A machine vision-based friction stir welding seam quality monitoring system and method

Abstract

The application relates to a machine vision-based friction stir welding seam quality monitoring system and method, and relates to the technical field of welding.The system comprises a data acquisition module, which acquires real-time welding seam image data during the welding process; an image processing module, which extracts feature information from the welding seam image based on an image processing algorithm; a quality analysis module, which establishes a welding seam quality analysis model according to the extracted feature information, and analyzes the welding seam quality in real time; and a feedback control module, which receives the quality analysis result and adjusts the welding parameters to optimize the welding seam quality.The application can efficiently and accurately evaluate the welding seam quality by acquiring real-time image data during the welding process and performing accurate image processing, real-time identification of welding seam defects and prediction of the quality score are realized by using advanced image enhancement, edge detection and deep learning algorithms, the welding process is optimized, the welding seam quality is improved, and the stability and reliability of the production process are ensured.

Description

TECHNICAL FIELD

[0001] The application relates to the field of welding technology, in particular to a friction stir welding seam quality monitoring system and method based on machine vision. BACKGROUND

[0002] Friction stir welding is a new type of welding technology that generates friction heat on the contact surface of the workpiece through a rotating tool, softens the workpiece locally, and causes plastic deformation under the action of the tool, ultimately completing the welding. This method has the advantages of no molten pool, low deformation, good mechanical properties, and has been widely used in the fields of aerospace, automobile manufacturing, shipbuilding industry, etc. However, real-time monitoring and control of welding quality during friction stir welding is still a challenge. The quality of the weld directly affects the strength, toughness and other important properties of the welded joint. Therefore, accurate monitoring and evaluation of the weld quality is crucial for improving product reliability and production efficiency. Traditional welding quality evaluation relies on manual detection or destructive testing methods, which not only takes time and effort, but also cannot achieve real-time monitoring and feedback.

[0003] Therefore, a welding seam quality monitoring system based on machine vision has emerged, which combines modern image processing technology, deep learning algorithms and feedback control systems, can monitor the quality of the weld in real time during the welding process, and provide feedback control to optimize the welding parameters, thereby effectively improving the welding quality. SUMMARY

[0004] The purpose of the present application is to provide a friction stir welding seam quality monitoring system and method based on machine vision.

[0005] In the first aspect, the friction stir welding seam quality monitoring system based on machine vision provided by the present application adopts the following technical solution:

[0006] A friction stir welding seam quality monitoring system based on machine vision, comprising: a data acquisition module, an image processing module, a quality analysis module and a feedback control module;

[0007] The data acquisition module is used to acquire weld image data in real time during the welding process;

[0008] The image processing module is used to extract feature information from the weld image based on image processing algorithms;

[0009] The quality analysis module is used to establish a weld quality analysis model according to the extracted feature information, and to analyze the weld quality in real time;

[0010] The feedback control module is used to receive the quality analysis results and adjust the welding parameters to optimize the weld quality.

[0011] Preferably, the image processing module specifically comprises:

[0012] The image preprocessing unit: adopts the Gaussian filter algorithm to remove the noise in the image and improve the definition of the image, wherein the Gaussian filter algorithm formula is:

[0013]

[0014] In the formula, represents the value of the pixel point in the image, is the Gaussian kernel function, is the size of the kernel, is the weight value of the Gaussian filter corresponding to the position , represents the pixel value of the original image at the offset position ;

[0015] The image enhancement unit: enhances the image quality through contrast enhancement and brightness adjustment technology, extracts the edge information of the weld seam using the edge detection algorithm, and identifies the contour of the weld seam.

[0016] Preferably, the image quality is enhanced through contrast enhancement and brightness adjustment technology, the edge information of the weld seam is extracted using the edge detection algorithm, and the contour of the weld seam is identified specifically include:

[0017] Enhance the contrast of the image through linear transformation and histogram equalization technology, adjust the overall brightness of the image, use the Canny edge detection algorithm to extract the edge information of the weld seam, and identify the contour of the weld seam by extracting and connecting the edges;

[0018] The Canny edge detection includes: Gaussian filtering, calculating the gradient of the image in x and y directions, performing local maximum suppression at the edge position, retaining the strongest edge, setting high and low threshold values to divide the edge into strong edge, weak edge and non-edge, and connecting the weak edge and the strong edge to finally form a complete edge contour.

