A feature extraction and identification method for phased array ultrasonic lack of fusion defects

Abstract

The invention discloses a feature extraction and recognition method for phased array ultrasonic unfusion defects. The method takes Olympus detection equipment as the main object, uses a phased array ultrasonic data analysis and storage algorithm, a data reconstruction algorithm and a three-dimensional imaging algorithm to present a weld area and corresponding defects in the form of a three-dimensional image, extracts data of the length and angle from the groove as feature parameters, and uses a support vector machine algorithm for learning. The invention has the following beneficial effects: eigenvalues ​​are easier to obtain, and eigenvalues ​​are defined for the morphological features of the unfusion defects without the need for wavelet decomposition, time-frequency transformation and other transformation methods, so that the efficiency is higher; fewer samples are required, and there is currently no unified public data set for weld phased array ultrasound, and the training samples required for welds are actually difficult to obtain, so the SVM algorithm is selected for defect recognition; the accuracy is higher, the selected eigenvalues ​​are more targeted, and the recognition result of the unfusion defect has a higher accuracy than that in other documents.

Description

Technical Field

[0001] The invention relates to the field of electronic information technology, and in particular to a method for extracting and identifying features of phased array ultrasonic unfusion defects. Background Art

[0002] With the development of artificial intelligence and big data technology, the method of industrial ultrasonic detection is moving towards automation and intelligence. Among them, phased array ultrasound, as a new technology in the field of non-destructive testing, has the advantages of high detection sensitivity and safe operation. At the same time, by using the characteristics of multi-angle scanning, the detection results can be displayed in three-dimensional form. Relevant personnel can directly judge based on the results such as images, which greatly improves the detection efficiency and accuracy.

[0003] Defect recognition is performed based on the texture features of the lack of fusion defect. There are many ways to represent texture features, among which the grayscale co-occurrence matrix is ​​a commonly used method, which represents the brightness distribution of pixels and the position distribution characteristics between pixels of the same brightness. According to the grayscale co-occurrence matrix, local stationarity, echo energy value, contrast, entropy value, correlation and other characteristics can be extracted. The characteristics of the lack of fusion defect are small energy value, small correlation change in four directions, weak directionality, and are similar to the characteristics of the lack of penetration defect in the weld. The experimental process is to calculate the correlation between the detected defect and the reference defect image to obtain the category of the detected defect. Since the lack of fusion and the lack of penetration defects in the weld are both area defects, and the extracted texture features are relatively similar, the recognition result will produce misjudgment, so that the final recognition rate is only about 71%, which is far from meeting the requirements of industrial production.

[0004] Since the echo amplitude of the unfused defect is closely related to the echo energy, the change of signal energy in a certain frequency band characterizes the type of defect. When the ultrasonic signal encounters a defect in the medium, defects with different boundaries have different degrees of enhancement and suppression on the amplitude-frequency and phase-frequency characteristics of each frequency band of the ultrasonic signal, so that the energy difference of different defects in the same frequency band will be very large. The wavelet packet decomposition method can completely decompose the defect signal into independent frequency bands, which is conducive to reflecting the nature and characteristics of the defect signal. At the same time, the use of fractal technology to detect irregular signals is also an effective method. The fractal feature is extracted based on the wavelet fractal as another important feature for defect classification in welds. However, when using multiple features for defect classification, due to the large number of selected features, genetic algorithms, dimensionality reduction algorithms, etc. are often required to simplify the feature parameters, which increases the complexity of recognition and reduces the efficiency of recognition. A three-layer neural network is used for training. By adjusting the network parameters, the final recognition rate reaches 75%, which is still insufficient to meet the recognition rate requirements for automatic identification of defects in industrial welds. The support vector machine was used for automatic defect identification, and the final recognition rate of the lack of fusion defect was 85.7%, which still did not meet the accuracy requirements of automatic defect identification in industry. The distribution of echo energy of the lack of fusion defect and the lack of penetration defect in the weld is relatively similar. The use of wavelet decomposition method to extract eigenvalues ​​for classification will result in a high proportion of misjudgment of these two defects. Summary of the invention

[0005] The object of the present invention is to provide a method for extracting and identifying features of phased array ultrasonic incomplete fusion defects to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution, the steps of which are as follows:

[0007] Step 1: Use Olympus monitoring equipment to scan the test sample, and use phased ultrasonic data analysis and storage algorithm, data reconstruction algorithm and three-dimensional imaging algorithm to jointly calculate and simulate modeling, and present the weld area and corresponding defects in three dimensions;

[0008] Step 2: Based on the three-dimensional model obtained in step 1, the length and angle of the unfused defect from the groove are selected as characteristic parameters for identifying the defect;

[0009] Step 3: Manually select the characteristic parameters of the unfused sample and the characteristic parameters of the non-unfused sample obtained in step 2, divide them into training set, validation set and test set in a ratio of 7:2:1, and use support vector machine (SVM) algorithm for learning and recognition.

