Propeller phase synchronous measurement device and method based on U-NET feature recognition

Abstract

The invention provides a propeller phase synchronous measurement device and method based on U-NET feature recognition, and belongs to the technical field of flow field measurement, and the device comprises a test water tank, a to-be-measured object is installed in the test water tank, the four sides of the test water tank are provided with light-transmitting windows, and optical observation and measurement are carried out; the feature capture high-speed camera is arranged on a window on any side of the test water tank, and the water prism is arranged on a window of an observation surface of the test water tank where the feature capture high-speed camera is located; the active control processing system controls synchronous triggering of the feature capturing high-speed camera and the measuring equipment through the synchronous signal controller and records collected data. According to the invention, the characteristic capturing high-speed camera after algorithm learning is used for accurately positioning and outputting required phase angle information, so that the purpose of testing and measuring a flow field at a certain specific angle is achieved; and an indispensable data basis is provided for verifying and improving related unsteady flow physical models, optimizing hydrodynamic performance of fluid machinery and inhibiting flow instability.

Description

Technical Field

[0001] This invention belongs to the field of flow field measurement technology, specifically relating to a propeller phase synchronization measurement device and method based on U-NET feature recognition. Background Technology

[0002] In flow field experiments, determining phase measurements is crucial for a deep understanding of the nature of periodic or unsteady flows. Many complex flow phenomena, such as vortex shedding, disturbances caused by blade passing frequencies, or periodic shock wave motion, exhibit strong unsteady characteristics. Simple time-averaging or frequency domain analysis can only provide amplitude information, losing the crucial temporal correlation. Phase measurements reveal the sequence and synchronicity of flow events at different spatial points, enabling researchers to reconstruct the spatiotemporal evolution of flow structures. Through precise phase-locked measurements, the entire process of vortex generation, evolution, pairing, and dissipation can be clearly captured, and the phase lag relationship between hydrodynamic loads and flow structures can be quantitatively analyzed. This provides an indispensable data foundation for verifying and improving unsteady flow physical models, optimizing the hydrodynamic performance of fluid machinery, and suppressing flow instability. Therefore, phase measurement is a core element in deciphering the physical mechanisms of complex flow fields from a dynamic perspective.

[0003] However, most current phase synchronization devices can only output a single trigger signal at the beginning of each cycle, meaning they cannot pinpoint the precise phase angle or determine the initial angle of the object under test. To determine the specific angle of the object under test, current practices involve adding an additional camera synchronized with the existing testing device to specifically capture the rotation of the object, achieving a certain degree of phase synchronization. However, this method of adding an extra camera not only increases the data volume but is also limited by the device's acquisition rate, making it impossible to measure flow field characteristics at certain specific angles. In summary, current experimental methods for flow field phase synchronization measurement are not perfect. Therefore, developing a propeller phase synchronization measurement device and method based on U-NET feature recognition that can accurately locate phase characteristics is of great significance for a deeper understanding of periodic or unsteady flows. Summary of the Invention

[0004] The purpose of this invention is to provide a propeller phase synchronization measurement device and method based on U-NET feature recognition. The device uses a high-speed camera to accurately locate and output the required phase angle information by learning features through an algorithm, so as to achieve the purpose of experimental measurement of the flow field under a specific angle condition.

[0005] The objective of this invention is achieved through the following technical solution:

[0006] A propeller phase synchronization measurement device based on U-NET feature recognition includes: a test tank, a high-speed feature capture camera, a synchronization signal controller, a water prism, and an active control processing system;

[0007] The object to be tested is installed inside the test tank, which has light-transmitting windows on all four sides for optical observation and measurement.

[0008] The feature-capturing high-speed camera is positioned in the viewing window on any side of the test tank, allowing direct observation of the object under test through the optical glass on that side of the test tank.

[0009] The water prism is positioned at the viewing window of the test tank where the high-speed feature capture camera is located.

[0010] The synchronization signal controller connects the high-speed feature capture camera, the active control processing system, and the measuring equipment. The active control processing system controls the synchronous triggering of the high-speed feature capture camera and the measuring equipment through the synchronization signal controller to ensure the consistency of the device triggering and acquisition time, and records the acquired data.

[0011] Furthermore, the measuring device is a particle image acquisition camera and a pulsed laser.

[0012] Furthermore, the measuring device is a high-speed image acquisition camera.