[0019] Preferably, the quality analysis module specifically comprises:

[0020] The weld seam quality analysis model establishment unit: based on principal component analysis, extracts the features of the weld seam quality, reduces the feature dimension, retains the key information, defines the specific indicators of the weld seam quality evaluation according to the standard of the welding quality, including the weld seam uniformity, the detection of the weld seam surface defects, and the strength and carrying capacity of the weld seam;

[0021] The model training unit: uses the support vector machine algorithm to train the weld seam quality analysis model, uses the convolutional neural network to classify the weld seam image, detects whether the weld seam has defects, and uses the regression model to predict the specific quality score of the weld seam;

[0022] Real-time quality analysis unit: through the trained model, real-time analysis of the weld defects in the image, combined with deep learning for automatic identification of defects, scoring of the extracted features and analysis results, output of the quality score of the weld according to the quality evaluation standard of the weld;

[0023] Weld quality trend analysis unit: combined with multiple image data, trend prediction of the weld quality, detection of the quality change trend, and provision of a warning for abnormal conditions in the welding process;

[0024] Weld quality report generation unit: display of the real-time analysis results on the operation interface, including the weld quality score, defect position, and weld shape information, and storage of the quality analysis results in the welding process.

[0025] Preferably, the weld quality analysis model is trained using a support vector machine algorithm, a convolutional neural network is used to classify the weld image, detect whether the weld has defects, and a regression model is used to predict the specific quality score of the weld, which specifically includes:

[0026] The extracted weld features are classified and regressed using a support vector machine algorithm, the weld image is classified into different quality categories, and the quality score of the weld is predicted based on the regression task, wherein the formula of the support vector machine algorithm is:

[0027]

[0028] In the formula, is the normal vector of the decision hyperplane, is the bias term of the decision plane, is the slack variable, is the regularization parameter, represents the total number of samples, is the square norm of the normal vector;

[0029] The CNN is used to classify the weld image and determine whether the weld has defects, the CNN is trained using a training data set, and the parameters of the convolutional neural network are optimized;

[0030] A regression model is used to predict the specific quality score of the weld, the input features include the geometric features and image features of the weld, the regression model is trained using sample data in the training set, and the prediction error is minimized.

[0031] Preferably, the regression model is used to predict the specific quality score of the weld, the input features include the geometric features and image features of the weld, the regression model is trained using sample data in the training set, and the prediction error is minimized, which specifically includes:

[0032] Obtain sample data containing weld geometry features and image features and corresponding weld quality scores, divide the data set into training set and test set, 80% of the data for training, 20% of the data for testing;

[0033] Establish a weld quality score prediction model based on a linear regression model;

[0034] Use mean square error as objective function to minimize prediction error.

[0035] Preferably, the weld quality score prediction model based on the linear regression model specifically includes:

[0036] Wherein, the weld quality score prediction model formula is:

[0037]

[0038] In the formula, is the predicted weld quality score, is the intercept term of the model, is the regression coefficient, is the input feature variable, is the error term.

[0039] Preferably, the use of mean square error as objective function to minimize prediction error specifically includes:

[0040] Wherein, the objective function formula is:

[0041]

[0042] In the formula, is the true weld quality score of the sample, is the model prediction result of the sample, is the mean square error, is the number of samples.

[0043] Preferably, the feedback control module specifically includes:

[0044] Quality analysis result receiving unit: receiving weld quality evaluation results from the quality analysis module;

[0045] Quality deviation analysis unit: according to the received quality analysis results, evaluate the gap between the current weld quality and the target quality, and identify the welding parameters that need to be adjusted;

[0046] Adjustment strategy determination unit: according to the size of the quality deviation, determine the welding parameters that need to be adjusted, and decide the adjustment amplitude and direction of the welding parameters based on the preset PID control algorithm.

[0047] In a second aspect, the application provides a machine vision-based friction stir welding seam quality monitoring method using the following technical solution:

[0048] A machine vision-based friction stir welding seam quality monitoring method, comprising:

[0049] Step 1: Real-time acquisition of weld seam image data during welding process;

[0050] Step 2: Using a pre-set image processing algorithm, feature information is extracted from the weld seam image;

[0051] Step 3: Based on the extracted feature information, real-time analysis of weld seam quality is performed to generate a quality report;

[0052] Step 4: The quality analysis results are fed back to the welding control unit to adjust the welding parameters to ensure the quality of the weld seam.

