A welding method for automobile steering parts

Abstract

The invention discloses a welding method for automobile steering parts, comprising step S1, a storage module storing welding images, arc spectrum images and average temperature collected at several historical moments; step S2, a processing module processing to obtain a spectrum shift index, a smoke shift index and a distortion shift index; step S3, a training module training to obtain a preliminary detection model; step S4, an optimization module retraining the preliminary detection model according to the spectrum shift index, the smoke shift index and the distortion shift index to obtain an optimized detection model; step S5, an image acquisition module real-time acquisition of real-time welding images, and inputting the real-time welding images into the optimized detection model to obtain defect annotation images; step S6, an image analysis module generating a welding adjustment instruction according to the defect annotation image, and a welding robot dynamically adjusting welding parameters in the process of ball pin welding according to the welding adjustment instruction. The invention improves the ball pin welding quality.

Description

Technical Field

[0001] The invention relates to the technical field of mechanical processing, in particular to a welding method for automobile steering parts. Background Art

[0002] A ball pin is a mechanical connection component, usually used to connect two parts with different movement directions. Its structure is similar to a ball and a groove, allowing one part to rotate and move relative to another part in multiple directions. Ball pins are widely used in automobile suspension systems and automobile steering systems. Ball pins can provide flexible movement and stable connection for automobile suspension and steering. The design of ball pins can effectively reduce friction, improve movement efficiency, and maintain good wear resistance under certain loads.

[0003] Due to the importance of ball pins in the automotive steering system, the welding process of ball pins is strictly controlled. At present, ball pin welding is generally performed automatically by welding robots. The welding robot is controlled by setting the welding program to complete the welding according to the predetermined steps. After the welding is completed, the surface defects need to be detected by an image detection device and the ball pins are screened. This process will screen out a batch of ball pins that fail to meet the welding standards and wait for subsequent re-welding repair or melting and recasting.

[0004] The ball pin welding method in the prior art cannot accurately detect the actual defects on the ball pin surface during the welding process due to the interference of arc light generated during the welding process. Therefore, the welding parameters cannot be adjusted during the welding process, resulting in that the welding quality of the ball pin cannot be well guaranteed. Summary of the invention

[0005] In view of the deficiencies in the prior art, the object of the present invention is to provide a welding method for automobile steering parts, which is used to dynamically adjust the welding parameters of a welding robot during the welding process and improve the welding quality of a ball pin.

[0006] To achieve the above object, the present invention provides the following technical solution: a welding method for automobile steering parts, comprising:

[0007] Step S1, the storage module saves welding images, arc spectrum images and average temperatures collected at several historical moments, and obtains arc spectrum data according to the arc spectrum images, wherein the welding images include smoke images and air distortion images;

[0008] Step S2, the processing module processes the arc light spectrum data to obtain a spectrum shift index, processes the smoke images to obtain a smoke shift index, and processes the air distortion images and the average temperature to obtain a distortion shift index;

[0009] Step S3, the training module annotates the welding defects in each welding image to obtain a plurality of annotated images, and uses each welding image before annotation as an input parameter and the corresponding annotated image as an output parameter to train and obtain a preliminary detection model;

[0010] Step S4, the optimization module adjusts the weight parameters between the input layer and the hidden layer of the preliminary detection model according to the spectrum shift index, the smoke shift index and the distortion shift index, and retrains the model to obtain an optimized detection model;

[0011] Step S5, the image acquisition module acquires a real-time welding image in real time during the welding process of the welding robot on the surface of the ball pin, and inputs the real-time welding image into the optimized detection model to obtain a defect annotation image;

[0012] Step S6, the image analysis module generates a welding adjustment instruction according to the defect annotation image, and the welding robot dynamically adjusts the welding parameters in the ball pin welding process according to the welding adjustment instruction.

[0013] Furthermore, the welding parameters include argon gas flow rate, argon gas flow rate, welding voltage, welding current, welding speed and welding angle.

[0014] Further, the step S6 comprises:

[0015] Step S61, the image analysis module performs defect recognition on the defect-annotated image to obtain a number of surface welding defects;

[0016] Step S62, the image analysis module analyzes each of the surface welding defects to obtain a number of defect parameters;

[0017] Step S63, the image analysis module matches the defect parameters in a preset defect matching database to obtain a corresponding welding adjustment range, and comprehensively generates the welding adjustment instruction according to each welding adjustment range;

[0018] Step S64: the welding robot dynamically adjusts the welding parameters of the ball pin according to the welding adjustment instruction.

