A machine vision-based online correction method for motor rotor dynamic balance

Abstract

The application discloses a motor rotor dynamic balance online correction method based on machine vision, comprising the following steps: acquiring the sequence visual data of the rotor in the running state through a high-speed visual acquisition module, calling a rotor visual feature coupling dynamics model to extract vibration and mass distribution correlation parameters; using a visual interframe rotor vibration differential algorithm to obtain real-time vibration acceleration and instantaneous speed parameters, constructing an unbalance amount analysis model based on a rotor mass visual analysis balance platform; using a visual sequence balance error correction algorithm to compensate and correct the analysis results, determining the specific position and value of the unbalance amount, and then performing targeted mass adjustment through an online correction execution module. The method realizes online accurate correction of rotor dynamic balance through the collaborative operation of multiple algorithms and step refinement and decomposition, does not need to disassemble the rotor, improves the correction efficiency and accuracy, and meets the efficient and stable operation requirements of the motor.

Description

Technical Field

[0001] This invention relates to the field of online motor rotor correction technology, and in particular to an online dynamic balance correction method for motor rotors based on machine vision. Background Technology

[0002] As the core transmission component of a motor, the dynamic balance performance of the motor rotor directly affects the motor's operational stability, energy consumption, and service life. In industrial production, rail transportation, and new energy equipment, motors are developing towards higher speeds, lighter weights, and greater precision, continuously increasing the requirements for rotor dynamic balance accuracy. Traditional dynamic balancing correction largely relies on offline testing equipment, requiring rotor disassembly for static or semi-dynamic testing. This approach suffers from long testing cycles, low production efficiency, and an inability to respond in real-time to changes in balance caused by rotor wear and deformation during operation. With the popularization of intelligent manufacturing technologies, online real-time correction has become a key requirement for solving rotor dynamic balance problems. Machine vision, with its advantages of non-contact measurement, high acquisition frequency, and simultaneous acquisition of multiple parameters, is gradually becoming the core technological support for online rotor dynamic balance detection and correction, driving the transformation of dynamic balancing correction from an offline static mode to an online dynamic intelligent mode.

[0003] Existing technologies have significant limitations in online dynamic balancing of motor rotors: On the one hand, traditional online detection methods are mostly based on data acquisition from a single vibration sensor, which can only obtain dynamic parameters such as rotor vibration amplitude and phase, lacking coupled analysis of visual information such as rotor surface features and mass distribution. This results in insufficient accuracy in locating the imbalance and difficulty in accurately matching the correlation between uneven mass distribution and vibration response. On the other hand, existing error correction mechanisms mostly rely on preset fixed coefficients or simple linear compensation algorithms, failing to fully consider the impact of dynamic factors such as environmental interference and inter-frame delay on the detection results during visual acquisition. Furthermore, they lack a dedicated correction model for the fusion analysis of visual sequence data and dynamic parameters, resulting in large errors in imbalance calculation. This, in turn, affects the stability and reliability of the correction effect, and cannot meet the dynamic balancing requirements of high-precision motor rotors. Summary of the Invention

[0004] To overcome the shortcomings and deficiencies of existing technologies, this invention provides an online dynamic balance correction method for motor rotors based on machine vision.

[0005] The technical solution adopted in this invention is an online dynamic balance correction method for motor rotors based on machine vision, comprising the following steps: S1, continuously acquiring visual frames of the motor rotor in operation through a high-speed vision acquisition module to obtain sequential visual data including rotor surface feature points, contour edges, and vibration displacement information; S2, calling the rotor visual feature coupled dynamic model to perform coupled analysis on the feature information in the sequential visual data, extracting rotor vibration amplitude, phase shift, and mass distribution correlation parameters; S3, using the inter-frame rotor vibration differential algorithm to perform differential calculation on the rotor vibration displacement data in adjacent visual frames to obtain the rotor real-time vibration acceleration and instantaneous velocity parameters; S4, based on the rotor mass visual analysis balancing platform, combining the mass distribution correlation parameters extracted in S2 and the vibration dynamic parameters obtained in S3, constructing a rotor mass imbalance analysis model; S5, using a visual sequence balance error correction algorithm to compensate and correct the imbalance analysis results, determining the specific location and value of the rotor imbalance; S6, based on the imbalance data corrected in S5, performing targeted mass adjustment of the rotor through the online dynamic balance correction execution module to perform online dynamic balance correction of the rotor.

[0006] Furthermore, the expression for the rotor visual feature coupled dynamics model is as follows: ,in, The rotor visual characteristics are coupled with dynamic coefficients. The radial visual feature coordinates of the rotor. The axial visual feature coordinates of the rotor. The circumferential visual feature coordinates of the rotor. The rotor's angular velocity. The number of visual feature points, For the first Stiffness coupling coefficient of each feature point For the first A planar visual feature function for a feature point. For the first The circumferential vibration phase function of each feature point For the first Damping coupling coefficient at each characteristic point For the first The time derivative of the planar visual features of a feature point. For the first The time derivative of the circumferential vibration phase at each characteristic point.

[0007] Furthermore, the expression for the visual inter-frame rotor vibration differential algorithm is as follows: ,in, for Rotor vibration acceleration at all times for Rotor vibration displacement at all times for Rotor vibration displacement at all times for Rotor vibration displacement at all times The visual frame acquisition time interval. These are the inter-frame differential correction coefficients. This is the velocity correlation coefficient.

[0008] Furthermore, the expression for the visual sequence balance error correction algorithm is as follows: ,in, This is the corrected rotor imbalance. The initial unbalance value is calculated. For the number of frames in the visual sequence, For the first The phase error correction coefficient of the frame. For the first The phase deviation of the frame. For the first The unbalanced component corresponding to the frame, For the first Frame amplitude error correction coefficient, For the first Frame amplitude deviation, For the first The rate of change of frame imbalance over time.

[0009] Furthermore, the mass distribution calculation model of the rotor mass visual analysis balancing platform is as follows: ,in, For rotor mass distribution density, For rotor material density, The coordinates of the rotor's circumferential angle are... For the rotor axial coordinate, The effective volume of the rotor, For visual analysis area, Let be the material density distribution function at the feature point. The distance from the feature point to the rotor axis. This is the visual effective area function for feature points.

