KR20240032456A - A welding defect comfirmation method of a seat frame for vehicle - Google Patents

Abstract

In the present invention, the method of determining welding defects in a vehicle seat frame through machine learning based on image data sets the minimum number of cameras so that there are no oblique surfaces of the welded area, acquires image data for learning and comparison, and artificially processes the machine-learning data. Analyze, compare and judge through intelligence. Therefore, even without separate data, the same reliability can be obtained as when an inspector visually determines a defect in a welded area.

Description

{A welding defect comfirmation method of a seat frame for vehicle}

The present invention relates to a method for determining welding defects in a vehicle seat frame, and in particular, to a method for determining welding defects in a vehicle seat frame through machine learning based on image data.

Currently, vehicle seat frames are welded with various connecting members such as brackets, hinges, bolts, and nuts for mounting seat rails, seat belts, recliners, and seat cushions. Typically, a vehicle seat frame has about 100 welded areas. These welds are welded at the desired origin using a welding robot with set values of current, voltage, welding time, and pressure. If welding defects often occur in welded areas made in this way, the welder finds the defective welding area and repairs the welding defects directly. Defective items in CO2 arc welding include spatter, needles, high degree of sludge, side welding, defective bridges, punctures, porosity, drawing discrepancies (position, omission), overlaps, back bead, incorrect assembly (poor setting), etc. Defective items in welding include welding marks (dents, chips), overlapping parts, missing parts, uniformity (cracks, welding), spotting defects (position, half-hitting), separation, length defects, etc. Instead of directly determining the presence or absence of defects with the welder's naked eyes, various methods have recently been proposed to determine the presence or absence of defects in the welded area, and recently, sensor data inspection and image data inspection have been consistently announced.

In the title of the invention of patent registration number 10-2233498 (registration date March 23, 2021), “Welding defect determination device and method,” the welding defect determination device has a high correlation with porosity in learning welding current information and learning arc voltage information. a feature variable determination unit that determines the variables of the relationship as feature variables; A training unit that trains an artificial neural network using learning data as input data to calculate threshold values for each characteristic variable to determine whether pores have occurred; A detection unit that detects welding current information and arc voltage information of the welding device in operation; and a determination unit that determines whether pores have occurred using the values of characteristic variables extracted from the detected welding current information and arc voltage information and the calculated threshold. Unlike the sensor-based method, this patent is mainly based on welding current and arc voltage. This technology has the advantage of low additional facility costs and real-time monitoring, but the accuracy of determining defects is somewhat low.

As a different method, the title of the invention in Patent Publication No. 10-2021-0157253 (publication date December 28, 2021), "Learning device and method of welding quality evaluation model, and evaluation device using the same" is disclosed. . Here, the learning device for the quality evaluation model of welding is an interface for connecting to each of a welding device that performs gas tungsten arc welding and an image acquisition device, and, from the welding device and the image acquisition device, welding steps of the gas tungsten arc welding. Receiving welding data, extracting a plurality of characteristic variable values from the received welding data, and controlling learning of a quality evaluation model using the plurality of characteristic variable values and a label related to the quality of the gas tungsten arc welding. It includes a control unit and a memory that stores the learned quality evaluation model. The image acquisition device includes a plurality of image data acquired by the image acquisition device at a time before and after performing at least one pass welding. Since image data is generated for each pass, it can be used for parts with a small number of welded areas, but there is a problem in that it cannot be used in places where there are more than 10 welded areas.

Therefore, instead of judging welding defects in about 100 welded products such as vehicle seat frames using the inspector's eyes, it is necessary to use image data to machine learn the image data to conveniently, accurately and quickly determine it.

In order to determine welding defects using video data, a camera setting means is needed to photograph all welded areas from various angles without any blind sides of the vehicle seat frame.

In addition, there is a need to increase the accuracy in determining welding defects by integrating welding process operation data, sensor data, and image data that detect abnormal signs of welding defects.

Additionally, there is a need to quickly determine welding defects and improve production speed.

