A method for identifying cavitation in hydraulic turbines based on image recognition

Abstract

This application relates to the technical field of turbine cavitation identification and discloses a turbine cavitation identification method based on image recognition. This method collects an image of a target turbine runner to be identified, extracts features from the image, and constructs a target feature matrix of the target image. This feature matrix of the target turbine runner is first corrected and then fed into a cavitation identification model. This matrix is ​​then compared with the template-corrected feature matrix of a template image stored in the model, ultimately achieving cavitation identification of the turbine runner. Compared to existing methods, this method can not only increase the speed of turbine cavitation identification but also improve the identification quality.

Description

Technical Field

[0001] The present application relates to the technical field of turbine cavitation identification, and in particular to a method for distinguishing turbine cavitation based on image recognition. Background Art

[0002] Incipient cavitation in a turbine is a phenomenon that occurs when the local pressure within the liquid drops to a critical value, causing the gas nuclei contained in the liquid to rapidly grow, and cavitation begins. Incipient cavitation is often characterized by the incipient cavitation coefficient. The definition of the incipient cavitation coefficient of a turbine is: the cavitation coefficient when bubbles begin to form on the runner blades.

[0003] There are usually three main reasons for the generation of bubbles on blades. The first is the so-called airfoil cavitation, which mostly occurs on the back of the blade's water outlet edge and is greatly affected by changes in tailwater pressure (i.e., cavitation coefficient); the second is cavitation on the blade's water inlet edge (mainly on the back), and the main reason for this is flow separation; the third is cavitation bubbles formed in the gap flow at the end faces of axial and cross-flow turbine blades.

[0004] Observing incipient cavitation at the runner outlet of a hydraulic turbine is difficult. Currently, one method for detecting incipient cavitation is visual inspection. This involves observing the vortex bands and flow patterns at the runner outlet through a straight tapered section of a draft tube made of transparent plexiglass. During the experiment, a stroboscope (a frequency-adjustable light source that alternates between bright and dark) is used to illuminate the runner. The stroboscope frequency is adjusted to be equal to or close to the model turbine's rotational frequency. The apparently stationary or slowly rotating runner blades can be clearly seen with the naked eye, allowing the cavitation at the runner outlet to be observed and recorded with a camera. However, this method requires a high level of experience, typically at least ten years, to determine the presence of incipient cavitation in an image. Furthermore, this method is subject to significant observer subjectivity and suffers from low efficiency. Summary of the Invention

[0005] In order to solve the problems and shortcomings existing in the above-mentioned prior art, the present application proposes a turbine cavitation identification method based on image recognition. By collecting the image of the target turbine runner to be identified, feature extraction is performed on the image and the feature vector of the image is obtained, and then a feature matrix is ​​constructed. The feature matrix of the target turbine runner is first corrected and then sent to the cavitation recognition model, and it is compared with the corrected feature matrix of the template image stored in the model, and finally the cavitation recognition of the turbine runner is realized.

[0006] In order to achieve the above-mentioned invention objectives, the technical solutions of this application are as follows:

[0007] A method for identifying cavitation in a hydraulic turbine based on image recognition, comprising:

[0008] Acquire a template image, preprocess the template image, extract the complete outline of the cavitation bubble in the template image, select a number of pixel points on the outline as template feature points of the template image, and extract eigenvalues ​​of the template feature points. The eigenvalues ​​include the brightness of the feature point and the relative position relationship between the feature point and its adjacent feature points. The eigenvalues ​​of each feature point constitute a template feature matrix B of the template image;

[0009] The template feature matrix B is modified to obtain a template modified feature matrix D containing several new eigenvalues, and the template modified feature matrix D is sent to the cavitation recognition model and stored;

