Ultrasonic wave and machine vision collaborative weld quality evaluation system and method

Abstract

The invention provides a weld quality evaluation system and method based on cooperation of ultrasonic waves and machine vision. The system comprises an ultrasonic detection system used for preliminary quality division and defect judgment of a weld and a machine vision system used for quality detection and quality evaluation of the weld. The ultrasonic detection system comprises an ultrasonic probe, an ultrasonic detector and a signal processing and analyzing module; the machine vision system comprises an image acquisition module, a data processing system and a user interface. The invention further provides a weld defect detection and evaluation method of the system. According to the method, ultrasonic detection and machine vision technologies are combined, ultrasonic waves are used for preliminarily detecting the defects of the welding seam, the defect detection process is gradually optimized, the machine vision system is combined for deep learning recognition of the surface defects of the welding seam, and finally independent operation of the machine vision system is achieved. Therefore, steel structure weld defect detection and weld quality evaluation are realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of welding quality detection, and in particular relates to a weld quality assessment system and method using ultrasonic waves and machine vision in collaboration. Background Art

[0002] Welding quality inspection is of great significance in industrial fields such as steel structures, pressure vessels, and pipelines. The quality of welds directly affects the strength and safety of the structure. Therefore, the detection and evaluation of weld defects is crucial. At present, commonly used weld inspection methods include ultrasonic inspection, X-ray inspection, magnetic particle inspection, etc. Among them, ultrasonic inspection is widely used because it can efficiently and accurately detect internal defects in welds (such as cracks, pores, inclusions, etc.). Ultrasonic waves can clearly reveal deep defects in welds by emitting sound waves and analyzing the signals of reflected waves. They are widely used in preliminary evaluation of welding quality and defect location.

[0003] In contrast, machine vision technology is mainly used to detect surface defects of welds. By taking weld images with a high-resolution camera and combining them with image processing algorithms, defects such as cracks, pores, and spatter on the weld surface can be identified. With the rapid development of deep learning technology, deep learning methods such as convolutional neural networks (CNNs) have been widely used in the automatic detection of weld surface defects. The advantages of machine vision systems lie in their high efficiency and automation, especially in the rapid detection of surface defects. However, there are certain challenges in the practical application of machine vision systems. The most important problem is that the accuracy of defect classification depends on the quality of the training data set. In the training process of machine vision, manually annotated weld defect data sets often face the problems of misclassification and inaccurate annotation. This is because manual annotation is affected by subjective factors, experience limitations, and the diversity of weld defect types, which may lead to errors or inconsistencies in the labels in the data set, which in turn affects the training effect and recognition accuracy of the deep learning model. Summary of the invention

[0004] In view of the shortcomings of the prior art such as manual classification errors and inaccurate deep learning model training data, the present invention provides a weld quality assessment system and method that cooperates with ultrasound and machine vision. Its purpose is to combine ultrasonic detection with machine vision technology, use ultrasound to perform preliminary detection of weld defects, gradually optimize the defect detection process, combine the machine vision system to perform deep learning identification of weld surface defects, and finally realize the independent operation of the machine vision system, thereby realizing steel structure weld defect detection and weld quality assessment.

[0005] In order to achieve the above object, the specific scheme of the present invention is as follows:

[0006] Ultrasonic and machine vision-coordinated weld quality assessment system, including an ultrasonic detection system for preliminary quality classification and defect judgment of welds and a machine vision system for weld quality detection and quality assessment;

[0007] The ultrasonic detection system comprises:

[0008] Ultrasonic probes, which are used to detect types of defects in welds;

[0009] Ultrasonic detector, used to transmit and receive signals from ultrasonic probes, process echo signals and calculate time delay and intensity to determine the quality grade of welds;

[0010] Signal processing and analysis module, used for digital signal processing, echo signal analysis and defect judgment, and determines the quality grade of the weld according to the echo intensity and shape;

[0011] The machine vision system includes an image acquisition module, a data processing system and a user interface;

[0012] An image acquisition module, used to acquire weld defect images that have been quality-classified and defect-detected by an ultrasonic detection system;

[0013] The data processing module is used to preprocess the weld defect images collected by the image acquisition module, perform deep learning model training, defect classification and quality assessment;

[0014] The user interface is used to adjust system settings, view historical data, and view and display test results processed by the data processing system and generate reports.

