Method and system for detecting critical position of hoisting rotor of hydropower plant

Abstract

The application discloses a hydroelectric plant rotor hoisting critical position detection method and system, belongs to the critical position detection technical field, the method includes that the image acquisition module collects the hydroelectric plant rotor and stator picture, is photographed from multiple angles and is collected, and the data set of training neural network is obtained by classifying and labeling the picture; the data set is subjected to data enhancement, the data set is expanded, the training environment is configured, the neural network is trained by using the data set, and the rotor and stator neural network identification model are constructed; the application introduces the neural network to identify and position the target, thereby assisting the bridge machine hoisting operation, the application realizes the non-contact rotor hoisting critical position detection by using video identification and positioning, improves the hoisting operation efficiency, shortens the whole hoisting process time, reduces the consumption of manpower and material resources, the application can realize real-time updating and feedback of target information while obtaining the monitoring picture, and continuously obtains the latest target information.

Description

Technical Field

[0001] The present invention relates to the technical field of critical position detection, and in particular to a method and system for detecting the critical position of a rotor hoisting in a hydropower plant. Background Art

[0002] During the installation and maintenance of hydropower plant units, the gap between the turbine rotor and stator is typically 2-5 cm. When the crane hoists the rotor into the generator pit, it is crucial to ensure that there is no collision between the rotor and stator during operation. Currently, an on-site operator is required to monitor every two magnetic poles in the stator pit. This monitor and ground control personnel coordinate to determine the rotor position, and the crane operator operates the crane to adjust the position, requiring significant human resources to ensure installation accuracy.

[0003] Industrial video surveillance is a mature and widely used technology in the field. Video recognition and positioning technology can detect and filter abnormal behavior in real-time videos and issue timely alerts, transforming the traditional passive mode of industrial video surveillance from post-incident verification to proactive prevention. This revolutionizes the traditional passive mode of industrial video surveillance, which can only monitor but not control. Currently, the hydropower plant's monitoring system only monitors and records the rotor hoisting process and lacks more direct and effective means of obtaining information on the rotor's positioning during the hoisting process. Summary of the Invention

[0004] In order to address the deficiencies in the prior art, the present invention provides a method and system for detecting the critical position of a hydropower plant rotor hoisting to solve the technical problem that the prior art only monitors and records the rotor hoisting process, but lacks a more direct and effective means of obtaining information on the rotor hoisting positioning.

[0005] The present invention adopts the following technical solutions.

[0006] A first aspect of the present invention discloses a method for detecting a critical position of a rotor during hoisting in a hydropower plant, comprising the following steps:

[0007] The image acquisition module collects images of the hydropower plant's rotor and stator from multiple angles, classifies and labels the images to obtain a data set for training the neural network.

[0008] Perform data enhancement on the data set, expand the data set, configure the training environment, use the data set to train the neural network, and build the rotor and stator neural network recognition model;

[0009] An image acquisition module is installed under the bridge beam and main beam of a double-girder bridge crane to obtain original images of the rotor and stator. The original images are transmitted to the processor and host computer for image preprocessing and model recognition. The preprocessed target image is obtained and the rotor and stator neural network recognition model is used to identify the rotor and stator in the target image.

[0010] After successfully recognizing the target image, the pixel coordinates of the image are further obtained. The classification information of the rotor and stator is determined based on the confidence level. A visual anchor frame is formed at the edges of the rotor and stator. The pixel coordinates of the center of the anchor frame, i.e., the center point of the rotor and stator, are obtained. The pixel coordinates of the two-dimensional images of the rotor and stator are obtained, and the three-dimensional pixel coordinates are obtained by combining the depth information.

[0011] The three-dimensional pixel coordinates of the rotor and stator images in the video, as well as the camera's focal length, principal point coordinates, pixel scaling factor, rotation matrix, and translation vector are converted into the three-dimensional coordinate data of the rotor and stator centers in the world coordinate system. After pixel coordinate conversion is performed in the processor and host computer, the world coordinate data is obtained, and the rotor and stator anchor frames and three-dimensional coordinate information are displayed;

[0012] The difference between the three-dimensional coordinates of the rotor and the stator is calculated in the processor and the host computer, and a critical position detection program is set to determine the gap between the rotor and the stator. When the difference between the two approaches or exceeds the preset gap value, an alarm prompt is issued.