[0010] A turbine runner image to be identified is collected as a target image, the target image is preprocessed, an area where cavitation may occur is determined in the preprocessed target image, a number of pixel points near the area where cavitation may occur are selected as target feature points of the target image, the selected target feature points constitute a target contour of the cavitation bubble, and eigenvalues ​​of the feature points are extracted. The eigenvalues ​​include the brightness of the feature point and the relative positional relationship between the feature point and its adjacent feature points. The eigenvalues ​​of the feature points constitute a target feature matrix A of the target image;

[0011] Modify the target feature matrix A to obtain a target modified feature matrix C containing several new eigenvalues;

[0012] The target correction feature matrix C is sent to the cavitation recognition model, and the cavitation discrimination result is output; if the cavitation recognition model determines that the target correction feature matrix C is equal to the template correction feature matrix D, it means that cavitation occurs, otherwise no cavitation occurs.

[0013] Preferably, in the present application, the number of target feature points selected in the region where cavitation may occur in the target image is the same as the number of template feature points selected in the template image.

[0014] Preferably, in the present application, the adjacent feature point refers to the next feature point adjacent to the feature point on the contour in a clockwise or counterclockwise direction.

[0015] Preferably, in the present application, the image preprocessing includes image cropping, image masking, image sharpening and binarization processing.

[0016] Preferably, in the present application, the pre-processing of the target image further includes intercepting the ROI region of the target image, and performing angle and scale transformation on the ROI region of the target image with reference to the template image.

[0017] Preferably, in the present application, the modification of the template feature matrix B to obtain the template modified feature matrix D containing several new eigenvalues ​​refers to modifying the template feature matrix B using the modified values.

[0018] Preferably, in the present application, the correction of the target characteristic matrix A to obtain the target corrected characteristic matrix C containing a number of new eigenvalues ​​refers to correcting the target characteristic matrix A using the correction values.

[0019] Preferably, in the present application, the correction value is a discrimination coefficient or interval, and the correction value is determined by combining empirical parameters and turbine cavitation test data.

[0020] Beneficial effects of this application:

[0021] (1) The turbine cavitation identification of the present application is divided into two stages, one is the identification benchmark establishment stage before cavitation identification, and the other is the cavitation identification stage for the target turbine runner. In the identification benchmark establishment stage, firstly, a typical image of an existing turbine runner with cavitation is used as a template image, and a template feature matrix for cavitation identification is constructed through a series of image processing methods. Then, the template feature matrix is ​​corrected, and finally, the corrected template feature matrix is ​​sent to the cavitation identification model for storage as a benchmark for turbine cavitation identification judgment; in the cavitation identification stage, an image of the turbine runner to be identified with cavitation is collected as a target image, and the target image is processed in the same way as the template image to obtain a target feature matrix. The target feature matrix is ​​also corrected, and finally, the corrected target feature matrix is ​​sent to the cavitation identification model. The cavitation identification model determines whether cavitation occurs in the turbine runner by judging whether the corrected target feature matrix is ​​equal to the corrected template feature matrix. This method ultimately realizes the identification and judgment of turbine cavitation through feature matching. The whole method is simple and easy to implement, which can not only improve the speed of turbine cavitation judgment, but also improve the judgment quality.

[0022] (2) When pre-processing the target image, the present application also cuts out the ROI area and performs corresponding angle and scale transformation on the image with reference to the template image. The transformed image is in the same standard as the template image. Therefore, when comparing features and performing feature matching, the accuracy is higher and the output cavitation recognition judgment result is also more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The foregoing and following detailed description of the present application will become more apparent when read in conjunction with the following drawings, in which:

[0024] Figure 1 This is a flow chart of the application method. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the technical solutions in this application, the following will further illustrate the technical solutions for achieving the invention objectives of this application through several specific embodiments. It should be noted that the technical solutions claimed for protection in this application include but are not limited to the following embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making any creative efforts should fall within the scope of protection of this application.