[0015] Furthermore, the data processing system comprises:

[0016] Image processing unit for image noise reduction and image enhancement;

[0017] A deep learning training unit, used for training and learning the preprocessed weld defect images using a deep learning model;

[0018] Defect classification and quality assessment unit, used for weld defect classification and quality assessment based on deep learning.

[0019] Furthermore, the user interface includes:

[0020] An image display unit, used to display the weld defect images and analysis results collected by the machine vision system in real time;

[0021] Defect marking and feedback unit, used to display weld defect images with identified and marked defect locations and defect types;

[0022] A quality assessment display unit, used to display the weld defect quality grade and defect information in the weld image;

[0023] The report generation and export unit is used to record, generate, export or print defect detection reports of weld defect images.

[0024] A method for weld defect detection and quality assessment using the system in collaboration with ultrasonic and machine vision comprises the following steps:

[0025] S1, perform preliminary inspection through ultrasonic detection system to determine whether there are defects such as pores, cracks or inclusions in the weld, and classify the weld quality grade according to the intensity and echo shape of the echo signal, and judge the weld defects according to the echo intensity comparison, echo shape analysis and time delay;

[0026] S2, collecting, by an image acquisition module, the weld defect image after weld quality classification and defect detection by the ultrasonic detection system in step S1;

[0027] S3, preprocessing the weld defect image collected in step S2 through the data processing module, labeling the defect area of ​​the weld defect image and inputting it into the deep learning model for training and recognition, automatically extracting features, classifying and identifying defects and evaluating the quality of the weld defect image, and then analyzing the results through the deep learning model to output the weld quality grade and defect type;

[0028] S4, by establishing a user interface to display the analysis results, defect labels, weld quality grades and defect types of the weld defect images described in step 3 in real time, and automatically generate a test report.

[0029] Furthermore, the training and identification of the weld defect image includes the following steps:

[0030] Step 1. Input the weld defect image preprocessed in step 3 into the deep learning model, and divide the weld defect image into several grids, each grid predicting the category and location of a weld defect;

[0031] Step 2. Each convolution layer learns according to the features of the input weld defect image, gradually optimizes the convolution kernel through training, and extracts effective features;

[0032] Step 3. After multiple convolutional layers, pooling layers, and activation layers, the prediction information of each grid is output through the deep learning model, and the weld quality grade and defect type are finally divided through the deep learning model output.

[0033] Furthermore, the weld quality grades are divided into Grade A, Grade B, Grade C and Grade D, Grade A indicates no defects or very small defects, Grade B indicates slight defects, Grade C indicates medium defects, and Grade D indicates severe defects; the defect labels are 1D and 0C, 1D indicates that the quality grade of the weld during inspection is severe defect, and the defect type is crack; 0C indicates that the quality grade of the weld during inspection is medium defect, and the defect type is pores.

[0034] Advantages of the present invention

[0035] Compared with the prior art, the advantages of the present invention are:

[0036] 1. The ultrasonic and machine vision-coordinated weld quality assessment system and method of the present invention provides more accurate training data for the deep learning model through the detection results of the ultrasonic detection system, eliminating the errors and subjective bias of manual annotation, thereby optimizing the deep learning training process. The accurate quality division provided by the ultrasonic detection system is used as a supervisory signal to help the deep learning model converge quickly, avoiding the negative impact of low-quality or erroneous data on the training effect, thereby improving the detection efficiency.

[0037] 2. With the continuous optimization of the deep learning model, the present invention has achieved a transition from relying on ultrasonic detection system detection to relying entirely on machine vision system for weld defect detection. In the early stage, the ultrasonic detection system provides accurate internal defect division and high-quality training data; with the improvement of the performance of the improved deep learning model, the machine vision system is gradually able to independently complete defect identification, reduce dependence on ultrasound, and reduce system cost and complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The figure is a flow chart of the method of the present invention.

