Method, processor and system for detecting stud welding quality

Abstract

The invention discloses a method, a processor and a system for detecting the welding quality of a stud. The method comprises the following steps: acquiring an image of a target detection area; the obtained target detection area image is input into a pre-trained detection model, and target stud welding information of the target detection area is recognized; respectively comparing the welding information of the identified target stud with corresponding standard stud welding information preset in a detection model; whether the target stud is qualified or not is determined according to the comparison result of the welding information of the target stud and the welding information of the standard stud; and outputting the quality state information of the target stud in the target detection area. The method has the advantages of being low in cost, high in efficiency, large in detection coverage rate, high in accuracy and good in human-to-effect ratio.

Description

Technical Field

[0001] The present invention relates to the technical field of stud welding, and in particular to a method, a processor and a system for detecting stud welding quality. Background Art

[0002] Stud welding is a common welding process in the body manufacturing process. The stability of stud welding quality is a key component of body quality. Its quality dimensions include: whether the number of welds meets the requirements, whether the welds have obvious qualified characteristics, and whether the position of the welds deviates. The product design drawings will specify the number of welds, the qualified welds have weld nodules and black radial shading visually, and the product design drawings will specify the positioning of each stud.

[0003] At present, the quality of body chassis stud welding, i.e. welding quantity, welding characteristics, and welding position, is usually inspected by manual visual inspection, traditional 2D visual tracing inspection, or 3D inspection camera. However, these inspection methods have the following disadvantages:

[0004] Manual visual inspection and the cost of repairing stud welding defects are extremely high. When a single person is assigned to a single vehicle, the inspection time exceeds the working hours of the workstation (35 seconds), and the inspection coverage cannot reach 100%. In order to inspect the stud welding process, 2-3 people are often required, and the labor efficiency ratio is low. At the same time, manual inspection has errors. Once the defects are leaked, the cost of repair is extremely high, which greatly increases the company's vehicle production costs and causes losses and waste.

[0005] Traditional 2D visual tracing inspection takes a long time for a single recognition inspection and is limited by the camera's field of view, making it impossible to perform a one-time inspection of the stud welds on most vehicle body chassis.

[0006] 3D inspection camera inspection. The 3D inspection camera is generally installed on a robotic arm. As the robotic arm inspects the welding samples point by point, the inspection method determines that the inspection time of a single vehicle is relatively long, which greatly exceeds the production rhythm of a single vehicle. Therefore, 3D inspection needs to be carried out point by point away from the production line and cannot adapt to the production rhythm. Summary of the invention

[0007] In order to solve at least one of the above technical problems, the present invention provides a method, processor and system for detecting the quality of stud welding. Through a pre-trained detection model, the target stud welding information of the target detection area is identified, and the welding information of the identified target stud is compared with the corresponding standard stud welding information pre-set in the detection model to determine whether the target stud is qualified.

[0008] The present invention solves the technical problem by adopting the following technical solution:

[0009] A method for detecting stud welding quality, comprising:

[0010] Step 100: Acquire an image of the target detection area;

[0011] Step 200: input the acquired target detection area image into a pre-trained detection model to identify target stud welding information of the target detection area, wherein the target stud welding information includes the number of studs, the coordinates of each stud, the column state of each stud, and the welding trace state of each stud;

[0012] Step 300, comparing the identified welding information of the target stud with the corresponding standard stud welding information preset in the detection model;

[0013] Step 400: Determine whether the target stud is qualified based on the result of comparing the welding information of the target stud with the welding information of the standard stud;

[0014] Step 500: Output target stud quality status information of the target detection area.

[0015] Furthermore, the method for obtaining a pre-trained detection model comprises the following steps:

[0016] Step 201: Acquire a true value data set of stud samples, wherein the true value data set of stud samples includes a standard coordinate set and a standard pixel set of the stud samples;

[0017] Step 202: input the acquired stud sample true value data set and multiple training images into a pre-configured original image recognition neural network model, perform training optimization on the multiple training images, and obtain an optimized image recognition neural network model;

[0018] Step 203, perform functional verification on the optimized image recognition neural network model. If the verification is qualified, the optimized image recognition neural network model is used as a detection model. If the verification is unqualified, repeat step 202 to continue training optimization.

