A machine vision-based industrial robot motor dynamic performance testing device and method

Abstract

The application discloses a kind of industrial robot motor dynamic performance testing device and method based on machine vision, the device includes: machine vision module, for obtaining the image information of the end of the motor to be measured, and the structural features of the end are identified by image processing, to generate the three-dimensional coordinates of measuring point;Mechanical arm module, for receiving the three-dimensional coordinates, and driving the end of the actuator to move to target measuring point position and carry out dynamic excitation;Self-adaptive adjustment module, for receiving the position information fed back by the machine vision module and the force information fed back by the actuator in real time, and compared with the preset target value, when deviation is out of limit, generate adjustment instruction, dynamically correct the motion trajectory of the mechanical arm module and / or the output parameter of the actuator;Data acquisition and analysis module, for synchronously collecting excitation and response data in testing process, and based on multi-source data fusion analysis, generate dynamic characteristic evaluation report.

Description

Technical Field

[0001] This invention relates to the field of motor product technology, specifically to a machine vision-based device and method for testing the dynamic performance of industrial robot motors. Background Technology

[0002] The stator winding ends of large steam turbine generators are subjected to a 100Hz second harmonic electromagnetic force during operation. If their overall natural frequency is close to 100Hz (especially for elliptical mode shapes), resonance can easily occur, leading to structural loosening, insulation wear, or even breakage. Traditional manual testing relies on manually arranging test points (e.g., 32 test points for a 4-pole motor) and striking them with a hammer. This method has the following problems: low efficiency: a single test takes 3-5 hours, manual operation is highly repetitive, and the position of the reference object needs to be repeatedly adjusted; poor accuracy: the deviation of the striking force and position (error of ±2mm or more) leads to distortion of the frequency response function curve and large damping ratio error; experience dependence: mode shape judgment (elliptical, four-lobed) depends on human experience, which easily introduces subjective errors; insufficient compatibility: the end structure of large-capacity units is complex (e.g., lead wires, nose connectors), and traditional methods are difficult to use for multi-dimensional dynamic characteristic testing.

[0003] Current standards (such as DL / T 735-2000 and GB / T 20140-2016) clearly require the assessment of end-structure safety through dynamic characteristic testing, but they do not impose mandatory specifications for automated testing, leading most companies to still use traditional manual methods. Although the standards emphasize the necessity of testing, manual methods are difficult to meet the high-precision and high-efficiency testing requirements of large-capacity units. Summary of the Invention

[0004] The purpose of this invention is to provide a machine vision-based device and method for testing the dynamic performance of industrial robot motors, in order to solve at least one related technical problem in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, this application provides a machine vision-based dynamic performance testing device for industrial robot motors, comprising:

[0007] The machine vision module is used to acquire image information of the end of the motor under test, and to identify the structural features of the end through image processing in order to generate the three-dimensional coordinates of the measurement point;

[0008] The robotic arm module is used to receive the three-dimensional coordinates and drive the end effector to move to the target measurement point position for dynamic excitation.

[0009] An adaptive adjustment module is used to receive position information fed back by the machine vision module and force information fed back by the actuator in real time, and compare them with preset target values. When the deviation exceeds the limit, an adjustment command is generated to dynamically correct the motion trajectory of the robotic arm module and / or the output parameters of the actuator.

[0010] The data acquisition and analysis module is used to synchronously collect stimulus and response data during the testing process, and generate dynamic characteristic evaluation reports based on multi-source data fusion analysis.

[0011] The machine vision module, robotic arm module, adaptive adjustment module, and data acquisition and analysis module constitute a closed-loop control system to realize automated testing and intelligent diagnosis of the testing device.

[0012] In order to pursue full automation while retaining necessary human intervention to cope with complex and ever-changing on-site situations, and to improve the robustness and practicality of the system, in one optional implementation, the machine vision module includes an industrial camera and an image processing unit.

[0013] The image processing unit identifies the winding nose and / or binding ring contour of the end through an image feature matching algorithm, and calculates the three-dimensional coordinates of the measuring point based on a stereo vision algorithm, with a positioning accuracy within ±1mm; the image processing unit also provides a human-computer interaction interface, which supports the operator to manually adjust or confirm the automatically generated measuring point coordinates.

[0014] To address the technical problems of image quality being affected by lighting, reflection, and shadow in industrial settings, and the inability of pixel-level positioning accuracy to meet the high-precision operation requirements of robotic arms, in one optional embodiment, the image processing unit is further configured to:

[0015] Before generating the three-dimensional coordinates, the acquired image is preprocessed, including grayscale conversion, image enhancement, and Gaussian filtering; a shape-based template matching algorithm and a sub-pixel edge detection algorithm are used to extract and accurately locate the structural features.

[0016] To address the fundamental technical problem that machine vision systems and robot motion systems cannot work together due to incompatible coordinate systems, in one optional implementation, the machine vision module establishes a transformation relationship between the camera coordinate system and the robotic arm coordinate system through hand-eye calibration.

[0017] The hand-eye calibration converts the pixel coordinates of the identified feature points into motion coordinates that the robotic arm module can recognize by solving the affine transformation matrix.

[0018] To address the technical problem of inconsistent and unquantifiable striking force in traditional hammers, leading to poor repeatability of test data, in one optional embodiment, the end effector of the robotic arm module is a striking device, which includes a servo motor and a force sensor; the servo motor drives the striking device and provides an adjustable striking force ranging from 0.5N to 50N; the force sensor is a piezoelectric force sensor, used to detect the striking force in real time and feed it back to the adaptive adjustment module.