[0019] Furthermore, the step S2 comprises:

[0020] Step S21, the processing module acquires a plurality of standard spectrum images, wherein the standard spectrum images are acquired by the image acquisition module before welding the ball pin;

[0021] Step S22, the processing module processes each of the standard spectrum images to obtain standard spectrum data;

[0022] Step S23: the processing module obtains the spectrum shift index according to each of the arc spectrum data and each of the standard spectrum data.

[0023] Further, the step S23 includes:

[0024] Step S231, the processing module respectively selects a plurality of first spectrum characteristic data from each of the arc spectrum data and a plurality of second spectrum characteristic data from each of the standard spectrum data;

[0025] In step S232, the processing module inputs each of the first spectral feature data and each of the second spectral feature data into a preset spectral shift calculation formula to calculate the spectral shift index.

[0026] Furthermore, the spectrum shift calculation formula is configured as:

[0027]

[0028] Among them, S si Used to represent the spectral shift index; k i Used to represent the preset configuration coefficients, each of which is a constant; x i Used to represent each of the first spectrum characteristic data in the arc spectrum data; y i Used to represent each of the second spectrum feature data in the standard spectrum data.

[0029] Furthermore, the step S2 further includes:

[0030] Step S24, the processing module inputs each of the welding smoke images into a preset three-dimensional convolutional network, collects the time series features in each of the welding smoke images, and outputs smoke concentration data;

[0031] Step S25: the processing module obtains the smoke deviation index according to the smoke concentration data.

[0032] Furthermore, the step S2 further includes:

[0033] Step S26, the processing module inputs the average temperature and each of the air distortion images into a pre-trained distortion repair model to obtain a plurality of distortion repair images;

[0034] Step S27, the processing module calculates the difference between each repaired feature point on the distorted repaired image and each distorted feature point on the corresponding air distorted image to obtain a comprehensive difference;

[0035] Step S28: the processing module obtains the distortion offset index according to the comprehensive difference.

[0036] Furthermore, the optimized detection model is trained based on the YOLOV7 network model.

[0037] Beneficial effects of the present invention:

[0038] The present invention obtains a spectral shift index, a smoke shift index and a distortion shift index by processing an arc spectrum image, a smoke image and an air distortion image, and uses the spectral shift index, the smoke shift index and the distortion shift index to adjust the weight parameters of a preliminary detection model and retrain to obtain an optimized detection model. The optimized detection model filters out the influence of arc spectrum, smoke and air distortion caused by high temperature in the welding process on welding defect recognition, so that the recognition accuracy of welding defects in the welding process can be effectively improved; finally, a welding adjustment instruction is generated according to the defect annotation image output by the optimized detection model, and a welding robot dynamically adjusts the welding parameters in the ball pin welding process according to the welding adjustment instruction. The process realizes dynamic adjustment of argon gas flow rate, argon gas flow rate, welding voltage, welding current, welding speed and welding angle in the welding process of the ball pin according to the defects on the surface of the ball pin, so that the welding quality of the ball pin can be significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of steps of a welding method for automobile steering parts in the present invention;

[0040] Figure 2 is a step sub-flow chart of step S6 in the present invention;

[0041] Figure 3 is a step sub-flow chart of step S2 in the present invention;

[0042] Figure 4 It is a step sub-flow chart of step S23 in the present invention. DETAILED DESCRIPTION

[0043] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. The same parts are represented by the same reference numerals. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to directions in the accompanying drawings, and the words "bottom surface" and "top surface", "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0044] Example 1, reference Figure 1 to Figure 2 , which is the first embodiment of the present invention, provides a welding method for automobile steering parts, which can improve the welding quality of ball pins, including:

[0045] Step S1, the storage module saves welding images, arc spectrum images and average temperatures collected at several historical moments, and obtains arc spectrum data according to the processing of each arc spectrum image, wherein the welding image includes a smoke image and an air distortion image;

