Online flaw detection method and system combined with image analysis and processing technology

Abstract

The invention relates to the technical field of flaw detection, in particular to an online flaw detection method and system for flaw detection by combining an image analysis and processing technology. Comprising the following steps: S1, obtaining detection image data, preprocessing the detection image data, carrying out grid division on the preprocessed image, and extracting features; s2, performing odd-even division on the gridded image, and calculating an angle according to a division result; and S3, constructing a prediction model and performing real-time prediction analysis. According to the method, the image data are collected at equal intervals and preprocessed, so that the calculation complexity is reduced, and the processing efficiency is improved; performing odd-even grid division, calculating an included angle, and refining defect types and severity; an LSTM model is constructed for real-time prediction, and the prediction precision and reliability are improved in combination with historical data; and finally, a closed-loop system is formed, real-time online monitoring is realized, the defect type and severity are accurately judged, the continuity and reliability of detection are ensured, and powerful technical support is provided for defect damage detection.

Description

Technical Field

[0001] The present invention relates to the field of flaw detection technology, and in particular to a flaw detection online defect detection method and system combined with image analysis and processing technology. Background Art

[0002] The main directions of online flaw detection include ultrasonic testing, eddy current testing, magnetic particle testing, radiographic testing and visual testing. The advantages of existing online flaw detection methods include high sensitivity, non-destructive, real-time detection, high degree of automation, wide application range and reliable detection results. However, these methods also have some disadvantages, such as high cost, high requirements for operators, limited detection speed, great influence of environmental factors, and some methods have limitations on defect types. Existing online flaw detection methods are insufficient in accuracy, mainly manifested in the lack of fine recognition and classification of defect features, resulting in common misjudgments and missed detections. Image analysis and processing technology can improve detection speed and accuracy, reduce dependence on operators, reduce the impact of environmental factors, and expand the scope of application of detection methods. Image analysis and processing technology can significantly improve detection accuracy in online flaw detection, effectively reduce false detections and missed detections through advanced algorithms and pattern recognition technologies, and improve detection reliability and stability. Summary of the invention

[0003] In order to overcome the shortcoming of low accuracy of online flaw detection, the present invention provides an online flaw detection method and system combining image analysis and processing technology.

[0004] The technical real-time solution of the present invention is: an online defect detection method for flaw detection combined with image analysis and processing technology, comprising the following steps: S1: Acquire the detection image data and perform preprocessing, grid the preprocessed image and extract features; S2: divide the gridded image into even and odd parts, and calculate the angle according to the division result; S3: Build a prediction model and perform real-time prediction analysis to determine the damage degree of the defect based on the prediction results; S4: The entire system is designed as a closed loop to achieve real-time online detection of flaws.

[0005] Preferably, the acquisition of detection image data and preprocessing, gridding the preprocessed image and extracting features, includes: collecting image data at equal intervals t, cleaning and standardizing the image data; evenly dividing the image data into n*n grids, using a CNN model to extract image defect features, calculating the mean and variance of each grid defect feature, and recording the position of each defect feature in the grid.

[0006] Preferably, the gridded image is divided into odd and even numbers, and the angle is calculated according to the division result, including: marking the horizontal grids and vertical grids of n*n grids in odd and even order, the marking result is that the horizontal grids are marked in the order of 1, 2, 1, 2..., and the vertical grids are marked in the order of 1, 2, 1, 2...; obtaining the defect feature distribution, extracting the defect features 1 and 2 with the closest interval in the horizontal direction, and at the same time extracting the defect feature 3 with the closest interval in the vertical direction, taking the defect features 1 and 2 as the horizontal line, connecting the defect feature 3 with the defect feature 1, forming an angle between the connecting lines of the defect features 1, 2, and 3, calculating the angle angle, and classifying the angle angle.