[0039] Figure 2 It is a system composition diagram of the present invention.

[0040] Figure 3 for Figure 2 The ultrasonic detection system workflow diagram in .

[0041] Figure 4 for Figure 2 Schematic diagram of the structural composition of the machine vision system.

[0042] Figure 5 for Figure 2 Flowchart of weld defect image acquisition and preprocessing of the machine vision system in FIG. DETAILED DESCRIPTION

[0043] The present invention is further explained and illustrated below in conjunction with the accompanying drawings and specific implementation methods. It should be noted that this specific embodiment is not intended to limit the scope of rights of the present invention.

[0044] like Figures 1 to 5 As shown, this specific embodiment provides a weld quality assessment system and method using ultrasonic waves and machine vision in collaboration, wherein the weld quality assessment system using ultrasonic waves and machine vision in collaboration includes an ultrasonic detection system for preliminary weld quality classification and defect judgment and a machine vision system for weld quality detection and quality assessment;

[0045] The ultrasonic detection system comprises:

[0046] Ultrasonic probes are used to detect the types of defects in welds; the appropriate frequency and type of ultrasonic probes are selected according to the weld material and the requirements of the welding process.

[0047] Ultrasonic detector, used to transmit and receive signals from ultrasonic probes, process echo signals and calculate time delay and intensity to determine the quality grade of welds;

[0048] Signal processing and analysis module, used for digital signal processing, echo signal analysis and defect judgment, and determines the quality grade of the weld according to the echo intensity and shape;

[0049] The machine vision system includes an image acquisition module, a data processing system and a user interface;

[0050] An image acquisition module, which is a high-definition camera, is used to acquire weld defect images that have been quality-classified and defect-detected by an ultrasonic testing system;

[0051] The data processing module is used to preprocess the weld defect images collected by the image acquisition module, perform deep learning model training, defect classification and quality assessment;

[0052] The data processing system comprises:

[0053] Image processing unit for image noise reduction and image enhancement;

[0054] A deep learning training unit, used to train and learn the preprocessed weld defect images using the improved YOLOv11 deep learning model;

[0055] Defect classification and quality assessment unit, used for weld defect classification and quality assessment based on deep learning.

[0056] The user interface is used to adjust system settings, view historical data, and view and display test results processed by the data processing system and generate reports.

[0057] The user interface comprises:

[0058] An image display unit, used to display the weld defect images and analysis results collected by the machine vision system in real time;

[0059] Defect marking and feedback unit, used to display weld defect images with identified and marked defect locations and defect types;

[0060] A quality assessment display unit, used to display the weld defect quality grade and defect information in the weld image; the weld quality grade is divided into grade A, grade B, grade C and grade D, grade A indicates no defect or very small defect, grade B indicates slight defect, grade C indicates medium defect, and grade D indicates severe defect;

[0061] The report generation and export unit is used to record, generate, export or print defect detection reports of weld defect images.

[0062] The method for weld defect detection and quality assessment using the above system by combining ultrasonic wave and machine vision is characterized by comprising the following steps:

[0063] S1, perform preliminary inspection through ultrasonic detection system to determine whether there are defects such as pores, cracks or inclusions in the weld, and classify the weld quality grade according to the intensity and echo shape of the echo signal, and judge the weld defects according to the echo intensity comparison, echo shape analysis and time delay;

[0064] Specifically, an ultrasonic detection system is used to conduct preliminary inspection of the weld. High-frequency sound waves are emitted by the ultrasonic probe, and the echo signal reflected from the weld defect is received. The signal processing and analysis module analyzes the echo signal through the receiver of the ultrasonic detector, calculates the signal's intensity, shape, time delay and other characteristics, and determines the quality of the weld and the type of defect.

[0065] According to the intensity and shape of the echo signal, the weld quality is divided into four levels:

[0066] Grade A means no defects or very small defects, which means the echo signal is weak and the time delay is long, indicating that the weld quality is good.

[0067] Grade B indicates a slight defect, which means that the echo signal is obvious, but the defect is small and the impact is minor.

