阅读说明：本技术 一种同轴零件装配装置及方法 (coaxial part assembling device and method ) 是由 李子豪 严小军 惠宏超 何江涛 杨文超 潘忠诚 于 2019-08-15 设计创作，主要内容包括：本发明一种同轴零件装配装置及方法,装置由机械部分、图像采集部分、点胶部分、图像处理部分和控制部分,共五部分组成。机械部分包括大理石台、滑动导轨、料盘、三维滑台和真空吸嘴。图像采集部分包括相机、环形光源、远心镜头和滑台。点胶部分包括装有两种不同胶体的针管和点胶机组成。图像处理部分包括计算机和其中的图像处理程序。控制部分包括计算机和其中的控制程序以及各部件控制器。料盘上装有四种待装配的零件,分别是圆柱形零件、圆环形零件、圆形薄片和中空圆柱形零件。本装置具有非接触、高效、高精度和全自动化的特点。(The invention relates to a coaxial part assembling device and a method. The mechanical part comprises a marble table, a sliding guide rail, a material tray, a three-dimensional sliding table and a vacuum suction nozzle. The image acquisition part comprises a camera, an annular light source, a telecentric lens and a sliding table. The glue dispensing part comprises a needle tube filled with two different colloids and a glue dispenser. The image processing section includes a computer and an image processing program therein. The control section includes a computer and a control program therein, and each component controller. The charging tray is provided with four parts to be assembled, namely a cylindrical part, a circular ring part, a circular slice and a hollow cylindrical part. The device has the characteristics of non-contact, high efficiency, high precision and full automation.)

1. The utility model provides a coaxial part assembly quality, includes mechanical unit, two sets of the same image acquisition units, some glue units, image processing unit and the control unit, its characterized in that: the mechanical part comprises: the device comprises a marble table (1), a first sliding guide rail (2), a second sliding guide rail (3), a first material tray (4), a second material tray (5), a first three-dimensional sliding table (6), a first vacuum suction nozzle (7), a second vacuum suction nozzle (8), a guide rail controller (9) and a first three-dimensional sliding table controller (10); the image acquisition unit is used for shooting and acquiring images of a measured object, and the image processing unit (12) is used for processing the images acquired by the image acquisition unit; the dispensing unit moves the needle head to a specified position, and parts in the feeding disc (5) are dispensed; the control unit controls the sliding guide rail, the three-dimensional sliding table, the light source and the needle head;

The marble table (1) is used as a base of the whole system, so that the overall stability of the system is ensured; four parts, namely a circular sheet part (i), a cylindrical part (ii), a circular ring part (iii) and a hollow cylindrical part (iv), are respectively placed on the first tray (4) and the second tray (5); the lower parts of the first material tray (4) and the second material tray (5) are fixed on the sliding tables through positioning pins, and the two sliding tables are respectively placed on the first sliding guide rail (2) and the second sliding guide rail (3); the first sliding guide rail (2) and the second sliding guide rail (3) are controlled by a sliding guide rail controller (9) so that the first material tray (4) and the second material tray (5) reach designated positions; the three-dimensional sliding table controller (10) controls the first three-dimensional sliding table (6), and the first three-dimensional sliding table (6) drives the vacuum suction nozzle to move in any direction in a three-dimensional space, so that the first vacuum suction nozzle (7) and the second vacuum suction nozzle (8) reach the upper part of the part, and the part is sucked up and placed into a specified position.

2. a coaxial parts assembling apparatus according to claim 1, wherein: the image acquisition unit comprises: the device comprises a first camera (11), a second camera (15), a first telecentric lens (12), a second telecentric lens (16), a first annular light source (13), a second annular light source (17), a first sliding table (14), a second sliding table (18), a sliding table controller (24) and a light source controller (25);

the first camera (11) is connected with the first telecentric lens (12) and fixed on the sliding table (14), and the annular light source (13) is arranged between the camera lens connector and the tray part and is used for polishing the part to be detected; before the image is collected, the distance between a camera lens and a light source and the distance between the light source and a material tray part are adjusted, and after the optimal position is determined, the relative distance is kept unchanged; the first telecentric lens (12) is placed at a position which satisfies the minimum distortion of the image edge; the resolution of the first camera (11) is more than ten million pixels and is used for collecting images of all parts on the material tray (4); the sliding module (14) can move up and down, left and right, so that the camera reaches an accurate position; the sliding module (14) and the first camera (11) are controlled by a sliding table controller (24); the first annular light source (13) and the second annular light source (17) are connected with the first three-dimensional sliding table (6) and reach the designated positions; the light source controller (25) is used for controlling the brightness of the first annular light source.

3. A coaxial parts assembling apparatus according to claim 1, wherein: the dispensing unit comprises a dispensing machine (27), a first inclined 45-degree needle head (19), a first vertical needle head (20), a second inclined 45-degree needle head (21), a second vertical needle head (22) and a second three-dimensional sliding table (23); the four needle heads are moved to the parts in the feeding disc (5) at the appointed position through the second three-dimensional sliding table (23) for dispensing through the second three-dimensional sliding table (23); the glue dispenser (27) is used for controlling the needle head to dispense glue.