[0019] Furthermore, the method for obtaining a true value dataset of stud samples comprises the following steps:

[0020] Step 2011, using an image acquisition device to acquire multiple stud sample images;

[0021] Step 2012: identifying qualified studs in the stud sample image;

[0022] Step 2013: use annotation software to annotate the coordinates of qualified studs, the pixel features of the studs, and the pixel features of the weld traces until all studs welded on the target assembly line parts are annotated to obtain a standard coordinate set and a standard pixel set of the stud samples.

[0023] Furthermore, the training images include stud sample images, background sample images and stud-free noise sample images.

[0024] Furthermore, the method for obtaining training images includes: using an image acquisition device to shoot and sample images of target detection areas of assembly line parts, the shooting angle includes 45°-135°, the shooting time includes 7 o'clock, 12 o'clock and 19 o'clock, and the sampling quantity includes 1000-2000 images.

[0025] Furthermore, the method for obtaining an optimized image recognition neural network model comprises the following steps:

[0026] Step 2021, constructing a basis function f(X) of the prediction value of the stud sample image, and obtaining the true value Y of the stud sample according to the true value data set of the stud sample;

[0027] Step 2022: construct a loss function L(X) of the deviation between the true value and the predicted value, L(X)=Yf(X);

[0028] Step 2023, setting the learning rate lr, and obtaining the iterative function fi(X) of the prediction value of the stud sample image according to the formula fi(X)=(1-lr)^(i-1)*f(X), where i=1, 2, 3, ..., n;

[0029] Step 2024, iteratively calculate the stud sample image according to the formula Li(X)=Y-fi(X) to obtain the loss value Li(X);

[0030] Step 2025, obtaining the absolute minimum value Lmin(X) of Li(X) within the preset cycle round n times, and using the iterative function f′(X) of the predicted value of the stud sample image corresponding to Lmin(X) as the optimized image recognition neural network model.

[0031] Further, the method for determining whether the target stud is qualified includes:

[0032] When the offset of the target stud does not exceed 10%, the coordinates of the target stud are judged to be qualified;

[0033] When the target stud has a column, the column of the target stud is determined to be qualified;

[0034] When the target stud has a black weld mark, the weld mark of the target stud is determined to be qualified;

[0035] When the coordinates, cylinder and weld trace of the target stud are all qualified, the target stud is judged to be qualified.

[0036] Further, the method of outputting the target stud quality status information of the target detection area includes:

[0037] The target assembly line part image and each target stud set on the target assembly line part image are displayed on the display screen. When the target stud is qualified, the target stud is marked with a first frame; when the target stud is unqualified, a second frame different from the first frame is used to display the target stud position.

[0038] Furthermore, a processor for detecting stud welding quality is applied to the above-mentioned method for detecting stud welding quality, comprising:

[0039] An input module, used to input the acquired target detection area image into a pre-trained detection model;

[0040] An identification module, used to identify target stud welding information of a target detection area, wherein the target stud welding information includes the number of studs, the coordinates of each stud, the column state of each stud, and the welding trace state of each stud;

[0041] A comparison module, used to compare the welding information of the identified target stud with the corresponding standard stud welding information pre-set in the detection model;

[0042] A determination module, used to determine whether the target stud is qualified according to the result of comparing the welding information of the target stud with the welding information of the standard stud;

[0043] The output module is used to output the target stud quality status information in the target detection area.

[0044] Furthermore, a system for detecting the quality of stud welding includes the processor described above, and also includes an image acquisition device, a light, a display screen and a controller, wherein the image acquisition device and the processor are electrically connected to the controller, and the image acquisition device and the display screen are electrically connected to the processor; when the output module outputs that the target stud has serious quality abnormalities, the processor sends a stop line signal to the controller, and the controller controls to stop production; the serious quality abnormality includes at least one of leaking welding, welding deviation or poor welding.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] This method uses a pre-trained detection model to identify the target stud welding information in the target detection area, and compares the welding information of the identified target studs with the corresponding standard stud welding information pre-set in the detection model to determine whether the target studs are qualified; it has low cost, high efficiency, large detection coverage, high accuracy, and good labor efficiency. Compared with traditional manual detection solutions, the detection error is reduced to 0.35%; compared with traditional 2D visual tracing detection, the recognition time is greatly shortened; compared with 3D detection camera detection, this method uses overall part detection rather than point-by-point, offline detection, which greatly meets the production rhythm requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to better understand the above and other purposes, features, advantages and functions of the present invention, reference may be made to the embodiments shown in the accompanying drawings. The same reference numerals in the accompanying drawings refer to the same components. It should be understood by those skilled in the art that the accompanying drawings are intended to schematically illustrate the preferred embodiments of the present invention and have no limiting effect on the scope of the present invention, and the components in the drawings are not drawn to scale.