[0068] Grade C indicates a medium defect, which means that the echo signal is strong and the defect is large, which may affect the load-bearing capacity of the weld.

[0069] Grade D indicates a serious defect, which means that the echo signal is extremely strong and the defect has a significant impact on the structural strength.

[0070] Defect judgment is made based on echo intensity contrast, echo morphology analysis and time delay. If the defect is judged to be a pore, it means that the echo is strong and concentrated, the time is short, and the shape is sharp; if the defect is judged to be a crack, it means that the echo is complex, there are multiple reflections or scatterings, and the echo time is long; if the defect is judged to be an inclusion, it means that the echo signal is relatively uniform and of medium intensity, the echo shape is relatively regular, and the time delay is irregular.

[0071] S2, collecting, by an image acquisition module, the weld defect image after weld quality classification and defect detection by the ultrasonic detection system in step S1;

[0072] Specifically, use a high-definition camera to capture weld defect images to ensure that the image resolution is high and at least able to distinguish tiny weld defects. Common resolution options are 12MP or above. Set the appropriate focal length and exposure time to ensure image clarity. The focal length selection should be adapted to the shooting distance to avoid out-of-focus blur caused by the image being too far or too close. When capturing weld defect images, you must first select a good viewing angle. For example: for butt welds or flat welds, choose flat shooting, for fillet welds or three-dimensional structure welds, choose side shooting, and for some smaller weld defects, take multiple angles. During the weld defect image acquisition process, mark the quality classification and defect type corresponding to each image to facilitate subsequent training and learning using deep learning models. After completing the weld defect image acquisition, ensure that the image file is stored safely.

[0073] S3, the weld defect image collected in step S2 is preprocessed through the data processing module to optimize the image quality and improve the accuracy of defect detection. The defect area of ​​the weld defect image is labeled and input into the deep learning model for training and recognition. After automatically extracting features, classifying and identifying defects, and evaluating the quality of the weld defect image, the deep learning model is used to analyze the results and output the weld quality grade and defect type.

[0074] Specifically, image preprocessing includes image denoising and image enhancement. Image denoising uses a bilateral filtering algorithm to eliminate environmental noise and defects in the image acquisition process. Image enhancement uses a histogram equalization algorithm to enhance the visibility of the image and make the defect area visually clearer. The effect is as follows: Figure 4 shown.

[0075] When labeling the weld defect images, the weld surface defects are classified into cracks, pores, and inclusions; "0" represents pore defects, "1" represents crack defects, and "2" represents inclusion defects; "A" represents no defects or very small defects, "B" represents slight defects, "C" represents medium defects, and "D" represents severe defects. If the quality level of the weld is severe defects during inspection and the defect type is pores, the defect label is "0D"; if the quality level of the weld is medium defects during inspection and the defect type is cracks, the defect label is "1C", and so on.

[0076] The weld defect images after image preprocessing will be input into the improved YOLOv11 deep learning model for training and recognition to perform feature extraction, defect classification and quality assessment of the image weld defects.

[0077] The training and identification of the weld defect image comprises the following steps:

[0078] Step 1. Input the weld defect image after image preprocessing into the improved YOLOv11 deep learning model. The weld defect image is divided into several grids, and each grid predicts the category and location of a weld defect;

[0079] Step 2. Each convolution layer learns according to the features of the input weld defect image, gradually optimizes the convolution kernel through training, and extracts effective features;

[0080] Step 3. After multiple convolutional layers, pooling layers, and activation layers, the improved YOLOv11 deep learning model outputs the prediction information for each grid; and the improved YOLOv11 deep learning model finally outputs the weld quality grade and accurately labels the defect type.

[0081] The improved YOLOv11 deep learning model reduces computational redundancy by applying single-head attention on a portion of the input channel while retaining the combination of global and local information, thereby improving efficiency and accuracy. Figure 5 As shown in the figure, its structural features include: (1) Partial channel processing: the attention layer is applied only to some input channels, and the remaining channels remain unchanged; (2) Full channel projection: the final projection is applied to all channels to ensure that the attention features are effectively propagated to all channels; (3) Simplified operation: reducing memory-bound operations and improving computational efficiency.