4. a coaxial parts assembling apparatus according to claim 1, wherein: the image processing unit comprises a computer (26) and an image processing algorithm unit; the internal work of the image processing algorithm unit is divided into three parts: and calibrating and detecting the circle center of the upper surface of each part and identifying the glue drops by a camera.

5. a coaxial parts assembling apparatus according to claim 1, wherein: the control unit comprises a computer (26), a control program in the computer, a sliding guide rail controller (9), a first three-dimensional sliding table controller (10), a second three-dimensional sliding table controller (25), a sliding table controller (22) and a light source controller (23); the computer (26) controls the sliding guide rail, the three-dimensional sliding table, the light source and the needle head through the controller.

6. The coaxial parts assembling apparatus according to claim 4, wherein: the camera calibration is used for calibrating a first camera (11) and a second camera (15) in an image acquisition unit, and the specific method comprises the following steps:

firstly, placing a checkerboard calibration board with the length of m grids and the width of N grids below a first camera (11) and a second camera (15), collecting N calibration board images at different angles, calibrating the first camera (11) and a first telecentric lens (12) by using a calibration algorithm in a computer (26) to obtain a camera matrix and a distortion coefficient, and correcting the images by using the parameters; the size of each checkerboard of the calibration plate is smaller than A/m, wherein A represents the field width of the camera; the whole calibration plate needs to be completely appeared in the picture; m > N, N > 15.

7. The coaxial parts assembling apparatus according to claim 4, wherein: the upper surface of four kinds of parts is circular, seeks the upper surface center of part promptly and seeks the centre of a circle in the part image, specifically includes: gray processing, histogram equalization, gaussian filtering, canny edge detection, and Hough circle detection.

8. a coaxial parts assembling apparatus according to claim 1, wherein: the first inclined 45-degree needle head (19) and the second inclined 45-degree needle head (21) are inclined at an angle of 45 degrees downwards and are used for rotary dispensing on the inner wall of the circular part (iii) in the material tray; the first vertical needle (20) and the second vertical needle (22) are vertically downward and are used for dispensing the appearance of the cylindrical part (i) and the hollow cylindrical part (iv) in the tray;

The repeated precision of the glue discharging track is controlled within 0.05mm, and the repeated precision of the glue discharging is controlled within 0.1mm³the content of the compound is less than the content of the compound; in the process of dispensing the circular ring-shaped part (iii), the center (x) of the part is obtained through a circle detection algorithm₀，y₀) And radius r₀aligning the first 45-degree inclined needle head (19) and the second 45-degree inclined needle head (21) to the inner wall of the circular ring-shaped part (iii) for dispensing; in dispensing the upper surfaces of the cylindrical part (ii) and the hollow cylindrical part (iv), the center (x, y) and the radius r of the parts are obtained by a circle detection algorithm, and dispensing is performed at the (x-0.7r, y-0.7r), (x +0.7r, y-0.7r), (x-0.7r, y +0.7r), (x +0.7r, y +0.7r) positions by using the first vertical needle (20); dispensing is performed at the (x +0.7r, y-0.7r), (x-0.7r, y +0.7r), (x +0.7r, y +0.7r) positions using a second vertical needle (22).

9. The component glue drop detection algorithm for the coaxial component assembly device according to claim 5, wherein: when the upper surfaces of the cylindrical part (ii) and the hollow cylindrical part (iv) are subjected to glue dispensing, two different kinds of glue are required to be dripped at the designated positions of the upper surfaces of the parts, and the total amount of the glue is 8 drops; after dispensing, detecting the glue drops by using a glue drop identification algorithm; the glue drop detection algorithm uses a deep learning-based method, and detects a boundary rectangular frame of each glue drop by training a deep neural network; the detection requirements are as follows:

a) 8 drops of glue;

b) The center of the bounding rectangular frame of each glue drop is positioned in a circle with the designated coordinate and the radius of 50 pixels;

c) the length-width ratio of the rectangular frame of each glue drop is within 1.5;

The three detection requirements are met, and the dispensing is qualified.

10. A coaxial part assembling method is characterized by comprising the following steps:

1) Moving the circular ring-shaped part (iii) of the material tray (4) to the lower part of the first camera (11), carrying out image acquisition on the circular ring-shaped part (iii), and detecting the relative coordinate (x) of the circle center of the circular ring-shaped part (iii) by using an image processing algorithm₀，y₀) And radius r₀；

2) moving the circular ring-shaped part (iii) of the tray (4) to the position below a first 45-degree inclined needle head (19) and a second 45-degree inclined needle head (21), and respectively dropping two glues on the inner wall of the circular ring-shaped part (iii);

3) moving the cylindrical part (ii) of the tray (4) to the lower part of the first camera (11), acquiring the image of the cylindrical part (ii), and detecting the relative coordinate (x) of the center of the circle₁，y₁) Sucking up the cylindrical part (ii) to move to the coordinate (x) using the vacuum suction nozzle (7)₀，y₀) (iv) placing into the circular ring-shaped part (iii) to obtain a combination of the cylindrical part (ii) and the circular ring-shaped part (iii);

4) Moving the hollow cylindrical part (iv) of the material tray (4) to the lower part of the first camera (11), collecting the image of the hollow cylindrical part (iv), and detecting the relative coordinates (x) of the center of the circle of the hollow cylindrical part (iv) by using an image processing algorithm₂，y₂)；