[0010] As a preferred technical solution of the present invention, the length of the unfusion defect from the groove in step 2 is the distance information of the side of the unfusion defect close to the groove from the groove.

[0011] As a preferred technical solution of the present invention, a method for measuring the distance information of the side of the unfusion defect close to the groove from the groove is as follows: sampling is performed at an even interval along the contour of the unfusion defect close to the groove, and the sampling points are 20-40. After the sampling is completed, the sampling is sorted and the middle 3 data values ​​are retained.

[0012] As a preferred technical solution of the present invention, the angle of the unfusion defect from the groove in step 2 is the angle between the side of the unfusion defect close to the groove and the groove in the same plane.

[0013] As a preferred technical solution of the present invention, a method for measuring the angle between a side of an unfusion defect close to the groove and the groove in the same plane is as follows: sampling is performed at average intervals along the contour of the unfusion defect close to the groove, and the acquired sampling points of the unfusion defect close to the groove are linearly fitted to obtain a straight line where the defect boundary is located, and then the angle between the boundary straight line of the unfusion defect and the straight line where the corresponding groove is located is calculated by a translation method.

[0014] As a preferred technical solution of the present invention, in step three, the ratio of the characteristic parameters of the unfused sample to the characteristic parameters of the non-unfused sample is 1:1.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] 1. Eigenvalues ​​are easier to obtain. Eigenvalues ​​are defined based on the morphological characteristics of unfused defects, without the need for wavelet decomposition, time-frequency transform and other transformation methods, which is more efficient.

[0017] 2. Fewer samples are required. There is currently no unified public data set for weld phased array ultrasound, and the training samples required for welds are actually difficult to obtain, so the SVM algorithm is selected for defect recognition.

[0018] 3. Higher accuracy. The selected feature values ​​are more targeted, and the recognition results of unfused defects are more accurate than those in other literature. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of groove information of the side wall unfused defect of the present invention;

[0020] Figure 2 It is a schematic diagram of the distance from the side wall unfusion defect to the corresponding groove of the present invention;

[0021] Figure 3 It is a schematic diagram of the side wall unfusion defect and the corresponding groove angle of the present invention; DETAILED DESCRIPTION

[0022] Example 1

[0023] like Figures 1 to 3 As shown, the present invention discloses a method for extracting and identifying features of phased array ultrasonic unfusion defects, and the steps are as follows:

[0024] Step 1: Use Olympus monitoring equipment to scan the test sample, and use phased ultrasonic data analysis and storage algorithm, data reconstruction algorithm and three-dimensional imaging algorithm to jointly calculate and simulate modeling, and present the weld area and corresponding defects in three dimensions;

[0025] Step 2: Based on the three-dimensional model obtained in step 1, the length and angle of the unfused defect from the groove are selected as characteristic parameters for identifying the defect;

[0026] Step 3: Manually select the characteristic parameters of the unfused sample and the characteristic parameters of the non-unfused sample obtained in step 2, divide them into training set, validation set and test set in a ratio of 7:2:1, and use support vector machine (SVM) algorithm for learning and recognition.

[0027] As a preferred technical solution of the present invention, the length of the unfusion defect from the groove in the step 2, that is, the distance information of the side of the unfusion defect close to the groove from the groove, is measured as follows: along the contour of the unfusion defect close to the groove, sampling is performed at an average interval, and the number of sampling points is 20. After the sampling is completed, the sampling is sorted and the middle 3 data values ​​are retained.

[0028] As a preferred technical solution of the present invention, the angle between the unfusion defect and the groove in the step 2, that is, the angle between the side of the unfusion defect close to the groove and the groove in the same plane, is measured as follows: along the contour of the unfusion defect close to the groove, sampling is performed at an average interval, and the obtained sampling points of the unfusion defect close to the groove are linearly fitted to obtain the straight line where the defect boundary is located, and then the angle between the boundary straight line of the unfusion defect and the straight line where the corresponding groove is located is calculated by translation.

[0029] As a preferred technical solution of the present invention, in step three, the ratio of the characteristic parameters of the unfused sample to the characteristic parameters of the non-unfused sample is 1:1.

[0030] Example 2

[0031] like Figures 1 to 3 As shown, the present invention discloses a method for extracting and identifying features of phased array ultrasonic unfusion defects, and the steps are as follows:

[0032] Step 1: Use Olympus monitoring equipment to scan the test sample, and use phased ultrasonic data analysis and storage algorithm, data reconstruction algorithm and three-dimensional imaging algorithm to jointly calculate and simulate modeling, and present the weld area and corresponding defects in three dimensions;

[0033] Step 2: Based on the three-dimensional model obtained in step 1, the length and angle of the unfused defect from the groove are selected as characteristic parameters for identifying the defect;

[0034] Step 3: Manually select the characteristic parameters of the unfused sample and the characteristic parameters of the non-unfused sample obtained in step 2, divide them into training set, validation set and test set in a ratio of 7:2:1, and use support vector machine (SVM) algorithm for learning and recognition.