[0082] S4, in order to intuitively and comprehensively display the weld defect detection results and facilitate users to evaluate and trace the weld quality, a user interface is established to display the analysis results, defect labels, weld quality grades and defect types of the weld defect images described in step 3 in real time, and an inspection report is automatically generated to provide users with an efficient and easy-to-operate inspection experience.

[0083] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. Ultrasonic and machine vision-based weld quality assessment system, characterized by: Including ultrasonic testing system for preliminary quality classification and defect judgment of welds and machine vision system for weld quality inspection and quality assessment; The ultrasonic detection system comprises: Ultrasonic probes, which are used to detect types of defects in welds; Ultrasonic detector, used to transmit and receive signals from ultrasonic probes, process echo signals and calculate time delay and intensity to determine the quality grade of welds; Signal processing and analysis module, used for digital signal processing, echo signal analysis and defect judgment, and determines the quality grade of the weld according to the echo intensity and shape; The machine vision system includes an image acquisition module, a data processing system and a user interface; An image acquisition module, used to acquire weld defect images that have been quality-classified and defect-detected by an ultrasonic detection system; The data processing module is used to preprocess the weld defect images collected by the image acquisition module, perform deep learning model training, defect classification and quality assessment; The user interface is used to adjust system settings, view historical data, and view and display test results processed by the data processing system and generate reports.

2. The system according to claim 1, characterized in that The data processing system comprises: Image processing unit for image noise reduction and image enhancement; A deep learning training unit, used for training and learning the preprocessed weld defect images using a deep learning model; Defect classification and quality assessment unit, used for weld defect classification and quality assessment based on deep learning.

3. The system according to claim 1, characterized in that The user interface comprises: An image display unit, used to display the weld defect images and analysis results collected by the machine vision system in real time; Defect marking and feedback unit, used to display weld defect images with identified and marked defect locations and defect types; A quality assessment display unit, used to display the weld defect quality grade and defect information in the weld image; The report generation and export unit is used to record, generate, export or print defect detection reports of weld defect images.

4. A method for weld defect detection and quality assessment using the system described in any one of claims 1 to 3 by using ultrasonic and machine vision collaboration, characterized in that: The following steps are involved: S1, conduct preliminary inspection through ultrasonic detection system to determine whether there are defects such as pores, cracks or inclusions in the weld, and classify the weld quality grade according to the intensity and echo shape of the echo signal, and judge the weld defects according to the echo intensity comparison, echo shape analysis and time delay; S2, collecting, by an image acquisition module, a weld defect image obtained by the ultrasonic detection system in step S1 after weld quality classification and defect detection; S3, performing image preprocessing on the weld defect image collected in step S2 through a data processing module, and labeling the defect area of ​​the weld defect image and inputting it into a deep learning model for training and recognition, automatically extracting features, classifying and identifying defects, and evaluating the quality of the weld defect image, and then analyzing the results through a deep learning model to output the weld quality grade and defect type; S4, by establishing a user interface to display the analysis results, defect labels, weld quality grades and defect types of the weld defect images described in step 3 in real time, and automatically generate a test report.

5. The method according to claim 4, characterized in that: The training and identification of the weld defect image comprises the following steps: Step 1. Input the weld defect image preprocessed in step 3 into the deep learning model, and divide the weld defect image into several grids, each grid predicting the category and location of a weld defect; Step 2. Each convolution layer learns according to the features of the input weld defect image, gradually optimizes the convolution kernel through training, and extracts effective features; Step 3. After multiple convolutional layers, pooling layers, and activation layers, the prediction information of each grid is output through the deep learning model, and the weld quality grade and defect type are finally divided through the deep learning model output.

6. The method according to claim 4, characterized in that: The weld quality grades are divided into A, B, C and D, A indicates no defects or very small defects, B indicates slight defects, C indicates medium defects, and D indicates severe defects; the defect labels are 1D and 0C, 1D indicates that the quality grade of the weld is severe defect during detection, and the defect type is crack; 0C indicates that the quality grade of the weld is medium defect during detection, and the defect type is pores.

Images