[0048] Figure 1 The present invention is a flow chart of the method for detecting stud welding quality.

[0049] Figure 2 The block diagram of the processor for detecting the stud welding quality of the present invention.

[0050] Figure 3 The figure is a schematic diagram of the structure of the system for detecting the stud welding quality of the present invention.

[0051] Figure 4 The block diagram of the system for detecting stud welding quality of the present invention.

[0052] In the figure: 1-processor; 11-input module; 12-recognition module; 13-comparison module; 14-determination module; 15-output module; 2-body chassis; 3-camera; 4-light; 5-controller; 6-display screen. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the exemplary embodiments of the present invention are described below in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0054] In the description of the present invention, it should be noted that the term "including" and its variations represent open inclusion, that is, "including but not limited to". The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance. The term "and / or" when used to list two or more items means that any one of the listed items can be used, or any combination of two or more of the listed items can be used. In addition, in the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0055] The present invention provides a method for detecting the quality of stud welding, such as Figure 1 As shown, including:

[0056] Step 100: Acquire an image of the target detection area;

[0057] Step 200: input the acquired target detection area image into a pre-trained detection model to identify target stud welding information of the target detection area, wherein the target stud welding information includes the number of studs, the coordinates of each stud, the column state of each stud, and the welding trace state of each stud;

[0058] Step 300, comparing the identified welding information of the target stud with the corresponding standard stud welding information preset in the detection model;

[0059] Step 400: Determine whether the target stud is qualified based on the result of comparing the welding information of the target stud with the welding information of the standard stud;

[0060] Step 500: Output target stud quality status information of the target detection area.

[0061] The target detection area is the entire area of ​​the target flow part image or a partial area with the target stud. For example, the vehicle chassis 2 can be divided into two areas, each of which is provided with a camera 3. By inputting the target detection area image into the pre-trained detection model, it can be determined whether the target stud is qualified. The output target stud quality status information of the target detection area at least includes abnormal stud information. Preferably, the quality status information of each target stud is output, including qualified studs and abnormal studs. When the target stud is abnormal, a warning message can also be issued.

[0062] Specifically, the method for obtaining a pre-trained detection model includes the following steps:

[0063] Step 201: Acquire a true value data set of stud samples, wherein the true value data set of stud samples includes a standard coordinate set and a standard pixel set of the stud samples;

[0064] Step 202: input the acquired stud sample true value data set and multiple training images into a pre-configured original image recognition neural network model, perform training optimization on the multiple training images, and obtain an optimized image recognition neural network model;

[0065] Step 203, perform functional verification on the optimized image recognition neural network model. If the verification is qualified, the optimized image recognition neural network model is used as a detection model. If the verification is unqualified, repeat step 202 to continue training optimization.

[0066] After obtaining the optimized image recognition neural network model, multiple stud sample images can be detected through the model and compared with manual detection for verification; if the detection results are consistent, it means that the model has passed the verification and the optimized image recognition neural network model can be used as a detection model. If the verification fails, the original image recognition neural network model and / or learning rate need to be modified and optimized, and step 202 should be repeated to continue training optimization.

[0067] The stud sample images used for verification need to include various stud welding information, including images of all studs in the target detection area that are qualified, images of stud coordinates that are offset (images that change the welding position of the stud and weld the stud off-center), images of missed welds (images of studs not welded at the specified position), and images of poor welding (images of retaining black welds and removing silver studs, or retaining silver studs and removing black welds). If the welding information of each stud predicted by the pre-trained detection model is consistent with the actual welding information, then the optimized image recognition neural network model is accurate and reliable as a detection model.