[0046] Step S2, the processing module obtains a spectrum shift index according to each arc spectrum data, obtains a smoke shift index according to each smoke image, and obtains a distortion shift index according to each air distortion image and average temperature;

[0047] Step S3, the training module annotates the welding defects in each welding image to obtain a number of annotated images, and uses each welding image before annotation as an input parameter and the corresponding annotated image as an output parameter to train and obtain a preliminary detection model;

[0048] Step S4, the optimization module adjusts the weight parameters between the input layer and the hidden layer of the preliminary detection model according to the spectrum shift index, the smoke shift index and the distortion shift index, and retrains the model to obtain an optimized detection model;

[0049] Step S5, the image acquisition module acquires real-time welding images in real time during the welding process of the welding robot on the surface of the ball pin, and inputs the real-time welding images into the optimized detection model to obtain defect annotation images;

[0050] Step S6, the image analysis module generates a welding adjustment instruction according to the defect annotation image, and the welding robot dynamically adjusts the welding parameters in the ball pin welding process according to the welding adjustment instruction.

[0051] Among them, the welding parameters include argon gas flow rate, argon gas flow rate, welding voltage, welding current, welding speed and welding angle.

[0052] Specifically, in this embodiment, the storage module may include a non-volatile memory configured in the intelligent processing terminal, which is used to save the welding image, arc spectrum image and average temperature collected at the historical moment. An image recognition program based on the Python algorithm is configured in the intelligent processing terminal, and the image recognition program recognizes the smoke and distorted air in the welding image respectively to obtain the smoke image and the air distortion image. Among them, the welding image is collected by an industrial array camera at a historical moment. The model of the industrial array camera can be Basler ace 2Basic. The resolution of the industrial array camera can be up to 24 million pixels and the frame rate can be up to 160fps. It is used to realize continuous collection of several welding images according to a preset sampling period. The sampling period of the industrial array camera can be set to 100ms. The arc spectrum image is collected by the spectral imager according to the preset sampling period. The sampling period of the arc spectrum image can be 100ms. The average temperature is obtained by multiple high-precision temperature sensors. The model of the high-precision temperature sensor can be PT100. Among them, multiple high-precision temperature sensors are evenly distributed around the industrial area array camera, and temperature data around the industrial area array camera is collected according to a temperature sampling period. After the temperature data are collected separately, the average value is taken to obtain the average temperature.

[0053] The processing module, training module, optimization module and image analysis module are respectively a data calculation program, a model training program, a model optimization program and an analysis control program pre-configured on the intelligent processing terminal.

[0054] The processing module obtains the spectral shift index according to each arc spectrum data, obtains the smoke shift index according to each smoke image, and obtains the distortion shift index according to each air distortion image and the average temperature. The spectral shift index can be used to reflect the degree of interference of the arc spectrum in the welding process on the welding image, the smoke shift index can be used to reflect the degree of interference of the smoke generated in the welding process on the welding image, and the distortion shift index can be used to reflect the degree of interference of the air distortion in the welding process on the welding image. The training module first pre-labels the welding defects in each welding image to obtain a labeled image. The labeling of the above welding defects is carried out by manual labeling by technicians in this field. This method can ensure a high defect detection accuracy, but the efficiency is very low.

[0055] Before each welding image training module in the storage module, each welding image needs to be preprocessed, where the image preprocessing process includes image screening, image cropping, image position adjustment and image size adjustment. Image screening first filters out a portion of images that obviously do not meet the training requirements, including overexposed images, unclear images of welding defects, etc., to reduce the amount of subsequent calculations. Image cropping, image position adjustment and image size adjustment adjust each welding image to an image of the same size suitable for model training, which can reduce the model prediction deviation caused by the original training image.

[0056] The training module first needs to introduce an original model as the training basis. The original model can be a YOLOV7 network model. Therefore, the optimized detection model is obtained based on the optimized training of the YOLOV7 network model.