[0010] Furthermore, the calculation model for the adjustment amount of the online dynamic balance correction of the motor rotor is as follows: ,in, This is the rotor mass adjustment amount. The inscribed angle of the unbalanced quantity. This indicates the axial position of the unbalance. For correction factors, This is the corrected imbalance. The rotor's angular velocity. The distance from the point of imbalance to the axis. It represents the vibration phase angle.

[0011] Further, step S3 includes the following sub-steps: S31, extracting the coordinates of rotor contour feature points corresponding to two adjacent frames from the sequence visual data, determining the position offset of each feature point in the two frames, and establishing a feature point displacement mapping relationship; S32, based on the displacement mapping relationship, calculating the displacement change of each feature point in the inter-frame time interval, filtering out effective vibration displacement data and removing abnormal interference points; S33, performing first-order differential processing on the effective vibration displacement data using differential operation to obtain the instantaneous velocity parameters of rotor vibration, and recording the corresponding time node information at the same time; S34, performing second-order differential operation on the instantaneous velocity parameters, and combining them with the visual frame acquisition frequency parameters to obtain the real-time vibration acceleration data of the rotor, forming a complete set of vibration dynamic parameters.

[0012] Further, step S4 includes the following sub-steps: S41, importing the mass distribution correlation parameters extracted in S2 through the rotor mass visual analysis balancing platform, including feature point density distribution, contour symmetry, and coordinates of the mass concentration area; S42, aligning the vibration acceleration and instantaneous velocity parameters obtained in S3 with the mass distribution correlation parameters to establish a parameter mapping table; S43, based on the correlation data in the mapping table, constructing an unbalance analysis model framework with uneven mass distribution as the independent variable and vibration parameters as the dependent variable; S44, substituting rotor structural dimensions and material property parameters to optimize and adjust the model framework, determining the coefficients in the model, and forming an analysis model that can directly calculate the unbalance.

[0013] Further, step S5 includes the following sub-steps: S51, obtaining the imbalance calculation results of each frame in the visual sequence, statistically analyzing the dispersion of the calculation results, and determining the initial error range; S52, analyzing the impact of ambient light interference and frame acquisition delay factors on the calculation results during the visual acquisition process, and establishing an error influence factor database; S53, calling the visual sequence balance error correction algorithm, substituting the error influence factors into the algorithm formula, and calculating the error correction amount corresponding to each frame; S54, applying the error correction amount to the corresponding imbalance calculation results, smoothing the discrete calculation results, and obtaining the accurate imbalance position and value.

[0014] A machine vision-based online dynamic balancing method for motor rotors is implemented through different units, including: a high-speed vision multi-dimensional data acquisition unit, positioned opposite the motor rotor, for continuously acquiring visual data related to surface features, contour edges, and vibration displacement during rotor operation, and transmitting the data to a feature coupling analysis unit; a rotor visual feature coupling dynamic analysis unit, communicatively connected to the high-speed vision multi-dimensional data acquisition unit, for coupling the feature information in the visual data using a preset model to extract parameters related to vibration amplitude, phase shift, and mass distribution; and an inter-frame vibration differential calculation unit, coupled with the rotor visual feature coupling dynamic analysis unit. The system consists of a connection unit that performs differential calculations on vibration displacement data from adjacent frames, outputting instantaneous velocity and vibration acceleration parameters; a mass imbalance analysis unit that communicates with the rotor visual feature coupled dynamic analysis unit and the inter-frame vibration differential calculation unit, integrating various parameters to construct an analysis model and obtain initial imbalance data; a visual sequence error correction unit that connects to the mass imbalance analysis unit, using a dedicated algorithm to compensate for and correct errors in the initial imbalance data, outputting accurate imbalance information; and an online correction execution unit that communicates with the visual sequence error correction unit, adjusting the mass of the motor rotor based on the accurate imbalance information to perform online rotor dynamic balance correction.

[0015] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0016] This invention proposes an online correction method for motor rotor dynamic balance based on machine vision. Through high-speed visual acquisition and the collaborative operation of multiple algorithms and units, it achieves accurate online correction of rotor dynamic balance. It uses continuous visual frame acquisition to obtain rotor surface features, contours and vibration displacement data. By coupling the rotor visual features with a dynamic model, it deeply couples visual information with dynamic parameters, which makes up for the shortcomings of traditional methods that rely on a single vibration sensor and lack correlation analysis of mass distribution and vibration response, and greatly improves the accuracy of imbalance positioning.

[0017] Vibration velocity and acceleration parameters are accurately extracted by the visual inter-frame vibration differential algorithm. A dedicated unbalance analysis model is constructed by combining the rotor mass visual analysis balancing platform. Then, the error caused by dynamic interference factors is compensated in a targeted manner by the visual sequence balancing error correction algorithm. This solves the problems of simple error correction mechanism and weak anti-interference ability in the existing technology, and significantly reduces the calculation error of unbalance.

[0018] The entire process does not require disassembling the rotor. Through a closed-loop operation of online data acquisition, analysis, correction and calibration, it breaks through the limitations of traditional offline calibration, which is characterized by low efficiency and inability to respond to dynamic balance changes. At the same time, the functional units have clear division of labor and work together to ensure the real-time performance and reliability of the calibration process, meet the dynamic balance requirements of high-precision motor rotors, and provide a strong guarantee for the efficient and stable operation of the motor. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention.

[0020] Figure 2 This is a diagram showing the unit composition for implementing the method of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] like Figure 1 As shown, a machine vision-based online dynamic balance correction method for motor rotors includes the following steps:

[0023] S1, continuously acquires visual frames of the motor rotor in operation through the high-speed visual acquisition module, and obtains sequential visual data including rotor surface feature points, contour edges and vibration displacement information;

[0024] Specifically, in step S1, a high-speed vision acquisition module with a resolution of no less than 1920×1080 pixels and a frame rate of 500 to 1000 frames per second is used to continuously acquire visual frames while the motor rotor is running at its rated speed of 3000 to 15000 rpm. The acquisition time is set to 10 to 30 seconds depending on the rotor model, ensuring that at least 5000 to 30000 frames of sequential visual data are acquired. During the acquisition process, the high-speed vision acquisition module uses an industrial camera with a lens focal length of 16mm to 50mm, and an ISO sensitivity setting of 100 to 800, to accurately capture feature points (including processing marks, material texture differences, etc.) with a diameter of 0.1mm to 2mm on the rotor surface, as well as 0.05mm-level details of the contour edges and vibration displacement changes. The acquisition area covers more than 95% of the rotor's total length, focusing on areas prone to imbalance, such as the shaft ends and the middle cylindrical section. The acquisition mode is synchronized with the rotor's rotation, collecting 36 to 72 frames of data per revolution. This ensures that the sequence of visual data can fully reflect the surface feature distribution and vibration displacement trajectory of the rotor in each rotation cycle, providing high-density, high-resolution raw data support for subsequent feature extraction and dynamic analysis.