[0068] Specifically, the method for obtaining a true value dataset of stud samples includes the following steps:

[0069] Step 2011, using an image acquisition device to acquire multiple stud sample images;

[0070] Step 2012: identifying qualified studs in the stud sample image;

[0071] Step 2013: use annotation software to annotate the coordinates of qualified studs, the pixel features of the studs, and the pixel features of the weld traces until all studs welded on the target assembly line parts are annotated to obtain a standard coordinate set and a standard pixel set of the stud samples.

[0072] Acquiring the stud sample data set includes acquiring the standard coordinate set and the standard pixel set of the welding stud. The stud sample image is a target assembly line part image with a target stud. The sample is photographed and sampled on the production line. The coordinates are collected by a fixed-position image acquisition device. The first pixel coordinate in the upper left corner of the fixed-position photo is set as the coordinate 0 point. The coordinates of each stud can be annotated according to the coordinate 0 point. At the same time, the pixels of the column and weld mark of each stud are annotated by the RGB channel value.

[0073] The standard pixel set may also include pixel information of unqualified studs. When the qualified studs are annotated with RGB values, the unqualified studs may also be annotated with RGB values, providing reverse judgment comparison.

[0074] Preferably, the training images include stud sample images, background sample images and stud-free noise sample images, which can improve target recognition accuracy and enhance robustness. The stud sample images are vehicle chassis 2 images with target studs; the background sample images are environment images without vehicle chassis 2 under different lighting conditions; and the stud-free noise sample images are vehicle chassis 2 images without target studs.

[0075] Specifically, the method for obtaining training images includes: using an image acquisition device to shoot and sample images of target detection areas of assembly line parts, wherein the shooting angles include 45°-135°, and the shooting angle range is large, which can ensure that the training set contains various postures of studs and improve the target recognition accuracy; the shooting times include 7 o'clock, 12 o'clock and 19 o'clock, and shooting at different times can ensure the richness of training set pixels and improve the target recognition accuracy; the sampling number includes 1000-2000, and the quality of model training can be improved by increasing the sample capacity.

[0076] Specifically, the method for obtaining an optimized image recognition neural network model includes the following steps:

[0077] Step 2021, constructing a basis function f(X) of the prediction value of the stud sample image, and obtaining the true value Y of the stud sample according to the true value data set of the stud sample;

[0078] Step 2022: construct a loss function L(X) of the deviation between the true value and the predicted value, L(X)=Yf(X);

[0079] Step 2023, setting the learning rate lr, and obtaining the iterative function fi(X) of the prediction value of the stud sample image according to the formula fi(X)=(1-lr)^(i-1)*f(X), where i=1, 2, 3, ..., n;

[0080] Step 2024, iteratively calculate the stud sample image according to the formula Li(X)=Y-fi(X) to obtain the loss value Li(X);

[0081] Step 2025, obtaining the absolute minimum value Lmin(X) of Li(X) within the preset cycle round n times, and using the iterative function f′(X) of the predicted value of the stud sample image corresponding to Lmin(X) as the optimized image recognition neural network model.

[0082] The attention mechanism is used to extract the stud features of the image, such as the coordinates of each stud, the column state of each stud, and the welding trace state of each stud, and the various features are fused to generate a comprehensive target stud welding information set to improve processing efficiency; the dataset samples are changed by changing the image resolution through upsampling or downsampling, downsampling is performed during feature extraction, and the high-resolution image is downsampled to a low-resolution image; upsampling is performed during feature fusion, and the low-resolution image is upsampled to a high-resolution image, thereby improving the stud target accuracy and obtaining a stud training image sample dataset.

[0083] The stud sample true value data set includes at least two groups of content, content one input X, content two output Y, the relationship between the two can be one-to-one, such as: {(1,1), (2,3), (4,4), ..., (1012,999)}, or many-to-one, such as: {(1,2,3,4,5,6,7,8,9, ..., 9999, 10000, 10001, ..., 293475), (0,0,0,0,0,0,0,0,0, ..., 1,1,1, ..., 0)}, etc. In a cycle, the multiple txp files generated by the collected training images and the training images are simultaneously input into the function fi (X) for classification regression, and then compared with Y to obtain the loss value Li (X).