[0026] Currently, the primary method for detecting incipient cavitation in turbine runners is manual visual inspection. This involves observing the vortex bands and flow patterns at the runner outlet through a straight tapered section of a draft tube made of transparent plexiglass. During the test, a stroboscope (a frequency-adjustable light source that alternates between bright and dark) is used to illuminate the runner. The stroboscope frequency is adjusted to be equal to or close to the model turbine's rotational frequency. The apparently stationary or slowly rotating runner blades can be clearly seen with the naked eye, allowing the cavitation conditions at the runner outlet to be observed and recorded with a camera. However, this method requires a high level of experience, typically at least ten years, to determine the presence of incipient cavitation in an image. Furthermore, this method is subject to significant observer subjectivity and suffers from low efficiency.

[0027] Based on this, this embodiment provides a method for identifying cavitation in a turbine runner based on image recognition. This method divides turbine cavitation identification into two stages: a pre-cavitation benchmark establishment stage and a cavitation identification stage. In the benchmark establishment stage, a typical image of an existing turbine runner experiencing cavitation is first used as a template image. A series of image processing techniques are used to construct a template feature matrix for cavitation identification. This template feature matrix is ​​then corrected, and the corrected template feature matrix is ​​then stored in a cavitation identification model to serve as a benchmark for turbine cavitation identification. In the cavitation identification stage, an image of the turbine runner to be identified for cavitation is captured as a target image. The target image is processed using the same processing techniques as the template image to obtain a target feature matrix. This target feature matrix is ​​also corrected, and the corrected target feature matrix is ​​then fed into the cavitation identification model. The cavitation identification model determines whether cavitation has occurred in the turbine runner by determining whether the corrected target feature matrix is ​​equal to the corrected template feature matrix. This method ultimately realizes the identification and judgment of turbine cavitation through feature matching. The whole method is simple and easy to implement, which can not only improve the speed of turbine cavitation judgment, but also improve the judgment quality.

[0028] In this embodiment, once the cavitation identification reference is established, the identification reference can be used for a long time.

[0029] This embodiment discloses a method for distinguishing cavitation of a hydraulic turbine based on an image. Figure 1 , this method specifically comprises the following steps:

[0030] Step S101: Establishing a turbine runner cavitation identification benchmark.

[0031] Step S101.1. First, a large number of previously collected bubble sample images of turbine runner cavitation events are manually selected by experts. Images with typical features are selected from the sample images as template images, and the template images are fed into the cavitation recognition model.

[0032] Step S101.2. The cavitation recognition model first preprocesses the template image, extracts the complete outline of the cavitation bubble in the preprocessed template image, and then selects a number of pixel points on the outline as template feature points of the template image, performs feature extraction on the selected pixel points, and obtains the eigenvalues ​​of the selected pixel points. The eigenvalue is in a vector form, mainly including the brightness and grayscale of the feature point, the absolute position of the feature point in the image, and the relative position relationship between the feature point and its adjacent feature points. The eigenvalues ​​of each feature point form a PCIV eigenvector. These parameters are combined to form characteristic parameters representing the bubble image when cavitation occurs in the turbine runner. These image characteristic parameters are then mapped one by one to different row and column element positions of the matrix (the position can be changed), and finally form a template feature matrix B of the template image. The template feature matrix B constitutes the qualitative elements of machine intelligent recognition, and the template feature matrix B is saved in the cavitation recognition model.

[0033] In this embodiment, the PCIV feature vectors of all pixel points on the same curve segment form a corresponding vector chain structure.

[0034] In this embodiment, it should be noted that adjacent feature points refer to the direction of selecting feature points in a clockwise or counterclockwise direction. On the contour of the graphic, the next feature point away from the feature point is the adjacent feature point of the feature point.

[0035] In this embodiment, it should also be noted that the template image is pre-processed using conventional technical means in this field, including image clipping and graying, image masking, image sharpening, and binarization.

[0036] The image clipping and graying refers to clipping out portions of the image that are irrelevant to cavitation determination and converting the color image into a grayscale image, that is, each pixel has only "brightness" difference.