[0007] Preferably, the classification of the angles includes: obtaining image data before defect features are formed and defining it as Class A image data, and obtaining image data in the process of defect feature formation without defect features and defining it as Class B image data; recording image change data in the process from Class A image to Class B image, and classifying the angles according to the image change data, if the change is in the horizontal direction in the process from Class A image to Class B image, then modifying defect feature 1 and defect feature 2, and then forming a new angle and defining it as a new horizontal angle; if the change is in the vertical direction in the process from Class A image to Class B image, then modifying defect feature 3, and then forming a new angle and defining it as a new vertical angle; if the change is neither in the horizontal direction nor in the vertical direction, then modifying defect features 1, 2, 3 according to the angle change, and then forming a new angle and defining it as an oblique angle; taking the three new angles as a reference, forming three angle sequences.

[0008] Preferably, three angle sequences are formed based on the three new angles, including: in the horizontal direction, if defect feature 3 is located on the left side of defect feature 1 in the vertical direction or defect feature 3 is located within the vertical range of defect feature 1 and defect feature 2, then the new angle data caused by the change in the angle due to the change in the defect feature forms a horizontal sequence; in the vertical direction, the new angle data formed by the change in the vertical angle forms a vertical sequence; in the oblique direction, the new angle data are sorted by the size of the angle change so that they form an oblique sequence.

[0009] Preferably, the prediction model is constructed and real-time prediction analysis is performed to determine the degree of damage of the defect according to the prediction results, including: obtaining a training set of defect feature data, constructing an LSTM model, inputting the training set data into the LSTM model, obtaining a trained LSTM model, predicting the defect features with the LSTM model, obtaining predicted defect features, and marking them as first defect features; obtaining a historical defect feature data set, and simultaneously obtaining a predicted defect feature data set predicted by the LSTM model, using the historical defect feature data set as the denominator and the predicted defect feature data set as the numerator, constructing a prediction factor, predicting the defect features with the prediction factor, obtaining predicted defect features, and marking them as second defect features; analyzing three angle sequences with the first defect feature and the second defect feature; prediction factor = predicted defect feature data set / historical defect feature data set.

[0010] Preferably, the analysis of the three angle sequences with the first defect feature and the second defect feature includes: matching the first defect feature, the second defect feature and the three angle sequences; if the three angle sequences do not match the first defect feature and the second defect feature, marking the defect feature as a confirmed defect; if the three angle sequences match the second defect feature but do not match the first defect feature, marking the defect feature as a pending defect; if the three angles all match the first defect feature and the second defect feature, obtaining the angle parity change values ​​in the three angle sequences.

[0011] Preferably, the method of obtaining the angle parity change values ​​in the three angle sequences includes: obtaining a training set of angle parity change value data, constructing an SVM classification model, inputting the training set data into the SVM classification model, obtaining a trained SVM classification model, and classifying the angle parity changes with the SVM classification model; wherein the classification standard is, firstly, recording the parity ratio in the three angle sequences, and secondly using the relationship between the parity ratio and the angle change rate as the classification standard; obtaining the intersection of the three angle sequences, and if the determined defects, the defects to be determined and the characteristic defects are respectively located in different intersections, they are marked according to their respective classification rules; if the angle change rates in the three angle sequences do not exceed the threshold value (X), they are marked as non-defective areas.

[0012] Preferably, the entire system is constructed into a loop to realize real-time online monitoring of flaw detection defects, including: according to the classification criteria and intersection, all determined defects, defects to be determined and characteristic defects are detected, and finally the entire system is constructed into a loop system; the specific classification criteria are: if the angle change rate in the three angle sequences exceeds the threshold value X, it is marked as a determined defect.

[0013] An online flaw detection system combining image analysis and processing technology, comprising: Image acquisition module: used to collect detection image data; Image preprocessing module: used to preprocess the collected image data; Grid division and feature extraction module: used to evenly divide the preprocessed image data into n*n grids and extract features; Odd-even division and angle calculation module: used to perform odd-even division on the gridded image and calculate the angle according to the division result; Prediction model building and real-time prediction module: used to build prediction models and conduct real-time prediction analysis; Defect feature analysis and result output module: used to match the first defect feature with the second defect feature, determine the state of the defect feature, and output the final result.