[0019] To address the technical problem of large fluctuations in striking force and slow response due to factors such as mechanical transmission clearance and load changes, which prevent the force from quickly stabilizing at a preset value, in one optional implementation, the adaptive adjustment module performs closed-loop control of the servo motor using a PID control algorithm.

[0020] The PID control algorithm dynamically adjusts the output torque of the servo motor based on the deviation between the force feedback from the force sensor and the preset force, so that the striking force is kept stable within the preset range; the adaptive adjustment module has a response time of less than 0.01 seconds to the position and force deviation.

[0021] To address the technical problem of cumulative position deviation after long-term operation caused by factors such as absolute positioning error of the robotic arm, temperature drift, or workpiece deformation, in one optional implementation, the real-time feedback and correction function of the adaptive adjustment module includes:

[0022] The machine vision module performs secondary sampling and verification of the current position of the robotic arm module at predetermined time intervals;

[0023] If the position deviation exceeds ±1mm, the coordinate fine-tuning algorithm is triggered to generate a correction command to correct the motion trajectory of the robotic arm module.

[0024] To address the technical problem that unexpected interference during testing (such as personnel accidentally entering the testing area or momentary sensor malfunctions) may lead to equipment damage or invalidation of test data, in one optional implementation, the adaptive adjustment module further includes an exception handling mechanism for:

[0025] The received data is filtered, and abnormal data is automatically identified and marked. The abnormal data includes at least one of the following: coordinates that are outside the range of motion of the robotic arm, sudden changes in force that are outside the preset range, communication interruption, or data loss.

[0026] When the abnormal data is detected, the testing process is paused and awaits manual confirmation.

[0027] This mechanism operates as a software logic within the adaptive adjustment module. It sets safety thresholds for all input data (coordinates, force). If data exceeds limits or a communication anomaly is detected, the current task is immediately paused, an alarm message pops up on the host computer interface, and a log is recorded. The system awaits operator intervention for verification before deciding whether to continue, retry, or terminate the test.

[0028] This invention incorporates an anomaly handling mechanism into the adaptive adjustment module. This mechanism proactively identifies risks and pauses operations, preventing the robotic arm from colliding or performing invalid tests under erroneous data guidance, thus improving the safety and reliability of the entire system.

[0029] To address the potential blind spots or errors in single-sensor data, and the technical problems of test results relying on manual interpretation and inconsistent standards, in one optional implementation, the data acquisition and analysis module is further used for:

[0030] Vibration image data analyzed by integrating digital image correlation method and vibration signals collected by sensors are fused into multimodal data to enhance the reliability of mode shape identification;

[0031] Based on the frequency response function, natural frequency, damping ratio, and mode shape analysis results, an end-effector dynamic characteristic evaluation model is constructed, and a test report containing resonance risk warning is automatically generated.

[0032] Secondly, this application provides a machine vision-based method for testing the dynamic performance of a motor, employing any of the testing devices described in the previous application. The method includes the following steps:

[0033] Visual positioning steps: The machine vision module acquires an image of the motor end, identifies structural features, and generates the three-dimensional coordinates of the measurement points;

[0034] Coordinate transformation and command generation steps: The three-dimensional coordinates are converted into robotic arm motion coordinates through hand-eye calibration, and excitation parameters are matched to generate control commands;

[0035] Automated test execution steps: Control the robotic arm module to move to the target test point, and its end effector will provide dynamic excitation according to preset parameters;

[0036] Closed-loop control steps: Real-time acquisition of position and force feedback data during execution, and dynamic correction of the robotic arm's motion trajectory and / or excitation parameters through PID control and visual verification;

[0037] Intelligent diagnostic steps: Simultaneously collect excitation and vibration response data, combine multimodal data fusion analysis, and automatically generate dynamic characteristic assessment results and risk warnings.

[0038] The beneficial effects that the machine vision-based industrial robot motor dynamic performance testing device and method disclosed in this application may bring include, but are not limited to:

[0039] 1. Improved testing efficiency

[0040] By automatically identifying end structures, dynamically planning and generating test points through machine vision, the tedious process of manual point-by-point measurement and marking has been replaced. The time for a single test has been significantly reduced from the traditional 3-5 hours to less than 40 minutes, improving efficiency by more than 400%, which greatly meets the urgent need for efficient testing at generator set manufacturing and maintenance sites.

[0041] 2. Guarantee of test accuracy and consistency

[0042] High positioning accuracy: Visual positioning and hand-eye calibration technology ensure that the positioning accuracy of the measuring point reaches ±1mm, which is much higher than the error of ±2mm or more when manually tapping, thus guaranteeing the accuracy of the test from the source.

[0043] Excellent excitation consistency: The use of servo motor and PID closed-loop force control ensures that the force of each strike is stable at the preset value (adjustable from 0.5N to 50N, control accuracy ±0.5N), completely eliminating the force fluctuation caused by fatigue and inconsistent techniques in manual striking. This results in a highly consistent excitation signal, laying the foundation for obtaining real and reliable frequency response function data. The distortion rate of the frequency response function curve is <2%.

[0044] 3. Enhanced system intelligence and automation level

[0045] It achieves full-process automation: from "seeing" to "hitting" and then to "analyzing", the entire process requires no manual intervention, reducing the technical threshold and labor intensity for operators.