[0025] S2, call the rotor visual feature coupled dynamics model to perform coupled analysis on the feature information in the sequence visual data, and extract the rotor vibration amplitude, phase shift and mass distribution related parameters;

[0026] Specifically, step S2 calls a pre-built rotor visual feature coupling dynamics model to perform multi-dimensional coupling analysis on the feature information in the sequence visual data. The model has built-in feature point matching thresholds of 0.85 to 0.95 and contour edge extraction gradient thresholds of 10 to 30. The sliding window method (window size set to 5 to 10 frames) is used to compare and associate the rotor surface feature points and contour edge data in continuous visual frames frame by frame. During the analysis, the feature points are first located to obtain radial, axial, and circumferential three-dimensional coordinate data. Then, combined with information such as the curvature change of the contour edge and the fluctuation of the feature point spacing, the data is compared with the preset rotor standard feature database in the model. Valid feature information is screened out through correlation analysis (correlation coefficient threshold above 0.7). Based on the filtered feature data, the range of rotor vibration amplitude variation (accuracy up to 0.001 mm) and phase offset angle (accuracy up to 0.1°) are calculated. At the same time, mass distribution correlation parameters are extracted, including the standard deviation of feature point density distribution (controlled between 0.01 and 0.05), profile symmetry deviation (not exceeding 5%), area ratio of mass concentration region and center coordinates, etc. Through these parameters, a correlation mapping between visual features and rotor dynamic characteristics is established, providing core data support for subsequent unbalance analysis.

[0027] S3, the rotor vibration differential algorithm between visual frames is used to perform differential calculation on the rotor vibration displacement data in adjacent visual frames to obtain the real-time vibration acceleration and instantaneous velocity parameters of the rotor.

[0028] Specifically, step S3 employs a visual inter-frame rotor vibration differential algorithm to perform differential calculations on the vibration displacement data of the located rotor surface feature points in adjacent visual frames. During the calculation, the inter-frame time interval is set to 1ms to 2ms (adaptively adjusted according to the frame rate of the high-speed visual acquisition module). First, the difference between the three-dimensional coordinates of the same feature point in two adjacent frames is calculated to obtain the displacement change of that feature point between frames. Then, linear interpolation is used to supplement the missing displacement data between frames, ensuring the continuity of the displacement sequence. Based on the continuous displacement data, a first-order difference algorithm is used to calculate the instantaneous rotor vibration velocity, with a velocity calculation accuracy controlled at 0.01mm / s. Simultaneously, the time node corresponding to each velocity data point is recorded (timestamp accuracy reaches 1μs). Subsequently, a second-order difference algorithm is used to further calculate the instantaneous velocity data. Combined with the square of the inter-frame time interval, the real-time rotor vibration acceleration is obtained, with an acceleration calculation accuracy of 0.1mm / s². During the calculation process, an outlier threshold is set (velocity mutations exceeding 5 mm / s and acceleration mutations exceeding 10 mm / s² are considered outliers). Outlier data is smoothed, and finally, the vibration velocity and acceleration time series data of each feature point during the acquisition period are output, forming a vibration dynamic parameter set including at least 5,000 sets of valid data, providing accurate dynamic basis data for the subsequent construction of the unbalance analysis model.

[0029] S4. Based on the rotor mass visual analysis balancing platform, and combining the mass distribution correlation parameters extracted in S2 with the vibration dynamic parameters obtained in S3, a rotor mass imbalance analysis model is constructed.

[0030] Specifically, step S4 is based on a rotor mass visual analysis and balancing platform. This platform integrates a rotor structure parameter database (including different specifications such as rotor length from 50mm to 500mm and diameter from 10mm to 100mm), a material density database (including density parameters of commonly used rotor materials such as steel, aluminum, and copper), and a visual analysis algorithm library. First, the mass distribution correlation parameters extracted in step S2 are imported, including feature point density distribution data, contour symmetry deviation value, coordinates of mass concentration areas, and area ratio, etc. Then, the vibration acceleration and instantaneous velocity parameters obtained in step S3 are aligned with the mass distribution correlation parameters according to the timestamp to establish a one-to-one correspondence parameter mapping table. Each record in the mapping table includes visual feature parameters, vibration dynamic parameters, and rotor structure parameters at the same time node. Based on the associated data in the mapping table, an unbalance analysis model framework is constructed with uneven mass distribution as the independent variable and vibration parameters as the dependent variable. The model framework includes the correlation function between mass unbalance and vibration amplitude and phase shift. The model framework is fitted and optimized using the least squares method. The model coefficients are adjusted by substituting rotor structural dimensions (such as rotor radius, length, moment of inertia, etc.) and material properties (such as density, elastic modulus, etc.). The model fitting error threshold is set to not exceed 3%. The weight coefficients of each item in the model are determined through iterative optimization. Finally, an analysis model that can directly calculate rotor unbalance based on input parameters is formed. The model calculation response time is controlled within 100ms to ensure the real-time performance of online applications.

[0031] S5. The visual sequence balance error correction algorithm is used to compensate and correct the error in the unbalance analysis results, and to determine the specific location and value of the rotor unbalance.