5) Moving the hollow cylindrical part (iv) of the tray (4) under the first vertical needle (20) and the second vertical needle (22), using the first vertical needle (20)(x₂-0.7r,y₂-0.7r),(x₂+0.7r,y₂-0.7r),(x₂-0.7r,y₂+0.7r),(x₂+0.7r,y₂+0.7r) position for dispensing; using a second vertical needle (22) at (x)₂-0.7r,y₂-0.7r),(x₂+0.7r,y₂-0.7r),(x₂-0.7r,y₂+0.7r),(x₂+0.7r,y₂+0.7r) position for dispensing;

6) Moving the hollow cylindrical part (iv) of the tray (4) to the lower part of the first camera (11), carrying out image acquisition on the hollow cylindrical part (iv), detecting glue drops in the image by using a trained target detection neural network, and screening out parts qualified in glue dropping; if not, wiping off the glue before the glue drops are solidified, and dispensing again;

7) Moving the combination of the cylindrical part (ii) and the annular part (iii) in the tray (4) to the lower part of the first camera (11), acquiring images of the combination, and detecting the relative coordinates (x) of the circle center of the combination₃，y₃) Sucking up the assembly of the cylindrical part (ii) and the annular part (iii) to the relative coordinate (x) using a vacuum suction nozzle (7)₂，y₂) (iii) into a hollow cylindrical part (iv); in the assembling process, the centers of the hollow cylindrical part (iv) and the combination of the cylindrical part (ii) and the circular ring-shaped part (iii) are detected for multiple times, the coaxiality of the centers is within 20 micrometers, and the cylindrical part (ii), the circular ring-shaped part (iii) and the hollow cylindrical part (iv) are connected together to form a new combination;

8) after the glue is completely cured, moving the circular sheet part (i) in the tray (4) to the position below the first camera (11), carrying out image acquisition on the combined body, and detecting the relative coordinate (x) of the circle center of the combined body₄，y₄) Sucking up the circular thin-sheet part (i) by using a vacuum suction nozzle (7) and moving to a relative coordinate (x)₂，y₂) (iii) into a hollow cylindrical part (iv); in the assembling process, the centers of the combination of the circular sheet part (i) and the cylindrical part (ii), the circular ring part (iii) and the hollow cylindrical part (iv) are detected for multiple times, the coaxiality is enabled to be within 20 micrometers, and the circular sheet part (i), the cylindrical part (ii), the circular ring part (iii) and the hollow cylindrical part (iv) are connectedAnd (5) connecting together and finishing the assembly.

Technical Field

The invention relates to a coaxial part assembling device and method, and belongs to the field of machine vision and precision detection.

Background

In ballistic missiles and launch vehicles, inertial navigation systems play a very important role in ensuring that missiles (or rockets) hit targets (or accurately enter orbit) according to predetermined requirements. The inertial device and the inertial instrument are key parts in an inertial navigation system, and the quality of the performance of the inertial device and the inertial instrument directly influences the hit precision (or the orbit entering precision) of a missile (or a rocket). The inertia device and the inertia instrument are high-precision components which are formed by combining a plurality of precision parts. Some precision parts are assembled from coaxial parts, such as the torquer in a quartz accelerometer. The invention aims to detect the center of a coaxial part through a machine vision technology, and automatically assemble a precision component according to a specified step through a computer control program. The assembly precision of parts is improved while the full automation of the assembly is realized, and the performance of an inertia device and an inertia instrument is fundamentally improved.

at present, coaxial parts are still assembled manually by detecting the centers of the parts through naked eyes, and the efficiency, the repeatability and the accuracy are poor.

Disclosure of Invention

The technical problem solved by the invention is as follows: the defects of the prior art are overcome, and the coaxial part assembling device and the method are provided. The position of each part of the device is controlled by a computer, a camera is used for collecting pictures of each part, the central relative coordinates of each part are obtained by adopting an image processing algorithm, the parts are dispensed according to the position coordinates of each part, four coaxial parts are assembled together, the full automation of the assembly process is realized, and the accumulated coaxiality error of each component is ensured to be less than 20 micrometers in the process. The assembly method and the assembly device can realize the automatic assembly of any small precise coaxial parts; the method has high assembly efficiency, precision and automation degree.

The technical scheme of the invention is as follows: the coaxial part assembling device comprises a mechanical unit, two sets of same image acquisition units, a dispensing unit, an image processing unit and a control unit, wherein the mechanical part comprises: the device comprises a marble table, a first sliding guide rail, a second sliding guide rail, a first material tray, a second material tray, a first three-dimensional sliding table, a first vacuum suction nozzle, a second vacuum suction nozzle, a guide rail controller and a first three-dimensional sliding table controller; the image acquisition unit is used for shooting and acquiring images of a measured object, and the image processing unit is used for processing the images acquired by the image acquisition unit; the dispensing unit moves the needle head to a specified position, and the parts in the feeding disc are dispensed; the control unit controls the sliding guide rail, the three-dimensional sliding table, the light source and the needle head;

The marble table is used as a base of the whole system, so that the overall stability of the system is ensured; four parts, namely a circular sheet part (i), a cylindrical part (ii), a circular ring part (iii) and a hollow cylindrical part (iv), are placed in the first tray and the second tray; the lower parts of the first material tray and the second material tray are fixed on the sliding tables through positioning pins, and the two sliding tables are respectively placed on the first sliding guide rail and the second sliding guide rail; the first sliding guide rail and the second sliding guide rail are controlled by the sliding guide rail controller, so that the first material tray and the second material tray reach the designated positions; the three-dimensional sliding table controller controls the first three-dimensional sliding table, and the first three-dimensional sliding table drives the vacuum suction nozzle to move in any direction in a three-dimensional space, so that the first vacuum suction nozzle and the second vacuum suction nozzle reach the upper part of the part, and the part is sucked up and placed at a specified position.