[0013] As an improvement, the produced image samples are subjected to data enhancement by operations such as rotation, scaling, flipping, and cropping. On the basis of the original neural network, image segmentation is introduced and an attention mechanism is added to improve the recognition of local features of the rotor and stator. The network of the rotor and stator neural network recognition model is pruned, and the number of neural network convolution layers and the size of the convolution kernel are adjusted. The recognition effect of the rotor and stator neural network recognition model is verified using the validation set every training cycle.

[0014] As an improvement, the image preprocessing process includes: region of interest delineation, grayscale conversion and filtering noise reduction;

[0015] Delineate the region of interest and focus on processing specific parts of the image around the rotor and stator during rotor hoisting, thereby narrowing the video recognition range;

[0016] The three channels of the color image are reduced to one channel through grayscale processing, which reduces the number of pixels and color interference;

[0017] Filtering noise reduction is used to reduce the interference to the target image caused by insufficient or excessive light in the hydropower plant building, and to suppress the noise of the target image.

[0018] As an improvement, the rotor and stator neural network identification model identification process includes:

[0019] The preprocessed image is input into the neural network recognition model for processing, and the pixel values ​​are normalized from the range of [0, 255] to the range of [-1, 1]. The rotor and stator in the image are extracted through the backbone network to obtain feature maps of different scales. The feature maps are enlarged to the same size as the high-level feature maps and then fused with the high-level feature maps. The image is divided into grids in the prediction layer, and the bounding box of the target in each grid cell is predicted to obtain the confidence and category of the predicted target. The predicted bounding box coordinates are converted from normalized coordinates to image coordinates, and the actual width and height are calculated. After obtaining the target detection results, the overlapping bounding boxes are eliminated, and the bounding boxes with the highest confidence are retained by setting the confidence threshold. The bounding boxes below the threshold will be ignored.

[0020] As an improvement, the process of obtaining the three-dimensional pixel coordinates of the rotor and stator includes:

[0021] After successfully identifying the rotor and stator, the center pixel coordinates of the anchor frame, that is, the two-dimensional center position coordinates of the rotor and stator, are calculated, and the parallax of the rotor and stator images is obtained. The depth information of the rotor and stator is obtained by calculating the parallax, and the three-dimensional image coordinate information is obtained by combining the two-dimensional center position coordinates and displayed on one side of the anchor frame.

[0022] As an improvement, the pixel coordinate conversion process is to convert the pixel coordinates of the anchor frame obtained in the recognition process into the actual position coordinates, from the world coordinate system ( ) to the camera coordinate system ( ) is transformed according to the position and orientation of the camera, and the three-dimensional position of the object is mapped to the camera coordinate system through rigid body transformation, and the intrinsic parameters of the camera are obtained. The camera coordinate system ( ) is mapped to the image physical coordinate system (x, y), and the three-dimensional point is converted into a point on a two-dimensional plane. Taking into account parameters such as the focal length and optical center of the camera, the pixel coordinate position of the point in the image is obtained by mapping the image physical coordinates to the image pixel coordinates (u, v). The depth information is obtained by measuring the parallax of the corresponding points in the two camera images and calculating the depth in the three-dimensional space using the principle of triangulation to obtain the real-world coordinates of the rotor and stator centers.

[0023] As an improvement, the critical position detection determination method of the critical position detection program includes:

[0024] The rotor and stator are divided into four detection areas, and the corresponding images are obtained by multiple image acquisition modules under the double-girder bridge crane. According to the structural parameters of the double-girder bridge crane, the target position in the camera coordinate system is converted into the target position in the double-girder bridge crane's own coordinate system. In the computer processing platform, the gap distance between the rotor and the stator is preset, and the anchor frame color is uniformly set to green and red. Before and after the rotor and stator enter the pit, the deviation value between the rotor center and the stator center is calculated and the absolute value is taken. The three-dimensional coordinate information of the rotor and stator is compared in real time, and a safety range threshold is set. The calculated value is compared with the threshold to determine whether it meets the preset conditions.