[0046] It has intelligent diagnostic capabilities: through multimodal data fusion (sensor data + DIC visual mode shape) and built-in evaluation model, the system can automatically identify mode shape, judge resonance risk and generate report, transforming qualitative judgment that relies on the "experience" of experienced workers into objective and quantitative intelligent diagnosis based on data and models.

[0047] 4. Optimization of equipment compatibility and safety / reliability

[0048] Wide compatibility: Based on three-dimensional vision reconstruction technology, this device can adaptively identify generator ends with different pole numbers (2 poles, 4 poles), different capacities (supporting up to 1200MW units) and complex structures (such as those with lead wires and noses), making it highly versatile.

[0049] High security: The built-in anomaly handling mechanism can filter and mark abnormal data such as out-of-limit coordinates and sudden changes in force in real time, and automatically pause and wait for manual confirmation, effectively preventing safety accidents such as equipment collisions and ensuring the safety of the system and personnel.

[0050] High reliability: Sub-second dynamic correction capability (response <0.01s) ensures that the system can maintain high accuracy during long-term operation, with strong anti-interference ability and stable and reliable operation. Attached Figure Description

[0051] Figure 1 This is a system flowchart of the machine vision-based industrial robot motor dynamic performance testing device described in this invention. Detailed Implementation

[0052] Various exemplary embodiments, features, and aspects of this application will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0053] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, an indirect connection through an intermediate medium, or the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0054] like Figure 1 As shown, in a first aspect, this application provides a machine vision-based industrial robot motor dynamic performance testing device, comprising:

[0055] The machine vision module is used to acquire image information of the end of the motor under test, and to identify the structural features of the end through image processing in order to generate the three-dimensional coordinates of the measurement point;

[0056] The robotic arm module is used to receive the three-dimensional coordinates and drive the end effector to move to the target measurement point position for dynamic excitation.

[0057] An adaptive adjustment module is used to receive position information fed back by the machine vision module and force information fed back by the actuator in real time, and compare them with preset target values. When the deviation exceeds the limit, an adjustment command is generated to dynamically correct the motion trajectory of the robotic arm module and / or the output parameters of the actuator.

[0058] The data acquisition and analysis module is used to synchronously collect stimulus and response data during the testing process, and generate dynamic characteristic evaluation reports based on multi-source data fusion analysis.

[0059] The machine vision module, robotic arm module, adaptive adjustment module, and data acquisition and analysis module constitute a closed-loop control system to realize automated testing and intelligent diagnosis of the testing device.

[0060] The device consists of four core modules: a machine vision module, a robotic arm module, an adaptive adjustment module, and a data acquisition and analysis module. These modules are interconnected via electrical and communication lines. The output of the vision module is connected to the input of the adaptive adjustment module, and the output of the adaptive adjustment module controls the robotic arm module. The data acquisition module interacts with force sensors, vibration sensors, and the adaptive adjustment module to form a complete data loop.

[0061] The vision module first completes end-effector scanning and coordinate positioning; the positioning data is sent to the adaptive module, which plans instructions to drive the robotic arm to execute; the data acquisition module records data throughout the process; the adaptive module makes dynamic corrections based on real-time feedback; finally, the data acquisition and analysis module integrates and analyzes all the data to generate a report.

[0062] This invention integrates four functional modules—machine vision, robotic arm, adaptive adjustment, and data-driven intelligent diagnostics—into a cohesive whole. It solves the systemic technical problems of traditional testing, such as limited equipment functionality, broken data chains, and low levels of automation and intelligence. This architecture achieves seamless automation throughout the entire process from "identification-execution-monitoring-diagnosis," significantly improving testing efficiency and accuracy. Furthermore, it endows the system with intelligent analysis and decision-making capabilities, overcoming the pain points of traditional methods that rely on manual labor and cannot perform comprehensive diagnostics.

[0063] Optionally, communication between modules is not limited to wired connections; within permissible interference limits, industrial wireless networks can be used for data transmission. The functions of the data acquisition and analysis modules can be partially or fully deployed on edge computing gateways or cloud servers.

[0064] In order to pursue full automation while retaining necessary human intervention to cope with complex and ever-changing on-site situations, and to improve the robustness and practicality of the system, in one optional implementation, the machine vision module includes an industrial camera and an image processing unit.

[0065] The image processing unit identifies the winding nose and / or binding ring contour of the end through an image feature matching algorithm, and calculates the three-dimensional coordinates of the measuring point based on a stereo vision algorithm, with a positioning accuracy within ±1mm; the image processing unit also provides a human-computer interaction interface, which supports the operator to manually adjust or confirm the automatically generated measuring point coordinates.

[0066] Image acquisition quality directly affects the accuracy of visual positioning. This study uses a 5-megapixel industrial camera equipped with a ring LED light source to reduce glare interference from the winding surface. During image acquisition, the camera exposure time (set to 5ms) and gain value (set to 20dB) are adjusted to ensure uniform image brightness and clear details.

[0067] The industrial camera is fixed in place by a bracket and faces the test area. The image processing unit (host computer) is connected to the camera via a cable and runs the algorithm software. After generating the coordinates of the measurement points, they are displayed graphically in the software interface. Engineers can add, delete, and modify the automatically generated measurement points by clicking and dragging with the mouse.