[0032] Specifically, step S5 uses a visual sequence balance error correction algorithm to compensate and correct the imbalance analysis results. First, the imbalance calculation results corresponding to each frame in the visual sequence are obtained. The dispersion of the calculation results is statistically analyzed by calculating the standard deviation (the standard deviation threshold is set to 0.01 g·mm) to determine the initial error range. Then, the influence of factors such as ambient light interference (light intensity fluctuation range of 500 lux to 5000 lux), frame acquisition delay (delay time of 0.5 ms to 1 ms), and feature point matching deviation (matching error not exceeding 0.05 mm) on the calculation results is analyzed. The error contribution value of each factor is quantified by the control variable method, and a database including 10 to 20 core error influencing factors is established. The visual sequence balancing error correction algorithm is invoked, and error influence factors are substituted into the algorithm according to their weights. The weight allocation is determined based on the contribution value of each error factor (ambient light interference weight 0.3, frame acquisition delay weight 0.25, feature point matching deviation weight 0.2, etc.). The error correction amount corresponding to each frame is calculated. The error correction amount is applied to the corresponding imbalance amount calculation result according to the timestamp. The moving average method (window size 3 to 5 frames) is used to smooth the discrete calculation results to remove the influence of random errors. Finally, the specific location (circumferential angle accuracy 0.5°, axial position accuracy 0.1mm) and value (accuracy 0.001g・mm) of the imbalance amount are output to ensure that the accuracy of the imbalance amount data meets the requirements of subsequent correction.

[0033] S6, based on the unbalance data corrected in S5, performs targeted mass adjustment of the rotor through the online dynamic balance correction execution module of the motor rotor, and performs online dynamic balance correction of the rotor.

[0034] Specifically, step S6, based on the unbalance data corrected in step S5, performs a correction operation through the online dynamic balancing module for the motor rotor. This module includes a high-precision robotic arm (with repeatability accuracy ±0.01mm), a mass adjustment mechanism (supporting both grinding, drilling, and counterweight adjustment methods), and a real-time feedback unit. First, the specific location of the unbalance (circumferential angle, axial position) is converted into the motion coordinates of the robotic arm. The robotic arm moves to the designated correction position according to the coordinates, with the moving speed controlled between 5mm / s and 10mm / s to ensure accurate positioning. Then, the adjustment amount is determined based on the unbalance value (the adjustment amount matches the unbalance value in a 1:1 ratio, with an error not exceeding 2%). If grinding is used, the grinding speed is set to 30,000 rpm to 50,000 rpm, and the grinding depth to 0.01mm to 0.5mm. Excess mass is removed through successive grinding (each grinding depth not exceeding 0.1mm). If counterweight is used, a counterweight block with matching density (mass accuracy 0.001g) is selected and fixed to the designated position via laser welding or threaded connection. During the calibration process, the real-time feedback unit synchronously monitors the calibrated rotor vibration parameters through the visual acquisition module. 500 frames of data are collected for verification after each adjustment. When the vibration amplitude is less than 0.01mm and the phase offset is less than 1°, the calibration is deemed qualified. If it does not meet the standard, the above adjustment process is repeated until the dynamic balance requirements are met. The entire calibration process takes no more than 60 seconds, enabling efficient online calibration of rotor dynamic balance.

[0035] Preferably, the expression for the rotor visual feature coupled dynamics model is: ,in, The rotor visual characteristics are coupled with dynamic coefficients. The radial visual feature coordinates of the rotor. The axial visual feature coordinates of the rotor. The circumferential visual feature coordinates of the rotor. The rotor's angular velocity. The number of visual feature points, For the first Stiffness coupling coefficient of each feature point For the first A planar visual feature function for a feature point. For the first The circumferential vibration phase function of each feature point For the first Damping coupling coefficient at each characteristic point For the first The time derivative of the planar visual features of a feature point. For the first The time derivative of the circumferential vibration phase at each characteristic point.

[0036] Specifically, the rotor visual feature coupled dynamics model is used to achieve a deep correlation between rotor visual information and dynamic characteristics. During implementation, the specific value ranges and calibration methods of each model parameter are first defined. The number of visual feature points is set to 50 to 200 based on the rotor size and surface feature density. The stiffness coupling coefficient of each feature point is calculated using the rotor material's elastic modulus and the feature point distribution density, with a value ranging from 0.8 to 3.5. The damping coupling coefficient is determined based on the rotor's frictional characteristics and environmental damping during operation, with a value ranging from 0.01 to 0.15. The rotor's angular velocity is directly acquired from real-time motor operating data, covering a range of 3000 rpm to 15000 rpm. The radial, axial, and circumferential visual feature coordinates are converted into actual physical coordinates through pixel calibration by the high-speed visual acquisition module, achieving an accuracy of 0.001 mm. The planar visual feature function is constructed using the grayscale values, gradient changes, and spacing between adjacent feature points. The circumferential vibration phase function is determined by combining the rotor rotation period and the frame acquisition timestamp; the time derivatives of both are obtained through differential calculations of adjacent frame data. The model calculation adopts a sliding window summation method with a window size of 3 to 8 frames. The weighted summation is used to couple visual features with dynamic parameters. The final output rotor visual feature coupled dynamic coefficients can accurately reflect the correlation strength between rotor surface feature distribution and vibration response, providing a quantitative basis for subsequent vibration parameter extraction, ensuring the matching accuracy between feature information and dynamic characteristics, and improving the reliability of basic data for unbalance analysis.

[0037] Preferably, the expression for the visual inter-frame rotor vibration differential algorithm is: ,in, for Rotor vibration acceleration at all times for Rotor vibration displacement at all times for Rotor vibration displacement at all times for Rotor vibration displacement at all times The visual frame acquisition time interval. These are the inter-frame differential correction coefficients. This is the velocity correlation coefficient.

[0038] Specifically, the visual inter-frame rotor vibration differential algorithm obtains high-precision vibration dynamic parameters through inter-frame data differential calculation. During implementation, the inter-frame time interval is first determined based on the frame rate of the high-speed visual acquisition module. When the frame rate is set to 500 frames / second to 1000 frames / second, the corresponding inter-frame time interval is 1ms to 2ms. The time interval is synchronized and calibrated using the system clock, with the error controlled within 0.1μs. The inter-frame differential correction coefficient is dynamically adjusted based on the degree of environmental interference and data noise level during visual acquisition, ranging from 0.9 to 1.1. The speed correlation coefficient is determined based on the correlation analysis between rotor operating speed and vibration characteristics, ranging from 0.75 to 0.95. Both are calibrated through multiple experiments and stored in the algorithm parameter library. During the calculation, vibration displacement data of the same feature point is first extracted from visual data of three adjacent frames. The displacement change is calculated by the displacement difference between consecutive frames, and then substituted into the algorithm for first-order and second-order difference calculations. The first-order difference is used to obtain instantaneous velocity parameters, and the second-order difference, combined with the square of the inter-frame time interval, yields vibration acceleration. A data smoothing mechanism is employed during the calculation to remove abnormal displacement data exceeding three standard deviations of the normal fluctuation range, ensuring the stability of the calculation results. This algorithm requires no additional sensors and can achieve accurate calculation of vibration velocity and acceleration using only visual frame data. The velocity calculation accuracy reaches 0.01 mm / s, and the acceleration accuracy reaches 0.1 mm / s², providing key dynamic parameter support for unbalance analysis and significantly improving the real-time and synchronous nature of parameter acquisition.