[0014] Beneficial effects: The present invention proposes an online defect detection method for flaw detection combined with image analysis and processing technology. By collecting image data at equal intervals t and performing preprocessing, the temporal continuity and consistency of the data are ensured, and data deviation caused by inconsistent collection time intervals is avoided. Then, the preprocessed image is gridded and features are extracted. The image is evenly divided into n*n grids, and the image defect features are extracted using the CNN model. The mean and variance of each grid defect feature are calculated, and the position of each defect feature in the grid is recorded. This not only reduces the computational complexity, but also improves the processing efficiency. At the same time, local processing better captures small defect features. On this basis, the gridded image is divided into odd and even, and the angle is calculated according to the division result. The horizontal and vertical grids are marked in odd and even order, so that the distribution and changes of defects in different directions can be more easily identified and tracked, and the defect features with the closest intervals in the horizontal and vertical directions are extracted. The angle is formed and the angle is calculated, and the type and severity of the defect are further refined, providing more detailed information for subsequent defect analysis and processing. Subsequently, a prediction model is constructed and real-time prediction analysis is performed. By obtaining a training set of defect feature data, an LSTM model is constructed. The characteristics of the LSTM model are used to predict the defect features in the current detection image in real time. The prediction factor is constructed by combining historical data and prediction data. The difference between the prediction result and the historical data is quantified to further improve the accuracy and reliability of the prediction. By combining the first defect feature and the second defect feature, the spatial relationship and evolution pattern of the defect feature are analyzed more comprehensively to improve the accuracy of diagnosis. Finally, the entire system is constructed into a loop to realize real-time online monitoring of flaw detection defects. By analyzing the angle parity change value in the intersection, the type and severity of the defect can be more accurately judged to avoid misjudgment. The specific classification standard is: if the angle change rate in the three angle angle sequences exceeds the threshold X, it is marked as a confirmed defect, which ensures the continuity and real-time nature of defect detection and provides reliable technical support for the assessment of the degree of damage of the defect. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flowchart of an online defect detection method and system for flaw detection combined with image analysis and processing technology of the present invention; Figure 2 The present invention is a structural schematic diagram of an online flaw detection system combining image analysis and processing technology. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] Embodiment 1: A method for online defect detection using a combined image analysis and processing technique, such as Figure 1 As shown, the following steps are included: S1: Acquire the detection image data and perform preprocessing, grid the preprocessed image and extract features; S2: divide the gridded image into even and odd parts, and calculate the angle according to the division result; S3: Build a prediction model and perform real-time prediction analysis to determine the damage degree of the defect based on the prediction results; S4: The entire system is designed as a closed loop to achieve real-time online detection of flaws.

[0018] The detection image data is obtained and preprocessed, and the preprocessed image is gridded and features are extracted, including: collecting image data at equal intervals t, cleaning and standardizing the image data; dividing the image data evenly into n*n grids, using the CNN model to extract image defect features, calculating the mean and variance of each grid defect feature, and recording the position of each defect feature in the grid.

[0019] Further explanation is that the image data is collected at equal intervals t: ensuring the time continuity and consistency of the data, avoiding data deviation caused by inconsistent collection time intervals. The image preprocessing: cleaning and standardizing the collected image data, using image processing algorithms to remove noise, adjusting brightness and contrast, and making the defect features in the image more obvious. The grid division: evenly dividing the image into n*n grids, for example, dividing a 1024x1024 image into 64 16x16 small grids. This division reduces computational complexity and improves processing efficiency, while local processing better captures small defect features. The position recording: recording the position of each defect feature in the grid, including the center coordinates and range. For example, the center coordinates of a crack feature are (10, 10), and the range is a 16x16 grid. Accurately recording the location information of the defect helps to locate and track the development of the defect and provide guidance for repair and maintenance.