[0025] As an improvement, if the position meets the threshold, a green prompt box is displayed, indicating that the position is normal; if the position exceeds the threshold, a red error box is displayed and an alarm is issued through the speaker.

[0026] As an improvement, the detection results are fed back to the double-girder bridge crane operator in real time. The target object and its deviation information are displayed on the display screen in the bridge crane control room to guide the operator to adjust the position of the double-girder bridge crane.

[0027] A second aspect of the present invention discloses a hydropower plant rotor hoisting critical position detection system, comprising:

[0028] The image acquisition module includes a binocular camera. When receiving a detection instruction from the processor, the image acquisition module starts and collects raw image data;

[0029] The communication module is electrically connected to the image acquisition module and serves as a communication terminal between the processor and the image acquisition module to transmit image data and receive instruction information;

[0030] A processor, electrically connected to the communication module, performs preprocessing operations on the original image, runs a rotor and stator neural network recognition model, identifies and locates the rotor and stator in the image, and converts the coordinates;

[0031] The coordinate information feedback module is electrically connected to the processor and displays the rotor and stator anchor frames in real time. One side of the anchor frame dynamically displays the coordinate information of the rotor and stator in three dimensions: X, Y, and Z.

[0032] The critical position detection module is electrically connected to the processor, calculates the coordinate difference between the rotor and the stator, displays the relative position information, and dynamically detects whether the gap between the rotor and the stator meets the preset requirements;

[0033] An alarm module, including a speaker, is used to sound an alarm when the gap between the rotor and the stator does not meet the preset requirements;

[0034] The host computer is electrically connected to the processor.

[0035] The beneficial effect of the present invention is that, compared with the prior art,

[0036] 1. This invention incorporates a neural network to identify and locate targets, thereby assisting in bridge crane hoisting operations. Traditional hoisting operations require extensive manpower to determine the position. This invention utilizes video recognition and positioning to achieve non-contact detection of the rotor's critical position during hoisting, improving hoisting efficiency, shortening the entire hoisting process, and reducing both labor and material resources.

[0037] 2. Traditional hydropower plant monitoring systems have limited functionality, solely capturing and recording the hoisting operation scene. The rotor hoisting critical position detection system proposed in this invention not only captures monitoring footage but also provides real-time updates and feedback on target information, continuously acquiring the latest target information and ensuring real-time tracking and positioning of the target.

[0038] 3. It can provide early warning of potential rotor-stator collision risks, ensuring accurate positioning and safety of the rotor during the hoisting process. Furthermore, the positioning method of the present invention uses multi-point positioning, which can capture all angles of the rotor and stator, enabling blind spot identification and positioning during the hoisting operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a schematic structural diagram of a hydropower plant rotor hoisting critical position detection system according to the present invention;

[0040] Figure 2 This is a working diagram of the method for detecting the critical position of a rotor hoisting in a hydropower plant according to the present invention;

[0041] Figure 3 A flow chart of converting camera pixel coordinates into world coordinates for the method for detecting the critical position of a hydropower plant rotor hoisting according to the present invention;

[0042] Figure 4 This is a schematic diagram of the critical position detection system for rotor hoisting in a hydropower plant according to the present invention;

[0043] Figure 5 This is a diagram showing the rotor identification and positioning effects of the method for detecting the critical position of the rotor hoisting in a hydropower plant according to the present invention;

[0044] Figure 6 This is a stator identification and positioning effect diagram of the hydropower plant rotor hoisting critical position detection method of the present invention;

[0045] Figure 7 This is a diagram showing the rotor and stator identification and positioning effects of the hydropower plant rotor hoisting critical position detection method of the present invention.

