A motor shaft vibration signal detection method and system based on monocular vision

Abstract

The application discloses a motor shaft vibration signal detection method based on monocular vision, comprising: camera calibration and depth estimation: calibrating a monocular camera, obtaining intrinsic parameters, extrinsic parameters and distortion coefficients, and estimating depth information of a detection scene; video acquisition and correction: acquiring motor shaft rotation video, and correcting distortion by using the calibration result; motor shaft axis positioning and tracking: accurately positioning the motor shaft axis in the corrected initial frame and selecting ROI, tracking the axis pixel coordinates in the subsequent frames, and forming a displacement trajectory; vibration information calculation: calculating the actual vibration amplitude of the motor shaft based on the axis displacement trajectory, camera parameters and estimated depth information. The application realizes non-contact and high-precision circle center vibration measurement, avoids signal distortion of contact type measurement, and is suitable for industrial continuous production scenes.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of motor condition detection and computer vision, specifically to a method for detecting motor shaft vibration signals based on monocular vision, which is applicable to non-contact vibration signal measurement of rotating cylindrical rigid bodies such as motor shafts and rotors in industrial scenarios. Background Technology

[0002] Existing methods for measuring motor shaft vibration mainly include contact sensors and non-contact sensors. Contact sensor measurement technology (CN218156490U): This technology is based on "direct sensor acquisition + wireless data transmission." It involves installing a vibration sensor on a square shell structure around the motor shaft to acquire vibration signals in real time. The vibration sensor is electrically connected to an ARM processor, which stores the data using a storage module and transmits the data to a computer terminal via a wireless communication module, enabling real-time monitoring and recording of vibration data. Simultaneously, to adapt to the stability of motor shaft operation, this technology incorporates a wave-like device consisting of a circular block, vertical rod, spring, and ring plate outside the square shell. The spring elastically buffers the impact of motor shaft vibration on the sensor mounting structure, making it widely used in stationary scenarios such as generator sets. Non-contact gap sensing measurement technology (CN211317734U): Addressing the challenges of compact space and difficult sensor placement in rotor systems, this technology places a damper between the bearing and the stator. A first sensor (such as a piezoelectric ceramic sensor) is placed in the second gap between the damper and the stator (larger than the first gap between the bearing and the rotor). The relative vibration information between the rotor and bearing is indirectly obtained by measuring the damper compression. Some solutions also add a second sensor (displacement / velocity / acceleration sensor), installed on the rotor system housing or support frame, to collect the absolute vibration information of the bearing and rotor relative to the ground. Finally, a processing and analysis device combines the relative and absolute vibration information to reconstruct the rotor / bearing motion trajectory. This technology is suitable for high-speed rotor systems such as micro gas turbines.

[0003] The first type of contact sensor is difficult to install directly on rotating parts. The square shell structure installed on the outer periphery of the motor shaft can only indirectly collect the vibration signal transmitted from the motor shaft to the square shell through the bearing. It cannot directly obtain the vibration data of the core rotating parts of the motor shaft. It is prone to attenuation or distortion due to the signal transmission path and cannot reflect the true vibration state of the motor shaft. The second type of non-contact sensor is limited by the installation environment. The sensor layout and adaptability are not high. Moreover, the calculation model is not optimized for the geometric characteristics of the cylindrical rigid body. It is difficult to accurately restore the two-dimensional vibration trajectory of the motor shaft center and cannot directly obtain the displacement of the cylindrical rigid body center. Summary of the Invention

[0004] This invention addresses the aforementioned problems by providing a method and system for detecting motor shaft vibration signals based on monocular vision. This solves the problem in existing technologies where high-precision, non-contact measurement of the vibration displacement of the center of a rotating cylindrical rigid body is impossible. This invention achieves stable and accurate measurement of motor shaft vibration displacement by combining camera calibration, depth information acquisition, and target tracking algorithms.