[0020] The gridded image is divided into even and odd parts, and the angle is calculated according to the division result, including: marking the horizontal grids and the vertical grids of the n*n grids in an even and odd order, and the marking result is that the horizontal grids are marked in the order of 1, 2, 1, 2..., and the vertical grids are marked in the order of 1, 2, 1, 2...; obtaining the defect feature distribution, extracting the defect features 1 and 2 that are closest to each other in the horizontal direction, and extracting the defect feature 3 that is closest to each other in the vertical direction, taking the defect features 1 and 2 as the horizontal line, connecting the defect feature 3 with the defect feature 1, forming an angle between the connecting lines of the defect features 1, 2, and 3, calculating the angle angle, and classifying the angle angle.

[0021] Further explanation is that the gridded image is divided into odd and even numbers: it is easier to identify and track the distribution and changes of defects in different directions. Obtain the defect feature distribution, extract the defect features 1 and 2 with the closest interval in the horizontal direction, and extract the defect feature 3 with the closest interval in the vertical direction: by extracting the closest defect features, the relative position relationship between the defects is more accurately determined, which is helpful for subsequent angle calculation and classification. Take defect features 1 and 2 as horizontal lines, connect defect feature 3 with defect feature 1, form an angle between defect features 1 and 2, and 3, and calculate the angle: calculating the angle helps the system to more accurately describe the spatial relationship between defects, so as to better understand the morphology and distribution of defects. Classify the angle: by classifying the angle, the type and severity of the defect are further refined, providing more detailed information for subsequent defect analysis and processing. Assuming that it is used to detect surface defects of metal sheets, the gridded image is divided into odd and even numbers: the 16x16 image grid is divided into odd and even numbers, and the horizontal and vertical grids are marked in odd and even order respectively.

[0022] Classifying the angles includes: obtaining image data before defect features are formed and defining it as Class A image data, and obtaining image data in the process of defect feature formation without defect features and defining it as Class B image data; recording image change data in the process from Class A image to Class B image, and classifying the angles according to the image change data, if the change is in the horizontal direction in the process from Class A image to Class B image, then modifying defect features 1 and defect features 2, and then forming a new angle and defining it as a new horizontal angle; if the change is in the vertical direction in the process from Class A image to Class B image, then modifying defect feature 3, and then forming a new angle and defining it as a new vertical angle; if the change is neither in the horizontal direction nor in the vertical direction, then modifying defect features 1, 2, and 3 according to the angle change, and then forming a new angle and defining it as an oblique angle; and forming three angle sequences based on the three new angles.