[0037] The image mask refers to masking the image with a mask to remove most of the graphic information irrelevant to the discrimination and only retain the positions where the image summary is expected to produce cavitation.

[0038] The image sharpening refers to obtaining the gradient by using the differential method, obtaining the approximate value by using the difference operation, and calculating the pixel value of the edge of the image.

[0039] The binarization process refers to performing binarization on an image in combination with a threshold method, and the processed image is only related to the positions of points with pixel values ​​ranging from 0 to 255.

[0040] In this embodiment, the method of extracting pixel feature values ​​is a conventional technical means in the art and will not be described in detail here.

[0041] Step S102: Correct the cavitation identification benchmark.

[0042] After obtaining the template feature matrix B, since the turbine cavitation process is a real-time dynamic process with complete randomness, a correction coefficient (for example, 0 or 1) or correction interval (for example, [0.37, 1.2]) is preset for each characteristic parameter of the template bubble image. After these correction coefficients or intervals are multiplied by each PCIV vector in the qualitative model, several new PCIV vectors are obtained. These new PCIV vectors are then recombined to form a new matrix, namely the template correction feature matrix D; the new matrix D constitutes the machine-recognizable corrected qualitative elements of the turbine runner cavitation bubble image, and the template correction feature matrix D is saved in the cavitation recognition model.

[0043] Step S103: Acquire the target turbine runner image to be identified in real time, send the acquired target turbine runner image to the cavitation identification model, and output the cavitation identification result.

[0044] Step S103.1. Collect the turbine runner image to be identified as the target image, and send the target image into the cavitation recognition model; the cavitation recognition model first preprocesses the target image, and first determines the area where cavitation may occur in the preprocessed image based on experience, and then selects a number of pixel points near the area where cavitation may occur as target feature points of the target image. The selected target feature points constitute the target outline of the cavitation bubbles in the area where cavitation may be sent, and the eigenvalues ​​of the target feature points are extracted in the same way as the template feature points, and finally the brightness, grayscale, absolute position of the target feature point in the image and the relative position relationship between the target feature point and its adjacent target feature points are obtained. Similarly, the eigenvalue of each target feature point will also constitute a corresponding eigenvector, and then the eigenvectors of each selected target feature point are mapped one by one to different row and column element positions of the matrix (the position can be changed), and finally a target feature matrix A of the target image is formed, and the target feature matrix A is saved to the cavitation recognition model.

[0045] In this embodiment, similarly, for the target image, the PCIV feature vectors of all target feature points on the same curve segment form a corresponding vector chain structure.

[0046] In this embodiment, it should be noted that the number of target feature points selected in the target image is the same as the number of template feature points selected in the template image.

[0047] Similarly, for the target image, the meaning of adjacent feature points refers to the template image.

[0048] In this embodiment, the target feature matrix A is a real-time matrix, and the matrix changes continuously as the sampling changes.

[0049] In this embodiment, it should be noted that the target image preprocessing process is specifically as follows:

[0050] First, the ROI area of ​​the target image is captured, and then the angle and scale of the ROI area of ​​the target image are transformed with reference to the template image. Then, the target image is processed by image sharpening, image binarization, etc. with reference to the processing method of the template image.

[0051] In this embodiment, it should be noted that the method of transforming the angle and scale of the ROI region of the target image is a conventional method known to those skilled in the art.

[0052] Step S103.2. Correct the target feature matrix of the target image to obtain a corrected target corrected feature matrix C.

[0053] Similarly, since the cavitation process of the turbine is a real-time dynamic process with complete randomness, a correction coefficient (for example, 0 or 1) or correction interval (for example, [0.37, 1.2]) is preset for each characteristic parameter of the target bubble image. After these correction coefficients or intervals are multiplied by the eigenvectors in the target model, several new eigenvectors are obtained. These new eigenvectors are then recombined to form a new matrix, namely the target correction feature matrix C, which is saved in the cavitation recognition model.