[0005] This invention provides a method for detecting motor shaft vibration signals based on monocular vision, characterized by the following steps: Step S1, camera calibration and depth information estimation: calibrating the monocular camera, obtaining the camera's intrinsic and extrinsic parameters and distortion coefficients, and estimating the depth information of the detection scene based on the calibration results; Step S2, video acquisition and correction: acquiring video of motor shaft rotation, and correcting video stream distortion using the calibration results; Step S3, precise positioning and tracking of the motor shaft centerline: locating the physical centerline of the motor shaft in the corrected initial video frame, and selecting a Region of Interest (ROI) centered on this centerline; continuously tracking the pixel coordinates of the centerline in subsequent video frames to form a centerline displacement trajectory; Step S4, vibration information calculation: calculating the actual vibration amplitude of the motor shaft based on the centerline displacement trajectory, camera parameters, and depth information.

[0006] Preferably, in step S1, multi-pose checkerboard images are acquired, the camera is calibrated using the Zhang Zhengyou calibration method to obtain camera intrinsic parameters, extrinsic parameters and distortion coefficients, and depth information is calculated based on the mapping relationship between the pixel coordinates of the checkerboard corner points and world coordinates.

[0007] Preferably, in step S3, locating the center of the motor shaft specifically includes: selecting a region containing the cross-sectional outline of the motor shaft in the initial frame of the video, extracting the outline of the motor shaft through edge detection, and locating the center of the shaft by combining Hough circle detection technology.

[0008] Preferably, the pixel radius of the circle in the Hough circle detection algorithm is calculated based on the transformation formula between the world coordinate system and the pixel coordinate system, and by combining the actual radius of the motor shaft with the camera calibration parameters.

[0009] Preferably, in step S3, the selection of ROI is based on the following logic: (1) If there is a directly identifiable optical feature in the central region of the shaft surface, then a rectangular region containing the optical feature is selected; (2) If not, then it is further determined whether an auxiliary optical mark can be added. If so, then a rectangular region containing the auxiliary optical mark is selected; (3) If none of the above conditions are met, then the outer rectangular region of the outer edge of the motor shaft end face is selected.

[0010] Preferably, in step S3, forming the axis displacement trajectory specifically includes: taking the axis as the center of the ROI and the selected ROI as the initial target template, continuously tracking the pixel coordinates of the axis in subsequent video frames to form the axis displacement trajectory.

[0011] Preferably, in step S4, a mapping model from pixel coordinates to world coordinates is established based on the axis displacement trajectory, camera parameters, and depth information. The axis pixel displacement is converted into the actual vibration displacement of the motor shaft, and the vibration amplitude curves in the X and Y directions are output.

[0012] Preferably, the frame rate of the monocular camera is not less than 30fps, which is suitable for industrial scenarios where the motor shaft speed range is 330-1350r / min.

[0013] The present invention also provides a system for implementing the above method, comprising: a monocular camera for acquiring video sequences of motor shaft operation; and a processing unit configured to perform the following operations: calibrating the monocular camera and estimating depth information; performing distortion correction on the acquired video sequence; locating the axis center of the motor shaft and selecting a Region of Interest (ROI) in the initial frame of the video; tracking the displacement of the axis center in subsequent frames; and calculating the actual vibration amplitude of the motor shaft based on the axis center displacement, camera parameters, and depth information.

[0014] Preferably, the processing unit is integrated into an industrial computer and uses an image processing library to implement the steps of the method in this invention. Attached Figure Description

[0015] Figure 1 This is a flowchart of the motor shaft vibration signal detection method based on monocular vision in an embodiment of the present invention. Detailed Implementation

[0016] To make the technical means, creative features, objectives and effects of this invention easier to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate a method and system for detecting motor shaft vibration signals based on monocular vision.

[0017] Example 1 This embodiment provides a method for detecting motor shaft vibration signals based on monocular vision.

[0018] Figure 1 This is a schematic diagram of the motor shaft vibration signal detection method based on monocular vision in an embodiment of the present invention.