[0013] Furthermore, the measuring device is a sensor.

[0014] The present invention may also include:

[0015] A method for performing PIV testing using the aforementioned propeller phase synchronization measurement device based on U-NET feature recognition includes:

[0016] The propeller power unit is started, followed by the feature capture high-speed camera and the particle image acquisition camera. The feature capture high-speed camera begins to acquire propeller images.

[0017] Before conducting the experiment, network training is required. This involves training the algorithm based on the U-NET structure to perform feature recognition and pixel-level point localization training. Once this is completed, the experiment can officially begin.

[0018] Phase information was calibrated using historical experimental images, including the propeller blade profile, hub profile, circular region, the phase vector formed by the center region and the hub center point and the horizontal angle. The calibrated data was divided into training set, validation set and test set in a ratio of 8:1:1.

[0019] When network training is complete and test results are ideal, the images captured by the high-speed camera are input into the network for recognition. After relevant index calculations, the phase information is judged. When the average intersection-union ratio (mIoU) is greater than 0.95, the average pixel accuracy (mPA) is greater than 0.9, the average Euclidean distance of the center point positioning accuracy is less than 5 pixels, and the phase angle error (PAE) is less than 0.5°, and the feature detection success rate (DSR) is higher than 0.95, the training effect is considered good. At this time, the phase matching is considered good, and the corresponding pulse signal is output.

[0020] The laser illuminates the area to be tested and stores the data;

[0021] Determine whether the stopping criterion is met. If it is, the process ends; if not, it returns. The high-speed camera for feature capture then begins acquiring propeller images and continues network discrimination until the stopping criterion is met.

[0022] Furthermore, the average crossover ratio It is the average of the ratios of the intersection and union of the predicted regions and the actual regions for all categories, and is used to measure the accuracy of the blade edge, circular region, and hub region segmentation.

[0023] Average crossover ratio of a single image :

[0024] .

[0025] in, For the number of categories, For the true category And it was correctly predicted as The number of pixels; For true category, not category However, it was incorrectly predicted as The number of pixels; The true category is However, it was incorrectly predicted as not belonging to the category. The number of pixels.

[0026] The entire dataset :

[0027] .

[0028] Furthermore, the average pixel precision (mPA) is calculated by taking the average of the proportion of correctly classified pixels out of all pixels.

[0029]

[0030] For the true category not And it was correctly predicted as not The number of pixels.

[0031] Furthermore, the center point positioning accuracy is used to measure the distance between the contour region point and the actual contour region point, the distance between the predicted center point and the actual center point, and the distance between the predicted hub center point and the actual hub center point, calculated using Euclidean distance:

[0032] .

[0033] Furthermore, the phase angle error (PAE) is used to measure the accuracy of phase angle prediction:

[0034]

[0035] Furthermore, the algorithm network uses cosine annealing to adjust the learning rate, and the loss function used is:

[0036]

[0037] in, and It is used to balance detection accuracy and positioning accuracy.

[0038] The beneficial effects of this invention are as follows:

[0039] This invention utilizes an optical system to acquire the phase state of the object under test and uses a trained algorithm to accurately locate its phase characteristics, outputting precise synchronization information to achieve the experimental objective of accurately capturing the flow field characteristics at a set angle.

[0040] This invention uses algorithm-learned features to capture high-speed cameras for accurate positioning and output of the required phase angle information, so as to achieve the purpose of experimental measurement of the flow field under a specific angle condition. This provides an indispensable data foundation for verifying and improving relevant unsteady flow physical models, optimizing the hydrodynamic performance of fluid machinery, and suppressing flow instability.

[0041] This invention achieves precise identification and high-synchronization-accuracy measurement of propeller phase in unsteady flow fields. The device integrates a high-speed feature-capturing camera and the U-NET deep learning algorithm to identify the propeller's contour and feature points in real time, accurately outputting arbitrary preset phase angle signals. A synchronization controller enables high-precision triggering synchronization with measurement equipment such as PIV particle image cameras and pulsed lasers. This method overcomes the limitations of traditional phase measurement methods that rely on a single-period trigger signal and an additional camera, significantly improving phase positioning accuracy and temporal resolution. It can effectively capture transient flow structures such as vortex evolution and blade disturbances at specific angles. The system possesses good scalability and can be adapted to various flow field measurement configurations, providing a reliable data foundation for in-depth research on periodic flow mechanisms, verification of unsteady fluid models, and optimization of fluid machinery performance. Attached Figure Description

[0042] Appendix Figure 1 This is a schematic diagram of the structure of the present invention when taking pictures using PIV;

[0043] Appendix Figure 2 This is a schematic diagram of the object feature recognition of the present invention;

[0044] Appendix Figure 3 This is an application flowchart of the present invention;

[0045] Appendix Figure 4 This is a schematic diagram of the test setup for the Tomo-PIV of the present invention.