[0068] This invention configures the image processing unit to provide a human-computer interaction interface. This ensures an effective combination of machine intelligence and human experience. When deviations are automatically identified in complex structures, engineers can quickly intervene to correct them, avoiding test interruptions or errors caused by algorithm limitations and ensuring the smooth progress of the test.

[0069] It should be noted that human-computer interaction is not limited to PC software interfaces, but can also be achieved through devices such as touch screens, handheld terminals (PADs), or even augmented reality (AR) glasses connected to a host computer.

[0070] To address the technical problems of image quality being affected by lighting, reflection, and shadow in industrial settings, and the inability of pixel-level positioning accuracy to meet the high-precision operation requirements of robotic arms, in one optional embodiment, the image processing unit is further configured to:

[0071] Before generating the three-dimensional coordinates, the acquired image is preprocessed, including grayscale conversion, image enhancement, and Gaussian filtering; a shape-based template matching algorithm and a sub-pixel edge detection algorithm are used to extract and accurately locate the structural features.

[0072] Preprocessing is an essential step before algorithmic processing and is performed sequentially or selectively within the image processing unit. After acquiring the original image, preprocessing is required to improve image quality. First, the color image is converted to grayscale to reduce processing complexity. Second, an image enhancement algorithm (emphasize operator) is used to improve contrast and highlight key features of the winding. Finally, Gaussian filtering is used to eliminate noise interference (gauss_filter operator), laying the foundation for subsequent feature extraction.

[0073] Feature extraction involves first performing rapid coarse localization on the preprocessed image using shape template matching to identify similar regions, and then using sub-pixel edge detection operators within those regions for precise localization down to the pixel level.

[0074] Specifically, the complex structure of motor windings makes feature extraction a crucial step. This study employs a shape-based template matching algorithm to create a winding template and search for similar regions in the target image to achieve initial winding localization. Then, a sub-pixel edge detection algorithm is used to precisely locate key points of the winding (such as coil inflection points and center points) to obtain pixel-level coordinates. The specific steps are as follows:

[0075] 1) Create a shape template: Select a standard winding image as a template, extract its contour features, and generate a shape model. The create_shape_model operator in Halcon can be used to create templates. Set the parameters to a rotation angle of -0.39 to 0.79 radians, a minimum contrast of 5, and the optimization algorithm to "auto".

[0076] (2) Template matching: Search for similar regions of the template in the target image to obtain the approximate position of the winding. The find_shape_model operator in Halcon is used to implement template matching. The minimum matching score is set to 0.8 and the number of matches is 1. The least squares method is used for optimization.

[0077] (3) Subpixel edge detection: Within the matching region, the Canny edge detection algorithm is used to extract the winding edges, fit a straight line or curve, and calculate the coordinates of key points. The edges_sub_pix operator in Halcon can realize subpixel edge detection. Set the filter to "canny", the Alpha parameter to 1, the low threshold to 20, and the high threshold to 40.

[0078] (4) Intersection point calculation: When multiple feature points need to be located, the intersection point is calculated using the intersection_lines operator, which serves as the key point of the winding. Through the above processing, the pixel coordinates of the key points of the winding can be obtained, providing a data basis for coordinate transformation.

[0079] This invention establishes an algorithmic flow for the image processing unit that includes preprocessing and two-level (coarse-fine) feature extraction. Preprocessing improves the algorithm's environmental adaptability, while sub-pixel technology breaks through the positioning accuracy from the integer pixel level to the sub-pixel level, laying the algorithmic foundation for achieving a physical positioning accuracy of ±1mm.

[0080] Feature extraction is not limited to shape matching; for features with significant texture, correlation-based template matching can be used. Subpixel edge detection is not limited to the Canny filter; other edge extraction operators such as Sobel and Laplacian can also be used.

[0081] To address the fundamental technical problem that machine vision systems and robot motion systems cannot work together due to incompatible coordinate systems, in one optional implementation, the machine vision module establishes a transformation relationship between the camera coordinate system and the robotic arm coordinate system through hand-eye calibration.

[0082] The hand-eye calibration converts the pixel coordinates of the identified feature points into motion coordinates that the robotic arm module can recognize by solving the affine transformation matrix.

[0083] Hand-eye calibration is a crucial step in system integration. It is a core component of a vision-guided system, aiming to establish the transformation relationship between the camera coordinate system and the robotic arm coordinate system. The transformation from one coordinate system XOY to another X'O'Y' can be viewed as a combination of transformations including translation, rotation, and scaling. This transformation relationship can be represented by the following matrix:

[0084]

[0085] in , respectively, represent the scaling scale in the X and Y directions; a and b represent the translation amounts in the X and Y directions, respectively; and , respectively, represent the rotation angle of the coordinate system.

[0086] The specific process is as follows: A calibration board (such as a chessboard) is fixed to the end effector of the robotic arm, and the arm is moved to multiple (usually no fewer than nine) different poses. At each pose, the camera captures an image of the calibration board, and the image processing unit extracts the corner pixel coordinates using an algorithm. Simultaneously, the physical coordinates of the end effector in the base coordinate system are read from the robotic arm controller. .