[0019] like Figure 1 As shown, the steps of the motor shaft vibration signal detection method based on monocular vision involved in this embodiment are as follows: Step S1, Camera Calibration and Depth Information Acquisition: Zhang Zhengyou's calibration method is used to acquire multi-pose checkerboard images, obtaining the camera intrinsic parameter matrix, distortion coefficients, and extrinsic parameter matrix. Depth information is calculated based on the mapping relationship between the pixel coordinates of the checkerboard corner points and world coordinates, eliminating the need for the measured plane to be strictly perpendicular to the camera's optical axis, thus overcoming ranging errors in non-orthogonal scenes. The specific implementation method is as follows: Using a pre-made checkerboard calibration board (9×6 corner points, 8mm side squares), 25-40 calibration images of different poses were acquired. The average reprojection error was controlled to be ≤0.5 pixels. Camera calibration was performed using Zhang Zhengyou's camera calibration method to obtain the intrinsic parameter matrix. in , These represent the equivalent focal lengths of the camera in the horizontal and vertical directions in the image coordinate system, respectively. These correspond to the coordinates of the principal point in the image in the pixel coordinate system, and the distortion coefficients (radial distortion coefficients). , , Tangential distortion coefficient , The extrinsic parameter matrix is Where, the rotation matrix R = Translation vector , Let be the direction cosine between the world coordinate system and the camera coordinate system. They are The translation amount in the direction. The pixel coordinates of this plane and the world coordinate imaging model are as follows: in This is depth information. Referring to Zhang Zhengyou's calibration method, when the world coordinate system is defined on the plane of the calibration plate, Therefore, the above formula can be further simplified to: in Because a monocular camera can obtain the pixel coordinates of a point in a plane, but not the world coordinates of that point, it cannot directly obtain... The value of .

[0020] For two points on the calibration plate plane , The world coordinates are respectively ( , ( Their pixel coordinates are respectively , Expanding the above equation horizontally, we get: Simplifying the above expression by taking the difference, we obtain the depth information. The horizontal correspondence between the pixel coordinate difference and the world coordinate difference between the corner points of the chessboard grid and the corresponding pixel coordinate difference: Similarly, the vertical relationship can be obtained as shown in the equation: Therefore, when obtaining the depth information of the motor shaft, the checkerboard pattern is placed on the plane where the motor shaft is located, ensuring that the checkerboard plane is perpendicular to the motor shaft to the greatest extent possible. The difference in corner pixel coordinates is then extracted. Difference from world coordinates Depth information can be calculated by combining the parameters in the rotation matrix and the translation vector parameters. To reduce errors, the depth information obtained in the horizontal and vertical directions is averaged. This method makes full use of the camera's intrinsic and extrinsic parameters obtained from camera calibration, effectively suppressing rotational errors introduced by the non-orthogonality of the shooting plane.

[0021] Step S2 involves acquiring video of the motor shaft rotation and using the calibration parameters from step S1 combined with the distortion correction step in Zhang Zhengyou's calibration method to correct video frame distortion. In this embodiment, the acquired motor shaft rotation video is preprocessed for distortion correction using camera distortion coefficients and camera intrinsic parameter matrices to ensure image quality.

[0022] Step S3, precise positioning of the motor shaft center: In the initial frame, manually select the area containing the cross-sectional outline of the motor shaft, extract the motor shaft outline through Canny edge detection, and accurately locate the shaft center by combining Hough circle detection technology; construct an outer rectangle with the shaft center as the reference and a preset radius as the target tracking ROI area.

