Intelligent pushing and positioning device and method for magnetic poles of wind driven generator based on machine vision

Abstract

The invention discloses a wind driven generator magnetic pole intelligent pushing and positioning device and method based on machine vision, the wind driven generator magnetic pole intelligent pushing and positioning device comprises a gantry assembly, a tool frame assembly and a positioner assembly, the tool frame assembly is distributed on the inner side of the gantry assembly, a camera is installed on one side of the tool frame assembly, and a lens is connected to the end of the camera; one side of the tool frame assembly is connected with a light source. According to the invention, a novel rotor magnetic pole automatic pushing system is introduced into a visual and neural network, so that visual positioning, magnetic pole pushing and efficiency can be improved and optimized more efficiently; a dynamic visual system is introduced into the novel automatic rotor magnetic pole pushing system, during pushing and positioning, a positioning area can be accurately recognized, a magnetic pole hoisting mechanical arm and a rotor rotating platform are adjusted in real time, and the pushing accuracy and efficiency are ensured; when different products are produced, the novel rotor magnetic pole automatic pushing system can perform real-time identification, adjust the posture of a station system in real time, and realize intelligent pushing after full adjustment.

Description

Technical Field

[0001] This invention relates to the field of intelligent magnetic pole pushing and positioning technology for semi-direct drive wind turbines, specifically to a machine vision-based intelligent magnetic pole pushing and positioning device and method for wind turbines. Background Technology

[0002] In recent years, with the rapid development of machine vision algorithms and wind turbine technology, vision has played a significant role in manufacturing production lines, efficiently, accurately, and flexibly completing area positioning tasks. Furthermore, by combining with neural network algorithms, the positioning of the work area can be adjusted in a timely manner as the vision system learns, making the placement of magnetic poles into slots more precise and efficient. This greatly improves production flexibility and enterprise competitiveness. To promote the construction and implementation of automated magnetic pole placement production lines for semi-direct drive rotors, the automated placement of magnetic poles for semi-direct drive generators requires the application of vision systems in conjunction with hoisting mechanisms.

[0003] The existing automatic rotor pole pushing and placing system consists of a hoisting robotic arm and a rotor rotating device. The number of rotation steps of the rotor rotating device is determined by a single visual teaching, without the introduction of any intelligent algorithms, and the accuracy cannot be optimized during the pushing and placing process. In the process of pushing and placing the poles, the automatic rotor pole pushing and placing system guides the rotor rotating platform to rotate based on visual images, resulting in a single outcome. Simple position teaching cannot guarantee more accurate and safe pushing and placing. For visual errors caused by disturbances, the automatic rotor pole pushing and placing system uses a secondary manual teaching, which contradicts the original purpose of intelligent production lines. Summary of the Invention

[0004] The purpose of this invention is to provide a machine vision-based intelligent push-and-place positioning device and method for wind turbine magnetic poles, in order to solve the problems mentioned in the background art. These problems include: the automatic push-and-place system for rotor magnetic poles consists of a hoisting robotic arm and a rotor rotating device; the rotor rotating device's rotation steps are determined by a single visual teaching method without any intelligent algorithm, making it impossible to optimize accuracy during the push-and-place process; the automatic push-and-place system for rotor magnetic poles relies on visual imaging to guide the rotor rotating platform's rotation as a single result, and simple position teaching cannot guarantee more accurate and safe push-and-place; and the automatic push-and-place system for rotor magnetic poles uses manual secondary teaching to address visual errors caused by disturbances, which contradicts the original purpose of intelligent production lines.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a machine vision-based intelligent magnetic pole pushing and positioning device for wind turbines, comprising a gantry assembly, a tooling frame assembly, and a positioner assembly. The tooling frame assembly is distributed inside the gantry assembly. A camera is mounted on one side of the tooling frame assembly, and a lens is connected to the end of the camera. A light source is connected to one side of the tooling frame assembly. A hoisting robotic arm is mounted inside the gantry assembly. The gantry assembly includes a gantry unit column, and a longitudinal beam moving assembly is connected to the top of the gantry unit column. A first crossbeam slide is slidably connected to the top of the longitudinal beam moving assembly, and a second crossbeam slide is slidably connected to the top of the first crossbeam slide. A connecting beam is connected to the end of the gantry unit column.

[0006] Preferably, a rotor rotation platform is connected to the outer side of the tooling frame assembly.

[0007] Preferably, the positioner assembly includes an L-shaped positioner fixture assembly, and a connecting beam assembly is connected to the middle of the L-shaped positioner fixture assembly.

[0008] A machine vision-based intelligent push-and-place method for wind turbine magnetic poles includes a vision system and a neural network system. The vision system consists of a camera, lens, light source, and vision software. The neural network system is electrically connected to the vision system.