[0054] Similarly, since the target feature matrix A is a real-time matrix, the corrected target corrected feature matrix C is also a real-time feature matrix.

[0055] Step S103.3. Identification of turbine runner cavitation.

[0056] The cavitation recognition model compares the target correction feature matrix C obtained in real time with the internally stored template correction feature matrix D and outputs the cavitation discrimination result; if the cavitation recognition model determines that the target correction feature matrix C is equal to the template correction feature matrix D, it means that cavitation has occurred, otherwise no cavitation has occurred.

[0057] The above description is merely a preferred embodiment of the present application and does not constitute any form of limitation to the present application. Any simple modification or equivalent change made to the above embodiment based on the technical essence of the present application shall fall within the scope of protection of the present application.

Claims

1. A method for identifying cavitation in a hydraulic turbine based on image recognition, characterized in that: include: Acquire a template image, preprocess the template image, extract the complete outline of the cavitation bubble in the template image, select a number of pixel points on the outline as template feature points of the template image, and extract eigenvalues ​​of the template feature points. The eigenvalues ​​include the brightness of the feature point and the relative position relationship between the feature point and its adjacent feature points. The eigenvalues ​​of each feature point constitute a template feature matrix B of the template image; Modify the template feature matrix B to obtain a template modified feature matrix D containing several new eigenvalues, and save the template modified feature matrix D in the cavitation recognition model; A turbine runner image to be identified is collected as a target image, the target image is preprocessed, an area where cavitation may occur is determined in the preprocessed target image, a number of pixel points near the area where cavitation may occur are selected as target feature points of the target image, the selected target feature points constitute a target contour of the cavitation bubble, and eigenvalues ​​of the feature points are extracted. The eigenvalues ​​include the brightness of the feature point and the relative positional relationship between the feature point and its adjacent feature points. The eigenvalues ​​of the feature points constitute a target feature matrix A of the target image; Correct the target feature matrix A to obtain a target corrected feature matrix C containing several new eigenvalues, and save the target corrected feature matrix C to the cavitation identification model; The cavitation identification model outputs the cavitation identification results; If the cavitation recognition model determines that the target correction feature matrix C is equal to the stored template correction feature matrix D, it means that cavitation occurs, otherwise no cavitation occurs.

2. The method for identifying cavitation in a hydraulic turbine based on image recognition according to claim 1, characterized in that: The number of target feature points selected in the region where cavitation may occur in the target image is the same as the number of template feature points selected in the template image.

3. The method for identifying cavitation in a hydraulic turbine based on image recognition according to claim 1, characterized in that: The adjacent feature point refers to the next feature point adjacent to the feature point on the contour in a clockwise or counterclockwise direction.

4. The method for identifying cavitation in a hydraulic turbine based on image recognition according to claim 1, characterized in that: Image preprocessing includes image cropping, image masking, image sharpening and binarization.

5. The method for identifying cavitation in a hydraulic turbine based on image recognition according to claim 4, characterized in that: The pre-processing of the target image further includes intercepting the ROI region of the target image, and performing angle and scale transformation on the ROI region of the target image with reference to the template image.

6. The method for identifying cavitation in a hydraulic turbine based on image recognition according to claim 1, characterized in that: The modifying of the template feature matrix B to obtain the template modified feature matrix D containing a number of new eigenvalues ​​means modifying the template feature matrix B by using the modified values.

7. The method for identifying cavitation in a hydraulic turbine based on image recognition according to claim 1, characterized in that: The said correcting the target characteristic matrix A to obtain the target corrected characteristic matrix C containing a number of new eigenvalues ​​means correcting the target characteristic matrix A by using the correction values.

8. A method for identifying cavitation in a hydraulic turbine based on image recognition according to claim 6 or 7, characterized in that: The correction value is a discrimination coefficient or interval, and the correction value is determined by combining empirical parameters and turbine cavitation test data.

Images