[0053] In summary, the application includes at least one of the following beneficial technical effects:

[0054] The application uses a machine vision-based monitoring system, which uses high-precision cameras and image processing technology to real-time acquire weld seam images during the welding process and perform analysis, avoiding errors caused by manual intervention, and enabling more accurate control of weld seam quality;

[0055] Through contrast enhancement, brightness adjustment, and Canny edge detection in the image processing module, the feature information of the weld seam can be clearly extracted, and weld seam defects can be effectively identified, thereby real-time evaluating the welding quality and ensuring the consistency of the welding process;

[0056] Using the synergistic effect of the quality analysis module and the feedback control module, the system can automatically adjust the welding parameters based on real-time quality analysis results, thereby optimizing the welding process, avoiding the instability caused by manual intervention, and ensuring that the quality of the weld seam reaches the best level;

[0057] The system can identify the trend of weld seam quality changes through analysis of multiple image data, predict potential quality problems in advance, and issue warnings in a timely manner when abnormalities are found, reducing the occurrence of defective products in production;

[0058] The system realizes real-time monitoring and dynamic control of the welding process, which helps to reduce the scrap rate caused by improper welding parameters, avoids manual quality inspection of a large number of finished products, saves labor costs and time costs, and improves production efficiency;

[0059] The system has strong adaptability, can adjust the image processing algorithm and quality analysis model according to different welding processes and material types, meet different production needs, through deep learning and other technologies, the system can continuously learn and optimize, improve the stability and reliability of long-term operation. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 is the system framework diagram of the application;

[0061] Figure 2 is the step flow framework diagram of the application. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings Figure 1 -Appendix Figure 2 , make a further detailed description of the present application.

[0063] Example 1: a kind of based on machine vision's friction stir welding weld quality monitoring system, refer to Figure 1 , including: data acquisition module, image processing module, quality analysis module and feedback control module;

[0064] Data acquisition module, for real-time acquisition of weld image data in the welding process;

[0065] Image processing module, for extracting feature information from weld image based on image processing algorithm;

[0066] Quality analysis module, for establishing weld quality analysis model according to the extracted feature information, real-time analysis of weld quality;

[0067] Feedback control module, for receiving quality analysis results and adjusting welding parameters, optimizing weld quality.

[0068] Its features are that the image processing module specifically includes:

[0069] Image preprocessing unit: Gaussian filter algorithm is used to remove noise in the image, improve the definition of the image, wherein the Gaussian filter algorithm formula is:

[0070]

[0071] In the formula, Indicates the value of the pixel point In the image, Is the Gaussian kernel function, The size of the core, Is the weight value of Gaussian filter, corresponding to the position , Indicates the pixel value of the original image at the offset position ;

[0072] The image enhancement unit enhances the image quality through contrast enhancement and brightness adjustment techniques, extracts the edge information of the weld seam using an edge detection algorithm, and identifies the contour of the weld seam.

[0073] The image preprocessing unit uses a Gaussian filter algorithm to remove image noise, significantly improving image quality and reducing the impact of external environmental interference on weld seam image quality. At the same time, through image enhancement and edge detection algorithms, the weld seam contour is extracted, providing more accurate input data for subsequent weld seam quality analysis, and solving the problem of unclear weld seam images.

[0074] The image enhancement unit enhances the image quality through contrast enhancement and brightness adjustment techniques, extracts the edge information of the weld seam using an edge detection algorithm, and identifies the contour of the weld seam.

[0075] The image enhancement unit enhances the image quality through contrast enhancement and brightness adjustment techniques, extracts the edge information of the weld seam using an edge detection algorithm, and identifies the contour of the weld seam.

[0076] The Canny edge detection includes Gaussian filtering, calculating the gradient of the image in the x and y directions, performing local maximum suppression at the edge position, retaining the strongest edge, setting high and low threshold values to divide the edge into strong edge, weak edge and non-edge, and connecting the weak edge and strong edge to finally form a complete edge contour.

