KR20230091248A - System for tig welding of stator coil using deep learning algorithm - Google Patents

Abstract

Embodiments of the present invention can improve the quality of welding by accurately detecting the welding position when welding the terminal pin of the stator coil and the terminal block pin, and deep learning that can improve productivity and reduce manufacturing cost by preventing welding defects It relates to a TIG welding system for stator coils using An embodiment of the present invention is a system for TIG (Tungsten Inert Gas) welding between terminal pins of a stator coil and terminal block pins, comprising: a vision unit acquiring an image including both the terminal pins and the terminal block pins; a position detection unit that receives an image obtained from the vision unit and calculates a center of welding between the terminal pin and the terminal block pin on the input image using a detection model to be welded generated by learning through a deep learning algorithm; and a TIG welding unit for performing TIG welding between the terminal pin and the terminal block pin using the welding center calculated by the position detector.

Description

TIG welding system of stator coil using deep learning algorithm {SYSTEM FOR TIG WELDING OF STator COIL USING DEEP LEARNING ALGORITHM}

Embodiments of the present invention can improve the quality of welding by accurately detecting the welding position during welding between terminal pins and terminal block pins constituting the stator coil, improve productivity and reduce manufacturing cost by preventing welding defects. It relates to a stator coil TIG welding system using a deep learning algorithm.

In general, an electric vehicle called an eco-friendly vehicle may generate driving force by an electric motor that obtains rotational force with electrical energy.

An electric vehicle runs in an EV (Electric Vehicle) mode, which is a pure electric vehicle mode using only power from an electric motor, or in a hybrid electric vehicle (HEV) mode using both an engine and a drive motor as power.

Most electric motors used as a power source for electric vehicles use a permanent magnet synchronous motor (PMSM).

An electric motor as a permanent magnet synchronous motor used as a power source of such an electric vehicle is basically composed of a stator generating magnetic flux, and a rotor disposed with a predetermined air gap from the stator and provided with a permanent magnet to perform rotational motion. .

Meanwhile, it is known that the output of an electric motor is proportional to the number of turns of a coil wound around a stator core. However, when the number of turns of the coil is increased, the size of the stator core or the electric motor itself including the stator core inevitably increases, making miniaturization difficult.

Therefore, in order to improve the output without increasing the size of the electric motor, the area ratio of the coil wound around the stator core is increased. That is, the dead space between the stator core and the winding coil or the dead space between coils can be minimized to increase the coil occupancy factor.

Therefore, in recent years, instead of using an annular coil having a circular cross section as a coil winding, a rectangular coil (referred to as a “segment coil” in the industry) is used to reduce dead space and improve the occupancy rate, resulting in output compared to the same size. There is a trend to produce high electric motors.

In order to facilitate coil winding work of the prismatic coil, a plurality of separate hairpin-shaped (approximately U-shaped or V-shaped) stator coils (hereinafter referred to as hairpins) are inserted into slots of the stator core, and the slots A method of welding the protruding ends of adjacent hairpins in the inner radial direction to enable individual stator coils to be fully energized has been proposed.

Here, the stator coil may configure a terminal terminal for external power connection by welding a terminal pin and a terminal block pin among a plurality of hairpins, using TIG (Tungsten Inert Gas) welding between the terminal pin and the terminal block pin. so that they can be mutually welded.

At this time, in the prior art, since the operator must manually check and adjust the positions of the terminal pin and the terminal block pin, there is a problem in that the welding process takes a long time.

In order to solve this problem, an attempt was made to determine the location of the terminal pin and the terminal block pin with a vision technology applying a rule-based algorithm that inputs rules in advance, but by the shape, angle and size of the cut surface of the pins, etc. If the image is captured differently from the desired image, even if the vision technology is used, the position detection of the terminal pin and the terminal block pin is not properly performed, so welding is performed at the wrong location, such as being welded to another part or welded to a jig, etc. There was a problem that caused a defect and furthermore caused damage to the equipment.

In particular, when the positions of the terminal pin and the terminal block pin are not detected, the worker manually repositions them to proceed with the welding process, thereby reducing workability and reducing productivity due to increased working time. there was.

Matters described in this background art section are prepared to enhance understanding of the background of the invention, and may include matters that are not prior art already known to those skilled in the art to which this technique belongs.

Republic of Korea Patent Registration No. 10-2023012 (2019.09.11.)

An embodiment of the present invention provides a stator coil TIG welding system using a deep learning algorithm capable of minimizing welding defects by precisely detecting the welding position between the terminal pin and terminal block pin of the stator coil so that welding is performed at the correct position. It is to do.