[0046] In the attached diagram: 1. Test tank, 2. Test object, 3. Feature capture high-speed camera, 4. Synchronization signal controller, 5. Water prism, 6. Active control processing system, 7. Image acquisition camera, 8. Pulsed laser. Detailed Implementation

[0047] The present invention will now be further described with reference to the accompanying drawings.

[0048] This invention provides a propeller phase synchronization measurement device based on U-NET feature recognition, as shown in the attached figure. Figure 1 As shown, it includes: a test tank 1, an object under test 2, a high-speed camera for feature capture 3, a synchronous signal controller 4, a water prism 5, an active control processing system 6, and measuring equipment.

[0049] The test tank 1 has windows made of light-transmitting materials such as glass on all four sides, which can be used for optical observation and measurement;

[0050] The object under test 2 has periodic characteristics such as rotation and phase characteristics.

[0051] The high-speed feature-capturing camera 3 is positioned in the viewing window on any side of the test tank 1, allowing direct observation of the object under test through the optical glass of that side of the test tank 1. Figure 1 Taking the example of a high-speed feature-capturing camera 3 being positioned at the top of a test tank 1;

[0052] The high-speed feature-capturing camera 3 captures images of the object under test at a certain incident angle to ensure that feature information can be captured. Figure 1 Because the high-speed feature capture camera 3 is positioned above the test tank 1, the camera's incident angle is a top-down angle. Figure 2 In the figure, (a), (b), and (c) represent the contour of the object under test, the angle between the pre-marked feature points and the defined vector during the feature capture process, respectively. Figure 2 Image (d) shows the working status of the feature-capturing high-speed camera 3;

[0053] The water prism 5, together with the high-speed camera, is positioned at the observation window of the test water tank 1 where the feature capture high-speed camera 3 is located.

[0054] The synchronization signal controller 4 connects the feature capture high-speed camera 3, the measuring equipment and the active control processing system 6 to ensure the consistency of the triggering and acquisition time of the above devices;

[0055] The active control processing system 6 controls the synchronous triggering of the feature capture high-speed camera 3 and other measuring devices 7+8 through the synchronous signal controller 4, and records the acquired data.

[0056] The measuring device can be a flow field morphology or flow field quantitative characteristic measuring device. In the test environment of PIV as an example, it consists of a particle image acquisition camera 7 and a pulsed laser 8.

[0057] If a 2D3C or even 3D3C PIV test is required, it can be flexibly adjusted according to the test setup.

[0058] Appendix Figure 4 The experimental setup diagram of Tomo-PIV (a 3D3C flow field measurement method) is shown; if only the cavitation morphology needs to be observed, only a high-speed image acquisition camera needs to be equipped; if it is necessary to collect relevant excitation forces, noise and other physical characteristics, only the corresponding sensors need to be equipped.

[0059] In summary, different measuring devices can be used to suit different experimental purposes.

[0060] This embodiment uses the PIV test as an example to describe the detailed test procedure steps:

[0061] The algorithm is built on a U-Net architecture. Its encoder uses ResNet34 as the feature extraction module to fully extract multi-level features from the blades, circular markers, and hub. The decoder gradually recovers spatial detail information through upsampling and skip connections, and combines an attention gate mechanism to focus on key regions, ensuring pixel-level segmentation accuracy of the blade edges and the center point of the circular marker. This provides high-precision mask and coordinate information for subsequent phase angle calculation, ultimately achieving accurate phase recognition with an error of less than 0.5 degrees. The specific implementation steps are as follows (see attached). Figure 3 As shown:

[0062] Before conducting the experiment, the algorithm needs to be trained for feature recognition and pixel-level point localization. Once this is completed, the experiment can begin. The specific steps are as follows:

[0063] Phase information was calibrated using historical experimental images, including the propeller blade profile, hub profile, circular region, and the phase vector formed by the center region and the hub center point, along with the horizontal angle. The calibrated data was divided into training, validation, and test sets in an 8:1:1 ratio. The evaluation metrics used included mean intersection-over-union ratio (mIoU), mean pixel accuracy (mPA), feature detection success rate (DSR), center point localization accuracy, and phase angle error (PAE). Their specific definitions and calculation methods are as follows:

[0064] Average crossover ratio This refers to the average ratio of the intersection and union of the predicted regions and the actual regions for all categories. This metric is used to measure the accuracy of the blade edge, circular region, and hub region segmentation.