[0046] 1. Cab; 2. Image acquisition module; 3. Double-girder bridge crane; 4. Rotor; 5. Stator; 6. Host computer. DETAILED DESCRIPTION

[0047] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0048] like Figure 1-7 As shown, Example 1 of the present invention provides a method for detecting the critical position of a rotor hoisting in a hydropower plant, which specifically includes the following steps:

[0049] Step 1: The image acquisition module 2 first collects images of the hydropower plant rotor 4 and stator 5, including local features such as the rotor poles and stator windings. The images are taken from multiple angles, including vertical, horizontal, and oblique directions. The images are then classified and annotated to obtain a dataset for training the neural network.

[0050] Step 2: Perform data augmentation on the dataset to expand it to 3,000 images, including 2,500 training images and 500 validation images. Use a cloud server or a device with good GPU hardware to configure the training environment. Use the dataset to train the YOLOv8 neural network and build a rotor and stator neural network recognition model.

[0051] Step 3: Install an image acquisition module 2 beneath the bridge beam and main girder of the double-girder overhead crane 3 to capture raw images of the rotor 4 and stator 5. These images are then transmitted to the processor and host computer 6 located in the cab 1 for image preprocessing and model recognition. This preprocessed target image highlights target features and improves recognition efficiency. The rotor and stator neural network recognition model is then used to identify the rotor 4 and stator 5 in the target image.

[0052] Step 4: After successfully recognizing the target image, the pixel coordinates of the image are further obtained. The classification information of the rotor 4 and stator 5 is determined based on the confidence level. A visual anchor frame is formed at the edges of the rotor 4 and stator 5. The pixel coordinates of the center of the anchor frame (i.e., the center point of the rotor 4 and stator 5) are obtained. The pixel coordinates of the two-dimensional images of the rotor 4 and stator 5 are obtained and combined with the depth information to obtain the three-dimensional pixel coordinates.

[0053] Step 5: Convert the three-dimensional pixel coordinates of the rotor 4 and stator 5 in the video image, as well as the focal length, principal point coordinates, pixel scaling factor, rotation matrix, and translation vector of the camera, into the three-dimensional coordinate data of the centers of the rotor 4 and stator 5 in the world coordinate system. After performing pixel coordinate conversion in the processor and the host computer 6, the world coordinate data is obtained, and the anchor frames and three-dimensional coordinate information of the rotor 4 and stator 5 are displayed.

[0054] Step 6: The processor and host computer 6 calculate the three-dimensional coordinate difference between the rotor 4 and stator 5. A critical position detection program is set to determine the clearance between the rotor 4 and stator 5. When the difference approaches or exceeds a preset clearance value, an alarm is triggered. The operator can adjust the double-girder bridge crane 3 based on the coordinate information displayed by the processor and host computer 6 to complete the lifting operation.

[0055] In a preferred but non-limiting embodiment of the present invention, in step 2, the prepared image samples are subjected to data enhancement by operations such as rotation, scaling, flipping, and cropping. On the basis of the original neural network, image segmentation is introduced and an attention mechanism is added to improve the recognition of local features of the rotor 4 and the stator 5, such as rotor poles and stator windings. The rotor and stator neural network recognition model is pruned, the number of neural network convolution layers and the size of the convolution kernel are adjusted to improve the inference speed, and an appropriate number of training rounds is selected. The recognition effect of the rotor and stator neural network recognition model is verified using a validation set every training cycle to avoid overfitting or underfitting of the rotor and stator neural network recognition model. During the training process, the network weights are adjusted according to the feedback of the loss function to optimize the performance of the rotor and stator neural network recognition model.

[0056] In a preferred but non-limiting embodiment of the present invention, in step 3, the image preprocessing process includes: delineation of the region of interest, grayscale conversion, and filtering and noise reduction. The region of interest is delineated using the Hausdorff distance, and specific portions of the image surrounding the rotor 4 and the stator 5 during hoisting are processed centrally to narrow the video recognition range, thereby improving processing efficiency and accuracy and reducing the amount of computation.