[0057] The network structure of YoloV7 can be mainly divided into three parts: Backbone, FPN and Yolo Head. Among them, Backbone is responsible for feature extraction of the input image, and this structure will eventually output three feature layers of different sizes. The FPN network is responsible for strengthening feature extraction, and it will partially fuse the three feature layers extracted by Backbone (to fuse feature information of different sizes, mainly relying on upsampling and downsampling). Yolo Head is responsible for classifying and regressing the anchor and outputting the final prediction box. YOLOv7 is more accurate than YOLOv5 at the same volume, 120% faster (FPS), 180% faster (FPS) than YOLOX, 1200% faster (FPS) than Dual-Swin-T, 550% faster (FPS) than ConvNext, and 500% faster (FPS) than SWIN-L. In the range of 5FPS to 160FPS, YOLOv7 exceeds the currently known detectors in terms of speed and accuracy. In the test on GPU V100, the model with an accuracy of 56.8% AP can achieve a detection rate of more than 30FPS (batch=1). At the same time, this is the only detector that can still exceed 30FPS at such a high accuracy. Therefore, the use of the YOLOV7 network model can initially guarantee the accuracy of defect detection in welding images.

[0058] The preprocessed welding images and the corresponding annotated images are divided into training set, test set and validation set according to the proportion. The original model is retrained with the image data of the training set to obtain the preliminary detection model. After the training process is completed, the prediction accuracy of the preliminary detection model is tested with the image data in the test set. When the prediction accuracy is not higher than 93%, the hyperparameters of the preliminary detection model are adjusted and retrained until the prediction accuracy is higher than 93%. At this time, the defect detection of the preliminary detection model will still be affected by the arc spectrum, smoke and air distortion caused by high temperature. In order to improve the detection accuracy, the optimization module uses the processed spectral shift index, smoke shift index and distortion shift index to adjust the weight parameters between the input layer and the hidden layer of the preliminary detection model in the feature extraction stage and retrain to obtain the optimized detection model.

[0059] The image acquisition module can be an industrial area array camera, which acquires real-time welding images and then inputs the optimized detection model to output defect annotation images.

[0060] The image analysis module identifies and analyzes the defects in the defect annotation image and generates corresponding welding adjustment instructions to dynamically adjust the welding state of the welding robot to the ball pin, thereby significantly improving the welding quality of the ball pin.

[0061] Working principle of embodiment 1:

[0062] In this embodiment, the spectral shift index, smoke shift index and distortion shift index are obtained by processing the arc spectrum image, smoke image and air distortion image, and the spectral shift index, smoke shift index and distortion shift index are used to adjust the weight parameters of the preliminary detection model and retrain to obtain the optimized detection model. The optimized detection model filters out the influence of the arc spectrum, smoke and air distortion caused by high temperature in the welding process on the recognition of welding defects, so it can effectively improve the recognition accuracy of welding defects in the welding process; finally, welding adjustment instructions are generated according to the defect annotation image output by the optimized detection model, and the welding robot dynamically adjusts the welding parameters in the ball pin welding process according to the welding adjustment instructions. This process realizes the dynamic adjustment of argon gas flow rate, argon gas flow rate, welding voltage, welding current, welding speed and welding angle in the welding process of the ball pin according to the defects on the surface of the ball pin, so the welding quality of the ball pin can be significantly improved.

[0063] Preferably, Figure 2 As shown, step S6 includes:

[0064] Step S61, the image analysis module performs defect recognition on the defect-marked image to obtain a number of surface welding defects;

[0065] Step S62, the image analysis module analyzes each surface welding defect to obtain a number of defect parameters;

[0066] Step S63, the image analysis module matches the corresponding welding adjustment range in the preset defect matching database according to each defect parameter, and comprehensively generates a welding adjustment instruction according to each welding adjustment range;

[0067] Step S64: the welding robot dynamically adjusts the welding parameters of the ball pin according to the welding adjustment instruction.

[0068] Specifically, in this embodiment, the image analysis module performs feature recognition on the defects in the defect annotation image based on the defect detection algorithm of machine vision to obtain surface welding defects, wherein the surface welding defects include welding undercut, weld bead, arc pit, surface pore, slag inclusion and surface crack. Then, the image analysis of the identified surface welding defects is performed according to the defect detection algorithm to obtain the corresponding defect parameters, and the defect parameters include defect shape, defect depth and defect size. The defect matching database contains the association relationship between several standard defect parameters and the corresponding welding adjustment range. The defect similarity between the defect parameters obtained by calculation and analysis and the standard defect parameters in the defect matching database can be output as the matching result when the defect similarity is greater than 88%. The welding adjustment range includes argon adjustment range, welding gun voltage adjustment range, welding gun current adjustment range, welding speed adjustment range and welding angle adjustment range. The image analysis module generates corresponding welding adjustment instructions according to the above welding adjustment range, so that the welding robot dynamically adjusts the argon flow rate, argon flow rate, welding voltage, welding current, welding speed and welding angle in the process of ball pin welding, which significantly improves the welding quality of the ball pin.