[0084] In each cycle, the basic function f(X) can be fine-tuned by the learning rate lr according to the formula fi(X)=(1-lr)^(i-1)*f(X), that is, the original image recognition neural network model is optimized and adjusted to reduce the deviation. For example, set the cycle to 200 rounds, calculate L1(X)=Y-f1(X) for the first time, calculate L2(X)=Y-(1-lr)f(X) for the second time, until L200(X)=Y-(1-lr)^199*f(X). If the absolute value of the ideal state loss function L200(X) is the smallest at this time, the basic function has been iterated to f′(X)=(1-lr)^199*f(X). The overall prediction of the function for the input X has the smallest overall deviation from the true output value Y, and the function f′(X) is the best prediction function. When the absolute minimum value Lmin(X) of Li(X) appears before 200 rounds, the corresponding iterative function f′(X) is used as the optimized image recognition neural network model. For example, if Lmin(X) appears in the 185th round, f′(X)=(1-lr)^184*f(X) is used as the best prediction function.

[0085] Specifically, the method for determining whether the target stud is qualified includes:

[0086] When the offset of the target stud does not exceed 10%, the coordinates of the target stud are judged to be qualified;

[0087] When the target stud has a column, the column of the target stud is determined to be qualified;

[0088] When the target stud has a black weld mark, the weld mark of the target stud is determined to be qualified;

[0089] When the coordinates, cylinder and weld trace of the target stud are all qualified, the target stud is judged to be qualified.

[0090] Standard stud welding information includes quantity, coordinates, cylinder status, and weld mark status information. Qualified studs have accurate positions, and pixel features are silver cylinders and black weld marks. Unqualified studs have offset coordinates, and pixel features are missing silver cylinders or missing black weld marks.

[0091] When judging whether the stud coordinates are qualified, the pre-trained detection model is used to compare the stud coordinates of the real-time image with the pre-marked standard stud coordinates one by one. If each coordinate in the standard coordinate set can find a corresponding coordinate that does not exceed the deviation range in the coordinate result of the real-time image, then the number and coordinates of the studs of the currently detected body chassis 2 are qualified; if any coordinate in the standard coordinate set cannot find a corresponding coordinate in the coordinate result of the real-time image, or there is a corresponding coordinate that exceeds the deviation range, then the stud welding of the currently detected body chassis 2 is unqualified, and this coordinate in the standard coordinate set is output. If the X coordinate and / or Y coordinate of the stud is offset by more than 10%, it is considered that the stud coordinate is offset.

[0092] When judging whether the status of the cylinder and the weld mark is qualified, the real-time image can be input into the pre-trained detection model, and the optimized image recognition neural network model is used to judge whether the RGB channel value of the real-time image conforms to the function f′(X), that is, the RGB channel value of the real-time image is compared with the optimized image recognition neural network model f′(X). If it conforms, the status of the cylinder and the weld mark is qualified, and the anchor frame size and coordinates are output.

[0093] In one embodiment, the method of outputting target stud quality status information of a target detection area includes:

[0094] The target assembly line part image and each target stud set on the target assembly line part image are displayed on the display screen. When the target stud is qualified, the target stud is marked with a first frame; when the target stud is unqualified, a second frame different from the first frame is used to display the target stud position.

[0095] The target assembly line part image can display the complete target assembly line part image, or display the outline of the target assembly line part, and display the target stud at each corresponding position of the target assembly line part. When the target stud is qualified, the target stud is marked with the first frame, for example, it can be marked with a green rectangular frame. When the target stud is unqualified, it is displayed at the target stud position using a second frame different from the first frame. The second frame can be a different color from the first frame, and the second frame can also flash to remind, and a buzzer sound or other warning reminder can also be issued. For example, when the target stud is welded off-center, the target stud is displayed at the target stud position, and the target stud is marked with a red rectangular frame; when the target stud is missing, the target stud is not displayed at the target stud position, and the target stud position is still marked with a red rectangular frame. When the target stud is marked with the first frame and the second frame, the corresponding stud status information can also be displayed at the same time, such as qualified, missing, offset, etc. In this way, according to the quality status of the target studs, each target stud is marked with the first frame or the second frame, and the sum of the number of first frames and second frames is equal to the number of target studs. The operator can clearly and quickly understand the welding status of each stud and quickly locate the stud with quality problems.