The purpose of the present invention is to provide a method for determining welding defects in a vehicle seat frame using only image data, without sensor data or operational data, in order to further advance the above-described prior art and meet the above-mentioned needs.

Another object of the present invention is to provide a method for determining welding defects in a vehicle seat frame that can capture images of various welded areas with as little camera as possible when acquiring image data.

Another purpose of the present invention is to provide a method for determining welding defects in a vehicle seat frame that integrates welding process operation data, sensor data, and image data.

To achieve the above-described task, the method of determining welding defects in vehicle seat frames through machine learning based on image data includes the steps of acquiring learning image data using good welding data and/or bad welding data, and determining good or bad status according to the data through AI. A step of applying machine learning (deep learning) to the platform, a step of acquiring comparison target data, which is data of each welding area of the seat frame to determine welding defects, a step of analyzing and comparing the comparison target image data through an AI platform, the comparison It includes the step of determining whether the welded portion of the seat frame with target image data is defective or good, and the step of displaying the judgment result and, if defective, which portion is defective.

Preferably, in the step of acquiring image data to be studied or compared, the Autocoder algorithm and/or YOLOv5 algorithm is used.

Preferably, the step of acquiring the image data to be studied or compared includes rotating the seat frame, pantilting the camera, and using a camera capable of taking 360 degrees to an inner welded portion of the seat frame. It includes at least one step among the steps for setting up for shooting.

Additionally, it further includes the step of acquiring welding process operation data including welding pressure, current, voltage, and energization time at each welding site, and sensor data from temperature and vibration sensors.

When inspecting welding defects in a vehicle seat frame with multiple welded areas, the minimum number of cameras is set so that there are no blind sides of the welded areas to obtain image data for learning and comparison, so the inspector can visually inspect the welded area without any additional data. The same reliability as when determining defects in welded areas can be obtained. In addition, welding defects can be determined quickly in about 2 minutes per product with pre-setting, contributing to increased productivity.

1 is a diagram illustrating the concept of the present invention.
Figure 2 is a schematic illustration of acquiring data on each welding area from multiple cameras.
Figure 3 is a schematic diagram showing the operation of a camera incorporating the pantilt algorithm of the present invention.
Figure 4 is a diagram showing the operation of the inner camera of the present invention.
Figure 5 is a diagram illustrating the concept of integrating sensor data and operational data of the present invention with image data of the present invention.

Matters regarding the operational effects, including the technical configuration for the above-mentioned purpose of the method for determining welding defects in a vehicle seat frame according to the present invention, will be clearly understood through the detailed description below with reference to the drawings showing preferred embodiments of the present invention.

Additionally, in describing the present invention, if it is determined that a detailed description of related known technologies may unnecessarily obscure the gist of the present invention, the detailed description will be omitted.

The detailed description of the present invention described below refers to the accompanying drawings, which show by way of example specific embodiments in which the present invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It should be understood that the various embodiments of the invention are different from one another but are not necessarily mutually exclusive. For example, specific shapes, structures and characteristics described herein with respect to one embodiment may be implemented in other embodiments without departing from the spirit and scope of the invention. Additionally, it should be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the present invention.

Accordingly, the detailed description that follows is not intended to be taken in a limiting sense, and the scope of the invention is limited only by the appended claims, together with all equivalents to what those claims assert, if properly described. Similar reference numbers in the drawings refer to identical or similar functions across various aspects.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings in order to enable those skilled in the art to easily practice the present invention.

Figure 1 shows a conceptual diagram of the present invention. Referring to Figure 1, in order to implement the present invention, hardware such as various data, collection servers, AI platforms, cameras, and displays, and software to operate the above-described hardware are required.

The present invention utilizes such hardware and software and includes a step of acquiring image data from a data collection server, that is, a step of acquiring training image data. And the acquired data is data for welding defects and/or good welding, and these data are stored in the data collection server. It involves forming big data through an AI platform and performing machine learning (deep learning).