[0009] Preferably, the specific steps of the intelligent push-and-place positioning method are as follows: Step 1: The hoisting system, consisting of the first crossbeam slide, the longitudinal beam moving assembly, the connecting beam, the second crossbeam slide, the gantry unit column, and the hoisting robotic arm, hoists the material to the top of the rotor rotating platform.

[0010] Step 2: First, install the camera, lens, and light source onto the tooling assembly in sequence, so that they work together with the vision software to form a vision system; Step 3: Start the rotor rotation platform and use the rotor rotation platform in conjunction with the vision system to take pictures of the rotating rotor surface. The vision system calculates and fits the motor offset of the hoisting system based on the photographed landing rotor surface, the measured distance and the area of ​​the landing surface. The rotation platform under the rotor rotates simultaneously according to the result to calculate the offset. The vision system takes pictures and calculates and adjusts continuously every millisecond until the offset is reduced to the required value. Step 4: Next, the neural network system is connected to the aforementioned vision system, allowing it to continuously annotate and train the acquired images. This is combined with the vision system and the hoisting system comprised of the first crossbeam slide, longitudinal beam moving assembly, connecting beam, second crossbeam slide, gantry unit column, and hoisting robotic arm, all optimized in parallel. The hoisting system's grippers move downwards to assemble the magnetic poles onto the rotor surface. N sets of magnetic poles are required for hoisting and assembly on the rotor surface. After each assembly, the neural network collects the gantry motor's movement data, the rotating platform's rotation data, and the visual capture image calculation results, continuously learning and matching OK data, and feeding it back to the next set of assembly data. When the data volume is sufficient, switching between different products does not require manual intervention at the teaching points, avoiding mechanical errors and improving the reliability and accuracy of the production process.

[0011] Preferably, the vision system uses a Hikvision camera, a white light source, and the OpenCV library and secondary development for image processing and reading / writing. It also uses a Modbus TCP board to communicate with the neural network system.

[0012] Preferably, the neural network uses a CNN architecture, and the neural network includes: Convolutional blocks: ability to fit basic features, target tilt, and partial occlusion; Downsampling convolutional blocks: Expand the receptive field. After downsampling at each layer, the receptive field is approximately twice that of the previous layer. Gradually capture the "overall outline of the target" from the "local edges", reduce the amount of spatial dimension computation, and save resources for high-level feature extraction. Residual blocks: Deep feature propagation, solving the gradient vanishing problem in deep networks, ensuring that detailed features extracted by low-level convolutions, such as localization markers on the target, can be propagated to higher levels, avoiding feature loss due to network deepening; The coordinate regression layer outputs multi-objective bias values.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention introduces vision and neural networks into a novel automatic rotor magnetic pole pushing and placing system, resulting in more efficient and optimized visual positioning, magnetic pole pushing and placing, and overall efficiency. The novel automatic rotor magnetic pole pushing and placing system incorporates a dynamic vision system, which can accurately identify the positioning area during pushing and placing, and adjust the magnetic pole lifting robotic arm and rotor rotation platform in real time to ensure the accuracy and efficiency of pushing and placing. Furthermore, the novel automatic rotor magnetic pole pushing and placing system can perform real-time identification and adjust the posture of the workstation system in real time during the production of different products, achieving intelligent pushing and placing after full adjustment. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall distribution in this invention; Figure 2 This is a schematic diagram of the gantry assembly structure in this invention; Figure 3 This is a schematic diagram of the positioner assembly structure in this invention; Figure 4 This is a schematic diagram of the tooling frame assembly structure in this invention; Figure 5 This is an enlarged schematic diagram of point A in this invention; Figure 6 This is a schematic diagram of the vision system in this invention; Figure 7 This is a schematic diagram of the neural network in this invention.

[0015] In the diagram: 1. Gantry assembly; 101. First crossbeam slide; 102. Longitudinal beam moving assembly; 103. Connecting beam; 104. Second crossbeam slide; 105. Gantry unit column; 2. Tooling frame assembly; 3. Camera; 4. Lens; 5. Light source; 6. Hoisting robotic arm; 7. Rotor rotating platform; 8. Positioner assembly; 801. L-shaped positioner fixture assembly; 802. Connecting beam assembly. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figures 1-7 This invention provides a technical solution: a machine vision-based intelligent push-and-place positioning device for wind turbine magnetic poles, comprising a gantry assembly 1, a tooling frame assembly 2, and a positioner assembly 8. The tooling frame assembly 2 is distributed inside the gantry assembly 1. A camera 3 is installed on one side of the tooling frame assembly 2, and a lens 4 is connected to the end of the camera 3. A light source 5 is connected to one side of the tooling frame assembly 2. A hoisting robotic arm 6 is installed inside the gantry assembly 1. The gantry assembly 1 includes a gantry unit column 105, and a longitudinal beam moving assembly 102 is connected to the top of the gantry unit column 105. A first crossbeam slide 101 is slidably connected to the top of the longitudinal beam moving assembly 102, and a second crossbeam slide 104 is slidably connected to the top of the first crossbeam slide 101. A connecting beam 103 is connected to the end of the gantry unit column 105.