In addition, the embodiment of the present invention is a stator using a deep learning algorithm that can reduce the time required for the welding process and improve productivity by automatically performing welding at the exact welding position between the terminal pin of the stator coil and the terminal block pin. It is to provide a coil TIG welding system.

According to an embodiment of the present invention, a system for TIG (Tungsten Inert Gas) welding between a terminal pin of a stator coil and a terminal block pin, a vision unit that acquires an image including both the terminal pin and the terminal block pin. ; a position detection unit that receives an image obtained from the vision unit and calculates a center of welding between the terminal pin and the terminal block pin on the input image using a detection model to be welded generated by learning through a deep learning algorithm; and a TIG welding unit for performing TIG welding between the terminal pin and the terminal block pin using the welding center calculated by the position detector.

In this case, the vision unit may include a camera that acquires an image from above the terminal pin and the terminal block pin.

The TIG welding system of the present embodiment learns sample images including the terminal pin and the terminal block pin, generates the welding target detection model based on the learning result, and transfers the generated welding target detection model to the position detector. It may include a deep learning modeling unit that provides.

Here, the deep learning modeling unit receives the sample images, performs a labeling process of collecting a region to be welded from each of the sample images, and generates a detection model to be welded including the terminal pin and the terminal block pin from the target region to be welded. A training process of learning and a test process of classifying and generating a final welding target detection model by verifying the learned welding target detection model may be performed.

In this case, the deep learning modeling unit may perform the process with a convolutional neural network (CNN)-based deep learning algorithm.

Meanwhile, the position detection unit selects a welding area including the terminal pin and the terminal block pin on the input image using the welding object detection model, and sets the center of the selected welding area to the corresponding terminal pin and terminal. It can be calculated as the center of the weld between block pins.

The TIG welding part may perform spot welding at a welding center between the terminal pin and the terminal block pin.

The TIG welding system of the present embodiment receives a welding image between the terminal pin and the terminal block pin acquired through the vision unit after the TIG welding, and uses a welding part detection model generated by learning with a deep learning algorithm to detect the terminal pin It may further include an inspection unit that detects a welding defect between the terminal block pin and the terminal block pin.

According to the embodiment of the present invention, there is an advantage in that the welding position between the terminal pin of the stator coil and the terminal block pin can be accurately detected, and accordingly, product defects due to welding defects can be minimized by ensuring that welding is performed at the correct position. there is.

In addition, according to an embodiment of the present invention, the time required for the process can be shortened by automatically and accurately detecting the welding position between the terminal pin of the stator coil and the terminal block pin to perform welding, thereby improving productivity. There is an advantage to being

1 is a block diagram schematically showing a TIG welding system of a stator coil using a deep learning algorithm according to an embodiment of the present invention;
2A to 2C are exemplary diagrams schematically showing regions to be welded in the labeling process of the deep learning modeling unit of FIG. 1
Figure 3a is an exemplary view schematically showing an image acquired in the vision unit of Figure 1
FIG. 3B is an exemplary view schematically showing an image of a region to be welded selected from the image of FIG. 3A and a welding center thereof;
Figure 4a is an exemplary view schematically showing an image acquired in the vision unit of Figure 1
Figure 4b is an exemplary view schematically showing an image of a welding target area selected from the image of Figure 4a and the center of the welding.
Figure 5 is an exemplary view schematically showing a welding image having a normally welded weld

Hereinafter, specific embodiments of the present invention will be described with reference to the drawings. The detailed descriptions that follow are provided to provide a comprehensive understanding of the methods, devices and/or systems described herein. However, this is only an example and disclosed embodiments are not limited thereto.

In describing the embodiments, if it is determined that a detailed description of related known technologies may unnecessarily obscure the gist of the disclosed embodiments, the detailed description will be omitted. In addition, terms to be described later are terms defined in consideration of functions in the disclosed embodiments, which may vary according to intentions or customs of users or operators. Therefore, the definition should be made based on the contents throughout this specification. Terminology used in the detailed description is only for describing the embodiments and should in no way be limiting. Unless expressly used otherwise, singular forms of expression include plural forms. In this description, expressions such as "comprising" or "comprising" are intended to indicate any characteristic, number, step, operation, element, portion or combination thereof, one or more other than those described. It should not be construed to exclude the existence or possibility of any other feature, number, step, operation, element, part or combination thereof.

1 is a block diagram schematically showing a TIG welding system of a stator coil using a deep learning algorithm according to an embodiment of the present invention, and FIGS. 2A to 2C show a welding target area in the labeling process of the deep learning modeling unit of FIG. 1 These are schematic examples.