[0069] Preferably, Figure 3 As shown, step S2 includes:

[0070] Step S21, the processing module acquires a number of standard spectrum images, where the standard spectrum images are acquired by the image acquisition module before welding the ball pin;

[0071] Step S22, the processing module processes each standard spectrum image to obtain standard spectrum data;

[0072] Step S23: the processing module processes each arc spectrum data and each standard spectrum data to obtain a spectrum shift index.

[0073] Specifically, in this embodiment, a storage unit is provided inside the processing module for storing a number of standard spectrum images. The standard spectrum images are collected by the spectrum imager before welding the ball pin, and then the standard spectrum images are processed by the image processing software to obtain standard spectrum data, so there is no influence of welding arc in the standard spectrum images.

[0074] Preferably, Figure 4 As shown, step S23 includes:

[0075] Step S231, the processing module selects a plurality of first spectrum characteristic data from each arc spectrum data and a plurality of second spectrum characteristic data from each standard spectrum data;

[0076] In step S232, the processing module inputs each first spectral feature data and each second spectral feature data into a preset spectral shift calculation formula to calculate a spectral shift index.

[0077] Specifically, in this embodiment, the processing module is configured with a spectral feature recognition program for identifying each first spectral feature data in the arc spectral data and each second spectral feature data in the historical spectral data, and then bringing each first spectral feature data and each second spectral feature data into the spectral shift calculation formula to calculate the spectral shift index.

[0078] Preferably, the spectrum shift calculation formula is configured as:

[0079]

[0080] Among them, S si Used to indicate the spectral shift index; k i Used to represent the preset configuration coefficients, each of which is a constant; x i Used to represent each first spectrum characteristic data in the arc spectrum data; y i Used to represent each second spectrum characteristic data in the standard spectrum data.

[0081] Preferably, step S2 further includes:

[0082] Step S24, the processing module inputs each welding smoke image into a preset three-dimensional convolutional network, collects the time series features in each welding smoke image, and outputs smoke concentration data;

[0083] Step S25: the processing module processes the smoke density data to obtain a smoke deviation index.

[0084] Specifically, in this embodiment, the three-dimensional convolutional network is used to continuously collect the time series features in the welding smoke image and process it to obtain smoke concentration data. Since there is no smoke in the normal state, the processing module can directly multiply the smoke concentration data with the preset smoke conversion coefficient to obtain the smoke deviation index.

[0085] Preferably, step S2 further includes:

[0086] Step S26, the processing module inputs the average temperature and each air distortion image into a pre-trained distortion repair model to obtain a number of distortion repair images;

[0087] Step S27, the processing module calculates the difference between the repaired feature points on each distorted repaired image and each distorted feature point on the corresponding air distorted image to obtain a comprehensive difference;

[0088] Step S28: The processing module obtains a distortion offset index according to the comprehensive difference.

[0089] Specifically, in this embodiment, during the welding process, the high temperature generated at the welding head causes the temperature of the welding area to rise. Because of the thermal expansion and contraction of the air, the air near the welding point becomes thin. The uneven density of the air changes the refractive index of light, causing the air in the welding image collected by the industrial array camera to be distorted. The distortion repair model is a pre-trained image repair model, which can repair the distortion features in the air distortion image to obtain the corresponding distortion repair image. Then, the image processing software is used to identify the repair feature points on each distortion repair image and the distortion feature points on the air distortion image, wherein the repair feature points correspond to the distortion feature points in position. By calculating the difference between each repair feature point and the distortion feature point, the variance is calculated to obtain the comprehensive difference, and then the comprehensive difference is multiplied by the preset distortion offset coefficient to obtain the distortion offset index. The distortion offset index can be used to reflect the influence of air distortion on welding recognition.