[0023] In the initial frame, the motor axis is manually outlined, and the axis contour is extracted using the Canny edge detection algorithm, which employs Gaussian filtering, gradient calculation, non-maximum suppression, and double-threshold hysteresis. Given the motor axis radius, the radius is converted to a pixel radius value based on the pixel coordinate to world coordinate transformation relationship established in S1. This yields the preset radius pixel value for the motor axis. Then, Hough circle detection is performed within the contour area to accurately locate the motor axis center coordinates. Finally, the Region of Interest (ROI) is selected based on the following principles: (a) If it is determined that there are directly identifiable optical features in the central region of the axial surface, then select a minimum rectangular region centered on the geometric center that completely contains the optical features as the first ROI, and the size of the first ROI is not less than the preset tracking size lower limit. (b) If it is determined that there are no directly identifiable optical features in the central region of the axial surface, then it is further determined whether auxiliary optical marks can be added in this region; (c) If it is determined that an auxiliary optical mark can be added, then after the addition, select a minimum rectangular area centered on the geometric center that completely contains the auxiliary optical mark as the second ROI, and the size of the second ROI is not less than the lower limit of the tracking size; (d) If it is determined that no additional auxiliary optical mark can be added, then the outer rectangle of the outer edge of the motor shaft end face is selected as the third ROI; In this embodiment, an outer rectangle is constructed with the axis as the center and the radius of the motor shaft, and the ROI is selected.

[0024] Step S3 also includes a target tracking strategy: using the axis as the center of the ROI region, an ROI is selected according to the ROI selection strategy, and then the tracking algorithm is used to estimate the displacement trajectory of the axis in the image sequence frame by frame to avoid system errors caused by the misalignment of the bounding box center and the axis. The above ROI selection strategy follows the following principles: the selected region must contain sufficiently discriminative features to ensure tracking stability; while meeting the minimum size requirements of the tracking algorithm, the region should be minimized as much as possible to ensure real-time tracking.

[0025] In this embodiment, the STARK target tracking algorithm is activated, with the initial axis center as the center of the ROI region and the ROI as the tracking area. The axis center displacement is continuously tracked in subsequent frames, and the axis center pixel coordinate sequence is output frame by frame.

[0026] Step S4, Pixel-to-World Displacement Conversion: Using camera parameters and depth information, a mapping model from pixel coordinates to world coordinates is established, converting the axis pixel displacement into the actual vibration displacement of the motor shaft, and outputting the vibration amplitude curves in the X and Y directions.

[0027] In this embodiment, based on camera intrinsic and extrinsic parameters, depth information, and a pixel-world coordinate mapping model, the axisymmetric pixel displacement is converted into actual vibration displacement, and the vibration amplitude in the X and Y directions is calculated. Vibration amplitude curves in the X and Y directions are then output for fault early warning and condition monitoring.

[0028] In summary, as Figure 1 As shown, the specific process for detecting motor shaft vibration signals in this embodiment is as follows: Initialization: Start the motor shaft vibration detection system and complete the initial setting of system parameters.

[0029] Monocular camera calibration: Calibrate the monocular camera to obtain its intrinsic and extrinsic parameters and distortion coefficients.

[0030] Depth information acquisition: Following the aforementioned depth information acquisition process, the depth information of the detection scene is calculated.

[0031] Video capture: Capture video data of the motor shaft rotation process.

[0032] Video distortion correction: Based on camera intrinsic parameters and distortion coefficients, distortion correction is performed on the acquired video to eliminate the impact of camera distortion on subsequent detection.

[0033] Precise shaft center positioning: In the initial frame, the center of the motor shaft is precisely located using Canny edge detection and Hough circle detection technology.

[0034] Target tracking: Select a Region of Interest (ROI) centered on the aforementioned axis and use target tracking algorithms such as STARK to continuously track the ROI in subsequent frames, thereby directly obtaining the displacement trajectory of the axis pixel coordinates.

[0035] Pixel information acquisition: Obtain the pixel coordinate information of the motor shaft center after correction.

[0036] Coordinate transformation: Convert the pixel coordinates of the motor shaft centerline to coordinates in the camera coordinate system.

[0037] Vibration amplitude calculation: The vibration amplitude of the motor shaft is calculated based on the coordinate changes in the camera coordinate system, thus completing the detection process.