Next, it includes a step of acquiring data on the seat frame for determining welding defects and each welded portion of the seat frame to be compared, that is, a step of acquiring image data to be compared. It includes transmitting these video data to the AI platform.

It includes a step of comparing and analyzing the training video data and the video data to be compared through an AI platform to determine whether the weld is defective or good, and a step of visualizing the judgment result and which part is defective through a display and notifying the inspector.

In the present invention, when inspecting welding defects in a large number of welded areas, at least 50 or more products, there is a risk that the inspector may miss a certain area, but the present invention has no such concerns. Additionally, since a defective seat frame is delivered to the welder along with information on the defective part, the welder can save time and effort in finding the defective part again.

Additionally, it may include a step where the inspector rechecks the judgment of important welded areas such as seat belt brackets. The image data of the seat belt bracket is displayed automatically.

The present invention proceeds in the following steps: data collection → data processing and listing → data analysis → defect confirmation → display method, and the process can be stored at a certain period and analyzed at a certain period to determine the status of the welding equipment.

The step of acquiring the above-described image data from a data collection server typically includes setting a sheet frame at a designated location and setting a plurality of cameras to match the welded portion of the sheet frame to be photographed.

The image data acquisition method of the present invention will be described with reference to FIG. 2. Figure 2 is a schematic diagram for acquiring data on each welded area from multiple cameras. As shown in Figure 2, as an example, four cameras are set and fixed in advance to ensure that there are no blind spots at each welding area, the sheet frame is seated in a designated position and direction, and images are taken in sectors 1-4 directions. Then, you can obtain data by rotating the seat frame 180 degrees and photographing sectors 5 and 6. In contrast, data can be obtained by installing 6 cameras in each of sector 1-6 directions. However, in order to cover all welded areas with a minimum number of cameras, the present invention includes a means and method for rotating the seat frame, as shown in FIG. 2.

Unlike FIG. 2, FIG. 3 is a schematic perspective view for acquiring data on each welding area from a plurality of cameras. As shown, the distance to the seat frame to be photographed is initially set, and the camera has a built-in pantilt algorithm that can move at various angles, allowing multiple welding areas to be photographed with a single camera.

In the seat frame of the present invention, there are welded areas on the inner side, and in order to obtain image data of these welded areas, a camera installed on the inside and capable of shooting 360 degrees is provided. As shown in FIG. 4, the inner camera moves and is fixed in the empty space of the seat frame through a rod that can move up and down, and photographs the welded areas on the inside in a 360-degree direction. Since the transparent support plate of the seat frame fixes and rotates the seat frame, the movable rod preferably moves from top to bottom as shown.

As an example of the present invention, about 15-20 cameras are needed to photograph about 100 welded areas without blind spots. It also includes lighting equipment for camera shooting. The camera of the present invention may be a 2D or 3D imaging device.

In the treadmill step described above, the present invention uses the Autocoder algorithm or YOLOv5 algorithm. The Autocoder algorithm is an execution behavior of self-supervised learning, but it can search for important features. This is because the YOLOv5 algorithm is excellent at distinguishing data. In order to increase the accuracy of judgment in the YOLOv5 algorithm, the amount of labeling data is increased. In other words, the amount of learning data must be large.

In the case of the Autocoder algorithm in determining welding defect data, a simple Encoder and Decoder are implemented using the model Pytorch (i.e. Encoder: input layer, laten vector, Decoder: latent vector, output layer). The welded area of the seat frame is captured using the YOLOv5 model. In this case, the defects that frequently occur in the field are much more likely to be inappropriate bead size or location than puncture or sludge, so good results are obtained when inspecting the bead size and location.

In the case of the YOLOv5 algorithm, when taking pictures, the same position is always cropped from the data on the premise that the pictures are always taken at the same distance and at the same angle. The normal position of the bead is set at the exact center of the crop area, and if it deviates from the center line by a certain distance, it is judged as defective.