[0018] The outer side of the tooling frame assembly 2 is connected to the rotor rotation platform 7, which consists of a rotor surface, a clamping and rotating structure, a chuck positioning assembly, etc., to achieve the purpose of rotating and driving the magnetic poles of the semi-direct drive wind turbine that are pushed and placed.

[0019] The positioner assembly 8 includes an L-shaped positioner fixture assembly 801, and a connecting beam assembly 802 is connected to the middle of the L-shaped positioner fixture assembly 801.

[0020] This invention relates to a machine vision-based intelligent magnetic pole pushing and positioning method for wind turbines. The method comprises a vision system and a neural network system. The vision system consists of a camera (3), a lens (4), a light source (5), and vision software. The neural network system is electrically connected to the vision system. The vision system uses a Hikvision camera and a white light source, employing the OpenCV library and secondary development for image processing and reading / writing. It communicates with the neural network system via a Modbus TCP board. The neural network uses a CNN architecture and includes: convolutional blocks for basic feature extraction, target tilting, and fitting of partial occlusion; downsampling convolutional blocks to expand the receptive field (approximately twice the size of the previous layer after downsampling), gradually capturing the overall target contour from "local edges," reducing spatial dimension computation and saving resources for higher-level feature extraction; residual blocks for deep feature propagation, addressing the gradient vanishing problem in deep networks, ensuring that detailed features extracted by lower-level convolutions, such as positioning markers on the target, are propagated to higher layers, avoiding feature loss due to network deepening; and a coordinate regression layer that outputs multi-target deviation values. Example

[0021] The specific steps of the intelligent push-and-place positioning method are as follows: Step 1: The hoisting system, consisting of the first crossbeam slide 101, the longitudinal beam moving assembly 102, the connecting beam 103, the second crossbeam slide 104, the gantry unit column 105, and the hoisting robotic arm 6, hoists the material to the top of the rotor rotating platform 7.

[0022] Step 2: First, install the camera 3, lens 4, and light source 5 sequentially on the tooling assembly 2 so that they work together with the vision software to form a vision system; Step 3: Start the rotor rotation platform 7. Use the rotor rotation platform 7 in conjunction with the vision system to take pictures of the rotating rotor surface. The vision system calculates and fits the motor offset of the hoisting system based on the photographed landing rotor surface, the measured distance and the area of ​​the landing surface. The rotating platform under the rotor rotates simultaneously according to the result to calculate the offset. The vision system takes pictures and calculates and adjusts continuously for 5ms until the offset is reduced to the required value. Step 4: The neural network system is then connected to the aforementioned vision system, allowing it to continuously annotate and train the captured images. This is combined with the vision system and the hoisting system comprised of the first crossbeam slide 101, longitudinal beam moving assembly 102, connecting beam 103, second crossbeam slide 104, gantry unit column 105, and hoisting robotic arm 6, for parallel optimization. The hoisting system's grippers move downwards to assemble the magnetic poles onto the rotor surface. N sets of magnetic poles need to be hoisted and assembled onto the rotor surface. After each assembly, the neural network collects the gantry motor's movement data, the rotating platform's rotation data, and the visual capture image calculation results, continuously learning and matching OK data, and feeding it back to the next set of assembly data. When the data volume is sufficient, switching between different products does not require manual intervention at the teaching points, avoiding mechanical errors and improving the reliability and accuracy of the production process.

[0023] Working principle: First, the operator installs camera 3, lens 4, and light source 5 sequentially on the tooling assembly 2, forming a vision system with the vision software. During rotor placement, the vision system, using the rotor rotation platform 7, photographs the rotating rotor surface. Based on the photographed rotor surface, the vision system calculates and fits the motor offset of the hoisting system according to the measured distance and area. The rotating platform below the rotor simultaneously rotates to measure the offset. The vision system continuously photographs and calculates adjustments every 5ms until the offset is reduced to the required value. Then, a neural network system is integrated into the vision system. This allows the system to continuously label and train the captured images, combining vision and the lifting robotic arm 6 for parallel optimization. The lifting system's grippers move downwards to assemble the magnetic poles onto the rotor surface. The rotor surface requires N sets of magnetic poles for lifting and assembly. After each assembly, the neural network collects the gantry motor's movement data, the rotating platform's rotation data, and the visual capture image calculation results, continuously learning and matching OK data, and feeding it back to the next set of assembly data. When the amount of data is sufficient, when switching between different products, there is no need for manual intervention at the teaching points, avoiding mechanical errors and improving the reliability and accuracy of the production process. Ultimately, the intelligent pushing and placing of magnetic poles for wind turbines is completed.