[0038] This embodiment features an optimized design for industrial scenarios: with a camera frame rate of 30fps, it supports a rotational speed range of 330-1350r / min, and improves feature clarity through double image magnification; it eliminates the need for motor shaft surface preprocessing, resists interference from light fluctuations and motion blur, and is suitable for continuous industrial production scenarios.

[0039] Example 2 This embodiment provides a motor shaft vibration signal detection system based on monocular vision.

[0040] The motor shaft vibration signal detection system based on monocular vision involved in this embodiment includes a monocular camera and a processing unit.

[0041] A monocular camera is used to capture video sequences of the motor shaft's rotation.

[0042] The processing unit is configured to perform the following operations: Perform the process in step S1 of embodiment 1 to calibrate the monocular camera and estimate depth information; Perform the process in step S2 of Example 1 to perform distortion correction on the video frames; The process in step S3 of embodiment 1 is executed as follows: first, the motor shaft is manually framed in the initial frame, and the center of the motor shaft is automatically located; then, the ROI region is selected according to the ROI selection strategy, and finally, the displacement of the center of the motor shaft is tracked using a video target tracking algorithm. The process in step S4 of Example 1 is executed to calculate the actual vibration amplitude of the motor shaft based on the motor shaft center displacement, camera parameters, and depth information.

[0043] The above embodiments are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention.

[0044] Beneficial effects The present invention has the following technical effects: Existing technologies (such as traditional pinhole imaging) require the plane being measured to be strictly perpendicular to the camera's optical axis. This invention obtains depth information by mapping the camera calibration to the corner points of a checkerboard grid, avoiding the dependence of traditional pinhole imaging on orthogonality and improving ranging accuracy.

[0045] Existing target tracking algorithms often fail to align the bounding box center with the motor shaft center, introducing systematic errors. This invention addresses this issue with an improved strategy of "first-frame geometric constraints + subsequent shaft center tracking": In the initial frame, the motor shaft center is accurately located using Canny edge detection contours and Hough circle detection, resolving the error accumulation problem caused by the misalignment of the bounding box center and the shaft center. Then, an ROI selection strategy is designed to ensure both tracking accuracy and system real-time performance. Finally, existing target tracking algorithms are used for continuous tracking of the shaft center coordinates.

[0046] Existing detection technologies (contact sensors and non-contact sensors) are complex to install, have low adaptability to different operating conditions, and are costly. This invention uses a monocular camera for non-contact data acquisition, resulting in a non-contact, low-cost method suitable for complex industrial environments.

[0047] In summary, this invention achieves high-precision displacement measurement through the following methods: First, depth information is obtained based on Zhang Zhengyou's calibration method and checkerboard corner point mapping to establish an accurate depth benchmark required for measurement; then, Canny edge detection and Hough circle detection based on the true radius are fused in the initial frame to achieve precise positioning of the axis center; subsequently, a ROI selection strategy centered on the motor axis center is designed to improve system real-time performance while ensuring tracking accuracy; furthermore, a target tracking algorithm is used to stably track the axis center movement in consecutive frames, avoiding the system error caused by replacing the axis center with a manually defined bounding box center in traditional methods; finally, a mapping model from pixel coordinates to world coordinates is constructed by combining camera parameters and depth information to calculate the actual physical displacement of the motor axis.

[0048] This invention does not require physical contact with the motor shaft, avoiding the impact of the sensor's added mass on motor operation. It is wear-free and requires no downtime, making it suitable for continuous industrial production scenarios. It is low-cost and easy to deploy, directly outputting physical vibration amplitude curves, and can be integrated into motor health monitoring systems to provide quantitative early warning for faults such as bearing loosening and abnormal dynamic balance.