[0087] Subsequently, using these corresponding coordinate pairs, an affine transformation matrix HomMat2D is calculated through algorithms such as least squares (e.g., the vector_to_hom_mat2d operator in Halcon). This matrix contains rotation, translation, and scaling information. During operation, the system can calculate the target coordinates (MachineX, MachineY) in the robotic arm coordinate system in real time and accurately for any identified feature point pixel coordinates (Row, Column) by calling the affine_trans_point_2d operator. Specifically:

[0088]

[0089] By averaging multiple samples, the calibration error is controlled within ±1mm, ensuring conversion accuracy. After hand-eye calibration, the system can convert the pixel coordinates of winding feature points into motion coordinates recognizable by the robotic arm, providing a basis for generating tapping commands.

[0090] This invention associates the machine vision module with the robotic arm module through hand-eye calibration. This technology establishes a bridge from machine vision to the robotic arm, ensuring that the position seen by the vision system is the position executed by the robotic arm system.

[0091] Of course, hand-eye calibration is not limited to the nine-point method; more points can be used to improve accuracy. The calibration object is not limited to a checkerboard pattern; a circular array calibration board can also be used. The mathematical model is not limited to affine transformations; when camera lens distortion is significant, perspective transformations or more complex nonlinear models can be used for correction.

[0092] To address the technical problem of inconsistent and unquantifiable striking force in traditional hammers, leading to poor repeatability of test data, in one optional embodiment, the end effector of the robotic arm module is a striking device, which includes a servo motor and a force sensor; the servo motor drives the striking device and provides an adjustable striking force ranging from 0.5N to 50N; the force sensor is a piezoelectric force sensor, used to detect the striking force in real time and feed it back to the adaptive adjustment module.

[0093] The striking device is mounted on the end of a six-axis robotic arm via a flange. A servo motor receives control signals and drives a linear motion mechanism (such as a ball screw) to produce the striking action. A piezoelectric force sensor is installed at the point of contact between the striking rod and the target, transmitting real-time force signals back to the control system via a signal line.

[0094] This invention configures the end effector of a robotic arm as a closed-loop excitation unit consisting of a servo motor and a force sensor. The servo motor provides a precisely controllable power source, and the force sensor provides accurate feedback, thereby enabling precise and repeatable excitation of any force within the range of 0.5N to 50N.

[0095] The excitation method is not limited to impact; it can also be replaced by a vibrator for continuous frequency sweep excitation. The force sensor is not limited to piezoelectric type; strain gauge force sensors can also be used.

[0096] To address the technical problem of large fluctuations in striking force and slow response due to factors such as mechanical transmission clearance and load changes, which prevent the force from quickly stabilizing at a preset value, in one optional implementation, the adaptive adjustment module performs closed-loop control of the servo motor using a PID control algorithm.

[0097] The PID control algorithm dynamically adjusts the output torque of the servo motor based on the deviation between the force feedback from the force sensor and the preset force, so that the striking force is kept stable within the preset range; the adaptive adjustment module has a response time of less than 0.01 seconds to the position and force deviation.

[0098] The PID controller operates as a software algorithm within the adaptive control module. It receives feedback values ​​from the force sensor, compares them with a preset force value to obtain the deviation e(t), and then outputs an adjustment amount to the servo motor driver based on the calculation results of the proportional, integral, and derivative terms, thereby dynamically adjusting the motor's output torque. The response time of the entire "detection-calculation-output" cycle is controlled to within 10 milliseconds.

[0099] Specifically, when the robotic arm performs a striking motion, the piezoelectric force sensor at the end effector detects the striking force in real time and feeds the data back to the adaptive adjustment module built into the host computer. The adaptive adjustment module compares the detected force with a preset threshold. If the deviation exceeds ±0.5N, it triggers a PID control algorithm to adjust the output torque of the servo motor. The PID algorithm formula is as follows:

[0100]

[0101] in, This is the torque adjustment amount. Due to force deviation, These are the proportional, integral, and derivative coefficients, respectively. The system dynamically adjusts the PID parameters based on the force deviation to ensure the force remains stable within the preset range.

[0102] This invention configures the adaptive adjustment module to employ a PID control algorithm and ensures its rapid response. The PID algorithm guarantees precise and stable control, while the response speed of <0.01 seconds ensures rapid suppression of force deviations, preventing overshoot or oscillation. It is particularly suitable for high-frequency, continuous impact testing scenarios, ensuring the quality of transient excitation signals.

[0103] To address the technical problem of cumulative position deviation after long-term operation caused by factors such as absolute positioning error of the robotic arm, temperature drift, or workpiece deformation, in one optional implementation, the real-time feedback and correction function of the adaptive adjustment module includes:

[0104] The machine vision module performs secondary sampling and verification of the current position of the robotic arm module at predetermined time intervals;

[0105] If the position deviation exceeds ±1mm, the coordinate fine-tuning algorithm is triggered to generate a correction command to correct the motion trajectory of the robotic arm module.

[0106] During the movement of the robotic arm, the adaptive adjustment module periodically (e.g., every 100ms) triggers the vision module to take a quick picture and locate the target position. The actual visual position of the robotic arm's end effector is then compared with the target position. If the position exceeds the tolerance (±1mm), a tiny trajectory correction command is immediately generated and sent to the robotic arm controller, enabling it to correct its path before reaching the target.

[0107] The specific process is as follows:

[0108] (1) The vision module acquires the image of the current position and processes it with the Halcon algorithm to obtain the pixel coordinates of the current feature point;

[0109] (2) Convert pixel coordinates into robotic arm coordinates using the hand-eye calibration matrix, and calculate the offset from the target position;

[0110] (3) Generate correction instructions to control the robotic arm to fine-tune the trajectory and eliminate positional deviations.