[0023] Further explanation is that the image data before the defect features are formed is obtained and defined as Class A image data: a benchmark is established for comparison and analysis of the formation process of defect features. Class A image data provides a reference in a defect-free state, which helps to identify and quantify the occurrence and development of defects. The image data in the process of defect feature formation but without defect features is obtained and defined as Class B image data: the early stage of defect feature formation is recorded to help analyze the initial form and evolution of defects. Class B image data provides a transition stage from defect-free to defective, which helps to identify early signs of defects more finely. Record image change data in the process from Class A images to Class B images: by recording image change data, the formation process of defect features is tracked to provide detailed information for subsequent analysis. These change data help to understand the development trend and speed of defects. Classify the angle according to the image change data: by classifying the angle, the spatial relationship and evolution pattern of defect features are more accurately described, which helps to identify different types of defects. If the change from the A-type image to the B-type image is in the horizontal direction, the defect feature 1 and the defect feature 2 are modified, and then a new angle is formed and defined as the new horizontal angle: By modifying the position of the defect feature, the change of the defect in the horizontal direction is more accurately reflected, which is helpful to identify and analyze the defect extension in the horizontal direction. If the change from the A-type image to the B-type image is in the vertical direction, the defect feature 3 is modified, and then a new angle is formed and defined as the new vertical angle: By modifying the position of the defect feature, the change of the defect in the vertical direction is more accurately reflected, which is helpful to identify and analyze the defect extension in the vertical direction. If the change is neither in the horizontal direction nor in the vertical direction, the defect features 1, 2, and 3 are modified according to the angle change, and then a new angle is formed and defined as the oblique angle: By comprehensively considering the changes of the defect features in multiple directions, the complex morphology of the defect is more comprehensively described, which is helpful to identify and analyze the oblique defect extension. Based on the three new angles, three angle sequences are formed: By forming an angle sequence, the change process of the defect features is systematically recorded and analyzed, providing data support for subsequent prediction and diagnosis. Assuming that it is used to detect surface defects of metal sheets, the image data before the defect features are formed is obtained and defined as Class A image data: an image of a metal sheet without defects is collected and defined as Class A image data. Image data in the process of defect feature formation but not yet formed is obtained and defined as Class B image data: during the production process, several images of metal sheet surfaces where tiny cracks begin to appear but obvious defects have not yet been completely formed are collected and defined as Class B image data. Record the image change data from Class A images to Class B images: record the expansion process of cracks on the surface of metal sheets from Class A images to Class B images, including changes in crack length, width and distribution.Classify the angles according to the image change data: Analyze the recorded image change data, calculate the angle changes between defect features 1, 2, and 3, and classify them as horizontal angles, vertical angles, or oblique angles. If the change is in the horizontal direction from the A-type image to the B-type image: Assume that the horizontal distance between defect features 1 and 2 increases during the process from the A-type image to the B-type image, recalculate the new angle and define it as the new horizontal angle. If the change is in the vertical direction from the A-type image to the B-type image: Assume that the defect feature 3 moves in the vertical direction during the process from the A-type image to the B-type image, recalculate the new angle and define it as the new vertical angle. If the change is neither in the horizontal direction nor in the vertical direction: Assume that the defect features 1, 2, and 3 change in the oblique direction during the process from the A-type image to the B-type image, recalculate the new angle and define it as the oblique angle. Based on the three new angles, three angle sequences are formed: based on the new horizontal angle, the new vertical angle and the oblique angle, three angle sequences are formed, namely, a horizontal sequence, a vertical sequence and an oblique sequence.

[0024] Based on the three new angles, three angle sequences are formed, including: in the horizontal direction, if defect feature 3 is located on the left side of defect feature 1 in the vertical direction or defect feature 3 is located within the vertical range of defect feature 1 and defect feature 2, then the new angle data caused by the change in the angle due to the change in the defect feature forms a horizontal sequence; in the vertical direction, the new angle data formed by the change in the vertical angle forms a vertical sequence; in the oblique direction, the new angle data are sorted by the size of the angle change so that they form an oblique sequence.

[0025] Further explanation is that by recording the change of the angle in the horizontal direction, the expansion and evolution of the defect in the horizontal direction can be more accurately described, which helps to identify the defect characteristics in the horizontal direction. By recording the change of the angle in the vertical direction, the expansion and evolution of the defect in the vertical direction can be more accurately described, which helps to identify the defect characteristics in the vertical direction. By recording the change of the angle in the oblique direction, the expansion and evolution of the defect in the oblique direction can be more comprehensively described, which helps to identify the defect characteristics in the oblique direction.

[0026] Construct a prediction model and conduct real-time prediction analysis, and determine the damage degree of the defect according to the prediction results, including: obtaining a defect feature data training set, constructing an LSTM model, inputting the training set data into the LSTM model, obtaining a trained LSTM model, predicting the defect features with the LSTM model, obtaining the predicted defect features, and marking them as the first defect features; obtaining a historical defect feature data set, and at the same time obtaining a predicted defect feature data set predicted by the LSTM model, using the historical defect feature data set as the denominator and the predicted defect feature data set as the numerator, constructing a prediction factor, predicting the defect features with the prediction factor, obtaining the predicted defect features, and marking them as the second defect features; analyzing three angle sequences with the first defect feature and the second defect feature; prediction factor = predicted defect feature data set / historical defect feature data set.