Referring to FIG. 1, a stator coil TIG welding system 100 using a deep learning algorithm according to an embodiment of the present invention includes a terminal pin 22a constituting the stator coil (see FIGS. 3A and 3B) and a terminal block A system for performing an accurate welding process using a deep learning algorithm when TIG welding between pins (22b: see FIGS. 3A and 3B), which includes both the terminal pin 22a and the terminal block pin 22b A vision unit 110 that acquires an image, a terminal pin 22a and a terminal pin 22a on the input image using the welding target detection model generated by receiving the image acquired by the vision unit 110 and learning with a deep learning algorithm A position detecting unit 120 that calculates a welding center between the terminal block pins 22b, and between the terminal pins 22a and the terminal block pins 22b using the welding center calculated by the position detecting unit 120. It may include a TIG welding unit 130 that performs TIG welding, and a control unit 101 that controls overall operation and control of each component and data transmission and reception between the components.

At this time, the vision unit 110 may include a camera that acquires an image from above the terminal pin 22a and the terminal block pin 22b.

In addition, in the TIG welding system 100 of this embodiment, a welding image 243 between the terminal pin 22a and the terminal block pin 22b acquired through the vision unit 110 after the TIG welding (see FIG. 5) ), and an inspection unit 150 for detecting welding defects between the terminal pin 22a and the terminal block pin 22b using a weld detection model generated by learning with a deep learning algorithm. It may further include.

In addition, the TIG welding system 100 of this embodiment learns the sample images 14 including the terminal pin 12a and the terminal block pin 12b, and based on the learning result, generates the welding target detection model, It may include a deep learning modeling unit 140 that provides the generated welding object detection model to the position detection unit 120 through the control unit 101 .

At this time, the deep learning modeling unit 140 learns sample images of TIG welding between the terminal pin 12a and the terminal block pin 12b, generates the weld detection model based on the learning result, and generates the The welding part detection model may be provided to the inspection unit 150 through the control unit 101 .

2A to 2C , the deep learning modeling unit 140 performs a labeling process of receiving the sample images and collecting welding target regions 141a from each of the sample images 14, and the welding target A training process of learning a welding target detection model including both the terminal pin 12a and the terminal block pin 12b from the area 141a, and classification and generation of a final welding target detection model by verifying the learned welding target detection model. The testing process can be performed.

The sample images 14 in the labeling process are a plurality of images acquired in advance through a vision unit such as a camera under various conditions and environments, and a plurality of welds including both the terminal pin 12a and the terminal block pin 12b. It may include the target area 141a.

Further, the training process may include a process of recognizing and learning the welding target region 141a including both the terminal pin 12a and the terminal block pin 12b as a welding target detection model.

In addition, the test process may include a process of verifying a final welding target detection model to be applied by the position detection unit 120 among the learned welding target regions 142b and classifying and generating the model.

Eventually, the deep learning modeling unit 140 may receive large-scale sample images captured under various conditions and environments, and classify the welding target detection model through a labeling process, a training process, and a test process from the input sample images It is possible to generate, and the classification and generated welding target detection model can be provided to the position detection unit 120 . Of course, the deep learning modeling unit 140 acquires sample images welded between the terminal pin 12a and the terminal block pin 12b in advance, and from the acquired sample images, the terminal pin 12a and the It is possible to classify and generate a final weld detection model by learning and verifying a normally welded weld between the terminal block pins 12b, and the classified and generated weld detection model may be provided to the inspection unit 150.

Accordingly, the position detection unit 120 uses the welding object detection model provided from the deep learning modeling unit 140 to determine the perspective of the terminal pin 22a and the terminal block pin 22b input from the vision unit 110. The welding center can be accurately detected, and the inspection unit 150 uses the welding detection model provided from the deep learning modeling unit 140 to detect the terminal pin 22a input from the vision unit 110 and the terminal pin 22a. It is possible to determine whether or not TIG welding has been normally performed from the welded image between the block pins 22b.

In this case, the deep learning modeling unit 140 may classify and generate the welding target detection model and the welding part detection model by performing the above processes with a convolutional neural network (CNN)-based deep learning algorithm.

Next, with further reference to FIGS. 3A to 5 , a process of a TIG welding system using a deep learning algorithm according to an embodiment of the present invention will be described in detail.

FIG. 3A is an exemplary view schematically showing an image acquired from the non-vision unit of FIG. 1 , and FIG. 3B is an exemplary view schematically showing an image of a region to be welded selected from the image of FIG. 3A and the center of the welding, FIG. 4A is an exemplary diagram schematically showing an image obtained from the vision unit of FIG. 1 , and FIG. 4B is an exemplary diagram schematically showing an image of a region to be welded selected from the image of FIG. 4A and the welding center thereof. And, Figure 5 is an exemplary view schematically showing a welding image having a normally welded weld.