[0039] Preferably, the expression for the visual sequence balance error correction algorithm is: ,in, This is the corrected rotor imbalance. The initial unbalance value is calculated. For the number of frames in the visual sequence, For the first The phase error correction coefficient of the frame. For the first The phase deviation of the frame. For the first The unbalanced component corresponding to the frame, For the first Frame amplitude error correction coefficient, For the first Frame amplitude deviation, For the first The rate of change of frame imbalance over time.

[0040] Specifically, the visual sequence balance error correction algorithm provides targeted compensation for various errors in the unbalance calculation process. During implementation, the number of visual sequence frames is first determined, ranging from 5000 to 30000 frames based on the acquisition duration and frame rate, ensuring sufficient data samples to support error analysis. The phase error correction coefficient and amplitude error correction coefficient for the j-th frame are determined through a combination of offline calibration and online adaptive adjustment. The phase error correction coefficient ranges from 0.85 to 1.05, and the amplitude error correction coefficient ranges from 0.9 to 1.1. Both are dynamically updated based on factors such as ambient light intensity and frame acquisition delay, with an update cycle of 100 frames. The phase deviation of the j-th frame is calculated by the difference between the rotor vibration phase and the average phase of that frame, with an accuracy of 0.1°. The amplitude deviation is determined by the difference between the vibration amplitude of that frame and the reference amplitude, with an accuracy of 0.001 mm. The unbalance component corresponding to the j-th frame is calculated using the mass distribution parameters and vibration parameters of that frame, and its rate of change over time is obtained through differential calculation of the unbalance components in adjacent frames. During algorithm operation, the initial unbalance value is first calculated. Then, the error correction terms of each frame are accumulated through summation. The error correction terms are jointly determined by the phase error correction coefficient, phase deviation, unbalance components, and amplitude-related parameters, ultimately yielding the corrected rotor unbalance. This algorithm effectively compensates for errors caused by environmental interference, frame acquisition delay, and feature point matching deviation, controlling the unbalance calculation error to within 3%. It significantly improves the accuracy of unbalance location and value calculation, providing precise data for subsequent correction operations.

[0041] Preferably, the mass distribution calculation model of the rotor mass visual analysis balancing platform is as follows: ,in, For rotor mass distribution density, For rotor material density, The coordinates of the rotor's circumferential angle are... For the rotor axial coordinate, The effective volume of the rotor, For visual analysis area, Let be the material density distribution function at the feature point. The distance from the feature point to the rotor axis. This is the visual effective area function for feature points.

[0042] Specifically, the rotor mass visual analysis balancing platform's mass distribution calculation model is used to accurately quantify the rotor's mass distribution state. During implementation, rotor structural parameters are first imported. The effective rotor volume is calculated based on the rotor's actual dimensions, including all rotating components. The visual analysis area covers more than 95% of the rotor surface, focusing on the shaft head, cylindrical sections, and key mating surfaces. The distance from feature points to the rotor axis is obtained through geometric calculations of visual feature coordinates and rotor axis coordinates, with an accuracy of 0.001 mm. The effective visual area function of each feature point is determined based on the number of pixels and pixel calibration coefficients, with the effective area of ​​each feature point ranging from 0.01 mm² to 0.1 mm². The material density distribution function at the feature points is constructed by combining rotor material uniformity detection data with visual grayscale value distribution. For homogeneous materials, the density distribution function is 0.98 to 1.02 times the standard material density; for heterogeneous materials, local density adjustments are made based on visual feature differences. During model computation, the density, distance, and effective area parameters of all feature points within the visual analysis area are accumulated through integration. The integration step size is set to 0.01 mm based on the feature point density to ensure the accuracy of the integration results. This model directly calculates the rotor mass distribution density from visual data, achieving a calculation accuracy of 0.01 g / cm³. It can accurately identify areas of concentrated mass and areas of uneven density, providing core mass distribution parameters for the unbalance analysis model. This makes the unbalance calculation more closely reflect the actual mass state of the rotor, improving the reliability and accuracy of the analysis model.

[0043] Preferably, the calculation model for the adjustment amount of the online dynamic balance correction of the motor rotor is as follows: ,in, This is the rotor mass adjustment amount. The inscribed angle of the unbalanced quantity. This indicates the axial position of the unbalance. For correction factors, This is the corrected imbalance. The rotor's angular velocity. The distance from the point of imbalance to the axis. It represents the vibration phase angle.

[0044] Specifically, the online dynamic balancing adjustment calculation model for the motor rotor is used to determine the specific values ​​for rotor mass adjustment. During implementation, the corrected imbalance is first obtained. This data comes from the output of the visual sequence balancing error correction algorithm, with an accuracy of 0.001 g·mm. The correction coefficient is determined based on the rotor type, operating speed, and correction method. For grinding correction, the value ranges from 0.95 to 1.05, while for counterweight correction, it ranges from 0.98 to 1.02. The optimal value is obtained through fitting with a large amount of experimental data and stored in the system database. The circumferential angle and axial position of the imbalance are determined visually, with an accuracy of 0.5° for the circumferential angle and 0.1 mm for the axial position. The distance from the imbalance point to the axis is calculated using visual feature coordinates, with an accuracy of 0.001 mm. The rotor's angular velocity is collected from real-time motor operating data, ranging from 3000 rpm to 15000 rpm. The vibration phase angle is obtained through vibration parameter analysis, with an accuracy of 0.1°. During model calculation, the product of the square of the angular velocity and the distance from the imbalance point to the axis is first calculated. Then, the product of the correction coefficient and the corrected imbalance is divided by this result. Finally, the result is multiplied by the cosine of the difference between the vibration phase angle and the circumferential angle of the imbalance point to obtain the final rotor mass adjustment. The adjustment accuracy calculated by this model reaches 0.001g, which can accurately match the adjustment amount according to the actual imbalance state and operating parameters of the rotor, ensuring the pertinence of grinding or counterweight operations, avoiding over-adjustment or under-adjustment, so that the corrected rotor vibration amplitude is less than 0.01mm and the phase offset is less than 1°, meeting the requirements of high-precision dynamic balancing, while improving correction efficiency and shortening the correction cycle.