[0027] Further explanation is that LSTM is a special recursive neural network that is good at processing time series data, can capture temporal dependencies, and is suitable for predicting defect features. Through the trained LSTM model, the defect features in the current detection image are predicted in real time to provide preliminary results for subsequent analysis. By comparing historical data and predicted data, the accuracy and stability of the prediction model are evaluated and the model is further optimized. By constructing prediction factors and quantifying the difference between the prediction results and historical data, the accuracy of the prediction is further improved. By correcting the prediction factors, the accuracy and reliability of the prediction are further improved. By combining the first defect feature and the second defect feature, the spatial relationship and evolution pattern of the defect features are more comprehensively analyzed to improve the accuracy of diagnosis.

[0028] The three angle sequences are analyzed with the first defect feature and the second defect feature, including: matching the first defect feature, the second defect feature and the three angle sequences; if the three angle sequences do not match the first defect feature and the second defect feature, marking the defect feature as a confirmed defect; if the three angle sequences match the second defect feature but do not match the first defect feature, marking the defect feature as a pending defect; if the three angles all match the first defect feature and the second defect feature, obtaining the angle parity change values ​​in the three angle sequences.

[0029] It is further explained that by matching the first defect feature and the second defect feature with the three angle sequences, the spatial relationship and evolution pattern of the defect features are more comprehensively analyzed to improve the accuracy of diagnosis. If none of the three angle sequences match the first defect feature and the second defect feature, it means that these features are completely different from the known defect pattern and are marked as confirmed defects, requiring further detailed inspection and processing. If the three angle sequences match the second defect feature but do not match the first defect feature, it means that these features need to be further verified and marked as pending defects for more inspection and analysis. If all three angle sequences match the first defect feature and the second defect feature, it means that these features are highly consistent with the known defect pattern, and the characteristics of the defect can be more accurately described by further analyzing the angle parity change value.

[0030] Obtaining the angle parity change values ​​in three angle sequences, including: obtaining a training set of angle parity change value data, building an SVM classification model, inputting the training set data into the SVM classification model, obtaining a trained SVM classification model, and classifying the angle parity change with the SVM classification model; wherein the classification standard is, firstly, recording the parity ratio in the three angle sequences, and secondly taking the relationship between the parity ratio and the angle change rate as the classification standard; obtaining the intersection of the three angle sequences, if the determined defect, the to-be-determined defect and the characteristic defect are respectively located in different intersections, marking them according to their respective classification rules; if the angle change rates in the three angle sequences do not exceed a threshold value (X), marking them as non-defective areas.

[0031] Further explanation is that support vector machine is a powerful machine learning algorithm suitable for classification tasks, which can effectively handle high-dimensional data and complex classification problems. By analyzing the change value of the angle parity in the intersection, the type and severity of the defect can be judged more accurately to avoid misjudgment.

[0032] A further explanation is that the intersection of defects is determined: if the angle change rates in the three angle sequences all exceed the threshold value (X), the area is marked as a determined defect. Intersection of defects to be determined: if two of the angle change rates in the three angle sequences exceed the threshold value (X), the area is marked as a defect to be determined. Intersection of characteristic defects: if one of the angle change rates in the three angle sequences exceeds the threshold value (X), the area is marked as a characteristic defect. Non-defective area: if the angle change rates in the three angle sequences do not exceed the threshold value (X), the area is marked as a non-defective area.

[0033] A further explanation is that the intersection of three angle sequences is obtained, and the specific definitions are as follows: Determined defect intersection: In the three angle sequences, the area is marked as a determined defect. Intersection of defects to be determined: In the three angle sequences, the area is marked as a defect to be determined. Intersection of characteristic defects: In the three angle sequences, the area is marked as a characteristic defect. Non-defect area: In the three angle sequences, the area is marked as a non-defect. Classification rules: Determined defect: If the angle change rate in the three angle sequences exceeds the threshold (X), it is marked as a determined defect. Defect to be determined: If two of the angle change rates in the three angle sequences exceed the threshold (X), it is marked as a defect to be determined. Characteristic defect: If one of the angle change rates in the three angle sequences exceeds the threshold (X), it is marked as a characteristic defect. Non-defect area: If the angle change rates in the three angle sequences do not exceed the threshold (X), it is marked as a non-defect area.