[0077] The image enhancement unit enhances the image quality through contrast enhancement and brightness adjustment techniques, extracts the edge information of the weld seam using an edge detection algorithm, and identifies the contour of the weld seam.

[0078] The quality analysis module specifically includes:

[0079] The weld seam quality analysis model establishment unit extracts the features of the weld seam quality based on principal component analysis, reduces the feature dimension, retains key information, defines specific indicators for weld seam quality evaluation according to the welding quality standards, including weld seam uniformity, weld seam surface defect detection, and weld seam strength and carrying capacity.

[0080] The model training unit uses a support vector machine algorithm to train the weld seam quality analysis model, uses a convolutional neural network to classify the weld seam image, detects whether the weld seam has defects, and uses a regression model to predict the specific quality score of the weld seam.

[0081] Real-time quality analysis unit: through the trained model, real-time analysis of the weld defects in the image, combined with deep learning for automatic identification of defects, scoring of the extracted features and analysis results, output of the quality score of the weld according to the quality evaluation standard of the weld;

[0082] Weld quality trend analysis unit: combined with multiple image data, trend prediction of the weld quality, detection of the quality change trend, and provision of a warning for abnormal conditions in the welding process;

[0083] Weld quality report generation unit: display of the real-time analysis results on the operation interface, including the weld quality score, defect position, and weld shape information, and storage of the quality analysis results in the welding process;

[0084] The quality analysis module extracts key information and reduces feature dimension through principal component analysis, solves the data overfitting problem that may be caused in traditional analysis methods, and combines advanced algorithms such as support vector machine and convolutional neural network to realize automatic identification and quality scoring of weld defects, greatly improving the efficiency and accuracy of weld quality evaluation.

[0085] The support vector machine algorithm is used to train the weld quality analysis model, the convolutional neural network is used to classify the weld image, the presence or absence of defects in the weld is detected, and the regression model is used to predict the specific quality score of the weld, which specifically includes:

[0086] The support vector machine algorithm is used to classify and regress the extracted weld features, the weld image is classified into different quality categories, and the quality score of the weld is predicted based on the regression task, wherein the formula of the support vector machine algorithm is:

[0087]

[0088] In the formula, is the normal vector of the decision hyperplane, is the bias term of the decision plane, is the slack variable, is the regularization parameter, represents the total number of samples, is the square norm of the normal vector;

[0089] The CNN is used to classify the weld image to determine whether there are defects in the weld, the training data set is used to train the CNN, and the parameters of the convolutional neural network are optimized;

[0090] The regression model is used to predict the specific quality score of the weld, the input features include the geometric features and image features of the weld, the sample data in the training set is used to train the regression model, and the prediction error is minimized;

[0091] The welding quality is classified and the regression task is performed by a support vector machine algorithm, the welding image is divided into different quality categories, and a convolutional neural network classification is combined, two different models are innovatively combined for quality analysis, and the detection rate of welding defects and the accuracy of quality prediction are improved.

[0092] The specific quality score of the welding seam is predicted using a regression model, the input features include the geometric features and image features of the welding seam, the regression model is trained using sample data in the training set, and the prediction error is minimized, specifically including:

[0093] Obtain sample data containing welding seam geometric features and image features and corresponding welding seam quality scores, divide the data set into a training set and a test set, 80% of the data is used for training, and 20% of the data is used for testing;

[0094] A welding seam quality score prediction model is established based on a linear regression model;

[0095] The mean square error is used as the objective function to minimize the prediction error;

[0096] The quality score of the welding seam is predicted using a regression model, and geometric features and image features are combined for training, which can accurately quantify the quality of the welding seam. This method optimizes the traditional quality evaluation method, making the quality evaluation of each welded part more practical and accurate.

[0097] The welding seam quality score prediction model based on the linear regression model specifically includes:

[0098] The welding seam quality score prediction model formula is:

[0099]

[0100] In the formula, is the predicted welding seam quality score, is the intercept term of the model, is the regression coefficient, is the input feature variable, is the error term;

[0101] The welding seam quality score is predicted based on the linear regression model, which further improves the accuracy of quantitative evaluation of welding seam quality. The regression model can handle complex multi-dimensional data, ensuring accurate prediction of welding seam quality scores, and is particularly suitable for quality monitoring in large-scale welding production processes.