[0045] Preferably, step S3 includes the following sub-steps: S31, extracting the coordinates of rotor contour feature points corresponding to two adjacent frames from the sequence visual data, determining the position offset of each feature point in the two frames, and establishing a feature point displacement mapping relationship; S32, based on the displacement mapping relationship, calculating the displacement change of each feature point in the inter-frame time interval, filtering out effective vibration displacement data and removing abnormal interference points; S33, performing first-order differential processing on the effective vibration displacement data using differential operation to obtain the instantaneous velocity parameters of rotor vibration, and recording the corresponding time node information at the same time; S34, performing second-order differential operation on the instantaneous velocity parameters, and combining them with the visual frame acquisition frequency parameters to obtain the real-time vibration acceleration data of the rotor, forming a complete set of vibration dynamic parameters.

[0046] Specifically, step S3 achieves accurate extraction of rotor vibration dynamic parameters through four sub-steps. S31 first filters clear rotor contour feature points from adjacent frames in the sequence visual data. Pixel coordinate transformation and calibration techniques are used to determine the actual physical coordinates of each feature point, establishing a one-to-one displacement mapping relationship between the feature points and the two frames. The coordinate transformation accuracy reaches 0.001mm, ensuring the accuracy of the displacement data. S32, based on the established displacement mapping relationship, calculates the displacement change of each feature point within the inter-frame time interval (1ms to 2ms). A displacement change threshold of 0.005mm to 0.5mm is set, and abnormal interference points exceeding this range are removed, retaining valid vibration displacement data. Simultaneously, the proportion of valid feature points is recorded as not less than 90%, ensuring data validity. S33 uses first-order difference... The sub-algorithm performs calculations on the effective vibration displacement data and, combined with the inter-frame time interval, calculates the instantaneous vibration velocity parameters of the rotor. The velocity calculation results are retained to three decimal places, and the time node corresponding to each velocity data is recorded synchronously with a timestamp accuracy of 1μs, achieving precise matching between velocity parameters and time. Based on the instantaneous velocity parameters, S34 performs secondary processing using differential operations and, combined with the visual frame acquisition frequency (500 frames / second to 1000 frames / second), calculates the real-time vibration acceleration data of the rotor with an acceleration calculation accuracy of 0.1mm / s². Finally, the instantaneous velocity and vibration acceleration data are integrated to form a complete set of vibration dynamic parameters including three-dimensional information of time, velocity, and acceleration. This provides high-precision and highly synchronous dynamic foundation data for the subsequent construction of the unbalance analysis model, ensuring the reliability of the model analysis.

[0047] Preferably, step S4 includes the following sub-steps: S41, importing the mass distribution correlation parameters extracted in S2 through the rotor mass visual analysis balancing platform, including feature point density distribution, contour symmetry, and coordinates of the mass concentration area; S42, aligning the vibration acceleration and instantaneous velocity parameters obtained in S3 with the mass distribution correlation parameters to establish a parameter correspondence mapping table; S43, based on the correlation data in the mapping table, constructing an unbalance analysis model framework with uneven mass distribution as the independent variable and vibration parameters as the dependent variable; S44, substituting rotor structural dimensions and material property parameters to optimize and adjust the model framework, determining the coefficients in the model, and forming an analysis model that can directly calculate the unbalance.

[0048] Specifically, step S4 constructs a precise rotor mass imbalance analysis model through four sub-steps. S31 first starts the rotor mass visual analysis and balancing platform, importing the mass distribution correlation parameters extracted in step S2, including feature point density distribution data (standard deviation controlled between 0.01 and 0.05), contour symmetry deviation value (not exceeding 5%), and the coordinates (accuracy up to 0.01 mm) and area percentage of the mass concentration region. Simultaneously, it loads the corresponding structural dimension parameters (length, diameter, etc.) and material property parameters (density, elastic modulus, etc.) of the rotor, providing basic data for model construction. S32 precisely aligns the vibration acceleration (accuracy 0.1 mm / s²) and instantaneous velocity (accuracy 0.01 mm / s) parameters obtained in step S3 with the mass distribution correlation parameters according to the timestamp, establishing a parameter mapping table. Each record in the mapping table includes all correlation parameters at the same time node, with alignment error controlled. The response time is controlled within 1μs to ensure parameter consistency. S33, based on the associated data in the mapping table, uses uneven mass distribution as the core independent variable and vibration parameters as the dependent variable to build a nonlinear unbalance analysis model framework. The framework includes the correlation function between mass distribution and vibration response, and clarifies the initial range of weight allocation for each parameter. S34 substitutes rotor structural dimensions (length 50mm to 500mm, diameter 10mm to 100mm) and material properties (density 7.8g / cm³ to 2.7g / cm³, etc.) into the model framework, uses the least squares method for fitting optimization, sets the model fitting error threshold to no more than 3%, and determines the weight coefficients of each item in the model through 3 to 5 iterations. Finally, an analytical model that can directly input parameters to calculate the unbalance is formed, with the model response time controlled within 100ms to meet the requirements of online real-time analysis, ensuring the efficiency and accuracy of unbalance calculation.

[0049] Preferably, step S5 includes the following sub-steps: S51, obtaining the imbalance calculation results of each frame in the visual sequence, statistically analyzing the dispersion of the calculation results, and determining the initial error range; S52, analyzing the impact of ambient light interference and frame acquisition delay factors on the calculation results during the visual acquisition process, and establishing an error influence factor database; S53, calling the visual sequence balance error correction algorithm, substituting the error influence factors into the algorithm formula, and calculating the error correction amount corresponding to each frame; S54, applying the error correction amount to the corresponding imbalance calculation results, smoothing the discrete calculation results, and obtaining the accurate imbalance position and value.