First, a stator having a stator coil including terminal pins 22a and terminal block pins 22b is loaded into the TIG welding system of this embodiment.

And, as in FIG. 3A or 4A, an image 242 including the terminal pin 22a and the terminal block pin 22b is obtained through the vision unit 110 described above.

Next, the acquired image 242 is provided to the above-described position detection unit 120 through the above-described control unit 101 .

Then, as shown in FIG. 3B or 4B, the position detection unit 120 uses the welding target detection model provided from the aforementioned deep learning modeling unit 140 to detect the corresponding terminal pin on the image 242 provided. A region to be welded 242a including 22a and the terminal block pin 22b is selected.

At this time, in the TIG welding system 100 of this embodiment, the position detection unit 120 sets the center 242a-1 of the selected region to be welded 242a to the corresponding terminal pin 22a and terminal block pin 22b. ) can be calculated as the center of the weld between

Thereafter, TIG welding may be performed with the center 242a-1 of the region to be welded 242a as a welding center.

Then, as shown in FIG. 5, a welded image 243 having performed TIG welding is obtained through the vision unit 110, and the image obtained through the control unit 101 is provided to the inspection unit 150 described above. do.

Then, the above-described inspection unit 150 uses the weld detection model provided from the deep learning modeling unit 140 to determine whether the weld 23 is normally TIG welded in the welding image 243 input from the vision unit 110. It can be determined whether this has been done. That is, it is determined whether the welding part 23 between the terminal pin 22a and the terminal block pin 22b is normal.

Here, if it is determined that TIG welding has been abnormally performed on the welded portion 23 of the terminal pin 22a and the terminal block pin 22b, re-welding may be performed.

Although representative embodiments of the present invention have been described in detail above, those skilled in the art will understand that various modifications are possible to the above-described embodiments without departing from the scope of the present invention. . Therefore, the scope of the present invention should not be limited to the described embodiments and should not be defined, and should be defined by not only the claims to be described later, but also those equivalent to these claims.

12a: terminal pins in the sample image
12b: terminal block pin of sample image
14 : Sample image
22a: real terminal pin
22b: actual terminal block pin
23: welding part
100: TIG welding system
101: control unit
110: vision department
120: position detection unit
130: TIG welding part
140: deep learning modeling unit
141a: welding target area of the sample image
150: inspection unit
242: image acquired in vision department
242a: welding target area
242a-1: the center of the area to be welded

Claims (8)

A system for TIG (Tungsten Inert Gas) welding between terminal pins of stator coils and terminal block pins,
a vision unit acquiring an image including both the terminal pin and the terminal block pin;
a position detection unit that receives an image obtained from the vision unit and calculates a center of welding between the terminal pin and the terminal block pin on the input image using a detection model to be welded generated by learning through a deep learning algorithm; and
A TIG welding part for performing TIG welding between the terminal pin and the terminal block pin using the welding center calculated by the position detection part; TIG welding system of a stator coil using a deep learning algorithm.
The method of claim 1,
The vision department,
A stator coil TIG welding system using a deep learning algorithm, including a camera for acquiring images from above the terminal pin and the terminal block pin.
The method of claim 1,
A deep learning modeling unit that learns sample images including the terminal pin and the terminal block pin, generates the welding target detection model based on the learning result, and provides the generated welding target detection model to the position detection unit. , TIG welding system of stator coil using deep learning algorithm.
The method of claim 3,
The deep learning modeling unit,
A labeling process of receiving the sample images and collecting welding target areas from each of the sample images; a training process of learning a welding target detection model including the terminal pin and the terminal block pin from the welding target region; A TIG welding system for stator coils using a deep learning algorithm that performs a test process to classify and create a final welding target detection model by verifying the detected welding target detection model.
The method of claim 4,
The deep learning modeling unit performs the process with a convolutional neural network (CNN)-based deep learning algorithm, the TIG welding system of the stator coil using a deep learning algorithm.
The method of claim 1,
The position detection unit,
Using the welding object detection model, a welding area including the terminal pin and the terminal block pin is selected on the input image, and the center of the selected welding area is set as the center of the welding between the corresponding terminal pin and terminal block pin. TIG welding system of stator coil using deep learning algorithm.
The method of claim 1,
The TIG welding part,
A TIG welding system for a stator coil using a deep learning algorithm, wherein spot welding is performed at a center of a weld between the terminal pin and the terminal block pin.
The method of claim 1,
After the TIG welding, a welding image between the terminal pin and the terminal block pin obtained through the vision unit is received, and a welding part detection model generated by learning with a deep learning algorithm is used. Defective welding between the terminal pin and the terminal block pin A TIG welding system of a stator coil using a deep learning algorithm, further comprising an inspection unit for detecting.

Images