The image acquisition unit comprises: the device comprises a first camera, a second camera, a first telecentric lens, a second telecentric lens, a first annular light source, a second annular light source, a first sliding table, a second sliding table, a sliding table controller and a light source controller;

The first camera is connected with the first telecentric lens and fixed on the sliding table, and the annular light source is arranged between the camera lens connector and the tray part and is used for polishing the part to be detected; before the image is collected, the distance between a camera lens and a light source and the distance between the light source and a material tray part are adjusted, and after the optimal position is determined, the relative distance is kept unchanged; the placement position of the first telecentric lens meets the condition that the distortion of the image edge is minimum; the first camera has the resolution ratio of more than ten million pixels and is used for acquiring images of all parts on the material tray; the sliding module can move up and down, left and right, so that the camera reaches an accurate position; the sliding module and the first camera are controlled by the sliding table controller; the first annular light source and the second annular light source are connected with the first three-dimensional sliding table and reach the designated positions; the light source controller is used for controlling the brightness of the first annular light source.

The dispensing unit comprises a dispensing machine, a first inclination needle head, a first vertical needle head, a second inclination needle head, a second vertical needle head and a second three-dimensional sliding table; the four needle heads are moved to the designated position through the second three-dimensional sliding table to dispense the parts in the feeding disc through the second three-dimensional sliding table; the glue dispenser is used for controlling the needle head to dispense glue.

The image processing unit comprises a computer and an image processing algorithm unit; the internal work of the image processing algorithm unit is divided into three parts: and calibrating and detecting the circle center of the upper surface of each part and identifying the glue drops by a camera.

The control unit comprises a computer, a control program in the computer, a sliding guide rail controller, a first three-dimensional sliding table controller, a second three-dimensional sliding table controller, a sliding table controller and a light source controller; the computer controls the sliding guide rail, the three-dimensional sliding table, the light source and the needle head through the controller.

The camera calibration is used for calibrating a first camera and a second camera in an image acquisition unit, and the specific method comprises the following steps:

Firstly, placing a checkerboard calibration board with the length of m grids and the width of N grids below a first camera and a second camera, collecting N calibration board images at different angles, calibrating the cameras and a lens by using a calibration algorithm in a computer to obtain a camera matrix and a distortion coefficient, and correcting the images by using the parameters; the size of each checkerboard of the calibration plate is smaller than A/m, wherein A represents the field width of the camera; the whole calibration plate needs to be completely appeared in the picture; m > N, N > 15.

The upper surface of four kinds of parts is circular, seeks the upper surface center of part promptly and seeks the centre of a circle in the part image, specifically includes: gray processing, histogram equalization, gaussian filtering, canny edge detection, and Hough circle detection.

The first inclined 45-degree needle head and the second inclined 45-degree needle head are inclined at an angle of 45 degrees downwards and are used for carrying out rotary dispensing on the inner wall of the circular part (iii) in the material tray; the first vertical needle head and the second vertical needle head are vertically downward and are used for dispensing the appearance on the cylindrical part (i) and the hollow cylindrical part (iv) in the material tray;

there are two types of glue: 1) epoxy glue to realize bonding, wherein the bonding strength is more than 8 MPa, and 2) conductive glue to realize electric signal conduction between parts; epoxy glue is filled in the first inclined 45-degree needle head and the first vertical needle head; conductive adhesive is filled in the second 45-degree inclined needle head and the second vertical needle head;

The repeated precision of the glue discharging track is controlled within 0.05mm, and the repeated precision of the glue discharging is controlled within 0.1mm³the content of the compound is less than the content of the compound; in the process of dispensing the circular ring-shaped part (iii), the center (x) of the part is obtained through a circle detection algorithm₀，y₀) And radius r₀aligning the first 45-degree inclined needle head and the second 45-degree inclined needle head to the inner wall of the annular part (iii) for dispensing; in the dispensing of the upper surfaces of the cylindrical part (ii) and the hollow cylindrical part (iv), the center (x, y) and the radius r of the parts are obtained by a circle detection algorithm, and the dispensing is performed at the (x-0.7r, y-0.7r), (x +0.7r, y-0.7r), (x-0.7r, y +0.7r), (x +0.7r, y +0.7r) positions by using a first vertical needle; dispensing is performed at the (x-0.7r, y-0.7r), (x +0.7r, y-0.7r), (x-0.7r, y +0.7r), (x +0.7r, y +0.7r) positions using a second vertical needle.