[0050] Specifically, step S5 achieves precise correction of the imbalance analysis results through four sub-steps. S51 first collects the imbalance calculation results corresponding to all frames in the visual sequence, and uses statistical analysis methods to calculate the standard deviation of the results (setting a threshold of 0.01 g·mm). The initial error range is determined based on the standard deviation. When the standard deviation exceeds the threshold, the error analysis dimensions are expanded to ensure the comprehensiveness of the error range determination. S52 systematically analyzes various influencing factors in the visual acquisition process, including ambient light interference (light intensity fluctuation range of 500 lux to 5000 lux), frame acquisition delay (0.5 ms to 1 ms), feature point matching deviation (not exceeding 0.05 mm), etc. The contribution value of each factor to the calculation results is quantified using the controlled variable method, establishing a database including 10 to 20 core error influencing factors. Each factor corresponds to a specific error contribution ratio. S53 calls the visual sequence balance error correction algorithm, assigns the weight of each factor according to the contribution ratio in the error influence factor database, substitutes the factor data into the algorithm for calculation, and calculates the error correction amount corresponding to each frame. The correction amount accuracy reaches 0.0001 g·mm, ensuring the precision of the correction. S54 applies the error correction amount of each frame precisely to the corresponding imbalance amount calculation result according to the timestamp, and uses the moving average method (window size of 3 to 5 frames) to smooth the corrected discrete calculation result, remove random error interference, and finally outputs the specific location (circumferential angle accuracy 0.5°, axial position accuracy 0.1 mm) and value (accuracy 0.001 g·mm) of the imbalance amount, providing accurate target data for subsequent online correction and ensuring the pertinence and effectiveness of the correction operation.

[0051] like Figure 2As shown, an online dynamic balancing method for motor rotors based on machine vision is implemented through different units, including: a high-speed vision multi-dimensional data acquisition unit, positioned opposite the motor rotor, for continuously acquiring visual data related to surface features, contour edges, and vibration displacement during rotor operation, and transmitting the data to a feature coupling analysis unit; a rotor visual feature coupling dynamic analysis unit, communicatively connected to the high-speed vision multi-dimensional data acquisition unit, for coupling the feature information in the visual data through a preset model to extract parameters related to vibration amplitude, phase shift, and mass distribution; and an inter-frame vibration differential calculation unit, connected to the rotor visual feature coupling dynamic analysis unit. The system comprises four main components: a meta-connection unit, which performs differential calculations on vibration displacement data from adjacent frames and outputs instantaneous velocity and vibration acceleration parameters; a mass imbalance analysis unit, which communicates with the rotor visual feature coupled dynamic analysis unit and the inter-frame vibration differential calculation unit to integrate various parameters and construct an analysis model to obtain initial imbalance data; a visual sequence error correction unit, which connects with the mass imbalance analysis unit and uses a dedicated algorithm to compensate for and correct errors in the initial imbalance data, outputting accurate imbalance information; and an online correction execution unit, which communicates with the visual sequence error correction unit and performs mass adjustment of the motor rotor based on the accurate imbalance information, thus performing online rotor dynamic balance correction.

[0052] A machine vision-based online dynamic balancing correction method for motor rotors overcomes the limitations of traditional single-sensor detection by leveraging high-speed visual acquisition. It simultaneously acquires multi-dimensional data on rotor surface features, contours, and vibration displacement. By coupling rotor visual features with a dynamic model, it achieves deep fusion of visual information and dynamic parameters, accurately establishing the correlation between mass distribution and vibration response. This effectively solves the problem of insufficient unbalance positioning accuracy caused by the lack of multi-parameter coupling analysis in traditional methods. Furthermore, it accurately extracts key parameters such as vibration velocity and acceleration using a visual inter-frame vibration differential algorithm. Combined with a rotor mass visual analysis balancing platform, it constructs a dedicated unbalance analysis model. Finally, a visual sequence balancing error correction algorithm is used to specifically compensate for errors caused by dynamic factors such as environmental interference and inter-frame delay. This thoroughly improves upon the shortcomings of existing technologies, such as simple error correction mechanisms and weak anti-interference capabilities, significantly enhancing the accuracy of unbalance calculation.

[0053] This method operates entirely online, completing a closed-loop process from data acquisition, analysis, correction to calibration without disassembling the rotor. It perfectly overcomes the shortcomings of traditional offline calibration, such as long cycles, low efficiency, and inability to respond to dynamic balance changes during rotor operation. At the same time, each functional unit has a clear division of labor and works in concert. From multi-dimensional data acquisition to precise parameter extraction, and then to efficient error correction and targeted quality adjustment, a complete intelligent calibration system is formed. This system not only ensures the real-time performance and reliability of the calibration process, but also meets the stringent requirements of high-speed and precision motors for rotor dynamic balance, providing strong technical support for the long-term stable and efficient operation of the motor.

[0054] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0055] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A machine vision-based online dynamic balancing method for motor rotors, characterized in that, Includes the following steps: S1, continuously acquires visual frames of the motor rotor in operation through the high-speed visual acquisition module, and obtains sequential visual data including rotor surface feature points, contour edges and vibration displacement information; S2, call the rotor visual feature coupled dynamics model to perform coupled analysis on the feature information in the sequence visual data, and extract the rotor vibration amplitude, phase shift and mass distribution related parameters; S3, the rotor vibration differential algorithm between visual frames is used to perform differential calculation on the rotor vibration displacement data in adjacent visual frames to obtain the real-time vibration acceleration and instantaneous velocity parameters of the rotor. S4. Based on the rotor mass visual analysis balancing platform, and combining the mass distribution correlation parameters extracted in S2 with the vibration dynamic parameters obtained in S3, a rotor mass imbalance analysis model is constructed. S5. The visual sequence balance error correction algorithm is used to compensate and correct the error in the unbalance analysis results, and to determine the specific location and value of the rotor unbalance. S6, based on the unbalance data corrected in S5, performs targeted mass adjustment of the rotor through the online dynamic balance correction execution module of the motor rotor, and performs online dynamic balance correction of the rotor.