[0065] single image :

[0066]

[0067] in For the number of categories, For the true category And it was correctly predicted as The number of pixels; For true category, not category However, it was incorrectly predicted as The number of pixels; The true category is However, it was incorrectly predicted as not belonging to the category. The number of pixels.

[0068] The entire dataset :

[0069]

[0070] Mean Pixel Accuracy (mPA): Calculates the proportion of all pixels that are correctly classified, and then averages the proportions by category.

[0071]

[0072] For the true category not And it was correctly predicted as not The number of pixels.

[0073] Feature Detection Success Rate (DSR)

[0074]

[0075] To correctly detect the number of images with feature values, This represents the total number of images in the test set.

[0076] Center point positioning accuracy is used to measure the distance between points in the contour region and actual points in the contour region, the distance between the predicted center point and the actual center point, and the distance between the predicted hub center point and the actual hub center point. It is calculated using Euclidean distance.

[0077]

[0078] Phase Angle Error (PAE) is used to measure the accuracy of phase angle prediction.

[0079]

[0080] The algorithm network uses cosine annealing to adjust the learning rate, and the loss function used is:

[0081]

[0082] in, and It is used to balance detection accuracy and positioning accuracy.

[0083] After network training is completed and the test results are ideal, the images captured by the high-speed camera 3 are input into the network for recognition. After calculating relevant indicators, the phase information is judged. A good training effect is considered achieved when the following conditions are met: Intersection over Union (IoU) greater than 0.95, average pixel accuracy greater than 0.9, average Euclidean distance less than 5 pixels, absolute phase angle error less than 0.5°, and feature detection success rate (DSR) accuracy greater than 0.95. In this case, the phase matching is considered good, and a corresponding pulse signal is output.

[0084] In addition, it is necessary to determine the operating conditions and the test object. Among the operating conditions, besides the flow rate and rotational speed under normal conditions, the most important thing is to determine the phase angle to be photographed; and the test object in this example is a propeller with strong periodicity and phase characteristics.

[0085] After completing the attached document Figure 1 After the experimental platform is set up, all measuring equipment can be turned on. However, at this time, the high-speed camera 7, which is responsible for recording particle images, does not start acquiring data; both it and the pulsed laser 8 are in a state of waiting to be triggered. The task at this point is to accelerate the test object 2 to the operating conditions determined in the pre-experimental stage. Once its rotational speed stabilizes, the feature-capturing high-speed camera 3 begins operation, using a pre-trained pattern to identify the particle... Figure 2 The three identification features (a), (b), and (c) are used to determine that the previously preset phase is being executed. At this time, the pulse signals capturing the features are continuously transmitted to the active control processing system 6 for monitoring via the synchronous controller 4. It should be noted that the propeller blades all have at least two blades; therefore, a phase capture range of 0° to 180° for a specific blade is sufficient. The trained algorithm and the field of view determined by the placement of the high-speed feature capture camera 3 also meet this requirement. After the feature signals captured by the high-speed feature capture camera 3 stabilize, the active control processing system 6 issues a collection command, controlling the high-speed camera 7 and the pulsed laser 8 to begin synchronous operation via the synchronous controller 4, collecting the particle image of the previously preset specific phase. After a single collection is completed, the operating conditions can be changed or the phase angle to be measured can be altered for the next collection, repeating this cycle until the experimental objective is achieved.