[0090] The above are only preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A welding method for automobile steering parts, characterized in that: include: Step S1, the storage module saves welding images, arc spectrum images and average temperatures collected at several historical moments, and obtains arc spectrum data according to the arc spectrum images, wherein the welding images include smoke images and air distortion images; Step S2, the processing module obtains a spectrum shift index according to each of the arc spectrum data, obtains a smoke shift index according to each of the smoke images, and obtains a distortion shift index according to each of the air distortion images and the average temperature. The spectrum shift index can be used to reflect the degree of interference of the arc spectrum on the welding image during the welding process, the smoke shift index can be used to reflect the degree of interference of the smoke generated during the welding process on the welding image, and the distortion shift index can be used to reflect the degree of interference of the air distortion on the welding image during the welding process; The step S2 further comprises: Step S24, the processing module inputs each welding smoke image into a preset three-dimensional convolutional network, collects the time series features in each welding smoke image, and outputs smoke concentration data; Step S25, the processing module processes the smoke concentration data to obtain the smoke deviation index; The step S2 further comprises: Step S26, the processing module inputs the average temperature and each of the air distortion images into a pre-trained distortion repair model to obtain a plurality of distortion repair images; Step S27, the processing module calculates the difference between each repaired feature point on the distorted repaired image and each distorted feature point on the corresponding air distorted image to obtain a comprehensive difference; Step S28, the processing module obtains the distortion offset index according to the comprehensive difference; Step S3, the training module annotates the welding defects in each welding image to obtain a plurality of annotated images, and uses each welding image before annotation as an input parameter and the corresponding annotated image as an output parameter to train and obtain a preliminary detection model; Step S4, the optimization module adjusts the weight parameters between the input layer and the hidden layer of the preliminary detection model according to the spectrum shift index, the smoke shift index and the distortion shift index, and retrains the model to obtain an optimized detection model; Step S5, the image acquisition module acquires a real-time welding image in real time during the welding process of the welding robot on the surface of the ball pin, and inputs the real-time welding image into the optimized detection model to obtain a defect annotation image; Step S6, the image analysis module generates a welding adjustment instruction according to the defect annotation image, and the welding robot dynamically adjusts the welding parameters in the ball pin welding process according to the welding adjustment instruction.

2. The welding method for automobile steering parts according to claim 1, characterized in that: The welding parameters include argon gas flow rate, argon gas flow rate, welding voltage, welding current, welding speed and welding angle.

3. The welding method for automobile steering parts according to claim 1, characterized in that: The step S6 comprises: Step S61, the image analysis module performs defect recognition on the defect-annotated image to obtain a number of surface welding defects; Step S62, the image analysis module analyzes each of the surface welding defects to obtain a number of defect parameters; Step S63, the image analysis module matches the defect parameters in a preset defect matching database to obtain a corresponding welding adjustment range, and comprehensively generates the welding adjustment instruction according to each welding adjustment range; Step S64: the welding robot dynamically adjusts the welding parameters of the ball pin according to the welding adjustment instruction.

4. The welding method for automobile steering parts according to claim 1, characterized in that: The step S2 comprises: Step S21, the processing module acquires a plurality of standard spectral images, wherein the standard spectral images are collected before welding the ball pin; Step S22, the processing module processes each of the standard spectrum images to obtain standard spectrum data; Step S23: the processing module obtains the spectrum shift index according to each of the arc spectrum data and each of the standard spectrum data.

5. The welding method for automobile steering parts according to claim 4, characterized in that: The step S23 comprises: Step S231, the processing module respectively selects a plurality of first spectrum characteristic data from each of the arc spectrum data and a plurality of second spectrum characteristic data from each of the standard spectrum data; In step S232, the processing module inputs each of the first spectral feature data and each of the second spectral feature data into a preset spectral shift calculation formula to calculate the spectral shift index.

6. The welding method for automobile steering parts according to claim 5, characterized in that: The spectrum shift calculation formula is configured as: ， in, Used to represent the spectral shift index; Used to represent preset configuration coefficients, each of which is a constant; used to represent each of the first spectrum characteristic data in the arc spectrum data; Used to represent each of the second spectrum feature data in the standard spectrum data.

7. The welding method for automobile steering parts according to claim 1, characterized in that: The optimized detection model is trained based on the YOLOV7 network model.

Images