2. The online dynamic balancing method for motor rotors based on machine vision according to claim 1, characterized in that, The expression for the rotor visual feature coupled dynamics model is: ， in, The rotor visual characteristics are coupled with dynamic coefficients. The radial visual feature coordinates of the rotor. The axial visual feature coordinates of the rotor. The circumferential visual feature coordinates of the rotor. The rotor's angular velocity. The number of visual feature points, For the first Stiffness coupling coefficient of each feature point For the first A planar visual feature function for a feature point. For the first The circumferential vibration phase function of each feature point For the first Damping coupling coefficient at each characteristic point For the first The time derivative of the planar visual features of a feature point. For the first The time derivative of the circumferential vibration phase at each characteristic point.

3. The online dynamic balancing method for motor rotors based on machine vision according to claim 1, characterized in that, The expression for the visual inter-frame rotor vibration differential algorithm is as follows: ， in, for Rotor vibration acceleration at all times for Rotor vibration displacement at all times for Rotor vibration displacement at all times for Rotor vibration displacement at all times The visual frame acquisition time interval. These are the inter-frame differential correction coefficients. This is the velocity correlation coefficient.

4. The online dynamic balancing method for motor rotors based on machine vision according to claim 1, characterized in that, The expression for the visual sequence balance error correction algorithm is as follows: ， in, This is the corrected rotor imbalance. The initial unbalance value is calculated. For the number of frames in the visual sequence, For the first The phase error correction coefficient of the frame. For the first The phase deviation of the frame. For the first The unbalanced component corresponding to the frame, For the first Frame amplitude error correction coefficient, For the first Frame amplitude deviation, For the first The rate of change of frame imbalance over time.

5. The online dynamic balancing method for motor rotors based on machine vision according to claim 1, characterized in that, The mass distribution calculation model of the rotor mass visual analysis balancing platform is as follows: ， in, For rotor mass distribution density, For rotor material density, The coordinates of the rotor's circumferential angle are... For the rotor axial coordinate, The effective volume of the rotor, For visual analysis area, Let be the material density distribution function at the feature point. The distance from the feature point to the rotor axis. This is the visual effective area function for feature points.

6. The online dynamic balancing method for motor rotors based on machine vision according to claim 1, characterized in that, The calculation model for the adjustment amount of the online dynamic balance correction of the motor rotor is as follows: ， in, This is the rotor mass adjustment amount. The inscribed angle of the unbalanced quantity. This indicates the axial position of the unbalance. For correction factors, This is the corrected imbalance. The rotor's angular velocity. The distance from the point of imbalance to the axis. It represents the vibration phase angle.

7. The online dynamic balancing method for motor rotors based on machine vision according to claim 1, characterized in that, S3 includes the following steps: S31, extract the rotor contour feature point coordinates corresponding to two adjacent frames from the sequence visual data, determine the position offset of each feature point in the two frames, and establish the feature point displacement mapping relationship. S32, based on the displacement mapping relationship, calculate the displacement change of each feature point in the inter-frame time interval, filter out the effective vibration displacement data and remove abnormal interference points. S33, the effective vibration displacement data is processed by first-order differential operation to obtain the instantaneous velocity parameters of rotor vibration, and the corresponding time node information is recorded at the same time. S34 performs a second differential operation on the instantaneous velocity parameters and combines it with the frequency parameters acquired by the visual frame to obtain the real-time vibration acceleration data of the rotor, forming a complete set of vibration dynamic parameters.

8. The online dynamic balancing method for motor rotors based on machine vision according to claim 1, characterized in that, S4 includes the following steps: S41, import the mass distribution correlation parameters extracted in S2 through the rotor mass visual analysis balancing platform, including feature point density distribution, contour symmetry and mass concentration area coordinates; S42, Align the vibration acceleration and instantaneous velocity parameters obtained in S3 with the mass distribution correlation parameters and establish a parameter mapping table; S43. Based on the associated data in the mapping table, construct an unbalance quantity analysis model framework with uneven mass distribution as the independent variable and vibration parameters as the dependent variable. S44, substitute the rotor structure dimensions and material property parameters to optimize and adjust the model framework, determine the coefficients in the model, and form an analytical model that can directly calculate the unbalance.

9. The online dynamic balancing method for motor rotors based on machine vision according to claim 1, characterized in that, S5 includes the following steps: S51, obtain the imbalance calculation results of each frame in the visual sequence, statistically analyze the dispersion of the calculation results, and determine the initial error range; S52, analyze the impact of ambient light interference and frame acquisition delay on the calculation results during the visual acquisition process, and establish an error impact factor database; S53, call the visual sequence balance error correction algorithm, substitute the error influence factor into the algorithm formula, and calculate the error correction amount for each frame; S54 applies the error correction amount to the corresponding unbalance calculation result, smooths the discrete calculation result, and obtains the accurate unbalance location and value.

10. The online dynamic balancing method for motor rotors based on machine vision according to claim 1, characterized in that, This method is implemented through different units, including: The high-speed vision multi-dimensional data acquisition unit is set opposite to the motor rotor and is used to continuously acquire visual data related to surface features, contour edges and vibration displacement during rotor operation, and transmit the data to the feature coupling analysis unit. The rotor visual feature coupled dynamic analysis unit is connected to the high-speed visual multi-dimensional data acquisition unit. It uses a preset model to couple the feature information in the visual data and extract the vibration amplitude, phase shift and mass distribution related parameters. The inter-frame vibration differential calculation unit is connected to the rotor visual feature coupled dynamic analysis unit to perform differential calculations on the vibration displacement data of adjacent frames and output instantaneous velocity and vibration acceleration parameters. The mass imbalance analysis unit communicates with the rotor visual feature coupled dynamic analysis unit and the inter-frame vibration differential operation unit respectively, integrates various parameters to construct an analysis model, and obtains the initial imbalance data. The visual sequence error correction unit is connected to the mass imbalance analysis unit. It uses a dedicated algorithm to compensate and correct the initial imbalance data and outputs accurate imbalance information. The online calibration execution unit communicates with the visual sequence error correction unit to adjust the mass of the motor rotor based on the accurate imbalance information, and performs online dynamic balance calibration of the rotor.

Images