[0102] The mean square error is used as the objective function to minimize the prediction error, specifically including:

[0103] The objective function formula is:

[0104]

[0105] wherein, is the true weld quality score of the sample, is the model prediction result of the sample, is the mean squared error, is the number of samples;

[0106] Using the mean squared error as the objective function for optimization can effectively reduce the prediction error, ensuring the stability and accuracy of the regression model. This optimization strategy enhances the reliability of the weld quality prediction model and provides technical support for large-scale applications.

[0107] The feedback control module specifically includes:

[0108] Quality analysis result receiving unit: receives the weld quality evaluation results from the quality analysis module;

[0109] Quality deviation analysis unit: based on the received quality analysis results, evaluates the gap between the current weld quality and the target quality, and identifies the welding parameters that need to be adjusted;

[0110] Adjustment strategy determination unit: determines the welding parameters that need to be adjusted according to the size of the quality deviation, and decides the adjustment amplitude and direction of the welding parameters based on the preset PID control algorithm;

[0111] The feedback control module innovatively combines the quality analysis results and the PID control algorithm, which can adjust in real time according to the weld quality deviation, automatically optimize the welding parameters, and thus ensure the continuous quality control during the welding process, and avoid the hysteresis and errors of manual adjustment.

[0112] Embodiment 2: A machine vision-based friction stir welding weld quality monitoring method, referring to Figure 2 as shown, including:

[0113] Step 1: Real-time acquisition of weld image data during the welding process;

[0114] Step 2: Use a pre-set image processing algorithm to extract feature information from the weld image;

[0115] Step 3: Based on the extracted feature information, real-time analysis of the weld quality, and generation of a quality report;

[0116] Step 4: Feedback the quality analysis results to the welding control unit, adjust the welding parameters, and ensure the weld quality;

[0117] The method realizes quality monitoring and automatic feedback adjustment in the whole process by real-time acquisition and processing of the weld image data in the welding process, not only improves the stability of the welding quality, but also reduces manual intervention and improves production efficiency, in addition, the use of deep learning and automatic analysis method makes the welding quality detection more accurate and fast.

[0118] The embodiments of the present embodiment are the preferred embodiments of the present application, not limited to the protection scope of the present application, wherein the same parts are indicated by the same reference numerals. Therefore: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A machine vision-based quality monitoring system for friction stir welding, characterized in that, include: The system includes a data acquisition module, an image processing module, a quality analysis module, and a feedback control module. The data acquisition module is used to acquire weld seam image data in real time during the welding process; The image processing module is used to extract feature information from the weld image based on the image processing algorithm; The quality analysis module is used to establish a weld quality analysis model based on the extracted feature information and analyze the weld quality in real time. The feedback control module is used to receive quality analysis results and adjust welding parameters to optimize weld quality. The image processing module specifically includes: Image preprocessing unit: Uses Gaussian filtering algorithm to remove noise from the image and improve image clarity. The formula for Gaussian filtering algorithm is: In the formula, Represents pixels in an image The value, It is a Gaussian kernel function. For the size of the core, These are the weights of the Gaussian filter, corresponding to the position. , Indicates the original image at the offset position Pixel value at; Image enhancement unit: Improves image quality through contrast enhancement and brightness adjustment techniques, and uses edge detection algorithms to extract edge information of the weld and identify the contour of the weld; The quality analysis module specifically includes: Weld quality analysis model building unit: Based on principal component analysis, extract the features of weld quality, reduce the feature dimensions, retain key information, and define specific indicators for weld quality assessment according to welding quality standards, including weld uniformity, detection of weld surface defects, and weld strength and load-bearing capacity. Model training unit: The weld quality analysis model is trained using the support vector machine algorithm, the weld images are classified using the convolutional neural network to detect whether there are defects in the weld, and the specific quality score of the weld is predicted using the regression model. Real-time quality analysis unit: Through a trained model, it analyzes weld defects in images in real time, combines deep learning to automatically identify defects, scores the extracted features and analysis results, and outputs the quality score of the weld according to the weld quality assessment standard. Weld quality trend analysis unit: Combines multi-frame image data to predict weld quality trends, detect quality change trends, and provide warnings for abnormal situations in the welding process; Weld quality report generation unit: Displays real-time analysis results on the operation interface, including weld quality score, defect location and weld shape information, and stores the quality analysis results during the welding process.