[0034] The entire system is constructed into a loop to realize real-time online monitoring of flaw detection defects, including: according to the classification criteria and intersection, all confirmed defects, pending defects and characteristic defects are detected, and finally the entire system is constructed into a loop system; the specific classification criteria are: if the angle change rate in the three angle sequences exceeds the threshold value X, it is marked as a confirmed defect.

[0035] A further explanation is that by analyzing the change value of the angle parity in the intersection, the type and severity of the defect can be judged more accurately to avoid misjudgment; the method for determining the threshold value X includes but is not limited to historical data and statistical analysis, empirical rules, machine learning methods, and dynamic adjustment.

[0036] Embodiment 2: Based on embodiment 1, an online defect detection system combining image analysis and processing technology includes:

[0037] Image acquisition module: used to collect detection image data; Image preprocessing module: used to preprocess the collected image data; Grid division and feature extraction module: used to evenly divide the preprocessed image data into n*n grids and extract features; Odd-even division and angle calculation module: used to perform odd-even division on the gridded image and calculate the angle according to the division result; Prediction model building and real-time prediction module: used to build prediction models and conduct real-time prediction analysis; Defect feature analysis and result output module: used to match the first defect feature with the second defect feature, determine the state of the defect feature, and output the final result.

[0038] The above is a detailed introduction to the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for online flaw detection using a combined image analysis and processing technique, characterized in that: The following steps are involved: S1: Acquire the detection image data and perform preprocessing, grid the preprocessed image and extract features; S2: divide the gridded image into even and odd parts, and calculate the angle according to the division result; S3: Build a prediction model and perform real-time prediction analysis to determine the damage degree of the defect based on the prediction results; S4: The entire system is designed as a closed loop to achieve real-time online detection of flaws.

2. The method for online flaw detection using a combined image analysis and processing technique according to claim 1 is characterized in that: The method acquires the detection image data and performs preprocessing, gridding the preprocessed image and extracting features, including: collecting image data at equal intervals t, performing data cleaning and standardization on the image data; evenly dividing the image data into n*n grids, extracting image defect features using a CNN model, calculating the mean and variance of each grid defect feature, and recording the position of each defect feature in the grid.

3. The method for online flaw detection using a combined image analysis and processing technique according to claim 1 is characterized in that: The method performs odd-even division on the gridded image and calculates the angle according to the division result, including: marking the horizontal grids and the vertical grids of n*n grids in an odd-even order, wherein the marking result is that the horizontal grids are marked in the order of 1, 2, 1, 2..., and the vertical grids are marked in the order of 1, 2, 1, 2...; obtaining the defect feature distribution, extracting the defect features 1 and 2 which are closest to each other in the horizontal direction, and extracting the defect feature 3 which is closest to each other in the vertical direction, connecting the defect feature 3 with the defect feature 1 with the defect features 1 and 2 as the horizontal line, forming an angle between the connecting lines of the defect features 1, 2, and 3, calculating the angle, and classifying the angle.

4. The method for online flaw detection combined with image analysis and processing technology according to claim 3 is characterized in that: The classification of the angles includes: obtaining image data before defect features are formed and defining it as Class A image data, and obtaining image data in the process of defect feature formation without defect features and defining it as Class B image data; recording image change data in the process from Class A image to Class B image, and classifying the angles according to the image change data, if the change is in the horizontal direction in the process from Class A image to Class B image, then modifying defect features 1 and defect features 2, and then forming a new angle and defining it as a new horizontal angle; if the change is in the vertical direction in the process from Class A image to Class B image, then modifying defect feature 3, and then forming a new angle and defining it as a new vertical angle; if the change is neither in the horizontal direction nor in the vertical direction, then modifying defect features 1, 2, and 3 according to the angle change, and then forming a new angle and defining it as an oblique angle; and forming three angle sequences based on the three new angles.