In the above-described embodiment of the present invention, all bad and good data are used as learning data, but after training the model only with good data (normal data), fake data can be detected.

In the method of determining welding defects in vehicle seat frames using only image data of the present invention, products judged to be good are substantially 100% good, and among products judged to be defective, about 10% are good. Therefore, the reliability of judgment using the method of the present invention is quite high.

Figure 5 shows another embodiment of the present invention. Description of the same parts as in FIG. 1 will be omitted here. The present invention further includes the step of acquiring welding process operation data including welding pressure, current, voltage, and energization time at each welding site, and sensor data from temperature and vibration pressure sensors. Welding process operation data and sensor data are collected using methods used in the prior art, and detailed descriptions thereof will be cited from the prior art and detailed descriptions thereof will be omitted here.

As an example of the present invention, welding process operation data including welding pressure, current, voltage, and energization time are transmitted and stored as MOV commands through XG5000 by installing a communication unit in the welding robot signal PLC, and the data is processed and stored. Detect welding defects by saving in TXT format.

As an example of the present invention, vibration sensor data is obtained by attaching a 3-axis vibration sensor to a robot arm to detect vibration of the robot arm during a welding process, and the data is transmitted, processed, and stored wirelessly through Haduino coding. Vibration sensor data can detect deviation from the welding position and normal operation of the robot arm.

According to the present invention, a more accurate judgment can be made because welding process operation data, sensor data, and image data are integrated into machine learning to determine whether the welded part is defective. Defects in welded areas of the seat frame can be corrected up to almost 100%. Therefore, seat frames that have passed this test process have high reliability in quality. In addition, it contributes to increased productivity as it can be quickly judged in about 2 minutes per product in a pre-set state.

In the above, preferred embodiments of the present invention have been shown and described, but the present invention is not limited to the specific embodiments described above, and may be used in the technical field to which the invention pertains without departing from the gist of the invention as claimed in the claims. Of course, various modifications can be made by those skilled in the art, and these modifications should not be understood individually from the technical idea or perspective of the present invention.

Although the present invention is limited to a seat frame as an example, it can be widely applied to various products including multiple welded parts.

Claims (6)

In the method of determining welding defects in vehicle seat frames through machine learning based on image data,
Obtaining learning image data using good welding data or/and bad welding data,
A step of machine learning (deep learning) on the AI platform for good and bad status according to the data,
Obtaining comparative data, which is data on each welded part of the seat frame to determine welding defects,
Analyzing and comparing the video data to be compared through an AI platform,
A step of determining whether the welding portion of the seat frame with the image data to be compared is defective or good;
A method for determining welding defects in a vehicle seat frame including the step of displaying the judgment result and, if defective, which part is defective.
According to paragraph 1,
A method for determining welding defects in a vehicle seat frame using the Autocoder algorithm and/or YOLOv5 algorithm in the step of acquiring image data for learning or comparison.
According to paragraph 1,
The step of acquiring image data to be studied or compared includes rotating the seat frame, pantilting the camera, and using a camera capable of 360-degree shooting to photograph the inner welded portion of the seat frame. A method for determining welding defects in a vehicle seat frame that includes at least one of the setting steps.
According to any one of claims 1 to 3,
A method for determining welding defects in a vehicle seat frame further comprising the step of acquiring welding process operation data including welding pressure, current, voltage, and energization time at each welding site, and sensor data from temperature and vibration sensors.
According to clause 4,
The welding process operation data includes welding pressure, current, voltage, and energization time, and is transmitted and stored as a MOV command through XG5000 by installing a communication unit in the welding robot signal PLC. The data is processed and saved in TXT format. A method for determining welding defects in vehicle seat frames that detects welding defects.
According to clause 4,
The sensor data includes vibration sensor data, and the vibration sensor data is obtained by attaching a 3-axis vibration sensor to the robot arm to detect vibration of the robot arm during the welding process, and transmitting the data wirelessly through Haduino coding. Method for determining welding defects in vehicle seat frames that are processed and stored.