[0057] Grayscale processing reduces the three channels (red, green, and blue) of a color image to one channel, reducing a large number of pixels. Therefore, grayscale processing can reduce color interference, making the brightness changes in the image more obvious, reducing interference from other information, and improving recognition speed. The specific calculation method is as follows:

[0058] Gray=R×0.299+G×0.587+B×0.114

[0059] In the formula, Gray is the grayscale value of the corresponding pixel in the grayscale image, R, G, and B are the values ​​of the red, green, and blue channels of the corresponding pixel in the original image, respectively. The multiplication coefficient is the weight factor, which can be adjusted according to actual conditions.

[0060] Filtering noise reduction is used to reduce the interference caused by insufficient or excessive light in the hydropower plant building to the target image, suppress the noise of the target image, and thus improve the recognition accuracy of the image. The calculation formula for image noise reduction is as follows:

[0061]

[0062] ( )

[0063] Where (x, y) is the coordinate, σ is the standard deviation, G(x, y) is the Gaussian function, f(x, y) is the function of the original image, and N(x, y) is the function of the smoothed image.

[0064] In a preferred but non-limiting embodiment of the present invention, in step 3, the specific recognition process of the rotor and stator neural network recognition model is as follows: the preprocessed image is input into the YOLOv8 model for processing, the pixel value is normalized from the range of [0, 255] to the range of [-1, 1], the rotor 4 and stator 5 in the image are extracted by the convolutional neural network (CNN) in the backbone network (Backbone), downsampling is used to obtain feature maps of different scales, upsampling is used to enlarge the feature map to the same size as the high-level feature map, and then fused with the high-level feature map. Then, the image is divided into grids in the prediction layer, and the bounding box (including center point coordinates, width and height) of the target in each grid unit is predicted to obtain the confidence and category of the predicted target, the predicted bounding box coordinates are converted from normalized coordinates to image coordinates, and the actual width and height are calculated according to the size of the grid unit and the predicted width and height. After obtaining the target detection result, the YOLOv8 neural network adopts non-maximum suppression (Non-Maximum NMS (Nearest Mean Suppression) eliminates overlapping bounding boxes and retains the bounding boxes with the highest confidence by setting a confidence threshold. Bounding boxes below the threshold are ignored. Throughout the process, the YOLOv8 network quickly and accurately detects and identifies the rotor 4 and stator 5 in the image through a combination of convolutional layers and prediction layers.

[0065] In a preferred but non-restrictive embodiment of the present invention, in step 4, the process of obtaining the three-dimensional pixel coordinates of the rotor 4 and the stator 5 is as follows: after successfully identifying the rotor 4 and the stator 5, the non-maximum suppression algorithm is used to evaluate the anchor frame that best fits the boundary of the rotor 4 and the stator 5, and the width and height of the anchor frame are obtained from the pixel coordinates of the four corner points of the anchor frame. The center pixel coordinates of the anchor frame, that is, the two-dimensional center position coordinates of the rotor 4 and the stator 5, are further calculated, and the parallax of the images of the rotor 4 and the stator 5 is obtained by using a binocular camera. The depth information of the rotor 4 and the stator 5 is obtained by calculating the parallax, and the three-dimensional image coordinate information is obtained in combination with the two-dimensional center position coordinates, and it is displayed on one side of the anchor frame.

[0066] In a preferred but non-limiting embodiment of the present invention, in step 5, the pixel coordinate conversion process mainly converts the anchor frame pixel coordinates obtained in the recognition process into actual position coordinates, such as Figure 3 As shown, first, from the world coordinate system ( ) to the camera coordinate system ( ) is transformed according to the position and orientation of the camera, and the three-dimensional position of the object is mapped to the camera coordinate system through rigid body transformation. Then, the intrinsic parameters of the camera are obtained and the camera coordinate system ( ) is mapped to the image physical coordinate system (x, y), and the three-dimensional point is converted into a point on a two-dimensional plane, and parameters such as the focal length and optical center of the camera are taken into consideration. Finally, the pixel coordinate position of the point in the image is obtained by mapping the image physical coordinates to the image pixel coordinates (u, v). The depth information is obtained by measuring the parallax of the corresponding points in the two camera images and calculating the depth in the three-dimensional space using the principle of triangulation, and finally the real-world coordinates of the centers of the rotor 4 and the stator 5 are obtained.