[0111] This invention incorporates a visual position verification function into the adaptive adjustment module. This is equivalent to adding a real-time visual GPS to the robotic arm, enabling it to continuously correct its course and ensure accurate arrival at the target measurement point. This is a closed-loop compensation based on external sensors, which does not rely on the open-loop accuracy of the robotic arm itself, greatly improving the system's robustness in non-ideal environments.

[0112] To address the technical problem that unexpected interference during testing (such as personnel accidentally entering the testing area or momentary sensor malfunctions) may lead to equipment damage or invalidation of test data, in one optional implementation, the adaptive adjustment module further includes an exception handling mechanism for:

[0113] The received data is filtered, and abnormal data is automatically identified and marked. The abnormal data includes at least one of the following: coordinates that are outside the range of motion of the robotic arm, sudden changes in force that are outside the preset range, communication interruption, or data loss.

[0114] When the abnormal data is detected, the testing process is paused and awaits manual confirmation.

[0115] This mechanism operates as a software logic within the adaptive adjustment module. It sets safety thresholds for all input data (coordinates, force). If data exceeds limits or a communication anomaly is detected, the current task is immediately paused, an alarm message pops up on the host computer interface, and a log is recorded. The system awaits operator intervention for verification before deciding whether to continue, retry, or terminate the test.

[0116] For example, abnormal situations include:

[0117] (1) Invalid coordinates outside the range of motion of the robotic arm;

[0118] (2) Data on sudden changes in intensity (e.g., >50N);

[0119] (3) Communication interruption or data loss, etc.

[0120] This invention incorporates an anomaly handling mechanism into the adaptive adjustment module. This mechanism proactively identifies risks and pauses operations, preventing the robotic arm from colliding or performing invalid tests under erroneous data guidance, thus improving the safety and reliability of the entire system.

[0121] To address the potential blind spots or errors in single-sensor data, and the technical problems of test results relying on manual interpretation and inconsistent standards, in one optional implementation, the data acquisition and analysis module is further used for:

[0122] Vibration image data analyzed by integrating digital image correlation method and vibration signals collected by sensors are fused into multimodal data to enhance the reliability of mode shape identification;

[0123] Based on the frequency response function, natural frequency, damping ratio, and mode shape analysis results, an end-effector dynamic characteristic evaluation model is constructed, and a test report containing resonance risk warning is automatically generated.

[0124] This module processes two types of data simultaneously: first, sequential images acquired by a high-speed camera, which are used to calculate full-field displacement and strain using the DIC algorithm to visually display the mode shapes; second, accelerometer signals synchronously acquired by a data acquisition card, which are used to calculate the frequency response function, natural frequency, and damping ratio. Subsequently, it cross-validates and fuses the visual mode shapes with the modal results from sensor analysis, and automatically marks high-risk areas in the report based on standards (such as whether the natural frequency is close to 100Hz±Δf) and model rules.

[0125] Secondly, this application provides a machine vision-based method for testing the dynamic performance of a motor, employing any of the testing devices described in the previous application. The method includes the following steps:

[0126] Visual positioning steps: The machine vision module acquires an image of the motor end, identifies structural features, and generates the three-dimensional coordinates of the measurement points;

[0127] Coordinate transformation and command generation steps: The three-dimensional coordinates are converted into robotic arm motion coordinates through hand-eye calibration, and excitation parameters are matched to generate control commands;

[0128] Automated test execution steps: Control the robotic arm module to move to the target test point, and its end effector will provide dynamic excitation according to preset parameters;

[0129] Closed-loop control steps: Real-time acquisition of position and force feedback data during execution, and dynamic correction of the robotic arm's motion trajectory and / or excitation parameters through PID control and visual verification;

[0130] Intelligent diagnostic steps: Simultaneously collect excitation and vibration response data, combine multimodal data fusion analysis, and automatically generate dynamic characteristic assessment results and risk warnings.

[0131] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment takes the dynamic characteristic test of the stator winding end of a 1200MW steam turbine generator as the application scenario. The end diameter is about 3.5 meters, it is a 4-pole motor, and 32 measuring points need to be tested.

[0132] I. Device Configuration and Module Connection Relationship

[0133] The specific configurations of each module of the testing device in this embodiment are as follows:

[0134] Machine vision module:

[0135] Hardware: Employs a 5-megapixel industrial camera equipped with a ring-shaped LED soft light source to reduce glare. The camera connects to the host computer via a gigabit Ethernet cable.

[0136] Software and Algorithms: The host computer is an industrial computer running the Halcon 20.11 image processing library. Its built-in algorithms include: image grayscale conversion, `emphasize` for contrast enhancement, and `gauss_filter` for Gaussian filtering and other preprocessing operations; `create_shape_model` and `find_shape_model` for shape-based template matching (setting rotation angles of -0.39 to 0.79 radians, minimum contrast of 5, and minimum matching score of 0.8); and `edges_sub_pix` for subpixel edge detection (filter "canny", Alpha=1, low threshold of 20, high threshold of 40), etc.

[0137] robotic arm module:

[0138] Main body: A 6-axis collaborative robotic arm is selected, whose working radius meets the requirements for operation within a diameter of 3.5 meters.

[0139] End effector: A customized servo-driven electric striking device is mounted via a flange. Its core consists of a Panasonic 400W servo motor and its driver, which precisely controls the stroke (50mm) and speed of the striking rod, thereby achieving an adjustable striking force ranging from 0.5N to 50N. An integrated piezoelectric force sensor (accuracy ±0.1N) is used for real-time measurement of the striking force.