When the upper surfaces of the cylindrical part (ii) and the hollow cylindrical part (iv) are subjected to glue dispensing, two different kinds of glue are required to be dripped at the designated positions of the upper surfaces of the parts, and the total amount of the glue is 8 drops; after dispensing, detecting the glue drops by using a glue drop identification algorithm; the glue drop detection algorithm uses a deep learning-based method, and detects a boundary rectangular frame of each glue drop by training a deep neural network; the detection requirements are as follows:

1) 8 drops of glue;

2) The center of the bounding rectangular frame of each glue drop is positioned in a circle with the designated coordinate and the radius of 50 pixels;

3) The length-width ratio of the rectangular frame of each glue drop is within 1.5;

The three detection requirements are met, and the dispensing is qualified.

the method for training the deep neural network comprises the following steps:

1) Capturing images of the dispensed parts (ii) and (iv) using a first camera;

2) Using labelme to mark all the glue drops in the image and making a glue drop data set;

3) Constructing a target detection convolutional neural network by using a pytorech, wherein the neural network can use an improved neural network based on SSD, Faster R-CNN or YOLO;

4) The Adam algorithm is used as an optimization algorithm, the Xavier is used as an initialization parameter method, the cross entropy is used as a loss function, and the neural network is trained through the glue drop data set.

The first annular light source and the second annular light source adopt fluorescent lamps, halogen lamps, LED light sources or laser light sources, and the brightness range of the light sources is between 50 and 150; the first camera and the second camera are CMOS cameras or CCD cameras, and the number of pixels is more than 1000 ten thousand.

A coaxial part assembling method comprises the following steps:

1) moving the circular ring-shaped part (iii) of the material tray to the lower part of the first camera, carrying out image acquisition on the circular ring-shaped part (iii), and detecting the circle center relative coordinate (x) of the circular ring-shaped part (iii) by using an image processing algorithm₀，y₀) And radius r₀；

2) Moving the circular ring-shaped part (iii) of the material tray to the position below the first 45-degree inclined needle head and the second 45-degree inclined needle head, and respectively dropping the two glues on the inner wall of the circular ring-shaped part (iii);

3) moving the cylindrical part (ii) of the tray to the lower part of the first camera, acquiring an image of the cylindrical part (ii), and detecting the relative coordinate (x) of the circle center of the cylindrical part₁，y₁) Sucking up the cylindrical part (ii) using a vacuum nozzle and moving to a coordinate (x)₀，y₀) (iv) placing into the circular ring-shaped part (iii) to obtain a combination of the cylindrical part (ii) and the circular ring-shaped part (iii);

4) Moving the hollow cylindrical part (iv) of the material tray to the lower part of the first camera, collecting the image of the hollow cylindrical part (iv), and detecting the relative coordinates (x) of the circle center of the hollow cylindrical part (iv) by using an image processing algorithm₂，y₂)；

5) Moving the hollow cylindrical part (iv) of the tray below the first vertical needle and the second vertical needle, using the first vertical needle at (x)₂-0.7r,y₂-0.7r),(x₂+0.7r,y₂-0.7r),(x₂-0.7r,y₂+0.7r),(x₂+0.7r,y₂+0.7r) position for dispensing; using a second vertical needle in (x)₂-0.7r,y₂-0.7r),(x₂+0.7r,y₂-0.7r),(x₂-0.7r,y₂+0.7r),(x₂+0.7r,y₂+0.7r) position for dispensing;

6) Moving the hollow cylindrical part (iv) of the tray to the position below the first camera, carrying out image acquisition on the hollow cylindrical part (iv), detecting glue drops in the image by using a trained target detection neural network, and screening out parts with qualified glue drops; if not, wiping off the glue before the glue drops are solidified, and dispensing again;

7) Moving the combination of the cylindrical part (ii) and the annular part (iii) in the tray to the position below the first camera, acquiring images of the combination, and detecting the relative coordinates (x) of the center of the circle₃，y₃) Sucking up the combination of the cylindrical part (ii) and the annular part (iii) to a relative coordinate (x) using a vacuum nozzle₂，y₂) (iii) into a hollow cylindrical part (iv); in the assembling process, the centers of the hollow cylindrical part (iv) and the combination of the cylindrical part (ii) and the circular ring-shaped part (iii) are detected for multiple times, the coaxiality of the centers is within 20 micrometers, and the cylindrical part (ii), the circular ring-shaped part (iii) and the hollow cylindrical part (iv) are connected together to form a new combination;

8) after the glue is completely cured, moving the circular sheet part (i) in the material tray to a position below the first cameraAnd (3) acquiring images of the combination, and detecting the relative coordinates (x) of the circle center of the combination₄，y₄) Sucking up the chip part (i) using a vacuum nozzle and moving it to a relative coordinate (x)₂，y₂) (iii) into a hollow cylindrical part (iv); in the assembling process, the centers of the combination of the circular sheet part (i) and the cylindrical part (ii), the circular ring part (iii) and the hollow cylindrical part (iv) are detected for multiple times, the coaxiality is enabled to be within 20 micrometers, the circular sheet part (i), the cylindrical part (ii), the circular ring part (iii) and the hollow cylindrical part (iv) are connected together, and the assembling is completed.

the invention has the advantages that:

at present, the small coaxial parts are still assembled manually by detecting the centers of the parts through naked eyes, and the small coaxial parts have low efficiency, poor repeatability and poor accuracy. The invention effectively adjusts the position of each component of the device through computer control; meanwhile, the machine vision method is used for identifying the centers of various parts to give the position coordinates of the parts, and the parts are subjected to dispensing and assembling according to the position coordinates; applying a deep convolutional neural network for target detection in deep learning to glue drop detection for identifying parts qualified in glue drop; and finally, the coaxial parts are automatically assembled, and the coaxiality of all the parts can reach within 20 microns. The assembly method and the assembly device can realize automatic assembly of any small precise coaxial parts and have the characteristics of non-contact, high efficiency, high precision and full automation.