[0096] This method for detecting the quality of stud welding has low detection cost, high efficiency, large detection coverage, high accuracy, and good labor efficiency ratio. Compared with the traditional manual detection scheme, the detection error is reduced to 0.35%; compared with the traditional 2D visual tracing detection, the recognition time is greatly shortened; compared with the 3D detection camera 3 detection, this method uses overall part detection rather than point-by-point, offline detection, which greatly meets the production rhythm requirements.

[0097] The present invention also provides a processor 1 for detecting stud welding quality, which is applied to the above-mentioned method for detecting stud welding quality, such as Figure 2 As shown, the processor 1 includes:

[0098] An input module 11 is used to input the acquired target detection area image into a pre-trained detection model;

[0099] An identification module 12, used to identify target stud welding information of a target detection area, wherein the target stud welding information includes the number of studs, the coordinates of each stud, the column state of each stud, and the welding trace state of each stud;

[0100] A comparison module 13, used to compare the welding information of the identified target stud with the corresponding standard stud welding information preset in the detection model;

[0101] A determination module 14 is used to determine whether the target stud is qualified according to the result of comparing the welding information of the target stud with the welding information of the standard stud;

[0102] The output module 15 is used to output the target stud quality status information in the target detection area.

[0103] The processor 1 can use a GPU processor 1, reserve a real-time image transmission entrance in the stud welding image recognition neural network program, input the target detection area image into the pre-trained detection model, and run it in the GPU processor 1 to detect the number, coordinates, column state and welding trace state of the studs. After 0.03~0.05ms, the target stud quality state information can be output through the output module, and the detection is fast, accurate and efficient.

[0104] The present invention also provides a system for detecting the quality of stud welding, referring to Figure 3 and Figure 4 As shown, it includes the processor 1 mentioned above, and also includes an image acquisition device, a light 4, a display screen 6 and a controller 5. The image acquisition device and the processor 1 are both electrically connected to the controller 5, and the image acquisition device and the display screen 6 are both electrically connected to the processor 1; when the output module outputs a target stud with serious quality abnormalities, the processor 1 sends a stop line signal to the controller 5, and the controller 5 controls to stop production; the serious quality abnormality includes at least one of leaking welding, welding deviation or poor welding.

[0105] The image acquisition device is used to acquire the target stud image. An industrial camera or camera 3 can be used. At a shooting distance of 2.5m, the industrial camera or camera 3 is used to shoot and sample the 2m*4m body chassis 2 parts. The lighting 4 can use a light source with an illumination of more than 2000lux, installed at the detection station, to provide lighting so that the camera 4 can collect clear images. The controller 5 uses a PLC controller, and the camera 3 and the processor 1 establish communication with the controller 5. When the controller 5 receives a signal that the assembly line parts of the body chassis 2 have arrived at the station, the camera 3 is driven to shoot a real-time image of the stud welding of the body chassis 2 under the light intensity of 2000lux of the lighting 4; the real-time image is input into the pre-trained detection model of the processor 1 through the reserved real-time image transmission entrance for detection. According to the detection results, the target stud quality status information is directly output through the display screen 6. From the display screen 6, it can be directly observed whether there are abnormalities in the number and position of the studs, and whether there are abnormalities in the column state and the welding trace state. When the target studs in the target detection area have serious quality abnormalities, the processor 1 sends a stop line signal to the controller 5 to stop production. In this way, the stud welding condition of each assembly line part of each chassis 2 can be conveniently and quickly inspected. When unqualified studs are detected, they can be reworked in time to avoid defective products flowing into other processes and affecting the production rhythm.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting the quality of stud welding, characterized in that: include: Step 100: Acquire an image of the target detection area; Step 200: input the acquired target detection area image into a pre-trained detection model to identify target stud welding information of the target detection area, wherein the target stud welding information includes the number of studs, the coordinates of each stud, the column state of each stud, and the welding trace state of each stud; Step 300, comparing the identified welding information of the target stud with the corresponding standard stud welding information preset in the detection model; Step 400: Determine whether the target stud is qualified based on the result of comparing the welding information of the target stud with the welding information of the standard stud; Step 500: Output target stud quality status information of the target detection area.