[0140] Adaptive adjustment module:

[0141] This module is integrated into the host computer in software form and communicates in real time with the robotic arm controller and force sensor data acquisition unit via EtherCAT bus.

[0142] The module runs a deviation comparison algorithm to compare the visual feedback position with the target position and the force sensor feedback force with the preset force in real time.

[0143] An integrated digital PID control algorithm is used for force closed-loop control. Its parameters, after tuning, are: proportional coefficient Kp = 2.5, integral coefficient Ki = 0.1, and derivative coefficient Kd = 0.05. The overall response time of this module from detecting the deviation to outputting the correction command is less than 10 milliseconds (0.01s).

[0144] Data acquisition and analysis module:

[0145] Hardware: A National Instruments PXIe-1082 chassis is used, with a built-in PXIe-4499 high-speed acquisition card, which synchronously acquires signals from the force sensor and eight PCB 333B32 accelerometers located at key end positions at a sampling rate of 1MHz.

[0146] Software and Algorithms: Diagnostic software developed based on MATLAB runs on the host computer. This software calls the Halcon library to perform digital image correlation (DIC) analysis, processing vibration sequence images acquired by a high-speed camera. Simultaneously, it employs the PolyMAX algorithm for modal parameter identification, calculating the frequency response function, natural frequencies, and damping ratios. It then fuses and compares the DIC mode shapes with the sensor modal analysis results, ultimately automatically generating an evaluation report.

[0147] II. System Working Process

[0148] Combination Figure 1 The system flowchart and the specific working process of this embodiment are as follows:

[0149] System initialization and hand-eye calibration:

[0150] Deploy the device in front of the generator end. Start the system and execute the "nine-point calibration method": control the robotic arm end effector to carry a 9×9 chessboard calibration board and move it to 9 different spatial poses. In each pose, the camera captures an image, Halcon's find_chessboard_corners operator extracts the corner pixel coordinates, and the physical coordinates of the robotic arm base are recorded simultaneously.

[0151] The affine transformation matrix HomMat2D was calculated using the vector_to_hom_mat2d operator, with the calibration error controlled within ±0.8mm. This matrix will be used for coordinate transformations in all subsequent work.

[0152] Visual positioning and measurement point planning:

[0153] The robotic arm was controlled to drive the camera to perform multi-view scanning and photography of the generator end, acquiring approximately 50 images covering the entire structure.

[0154] After the host computer Halcon program preprocesses the image, it executes feature matching and stereo vision algorithms to automatically identify 32 key feature points such as the winding nose and binding ring, and generates a list of measurement points containing X / Y / Z three-dimensional coordinates with a positioning accuracy of ±1mm.

[0155] Automated test execution and closed-loop control:

[0156] The host computer calls the affine_trans_point_2d operator to convert the pixel coordinates of all 32 measuring points into the motion coordinates of the robotic arm, and matches them with the preset tapping parameter library (in this embodiment, the tapping force is set to 20N) to generate a "position-force" linkage instruction sequence, which is then sent to the robotic arm controller via the EtherCAT bus.

[0157] The robotic arm begins to move. When it moves to the vicinity of the 5th measurement point, the adaptive adjustment module triggers a secondary visual verification. The camera takes pictures rapidly at 100ms intervals, and after processing, it is found that there is a 1.5mm deviation between the actual position of the robotic arm's end effector and the target position (>±1mm tolerance).

[0158] The adaptive module immediately generates trajectory correction instructions, and the robotic arm completes fine-tuning before reaching the target for precise positioning.

[0159] The servo motor drives the striking rod to perform the striking action, and the force sensor reports an actual force of 19.8N. The PID controller calculates a deviation of -0.2N (within a tolerance of ±0.5N), determining that the excitation is valid and no adjustment is needed.

[0160] During the execution of the 15th measurement point, the force sensor suddenly reported a peak value of 60N. The adaptive module's anomaly handling mechanism was immediately triggered, identifying it as a "sudden change in force." The system automatically paused the test, displayed a red alert on the host computer interface, and logged the event. After on-site inspection, the engineer found that the striking rod had momentarily struck a tiny protrusion. After eliminating the interference, the engineer clicked "Confirm and Continue" on the interface, and the system resumed the test from the current measurement point.

[0161] Intelligent diagnosis and report generation:

[0162] The entire test lasted approximately 35 minutes. During the test, the data acquisition system recorded the excitation force and acceleration response data for all 32 measurement points.

[0163] After the test is completed, the data analysis module will start automatically:

[0164] First, based on accelerometer data, the frequency response function at each measuring point was calculated. The PolyMAX algorithm was then used to fit the first few natural frequencies, damping ratios, and mode shapes of the entire end body. The results show that the first-order elliptical natural frequency is 99.8 Hz.

[0165] Meanwhile, the DIC analysis software processes the vibration video captured by the high-speed camera to generate a full-field vibration displacement cloud map, which intuitively displays the elliptical vibration mode and corroborates the sensor analysis results, enhancing the reliability of the conclusions.

[0166] Finally, the diagnostic software invoked its built-in evaluation model (based on standards such as DL / T 735-2000) to determine that the 99.8Hz frequency was too close to the 100Hz operating electromagnetic force frequency, classifying it as "high" risk. The system automatically generated a detailed test report, clearly indicating the risk of elliptical mode resonance and providing a warning that "structural reinforcement is recommended."