5. The method for online flaw detection combined with image analysis and processing technology according to claim 4 is characterized in that: The three new angles are used as a reference to form three angle sequences, including: in the horizontal direction, if defect feature 3 is located on the left side of defect feature 1 in the vertical direction or defect feature 3 is located within the vertical range of defect feature 1 and defect feature 2, then the new angle data resulting from the change in the angle due to the change in the defect feature forms a horizontal sequence; in the vertical direction, the new angle data formed by the change in the vertical angle forms a vertical sequence; in the oblique direction, the new angle data are sorted by the size of the angle change so that they form an oblique sequence.

6. The method for online flaw detection combined with image analysis and processing technology according to claim 1 is characterized in that: The method comprises: obtaining a defect feature data training set, constructing an LSTM model, inputting the training set data into the LSTM model, obtaining a trained LSTM model, predicting the defect feature with the LSTM model, obtaining the predicted defect feature, and marking it as a first defect feature; obtaining a historical defect feature data set, and simultaneously obtaining a predicted defect feature data set predicted by the LSTM model, using the historical defect feature data set as a denominator and the predicted defect feature data set as a numerator, constructing a prediction factor, predicting the defect feature with the prediction factor, obtaining the predicted defect feature, and marking it as a second defect feature; analyzing three angle sequences with the first defect feature and the second defect feature; prediction factor = predicted defect feature data set / historical defect feature data set.

7. The method for online flaw detection combined with image analysis and processing technology according to claim 6 is characterized in that: The analyzing of the three angle sequences with the first defect feature and the second defect feature includes: matching the first defect feature, the second defect feature and the three angle sequences; if the three angle sequences do not match the first defect feature and the second defect feature, marking the defect feature as a confirmed defect; if the three angle sequences match the second defect feature but do not match the first defect feature, marking the defect feature as a to-be-determined defect; if the three angles all match the first defect feature and the second defect feature, obtaining the angle parity change values ​​in the three angle sequences.

8. The method for online flaw detection combined with image analysis and processing technology according to claim 7 is characterized in that: The method of obtaining the angle parity change values ​​in the three angle sequences includes: obtaining a training set of angle parity change value data, constructing an SVM classification model, inputting the training set data into the SVM classification model, obtaining a trained SVM classification model, and classifying the angle parity change with the SVM classification model; wherein the classification standard is, firstly, recording the parity ratio in the three angle sequences, and secondly taking the relationship between the parity ratio and the angle change rate as the classification standard; obtaining the intersection of the three angle sequences, and if the determined defect, the to-be-determined defect and the characteristic defect are respectively located in different intersections, marking them according to their respective classification rules; and if the angle change rates in the three angle sequences do not exceed a threshold value (X), marking them as non-defective areas.

9. The method for online flaw detection combined with image analysis and processing technology according to claim 1 is characterized in that: The entire system is constructed into a loop to realize real-time online monitoring of flaw detection defects, including: according to the classification standard and the intersection, all the determined defects, the defects to be determined and the characteristic defects are detected, and finally the entire system is constructed into a loop system; the specific classification standard is: if the angle change rate in the three angle angle sequences exceeds the threshold value X, it is marked as a determined defect.

10. An online defect detection system for flaw detection using the combined image analysis and processing technology described in any one of claims 1 to 9, characterized in that: Included are: Image acquisition module: used to collect detection image data; Image preprocessing module: used to preprocess the collected image data; Grid division and feature extraction module: used to evenly divide the preprocessed image data into n*n grids and extract features; Odd-even division and angle calculation module: used to perform odd-even division on the gridded image and calculate the angle according to the division result; Prediction model building and real-time prediction module: used to build prediction models and conduct real-time prediction analysis; Defect feature analysis and result output module: used to match the first defect feature with the second defect feature, determine the state of the defect feature, and output the final result.

Images