Claims

1. A method for detecting motor shaft vibration signals based on monocular vision, characterized in that, Includes the following steps: Step S1, Camera calibration and depth information estimation: The monocular camera is calibrated to obtain the camera's intrinsic parameters, extrinsic parameters and distortion coefficients, and the depth information of the detection scene is estimated based on the calibration results; Step S2, Video Acquisition and Correction: Acquire video of motor shaft rotation, and use the calibration results to correct video stream distortion; Step S3, precise positioning and tracking of the motor shaft center: In the corrected initial frame of the video, the physical center of the motor shaft is located, and an ROI is selected with the center as the center; in subsequent video frames, the pixel coordinates of the center are continuously tracked to form the center displacement trajectory; Step S4, Vibration information calculation: Based on the shaft center displacement trajectory, camera parameters and the depth information, the actual vibration amplitude of the motor shaft is calculated.

2. The method for detecting motor shaft vibration signals based on monocular vision according to claim 1, characterized in that: in, In step S1, multi-pose checkerboard images are acquired, the camera is calibrated using the Zhang Zhengyou calibration method to obtain the camera's intrinsic parameters, extrinsic parameters, and distortion coefficients, and the depth information is calculated based on the mapping relationship between the pixel coordinates of the checkerboard corner points and the world coordinates.

3. The method for detecting motor shaft vibration signals based on monocular vision according to claim 1, characterized in that: in, In step S3, locating the center of the motor shaft specifically includes: selecting a region containing the cross-sectional outline of the motor shaft in the initial frame of the video, extracting the outline of the motor shaft through edge detection, and locating the center of the shaft by combining Hough circle detection technology.

4. The method for detecting motor shaft vibration signals based on monocular vision according to claim 3, characterized in that: in, The pixel radius of the circle in the Hough circle detection algorithm is calculated based on the transformation formula between the world coordinate system and the pixel coordinate system, and by combining the actual radius of the motor shaft and the camera calibration parameters.

5. The method for detecting motor shaft vibration signals based on monocular vision according to claim 1, characterized in that: in, In step S3, the selection of the ROI is based on the following logic: (1) If there is a directly identifiable optical feature in the central region of the axial surface, then select a rectangular region containing the optical feature; (2) If not, then determine whether an auxiliary optical mark can be added. If so, select a rectangular area containing the auxiliary optical mark. (3) If none of the above conditions are met, then select the outer rectangular area of ​​the outer edge of the motor shaft end face.

6. The method for detecting motor shaft vibration signals based on monocular vision according to claim 1, characterized in that: in, In step S3, forming the axial displacement trajectory specifically includes: Using the axis as the center of the region of interest (ROI) and the selected ROI as the initial target template, the pixel coordinates of the axis are continuously tracked in subsequent video frames to form the axis displacement trajectory.

7. The method for detecting motor shaft vibration signals based on monocular vision according to claim 1, characterized in that: in, In step S4, based on the axis displacement trajectory, camera parameters and depth information, a mapping model from pixel coordinates to world coordinates is established, the axis pixel displacement is converted into the actual vibration displacement of the motor shaft, and the vibration amplitude curves in the X and Y directions are output.

8. The method for detecting motor shaft vibration signals based on monocular vision according to claim 1, characterized in that: in, The monocular camera has a frame rate of no less than 30fps and is suitable for industrial scenarios where the motor shaft speed range is 330-1350r / min.

9. A monocular vision-based motor shaft vibration signal detection system for implementing the method of any one of claims 1-8, characterized in that, include: A monocular camera is used to capture video sequences of the motor shaft's rotation. The processing unit is configured to perform the following operations: The monocular camera is calibrated and its depth information is estimated. Distortion correction is performed on the acquired video sequence; Locate the center of the motor shaft and select the ROI in the initial frame of the video; The displacement of the axis is tracked in subsequent frames; The actual vibration amplitude of the motor shaft is calculated based on the shaft displacement, camera parameters, and depth information.

10. A motor shaft vibration signal detection system based on monocular vision according to claim 9, characterized in that: in, The processing unit is integrated into an industrial computer and uses an image processing library to implement the steps of the method.

Images