[0024] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various 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 intelligent magnetic pole pushing and positioning device for wind turbines, comprising: The gantry assembly (1), the tooling frame assembly (2), and the positioner assembly (8) are characterized in that: the tooling frame assembly (2) is distributed inside the gantry assembly (1), a camera (3) is installed on one side of the tooling frame assembly (2), and a lens (4) is connected to the end of the camera (3), a light source (5) is connected to one side of the tooling frame assembly (2), a hoisting robot arm (6) is installed inside the gantry assembly (1), the gantry assembly (1) includes a gantry unit column (105), and a longitudinal beam moving assembly (102) is connected to the top of the gantry unit column (105), a first crossbeam slide (101) is slidably connected to the top of the longitudinal beam moving assembly (102), and a second crossbeam slide (104) is slidably connected to the top of the first crossbeam slide (101), and a connecting beam (103) is connected to the end of the gantry unit column (105).

2. The intelligent push-and-place positioning device for wind turbine magnetic poles based on machine vision according to claim 1, characterized in that: The tooling frame assembly (2) is connected to a rotor rotating platform (7) on its outer side.

3. The intelligent push-and-place positioning device for wind turbine magnetic poles based on machine vision according to claim 1, characterized in that: The positioner assembly (8) includes an L-shaped positioner fixture assembly (801), and a connecting beam assembly (802) is connected to the middle of the L-shaped positioner fixture assembly (801).

4. A machine vision-based intelligent pushing and positioning method for magnetic poles of wind turbines, characterized in that: The intelligent push-and-place method includes a vision system and a neural network system. The vision system consists of a camera (3), a lens (4), a light source (5), and vision software. The neural network system is electrically connected to the vision system.

5. The intelligent pushing and positioning method for wind turbine magnetic poles based on machine vision according to claim 4, characterized in that: The specific steps of the intelligent push-and-place positioning method are as follows: Step 1: The hoisting system, consisting of the first crossbeam slide (101), the longitudinal beam moving assembly (102), the connecting beam (103), the second crossbeam slide (104), the gantry unit column (105), and the hoisting robot arm (6), hoists the material to the top of the rotor rotating platform (7). Step 2: First, install the camera (3), lens (4), and light source (5) on the tooling assembly (2) in sequence so that they can work together with the vision software to form a vision system; Step 3: Start the rotor rotating platform (7). Use the rotor rotating platform (7) in conjunction with the vision system to take pictures of the rotating rotor surface. The vision system calculates the offset of the hoisting system motor based on the measured distance and the area of ​​the landing rotor surface in real time. The rotating platform under the rotor rotates according to the result and the offset is adjusted by taking pictures continuously for 5ms until the offset is reduced to the required value. Step 4: Then connect the neural network system to the above-mentioned vision system, so that it can continuously label and train the captured images. Combine the vision system with the hoisting system consisting of the first crossbeam slide (101), the longitudinal beam moving assembly (102), the connecting beam (103), the second crossbeam slide (104), the gantry unit column (105), and the hoisting robot arm (6) to perform parallel optimization. The hoisting system grippers move downward to assemble the magnetic poles onto the rotor surface. The rotor surface requires N sets of magnetic poles for hoisting and assembly. After each assembly, the neural network collects the gantry motor movement data, the rotating platform rotation data, and the visual capture image calculation results, continuously learns and matches OK data, and feeds it back to the next set of assembly data. When the amount of data is sufficient, when switching between different products, there is no need for manual intervention in the teaching points, avoiding mechanical errors and improving the reliability and accuracy of the production process.

6. The intelligent push-and-place positioning method for wind turbine magnetic poles based on machine vision according to claim 4, characterized in that: The vision system uses a Hikvision camera and a white light source. It uses the OpenCV library and secondary development for image processing and reading / writing, and uses a Modbus TCP board to communicate with the neural network system.

7. The intelligent push-and-place positioning method for wind turbine magnetic poles based on machine vision according to claim 4, characterized in that: The neural network uses a CNN architecture, and the neural network includes: Convolutional blocks: ability to fit basic features, target tilt, and partial occlusion; Downsampling convolutional blocks: Expand the receptive field. After downsampling at each layer, the receptive field is about twice that of the previous layer. Gradually capture the "overall outline of the target" from the "local edge", reduce the amount of spatial dimension computation, and save resources for high-level feature extraction. Residual blocks: Deep feature propagation, solving the gradient vanishing problem in deep networks, ensuring that detailed features extracted by low-level convolutions, such as localization markers on the target, can be propagated to higher levels, avoiding feature loss due to network deepening; The coordinate regression layer outputs multi-objective bias values.

Images