Drawings

Fig. 1 is a schematic structural view of a coaxial parts assembling apparatus of the present invention:

in fig. 1: 1-a marble table; 2-a first sliding guide; 3-a second sliding guide; 4-a first tray; 5-a second material tray; 6-a first three-dimensional sliding table; 7-a vacuum nozzle 1; 8-vacuum suction nozzle 2; 9-a sliding guide controller; 10-a three-dimensional slipway controller; 11-a first CMOS camera; 12-a first telecentric lens; 13-a first ring light source; 14-a first small slide; 15-a second CMOS camera; 16-a second telecentric lens; 17-a second annular light source; 18-a second small slide; 19-a first 45 degree angled needle; 20-a first vertical needle; 21-a second 45 degree angled needle; 22-a second vertical needle; 23-a second three-dimensional slide; 24-a small slide controller; 25-a light source controller; 26-a computer; 27-a dispenser; 28-three-dimensional slipway controller.

fig. 2 is a schematic diagram of four parts on a tray:

in fig. 2: i-circular sheet; ii-a cylindrical part; iii-a circular ring shaped part; iv-hollow parts (cylindrical parts with a cylindrical space in the middle).

Fig. 3 is an assembly flow of coaxial parts assembly.

Detailed Description

the invention is further illustrated with reference to the following examples and figures, without thereby limiting the scope of the invention. Referring to fig. 1, fig. 1 is a schematic structural diagram of a coaxial component assembling apparatus according to the present invention, and as can be seen from fig. 1, the apparatus includes a mechanical unit, an image collecting unit, a dispensing unit, an image processing unit, and a control unit.

the mechanical part comprises: the device comprises a marble table 1, a first sliding guide rail 2, a second sliding guide rail 3, a first material tray 4, a second material tray 5, a first three-dimensional sliding table 6, a first vacuum suction nozzle 7, a second vacuum suction nozzle 8, a guide rail controller 9 and a first three-dimensional sliding table controller 10. The marble platform 1 is used as a base of the whole system, and the overall stability of the system is ensured. Four parts are placed in the first tray 4 and the second tray 5, and fig. 2 is a schematic diagram of the 4 parts, namely a circular slice i, a cylindrical part ii, a circular ring part iii and a hollow cylindrical part iv; the lower parts of the first material tray 4 and the second material tray 5 are fixed on sliding tables through positioning pins, and the two sliding tables are respectively placed on the first sliding guide rail 2 and the second sliding guide rail 3; the sliding guide rail controller 9 controls the first sliding guide rail 2 and the second sliding guide rail 3 to enable the first tray 4 and the second tray 5 to reach the designated positions; the three-dimensional sliding table controller 10 controls the first three-dimensional sliding table 6, and the first three-dimensional sliding table 6 can drive the vacuum suction nozzle to move in any direction in a three-dimensional space, so that the first vacuum suction nozzle 7 and the second vacuum suction nozzle 8 reach the upper part of a part, and suck the part and place the part into a specified position.

The image acquisition unit comprises: the system comprises a first camera 11, a second camera 15, a first telecentric lens 12, a second telecentric lens 16, a first annular light source 13, a second annular light source 17, a first sliding table 14, a second sliding table 18, a sliding table controller 24 and a light source controller 25; the unit comprises two sets of same image acquisition systems, two batches of parts are assembled simultaneously through the beats of the two parts of the control device, and only one set of image acquisition system is introduced. The first camera 11 is connected with the first telecentric lens 12 and fixed on the sliding table 14, and the annular light source 13 is arranged between the camera lens connector and the tray part and is used for polishing the part to be detected. Before the image is collected, the distance between the camera lens and the light source and the distance between the light source and the tray part are adjusted, and after the optimal position is determined, the relative distance is kept unchanged. The first telecentric lens 12 satisfies image edge distortion minimization; the resolution ratio of the first camera 11 is 4096 x 3000, and the first camera is used for acquiring accurate images of all parts on the material tray 4; the sliding module 14 can move up and down, left and right, so that the camera reaches an accurate position; the sliding module 14 and the first camera 11 are controlled by a sliding table controller 24; the first annular light source 13 and the second annular light source 17 are connected with the first three-dimensional sliding table 6 and reach the designated positions. The light source controller 25 is used to control the brightness of the ring light source.

the dispensing unit comprises a dispensing machine 27, a first inclined 45-degree needle 19, a first vertical needle 20, a second inclined 45-degree needle 21, a second vertical needle 22 and a second three-dimensional sliding table 23; the needle head is moved to the designated position by the second three-dimensional sliding table 23, and the parts in the feeding tray 5 are dispensed. The dispenser 27 is used for controlling the needle head to dispense.

the image processing unit comprises a computer 26 and image processing algorithms therein. The image processing algorithm is divided into three parts: and calibrating and detecting the circle center of the upper surface of each part and identifying the glue drops by a camera.