[0067] In a preferred but non-limiting embodiment of the present invention, in step 5, the three-dimensional coordinate information is displayed. Taking the recognition effect of the rotor 4 and stator 5 models as an example, first, the position information after coordinate conversion is divided into three dimensions of X axis, Y axis, and Z axis for display, with the image acquisition module 2 as the coordinate origin. Figure 5 The confidence level for identifying and locating the magnetic pole of the rotor 5 is 0.93, and the distances between the center of the anchor frame of the rotor 4 and the image acquisition module 2 in the X, Y, and Z dimensions are 7.45 cm, 1.89 cm, and 40.34 cm, respectively. Figure 6 The winding identification and positioning effect of the stator 6 is Figure 7 For the recognition and positioning effect of the rotor 4 entering the stator pit, a better recognition effect is achieved when the gap is very small.

[0068] In a preferred but non-limiting embodiment of the present invention, in step 6, the critical position detection procedure performs the following method: the rotor 4 and stator 5 are divided into four detection areas. Multiple image acquisition modules 2 located below the double-girder bridge crane 3 capture corresponding images. Based on the structural parameters of the double-girder bridge crane 3, the target position in the camera coordinate system is converted to the target position in the double-girder bridge crane 3's own coordinate system. In the computer processing platform, a preset gap distance between the rotor 4 and stator 5 is set, and the anchor frame colors are uniformly set to green and red. Before and after the rotor 4 and stator 5 enter the pit, the deviations (ΔX, ΔY, ΔZ) between the centers of the rotor 4 and stator 5 are calculated and their absolute values ​​are taken. The three-dimensional coordinate information of the rotor 4 and stator 5 is then compared in real time. A safety range threshold is set (e.g., ΔX ≤ 5 cm, ΔY ≤ 5 cm). The calculated values ​​are compared with the thresholds to determine whether they meet the preset conditions. If the position meets the threshold, a green prompt box is displayed, indicating normal position. If the position exceeds the threshold, a red error box is displayed, and an alarm is issued through the speaker. The detection results are fed back to the operator of the double-girder bridge crane 3 in real time, and the target object and its deviation information are displayed on the display screen in the bridge crane control room to guide the operator to adjust the position of the double-girder bridge crane 3.

[0069] like Figure 1 As shown, Example 2 of the present invention provides a hydropower plant rotor hoisting critical position detection system, comprising:

[0070] Image acquisition module 2, including a binocular camera, starts and acquires raw image data upon receiving a detection instruction from the processor;

[0071] The communication module is electrically connected to the image acquisition module 2 and serves as a communication terminal between the processor and the image acquisition module 2 to transmit image data and receive instruction information;

[0072] The processor is electrically connected to the communication module, performs preprocessing operations on the original image, runs the rotor and stator neural network recognition model, identifies and locates the rotor 4 and stator 5 in the image, and converts the coordinates;

[0073] The coordinate information feedback module is electrically connected to the processor and displays the anchor frame of the rotor 4 and the stator 5 in real time. The coordinate information of the rotor 4 and the stator 5 in three dimensions, X, Y and Z, is dynamically displayed on one side of the anchor frame.

[0074] The critical position detection module is electrically connected to the processor, calculates the coordinate difference between the rotor 4 and the stator 5, displays the relative position information, and dynamically detects whether the gap between the rotor 4 and the stator 5 meets the preset requirements;

[0075] The host computer 6 is electrically connected to the processor.

[0076] The beneficial effect of the present invention is that, compared with the prior art,

[0077] 1. This invention incorporates a neural network to identify and locate targets, thereby assisting in bridge crane hoisting operations. Traditional hoisting operations require extensive manpower to determine the position. This invention utilizes video recognition and positioning to achieve non-contact detection of the rotor's critical position during hoisting, improving hoisting efficiency, shortening the entire hoisting process, and reducing both labor and material resources.

[0078] 2. Traditional hydropower plant monitoring systems have limited functionality, solely capturing and recording the hoisting operation scene. The rotor hoisting critical position detection system proposed in this invention not only captures monitoring footage but also provides real-time updates and feedback on target information, continuously acquiring the latest target information and ensuring real-time tracking and positioning of the target.