[0035] As a preferred technical solution of the present invention, the length of the unfusion defect from the groove in the step 2, that is, the distance information of the side of the unfusion defect close to the groove from the groove, is measured as follows: along the contour of the unfusion defect close to the groove, sampling is performed at an average interval, and the number of sampling points is 30. After the sampling is completed, the sampling is sorted and the middle 3 data values ​​are retained.

[0036] The angle between the unfusion defect and the groove in the step 2, i.e., the angle between the side of the unfusion defect close to the groove and the groove in the same plane, is measured as follows: sampling is performed at average intervals along the contour of the unfusion defect close to the groove, and the obtained sampling points of the unfusion defect close to the groove are linearly fitted to obtain the straight line where the defect boundary is located, and then the angle between the boundary straight line of the unfusion defect and the straight line where the corresponding groove is located is calculated by translation.

[0037] As a preferred technical solution of the present invention, in step three, the ratio of the characteristic parameters of the unfused sample to the characteristic parameters of the non-unfused sample is 1:1.

[0038] Example 3

[0039] like Figures 1 to 3 As shown, the present invention discloses a method for extracting and identifying features of phased array ultrasonic unfusion defects, and the steps are as follows:

[0040] Step 1: Use Olympus monitoring equipment to scan the test sample, and use phased ultrasonic data analysis and storage algorithm, data reconstruction algorithm and three-dimensional imaging algorithm to jointly calculate and simulate modeling, and present the weld area and corresponding defects in three dimensions;

[0041] Step 2: Based on the three-dimensional model obtained in step 1, the length and angle of the unfused defect from the groove are selected as characteristic parameters for identifying the defect;

[0042] Step 3: Manually select the characteristic parameters of the unfused sample and the characteristic parameters of the non-unfused sample obtained in step 2, divide them into training set, validation set and test set in a ratio of 7:2:1, and use support vector machine (SVM) algorithm for learning and recognition.

[0043] As a preferred technical solution of the present invention, the length of the unfusion defect from the groove in the step 2, that is, the distance information of the side of the unfusion defect close to the groove from the groove, is measured as follows: along the contour of the unfusion defect close to the groove, sampling is performed at an average interval, and the number of sampling points is 40. After the sampling is completed, the sampling is sorted and the middle 3 data values ​​are retained.

[0044] The angle between the unfusion defect and the groove in the step 2, i.e., the angle between the side of the unfusion defect close to the groove and the groove in the same plane, is measured as follows: sampling is performed at average intervals along the contour of the unfusion defect close to the groove, and the obtained sampling points of the unfusion defect close to the groove are linearly fitted to obtain the straight line where the defect boundary is located, and then the angle between the boundary straight line of the unfusion defect and the straight line where the corresponding groove is located is calculated by translation.

[0045] As a preferred technical solution of the present invention, in step three, the ratio of the characteristic parameters of the unfused sample to the characteristic parameters of the non-unfused sample is 1:1.

[0046] Components not described in detail herein are prior art.

[0047] Although the specific embodiments of the present invention are described in detail above, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in this field without departing from the purpose of the present invention, and modifications or deformations without creative labor are still within the protection scope of the present invention.

Claims

1. A method for extracting and identifying features of phased array ultrasonic incomplete fusion defects, characterized in that: The steps are as follows: Step 1: Use Olympus monitoring equipment to scan the test sample, and use phased ultrasonic data analysis and storage algorithm, data reconstruction algorithm and three-dimensional imaging algorithm to jointly calculate and simulate modeling, and present the weld area and corresponding defects in three dimensions; Step 2: Based on the three-dimensional model obtained in step 1, the length and angle of the unfused defect from the groove are selected as characteristic parameters for identifying the defect; Step 3: Manually select the characteristic parameters of the unfused sample and the characteristic parameters of the non-unfused sample obtained in step 2, divide them into a training set, a validation set, and a test set in a ratio of 7:2:1, and use a support vector machine (SVM) algorithm for learning and recognition; The length of the unfusion defect from the groove in step 2 is the distance information of the side of the unfusion defect close to the groove from the groove; the distance information is measured by sampling along the contour of the unfusion defect close to the groove at an average interval, and the sampling points are 20-40, and after the sampling is completed, the sampling is sorted and the middle 3 data values ​​are retained; The angle between the unfusion defect and the groove in step 2 is the angle between the side of the unfusion defect close to the groove and the groove in the same plane; the angle measurement method is as follows: along the contour of the unfusion defect close to the groove, sampling is performed at an average interval, and the obtained sampling points of the unfusion defect close to the groove are linearly fitted to obtain the straight line where the defect boundary is located, and then the angle between the boundary straight line of the unfusion defect and the straight line where the corresponding groove is located is calculated by translation.

2. According to the method for feature extraction and identification of phased array ultrasonic lack of fusion defects as described in claim 1, it is characterized in that: The ratio of the characteristic parameters of the unfused sample and the characteristic parameters of the non-unfused sample in step 3 is 1:1.

Images