The control unit comprises a computer 26, a control program in the computer, a sliding guide rail controller 9, a first three-dimensional sliding table controller 10, a second three-dimensional sliding table controller 25, a sliding table controller 22 and a light source controller 23. The computer 26 is connected with the first three-dimensional sliding table controller 10, the small sliding table controller 24 and the second three-dimensional sliding table controller 28 through USB interfaces to control the sliding table, and is also connected with the first CMOS camera 11 and the second CMOS camera 15 through the USB interfaces to take a picture; the light sources 13 and 17 are connected through a 485 interface to control the brightness of the light sources; the adhesive dispenser 27 is connected through the 232 interface, and the adhesive dispenser is controlled to dispense the parts.

the specific method for calibrating the first camera 11 and the second camera 15 in the image acquisition unit is as follows:

Firstly, a checkerboard calibration plate (m > N) with the length of m grids and the width of N grids is placed below a first camera 11 and a second camera 15, N (N >15) calibration plate images of various angles are collected, the camera 11 and a lens 12 are calibrated by using a calibration algorithm in a computer 26, a camera matrix and a distortion coefficient are obtained, and the parameters are used for correcting the images. The size of each checkerboard of the calibration board is smaller than the width/m of the camera visual field, and the whole calibration board needs to be completely appeared in the picture.

starting a control program and an image processing algorithm in the computer 25, placing a 19 × 15 checkerboard calibration board under the first camera 11 and the second camera 15, wherein the size of each checkerboard is 0.5mm, and acquiring 16 checkerboard images with different rotation angles. And finding out coordinates of all black and white angular points of the checkerboard through gray processing, angular point detection and sub-pixel level accurate processing, and obtaining a camera matrix, a distortion coefficient, a rotation vector and a translation vector through a Zhang calibration algorithm. And correcting the subsequently acquired part images by using the parameters so as to eliminate errors brought by the camera. And calculating to obtain the actual distance corresponding to each pixel by the average pixel number and the actual distance between the adjacent angular points of the corrected checkerboard until the actual distance is accurate to 0.001 mu m.

The upper surfaces of the four parts are all circular, and the center of the upper surface of the part is found, namely the circle center in the image of the part. The method for identifying the coordinates of the circle center of the part comprises the following steps: gray processing, histogram equalization, gaussian filtering, canny edge detection, and Hough circle detection.

Finding the center of each part has two purposes:

1) In the assembling process, the centers of all parts are overlapped together, and the coaxiality of the centers of all parts is ensured to be within 20 micrometers;

2) according to the center coordinates of each part, a more accurate dispensing position coordinate is provided for the dispensing process.

The first inclined 45-degree needle head 19 and the second inclined 45-degree needle head 21 are inclined at an angle of 45 degrees downwards and are used for rotationally dispensing the inner wall of the circular ring-shaped part iii in the material tray; the first vertical needle head 20 and the second vertical needle head 22 are vertically downward and used for dispensing the appearance of the cylindrical part ii and the hollow cylindrical part iv in the tray, and the specific position of dispensing is determined by the obtained part center coordinate position by the method; there are two types of glue: 1) epoxy glue to realize bonding with bonding strength higher than 8 MPa, and 2) conductive glue to realize electric signal conduction between parts. Epoxy glue is filled in the first inclined 45-degree needle head 19 and the first vertical needle head 20; conductive adhesive is filled in the second inclined 45-degree needle head 21 and the second vertical needle head 22.

The repeated precision of the glue discharging track is controlled within 0.05mm, and the repeated precision of the glue discharging is controlled within 0.1mm³Within. When the circular ring-shaped part iii is subjected to dispensing, the center (x) of the part is obtained through a circle detection algorithm₀，y₀) And radius r₀And the first inclined 45-degree needle 19 and the second inclined 45-degree needle 21 are aligned to the inner wall of the annular part iii for dispensing. In dispensing the upper surfaces of the cylindrical part ii and the hollow part iv, the center (x, y) and the radius r of the parts are obtained by a circle detection algorithm, and dispensing is performed at the (x-0.7r, y-0.7r), (x +0.7r, y-0.7r), (x-0.7r, y +0.7r), (x +0.7r, y +0.7r) positions using the first vertical needle 20; dispensing is performed at the (x-0.7r, y-0.7r), (x +0.7r, y-0.7r), (x-0.7r, y +0.7r), (x +0.7r, y +0.7r) positions using the second vertical needle 22.

and dispensing the upper surfaces of the parts ii and iv, wherein two different kinds of glue are required to be dispensed at specified positions on the upper surfaces of the parts, and the total number of the glue is 8. Since the freshly extruded glue is not cured, it is difficult to precisely control its position and amount of glue, sometimes the glue will elongate, and sometimes the amount of glue extruded will not meet the requirements.

In order to ensure that the assembly of the parts is stable and the glue drops cannot overflow, the glue drops are detected by using a glue drop identification algorithm after the glue is dispensed. The glue drop detection algorithm uses a deep learning-based method, and detects a bounding rectangular frame of each glue drop by training a deep neural network. The detection requirement is as follows: 1) 8 drops of glue; 2) the center of the bounding rectangular frame of each glue drop is positioned in a circle with the designated coordinate and the radius of 50 pixels; 3) the ratio of the length to the width of the bounding rectangular box of each glue drop is within 1.5. The detection is satisfied, and the dispensing is qualified.