[0079] 3. It can provide early warning of potential rotor-stator collision risks, ensuring accurate positioning and safety of the rotor during the hoisting process. Furthermore, the positioning method of the present invention uses multi-point positioning, which can capture all angles of the rotor and stator, enabling blind spot identification and positioning during the hoisting operation.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A method for detecting the critical position of a rotor during hoisting in a hydropower plant, characterized in that: The following steps are involved: The image acquisition module collects images of the hydropower plant's rotor and stator from multiple angles, classifies and labels the images to obtain a data set for training the neural network. Perform data enhancement on the data set, expand the data set, configure the training environment, use the data set to train the neural network, and build the rotor and stator neural network recognition model; An image acquisition module is installed under the bridge beam and main beam of a double-girder bridge crane to obtain original images of the rotor and stator. The original images are transmitted to the processor and host computer for image preprocessing and model recognition. The preprocessed target image is obtained and the rotor and stator neural network recognition model is used to identify the rotor and stator in the target image. After successfully recognizing the target image, the pixel coordinates of the image are further obtained. The classification information of the rotor and stator is determined based on the confidence level. A visual anchor frame is formed at the edges of the rotor and stator. The pixel coordinates of the center of the anchor frame, i.e., the center point of the rotor and stator, are obtained. The pixel coordinates of the two-dimensional images of the rotor and stator are obtained, and the three-dimensional pixel coordinates are obtained by combining the depth information. The three-dimensional pixel coordinates of the rotor and stator images in the video, as well as the camera's focal length, principal point coordinates, pixel scaling factor, rotation matrix, and translation vector are converted into the three-dimensional coordinate data of the rotor and stator centers in the world coordinate system. After pixel coordinate conversion is performed in the processor and host computer, the world coordinate data is obtained, and the rotor and stator anchor frames and three-dimensional coordinate information are displayed; The processor and host computer calculate the difference in the three-dimensional coordinates of the rotor and stator, set a critical position detection program to determine the gap between the rotor and stator, and issue an alarm when the difference between the two approaches or exceeds the preset gap value; The process of obtaining the three-dimensional pixel coordinates of the rotor and stator includes: After successfully identifying the rotor and stator, the center pixel coordinates of the anchor frame, i.e., the two-dimensional center position coordinates of the rotor and stator, are calculated. The disparity of the rotor and stator images is obtained, and the depth information of the rotor and stator is obtained by calculating the disparity. The three-dimensional image coordinate information is obtained by combining the two-dimensional center position coordinates and displayed on the side of the anchor frame. The critical position detection determination method of the critical position detection program includes: The rotor and stator are divided into four detection areas, and the corresponding images are obtained by multiple image acquisition modules under the double-girder bridge crane. According to the structural parameters of the double-girder bridge crane, the target position in the camera coordinate system is converted into the target position in the double-girder bridge crane's own coordinate system. In the computer processing platform, the gap distance between the rotor and the stator is preset, and the anchor frame color is uniformly set to green and red. Before and after the rotor and stator enter the pit, the deviation value between the rotor center and the stator center is calculated and the absolute value is taken. The three-dimensional coordinate information of the rotor and stator is compared in real time, and a safety range threshold is set. The calculated value is compared with the threshold to determine whether it meets the preset conditions.

2. The method for detecting the critical position of a rotor during installation in a hydropower plant according to claim 1, wherein: For the produced image samples, data enhancement is performed using operations such as rotation, scaling, flipping, and cropping. On the basis of the original neural network, image segmentation is introduced and an attention mechanism is added to improve the recognition of local features of the rotor and stator. The network of the rotor and stator neural network recognition model is pruned, and the number of neural network convolution layers and the size of the convolution kernel are adjusted. The recognition effect of the rotor and stator neural network recognition model is verified using the validation set every training cycle.

