations with adjacent acquisition times to obtain a matching point set, including: Calculate the Euclidean distance between each feature descriptor in any combination of local wheel images acquired at adjacent times, and use the image feature points corresponding to the feature descriptors with the closest Euclidean distance as the initial matching points of the current two local wheel images. The initial matching points of all combinations of local wheel images acquired at adjacent times constitute the initial matching point set. The RANSAC algorithm is used to remove mismatched points in the initial matching point set to obtain a final matching point set.

5. The method for monitoring the status of a hydraulic turbine without shutting down according to claim 1, characterized in that: Obtaining health status information of the turbine according to the current runner image includes: Using the SURF algorithm to perform feature extraction processing on the current wheel image to obtain local feature information of the current wheel image; Inputting the current wheel image into a pre-trained feature extraction model to obtain global feature information of the current wheel image; Fusing the local feature information and the global feature information of the current wheel image to obtain fused feature information of the current wheel image; The fused feature information of the current runner image is input into a pre-trained health status assessment model to obtain the health status information of the turbine.

6. A method for monitoring the status of a hydraulic turbine without stopping according to claim 5, characterized in that: The health status assessment model adopts a convolutional neural network.

7. The method for monitoring the status of a hydraulic turbine without stopping according to claim 5, characterized in that: After obtaining the fused feature information of the current wheel image, the method further includes: Acquire a factory runner image of the turbine, and perform feature extraction processing on the factory runner image using a SURF algorithm to obtain local feature information of the factory runner image; Inputting the factory runner image into the feature extraction model to obtain global feature information of the factory runner image; fusing the local feature information and the global feature information of the factory runner image to obtain fused feature information of the factory runner image; Performing difference analysis on the fused feature information of the current wheel image and the fused feature information of the factory wheel image to obtain a feature difference between the current wheel image and the factory wheel image; Determining whether the feature difference is greater than a preset feature difference threshold; If so, the fused feature information of the current runner image is input into the health status assessment model to obtain the health status information of the turbine.

8. A non-stop status monitoring system for a hydraulic turbine, characterized in that: include: An image acquisition module is used to obtain multiple local runner images at different rotation angles during the operation of the turbine; a feature extraction module, communicatively connected to the image acquisition module, for performing feature extraction processing on each local wheel image to obtain a feature descriptor of each local wheel image; a feature matching module, in communication with the feature extraction module, for performing feature matching on all combinations of local wheel images acquired adjacently in time based on feature descriptors of the local wheel images to obtain a matching point set; an image transformation module, communicatively connected to the feature matching module, for obtaining, based on the matching point set, a perspective transformation matrix corresponding to a combination of all local wheel images acquired adjacently at the same time; and for performing coordinate transformation processing on each local wheel image based on the perspective transformation matrix corresponding to the combination of all local wheel images acquired adjacently at the same time, to obtain a plurality of local wheel images after coordinate transformation in the same coordinate system; an image stitching module, communicating with the image transformation module, for stitching a plurality of local runner images after coordinate transformation to obtain a complete current runner image of the turbine; A health status detection module is in communication with the image stitching module and is used to obtain health status information of the turbine based on the current runner image.

9. An electronic device, characterized in that: include: a memory for storing computer program instructions; as well as, A processor is configured to execute the computer program instructions to thereby complete the operation of the method for monitoring the condition of a hydraulic turbine without shutting down as claimed in any one of claims 1 to 7.

10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or the instruction is executed by a computer, the method for monitoring the status of a hydraulic turbine without shutting down is realized.

Images