The method for training the deep neural network comprises the following steps:

1) Using a camera 11 to collect 500 images of the dispensed parts ii and iv;

2) In order to improve the training and detection speed and reduce the memory consumed during processing, preprocessing the image and downsampling the image to 2000 multiplied by 1000, marking all the glue drops in the image by using labelme, and manufacturing a glue drop data set;

3) Constructing a target detection convolutional neural network by using the pytorech, wherein the neural network can use an SSD-based improved neural network;

4) The Adam algorithm is used as an optimization algorithm, the Xavier is used as an initialization parameter method, the cross entropy is used as a loss function, and the neural network is trained through the glue drop data set.

The first annular light source 13 and the second annular light source 17 adopt annular LED light sources, and the brightness of the light sources is set to be 100; the first camera 11 and the second camera 15 are CMOS cameras, and the resolution is 4096 × 3000. And adjusting the relative distance between the CMOS camera 11, the annular light source 13 and the material tray to find the position for acquiring the clearest image.

as shown in fig. 2, the tray has four different small circular coaxial parts, namely a circular sheet i, a cylindrical part ii, a circular ring part iii and a cylindrical part iv with a cylindrical gap in the middle, and the purpose of the coaxial part assembling device is to assemble the four parts together according to the flow chart shown in fig. 3, and the coaxiality of the parts is within 20 microns.

The specific assembling steps are as follows:

1) Moving the circular ring-shaped part iii of the tray 4 to the lower part of the camera 11, collecting the image of the part iii, and detecting the relative coordinate (x) of the circle center of the part iii by using an image processing algorithm₀，y₀) And radius r₀；

2) Moving the circular ring-shaped part iii of the tray 4 to the lower part of the first inclined 45-degree needle head 19 and the second inclined 45-degree needle head 21, adjusting the needle heads to the position right above the part iii, moving the needle heads downwards to a proper position, rotating the needle heads for 360 degrees, uniformly discharging glue, and respectively dispensing the two kinds of glue on the inner wall of the part iii;

3) Moving the cylindrical part ii of the tray 4 to the lower part of the camera 11, collecting the image of the part ii, and detecting the relative coordinates (x) of the center of the circle₁，y₁) Sucking up the part ii using the vacuum nozzle 7 and moving to the coordinate x₀，y₀Placed in part iii to give a combination of parts ii and iii;

4) Moving the hollow cylindrical part iv of the tray 4 to the lower part of the camera 11, collecting the image of the part iv, and detecting the relative coordinates (x) of the center of the circle of the part iv by using an image processing algorithm₂，y₂)；

5) the hollow cylindrical part iv of the tray 4 is moved under the first vertical needle 20 and the second vertical needle 22, using the first vertical needle 20 at (x)₂-0.7r,y₂-0.7r),(x₂+0.7r,y₂-0.7r),(x₂-0.7r,y₂+0.7r),(x₂+0.7r,y₂+0.7r) position for dispensing; using a second vertical needle (22) at (x)₂-0.7r,y₂-0.7r),(x₂+0.7r,y₂-0.7r),(x₂-0.7r,y₂+0.7r),(x₂+0.7r,y₂+0.7r) position for dispensing;

6) Moving the hollow cylindrical part iv of the material tray 4 to the lower part of the camera 11, carrying out image acquisition on the part iv, detecting glue drops in the image by using a trained target detection neural network, and screening out parts with qualified glue dropping according to the above mentioned glue drop detection standard (if the parts are unqualified, wiping the parts before the glue drops are solidified, and dispensing again);

7) Moving the combined body of the parts ii and iii in the tray 4 to the lower part of the camera 11, collecting the image of the combined body, and detecting the relative coordinates (x) of the center of the circle₃，y₃) Sucking up the assembly of parts ii and iii to the opposite seat using the vacuum suction nozzle 7Label (x)₂，y₂) Into part iv. During the assembly process, the center of the part iv and the combination of the parts ii and iii is detected for a plurality of times, the coaxiality of the parts iv and the combination of the parts ii and iii reaches less than 20 microns, and the parts ii, iii and iv are connected together to form a new combination;

8) After the glue is completely cured, moving the part i in the tray 4 to the position below the camera 11, carrying out image acquisition on the combined body, and detecting the relative coordinate (x) of the circle center of the combined body₄，y₄) The part i is sucked up by the vacuum suction nozzle 7 and moved to the relative coordinate (x)₂，y₂) Into part iv. During the assembly process, the center of the assembly of the parts i and ii iii iv is detected for a plurality of times, the coaxiality of the assembly is within 20 microns, the parts i, ii, iii and iv are connected together, and the assembly is completed.

in conclusion, the device automatically assembles four parts into a precision assembly by detecting the coordinates of the centers of circles of the four parts with different sizes and shapes and taking the coaxiality error of the parts not more than 20 micrometers as a standard. The method has the characteristics of high measurement precision, automation, convenience, rapidness and the like. The method and the device can realize the measurement, the glue drop detection and the automatic assembly of any coaxial parts.