3. The method for detecting the critical position of a rotor hoisting in a hydropower plant according to claim 1, characterized in that: The image preprocessing process includes: delineation of the region of interest, grayscale conversion, and filtering and noise reduction; Delineate the region of interest and focus on processing specific parts of the image around the rotor and stator during rotor hoisting, thereby narrowing the video recognition range; The three channels of the color image are reduced to one channel through grayscale processing, which reduces the number of pixels and color interference; Filtering noise reduction is used to reduce the interference to the target image caused by insufficient or excessive light in the hydropower plant building, and to suppress the noise of the target image.

4. The method for detecting the critical position of a rotor hoisting in a hydropower plant according to claim 1, characterized in that: The rotor and stator neural network identification model identification process includes: The preprocessed image is input into the neural network recognition model for processing, and the pixel values ​​are normalized from the range of [0, 255] to the range of [-1, 1]. The rotor and stator in the image are extracted through the backbone network to obtain feature maps of different scales. The feature maps are enlarged to the same size as the high-level feature maps and then fused with the high-level feature maps. The image is divided into grids in the prediction layer, and the bounding box of the target in each grid cell is predicted to obtain the confidence and category of the predicted target. The predicted bounding box coordinates are converted from normalized coordinates to image coordinates, and the actual width and height are calculated. After obtaining the target detection results, the overlapping bounding boxes are eliminated, and the bounding boxes with the highest confidence are retained by setting the confidence threshold. The bounding boxes below the threshold will be ignored.

5. The method for detecting the critical position of a rotor during installation in a hydropower plant according to claim 1, wherein: The pixel coordinate conversion process is to convert the pixel coordinates of the anchor frame obtained in the recognition process into the actual position coordinates, from the world coordinate system ( ) to the camera coordinate system ( ) is transformed according to the position and orientation of the camera, and the three-dimensional position of the object is mapped to the camera coordinate system through rigid body transformation, and the intrinsic parameters of the camera are obtained. The camera coordinate system ( ) is mapped to the image physical coordinate system (x, y), and the three-dimensional point is converted into a point on a two-dimensional plane. Taking into account parameters such as the focal length and optical center of the camera, the pixel coordinate position of the point in the image is obtained by mapping the image physical coordinates to the image pixel coordinates (u, v). The depth information is obtained by measuring the parallax of the corresponding points in the two camera images and calculating the depth in the three-dimensional space using the principle of triangulation to obtain the real-world coordinates of the rotor and stator centers.

6. The method for detecting the critical position of a rotor during hoisting in a hydropower plant according to claim 1, wherein: If the position meets the threshold, a green prompt box is displayed, indicating that the position is normal; if the position exceeds the threshold, a red error box is displayed and an alarm is issued through the speaker.

7. The method for detecting the critical position of a rotor during installation in a hydropower plant according to claim 1, wherein: The detection results are fed back to the double-girder bridge crane operator in real time. The target object and its deviation information are displayed on the display screen in the bridge crane control room to guide the operator to adjust the position of the double-girder bridge crane.

8. A hydropower plant rotor hoisting critical position detection system, based on the hydropower plant rotor hoisting critical position detection method according to any one of claims 1 to 7, characterized in that: include: The image acquisition module includes a binocular camera. When receiving a detection instruction from the processor, the image acquisition module starts and collects raw image data; The communication module is electrically connected to the image acquisition module and serves as a communication terminal between the processor and the image acquisition module to transmit image data and receive instruction information; A processor, electrically connected to the communication module, performs preprocessing operations on the original image, runs a rotor and stator neural network recognition model, identifies and locates the rotor and stator in the image, and converts the coordinates; The coordinate information feedback module is electrically connected to the processor and displays the rotor and stator anchor frames in real time. One side of the anchor frame dynamically displays the coordinate information of the rotor and stator in three dimensions: X, Y, and Z. The critical position detection module is electrically connected to the processor, calculates the coordinate difference between the rotor and the stator, displays the relative position information, and dynamically detects whether the gap between the rotor and the stator meets the preset requirements; An alarm module, including a speaker, is used to sound an alarm when the gap between the rotor and the stator does not meet the preset requirements; The host computer is electrically connected to the processor.

Images