2. The machine vision-based friction stir welding weld quality monitoring system according to claim 1, characterized in that, The process of improving image quality through contrast enhancement and brightness adjustment techniques, and extracting edge information of the weld seam using an edge detection algorithm to identify the weld seam contour specifically includes: Image contrast is enhanced by linear transformation and histogram equalization techniques, and the overall brightness of the image is adjusted. The edge information of the weld is extracted using the Canny edge detection algorithm, and the outline of the weld is identified by extracting and connecting the edges. The Canny edge detection includes: Gaussian filtering, calculating the gradient of the image in the x and y directions, performing local maximum suppression at the edge location to retain the strongest edge, setting high and low thresholds to divide the edge into strong edge, weak edge and non-edge, and connecting the weak edge with the strong edge to finally form a complete edge contour.

3. The machine vision-based friction stir welding weld quality monitoring system according to claim 2, characterized in that, The process of training a weld quality analysis model using a support vector machine algorithm, classifying weld images using a convolutional neural network to detect weld defects, and using a regression model to predict the specific quality score of the weld includes: The Support Vector Machine (SVM) algorithm is used to classify and regress the extracted weld features, dividing the weld images into different quality categories. Based on the regression task, the quality score of the weld is predicted. The SVM algorithm formula is as follows: In the formula, It is the normal vector of the decision hyperplane. It is the bias term of the decision plane. It is a slack variable. It is a regularization parameter. This represents the total number of samples. It is the square norm of the normal vector; CNN is used to classify weld seam images to determine whether there are defects in the weld seam. The CNN is trained using a training dataset to optimize the parameters of the convolutional neural network. A regression model is used to predict the specific quality score of the weld. The input features include the geometric and image features of the weld. The regression model is trained using sample data from the training set to minimize the prediction error.

4. The machine vision-based friction stir welding weld quality monitoring system according to claim 3, characterized in that, The method of using a regression model to predict the specific quality score of the weld seam involves input features including the geometric and image features of the weld seam, training the regression model using sample data from the training set, and minimizing the prediction error. Specifically, this includes: Obtain sample data containing weld geometric features, image features, and corresponding weld quality scores. Divide the dataset into a training set and a test set, with 80% of the data used for training and 20% used for testing. A weld quality score prediction model was established based on a linear regression model. The mean squared error is used as the objective function to minimize the prediction error.

5. The machine vision-based friction stir welding weld quality monitoring system according to claim 4, characterized in that, The specific details of establishing a weld quality score prediction model based on a linear regression model are as follows: include: The formula for the weld quality fraction prediction model is as follows: In the 𝑦 formula, It is the predicted weld quality score. It is the intercept term of the model. It is the regression coefficient. For the input feature variables, This is the error term.

6. The machine vision-based friction stir welding weld quality monitoring system according to claim 4, characterized in that, The specific method of using mean squared error as the objective function to minimize prediction error is as follows: include: The objective function formula is as follows: In the formula, It is the first The actual weld quality score of the sample It is the first Model prediction results for each sample Mean square error, It refers to the number of samples.

7. The machine vision-based friction stir welding weld quality monitoring system according to claim 1, characterized in that, The feedback control module specifically includes: Quality analysis result receiving unit: Receives weld quality assessment results from the quality analysis module; Quality Deviation Analysis Unit: Based on the received quality analysis results, assess the gap between the current weld quality and the target quality, and identify the welding parameters that need to be adjusted; Adjustment strategy determination unit: Based on the magnitude of the quality deviation, it determines the welding parameters that need to be adjusted, and determines the adjustment range and direction of the welding parameters based on the preset PID control algorithm.

8. A machine vision-based method for monitoring the quality of friction stir welds, employing the machine vision-based friction stir weld quality monitoring system as described in any one of claims 1-7, characterized in that... include: Step 1: Acquire weld seam image data in real time during the welding process; Step 2: Extract feature information from the weld image using a preset image processing algorithm; Step 3: Based on the extracted feature information, analyze the weld quality in real time and generate a quality report; Step 4: Feed back the quality analysis results to the welding control unit, adjust the welding parameters, and ensure the quality of the weld.

Images