[0086] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A propeller phase synchronization measurement device based on U-NET feature recognition, characterized in that, include: Test tank (1), feature capture high-speed camera (3), synchronization signal controller (4), water prism (5), active control processing system (6); The object to be tested (2) is installed inside the test water tank (1), which has light-transmitting windows on all four sides for optical observation and measurement; The feature-capturing high-speed camera (3) is arranged in the viewing window on any side of the test tank (1) and the test object is directly observed through the optical glass on that side of the test tank (1); The water prism (5) is positioned at the viewing window of the test tank (1) where the feature-capturing high-speed camera (3) is located; The synchronization signal controller (4) is connected to the feature capture high-speed camera (3), the active control processing system (6), and the measuring device. The active control processing system (6) controls the synchronous triggering of the feature capture high-speed camera (3) and the measuring device through the synchronization signal controller (4) to ensure the consistency of the device triggering and acquisition time, and records the acquired data.

2. The propeller phase synchronization measurement device based on U-NET feature recognition according to claim 1, characterized in that, The measuring equipment is a particle image acquisition camera (7) and a pulsed laser (8).

3. The propeller phase synchronization measurement device based on U-NET feature recognition according to claim 1, characterized in that, The measuring device is a high-speed image acquisition camera.

4. The propeller phase synchronization measurement device based on U-NET feature recognition according to claim 1, characterized in that, The measuring device is a sensor.

5. A method for performing PIV testing using a propeller phase synchronization measurement device based on U-NET feature recognition as described in any one of claims 1-4, characterized in that, include: Start the propeller power unit, and then start the feature capture high-speed camera (3) and particle image acquisition camera (7). The feature capture high-speed camera (3) begins to acquire propeller images. Before conducting the experiment, network training is required. This involves training the algorithm based on the U-NET structure to perform feature recognition and pixel-level point localization training. Once this is completed, the experiment can officially begin. Phase information was calibrated using historical experimental images, including the propeller blade profile, hub profile, circular region, the phase vector formed by the center region and the hub center point and the horizontal angle. The calibrated data was divided into training set, validation set and test set in a ratio of 8:1:

1. When the network training is completed and the test results are ideal, the image captured by the high-speed camera (3) is input into the network for recognition. After relevant index calculation, the phase information is judged. When the average intersection-union ratio mIoU is greater than 0.95, the average pixel accuracy mPA is greater than 0.9, the average Euclidean distance of the center point positioning accuracy is less than 5 pixels, and the phase angle error PAE is less than 0.5°, the feature detection success rate DSR accuracy is higher than 0.95, the phase matching is considered good, and the corresponding pulse signal is output. The laser illuminates the area to be tested and stores the data; Determine whether the stopping criterion is met. If it is met, the process ends; if it is not met, the process returns. The high-speed camera (3) for feature capture begins to acquire propeller images and continues to perform network discrimination until the stopping criterion is met.

6. The method for PIV testing using the propeller phase synchronization measurement device based on U-NET feature recognition according to claim 5, characterized in that, The average crossover ratio It is the average of the ratios of the intersection and union of the predicted regions and the actual regions for all categories, and is used to measure the accuracy of the segmentation of the blade edge, circular region, and hub region. Average crossover ratio of a single image : in, For the number of categories, For the true category And it was correctly predicted as The number of pixels; For true category, not category However, it was incorrectly predicted as The number of pixels; The true category is However, it was incorrectly predicted as not belonging to the category. The number of pixels, The entire dataset : 。 7. The method for PIV testing using the propeller phase synchronization measurement device based on U-NET feature recognition according to claim 6, characterized in that, The average pixel precision (mPA) is calculated by taking the average of the proportion of correctly classified pixels out of all pixels. For the true category not And it was correctly predicted as not The number of pixels.

8. The method for PIV testing using the propeller phase synchronization measurement device based on U-NET feature recognition according to claim 7, characterized in that, The center point positioning accuracy is used to measure the distance between the contour area point and the actual contour area point, the distance between the predicted center point and the actual center point, and the distance between the predicted hub center point and the actual hub center point. It is calculated using Euclidean distance. 。 9. The method for PIV testing using the propeller phase synchronization measurement device based on U-NET feature recognition according to claim 8, characterized in that, The phase angle error (PAE) is used to measure the accuracy of phase angle prediction. The feature detection success rate (DSR) is: in, To correctly detect the number of images with feature values, This represents the total number of images in the test set.

10. The method for PIV testing using the propeller phase synchronization measurement device based on U-NET feature recognition according to claim 9, characterized in that, The algorithm network uses cosine annealing to adjust the learning rate, and the loss function used is: in, and It is used to balance detection accuracy and positioning accuracy.

Images