[0167] This embodiment fully demonstrates the workflow and technical effects of the device of the present invention. Through the coordinated operation of the four modules and the application of core technologies such as hand-eye calibration, PID force control, visual secondary verification, and multimodal data fusion, the present invention successfully achieves fully automatic, high-precision dynamic characteristic testing of the end of a large and complex generator within 35 minutes. Its inherent frequency detection error is ≤0.2Hz, and the positioning and force control accuracy meet the design requirements. It also automatically outputs an intelligent diagnostic report with clear risk warnings, comprehensively surpassing traditional manual methods.

[0168] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A machine vision-based dynamic performance testing device for industrial robot motors, characterized in that, include: The machine vision module is used to acquire image information of the end of the motor under test and to identify the structural features of the end through image processing to generate the three-dimensional coordinates of the measurement point. The machine vision module includes an industrial camera and an image processing unit. The image processing unit identifies the winding nose and / or binding ring contour of the end through an image feature matching algorithm and calculates the three-dimensional coordinates of the measurement point based on a stereo vision algorithm, with a positioning accuracy within ±1mm. The image processing unit also provides a human-machine interface to support the operator to manually adjust or confirm the automatically generated measurement point coordinates. The robotic arm module receives the three-dimensional coordinates and drives the end effector to move to the target measurement point for dynamic excitation. The end effector of the robotic arm module is a striking device, which includes a servo motor and a force sensor. The servo motor drives the striking device and provides an adjustable striking force ranging from 0.5N to 50N. The force sensor is a piezoelectric force sensor, which detects the striking force in real time and feeds it back to the adaptive adjustment module. An adaptive adjustment module is used to receive position information fed back by the machine vision module and force information fed back by the actuator in real time, and compare them with preset target values. When the deviation exceeds the limit, an adjustment command is generated to dynamically correct the motion trajectory of the robotic arm module and / or the output parameters of the actuator. The real-time feedback and correction function of the adaptive adjustment module includes: the machine vision module performing secondary sampling and verification of the current position of the robotic arm module at predetermined time intervals; if the position deviation exceeds ±1mm, a coordinate fine-tuning algorithm is triggered to generate a correction command to correct the motion trajectory of the robotic arm module. The data acquisition and analysis module is used to synchronously acquire excitation and response data during the test process, and generate a dynamic characteristic evaluation report based on multi-source data fusion analysis. The data acquisition and analysis module is also used to: integrate vibration image data analyzed by digital image correlation method with vibration signals acquired by sensors, perform multi-modal data fusion to enhance the reliability of mode shape identification; and construct an end dynamic characteristic evaluation model based on frequency response function, natural frequency, damping ratio and mode shape analysis results, and automatically generate a test report including resonance risk warning. The machine vision module, robotic arm module, adaptive adjustment module, and data acquisition and analysis module constitute a closed-loop control system to realize automated testing and intelligent diagnosis of the testing device.

2. The testing apparatus according to claim 1, characterized in that, The image processing unit is also used for: Before generating the three-dimensional coordinates, the acquired image is preprocessed, including grayscale conversion, image enhancement, and Gaussian filtering. The structural features are extracted and precisely located using a shape-based template matching algorithm and a sub-pixel edge detection algorithm.

3. The testing apparatus according to claim 1, characterized in that, The machine vision module establishes the transformation relationship between the camera coordinate system and the robotic arm coordinate system through hand-eye calibration; The hand-eye calibration converts the pixel coordinates of the identified feature points into motion coordinates that the robotic arm module can recognize by solving the affine transformation matrix.

4. The testing apparatus according to claim 1, characterized in that, The adaptive adjustment module performs closed-loop control of the servo motor using a PID control algorithm. The PID control algorithm dynamically adjusts the output torque of the servo motor based on the deviation between the force feedback from the force sensor and the preset force, so that the striking force is kept stable within the preset range. The adaptive adjustment module has a response time of less than 0.01 seconds to position and force deviations.

5. The testing apparatus according to claim 1, characterized in that, The adaptive adjustment module also includes an exception handling mechanism for: The received data is filtered, and abnormal data is automatically identified and marked. The abnormal data includes at least one of the following: coordinates that are outside the range of motion of the robotic arm, sudden changes in force that are outside the preset range, communication interruption, or data loss. When the abnormal data is detected, the testing process is paused and awaits manual confirmation.

6. A method for testing the dynamic performance of a motor based on machine vision, characterized in that, The method, employing the testing apparatus as described in any one of claims 1 to 5, comprises the following steps: Visual positioning steps: The machine vision module acquires an image of the motor end, identifies structural features, and generates the three-dimensional coordinates of the measurement points; Coordinate transformation and command generation steps: The three-dimensional coordinates are converted into robotic arm motion coordinates through hand-eye calibration, and excitation parameters are matched to generate control commands; Automated test execution steps: Control the robotic arm module to move to the target test point, and its end effector will provide dynamic excitation according to preset parameters; Closed-loop control steps: Real-time acquisition of position and force feedback data during execution, and dynamic correction of the robotic arm's motion trajectory and / or excitation parameters through PID control and visual verification; Intelligent diagnostic steps: Simultaneously collect excitation and vibration response data, combine multimodal data fusion analysis, and automatically generate dynamic characteristic assessment results and risk warnings.

Images