2. The method for detecting stud welding quality according to claim 1, characterized in that: The method for obtaining a pre-trained detection model includes the following steps: Step 201: Acquire a true value data set of stud samples, wherein the true value data set of stud samples includes a standard coordinate set and a standard pixel set of the stud samples; Step 202: input the acquired stud sample true value data set and multiple training images into a pre-configured original image recognition neural network model, perform training optimization on the multiple training images, and obtain an optimized image recognition neural network model; Step 203, perform functional verification on the optimized image recognition neural network model. If the verification is qualified, the optimized image recognition neural network model is used as a detection model. If the verification is unqualified, repeat step 202 to continue training optimization.

3. The method for detecting stud welding quality according to claim 2, characterized in that: The method for obtaining the true value dataset of stud samples includes the following steps: Step 2011, using an image acquisition device to acquire multiple stud sample images; Step 2012: identifying qualified studs in the stud sample image; Step 2013: use annotation software to annotate the coordinates of qualified studs, the pixel features of the studs, and the pixel features of the weld traces until all studs welded on the target assembly line parts are annotated to obtain a standard coordinate set and a standard pixel set of the stud samples.

4. The method for detecting stud welding quality according to claim 3, characterized in that: The training images include stud sample images, background sample images and stud-free noise sample images.

5. The method for detecting stud welding quality according to claim 4, characterized in that: The method for acquiring training images includes: using an image acquisition device to shoot and sample the target detection area image of the assembly line parts, the shooting angle includes 45°-135°, the shooting time includes 7 o'clock, 12 o'clock and 19 o'clock, and the sampling quantity includes 1000-2000 images.

6. The method for detecting stud welding quality according to claim 2, characterized in that: The method for obtaining an optimized image recognition neural network model comprises the following steps: Step 2021, constructing a basis function f(X) of the prediction value of the stud sample image, and obtaining the true value Y of the stud sample according to the true value data set of the stud sample; Step 2022: construct a loss function L(X) of the deviation between the true value and the predicted value, L(X)=Yf(X); Step 2023, setting the learning rate lr, and obtaining the iterative function fi(X) of the prediction value of the stud sample image according to the formula fi(X)=(1-lr)^(i-1)*f(X), where i=1, 2, 3, ..., n; Step 2024, iteratively calculate the stud sample image according to the formula Li(X)=Y-fi(X) to obtain the loss value Li(X); Step 2025, obtaining the absolute minimum value Lmin(X) of Li(X) within the preset cycle round n times, and using the iterative function f′(X) of the predicted value of the stud sample image corresponding to Lmin(X) as the optimized image recognition neural network model.

7. The method for detecting stud welding quality according to claim 1, characterized in that: Methods for determining whether a target stud is qualified include: When the offset of the target stud does not exceed 10%, the coordinates of the target stud are judged to be qualified; When the target stud has a column, the column of the target stud is determined to be qualified; When the target stud has a black weld mark, the weld mark of the target stud is determined to be qualified; When the coordinates, cylinder and weld trace of the target stud are all qualified, the target stud is judged to be qualified.

8. The method for detecting stud welding quality according to claim 1, characterized in that: The method for outputting target stud quality status information of a target detection area includes: The target assembly line part image and each target stud set on the target assembly line part image are displayed on the display screen. When the target stud is qualified, the target stud is marked with a first frame; when the target stud is unqualified, a second frame different from the first frame is used to display the target stud position.

9. A processor for detecting stud welding quality, applied to the method for detecting stud welding quality according to any one of claims 1 to 8, characterized in that: include: An input module, used to input the acquired target detection area image into a pre-trained detection model; An identification module, used to identify target stud welding information of a target detection area, wherein the target stud welding information includes the number of studs, the coordinates of each stud, the column state of each stud, and the welding trace state of each stud; A comparison module, used to compare the welding information of the identified target stud with the corresponding standard stud welding information pre-set in the detection model; A determination module, used to determine whether the target stud is qualified according to the result of comparing the welding information of the target stud with the welding information of the standard stud; The output module is used to output the target stud quality status information in the target detection area.

10. A system for detecting the quality of stud welding, characterized in that: It includes the processor as claimed in claim 9, and also includes an image acquisition device, a light, a display screen and a controller, wherein the image acquisition device and the processor are electrically connected to the controller, and the image acquisition device and the display screen are electrically connected to the processor; when the output module outputs a target stud with serious quality abnormalities, the processor sends a stop line signal to the controller, and the controller controls to stop production; the serious quality abnormality includes at least one of leaking welding, welding deviation or poor welding.

Images