Similarity weld defect grade evaluation method based on multi-source information fusion

Abstract

The present invention provides a similarity weld defect grade evaluation method based on multi-source information fusion, relating to the technical field of weld defect grade evaluation. This similarity weld defect grade evaluation method based on multi-source information fusion addresses the current issues of defect grade evaluation using TOFD atlas data, which is primarily manual, highly subjective, inefficient, and lacks unstructured data such as standard defect grade atlases. By analyzing the characteristics of TOFD defect detection data and establishing an evaluation system based on detection standards and expert domain knowledge, this method achieves the construction of a defect standard database. Furthermore, by identifying and judging defect types, a Mahalanobis distance algorithm is used to establish a defect grade evaluation method based on the fusion similarity metric of Gabor grayscale co-occurrence matrix features. This method has high accuracy, expands existing image defect grade evaluation methods, and can be extended to other image grade determination fields, demonstrating its high versatility.

Description

Technical Field

[0001] The present invention relates to the technical field of weld defect grade evaluation, and in particular to a similarity weld defect grade evaluation method based on multi-source information fusion. Background Art

[0002] Welding processes are extensively used in the manufacturing of large-scale equipment in sectors such as energy and power, aerospace, and shipbuilding. The detection and evaluation of weld defects has become a bottleneck restricting the efficient manufacturing of current equipment and a crucial means of ensuring the safe and reliable operation of equipment during service. Ultrasonic time-of-flight diffraction (TOFD), with its advantages of rich detection information, strong noise immunity, high efficiency, and accurate positioning and quantitative analysis, has become one of the most widely used nondestructive testing methods for welds. For example, the Shaanxi Provincial Special Inspection Institute uses TOFD technology in its special equipment weld inspection and testing, accumulating over 50,000 meters of TOFD weld image data throughout 2020. Currently, defect assessment using data primarily relies on manual methods, referencing standards for defect grade. This involves technicians making judgments based on their own experience and professional background. This process is labor-intensive, highly subjective, and results lack consistency, making it difficult to achieve standardized and accurate defect assessment. Therefore, developing a weld defect grade assessment method based on TOFD image data is of great significance for improving defect analysis and determination capabilities and enhancing the safe and serviceable quality of equipment.

[0003] With the increasing application of TOFD inspection technology and the accumulation of inspection data, the use of artificial intelligence to intelligently assess defect grades, improve defect identification efficiency, and reduce inconsistencies in manual identification has become a key concern. Intelligent, automated assessment of defect grades in welds for nondestructive testing of carbon steel, low-alloy steel, and especially titanium alloys has long been a technology of great interest to companies, but research is currently limited both domestically and internationally. These methods primarily focus on constructing labeled data with defect grades and using existing AI models to classify them. However, as demonstrated by table lookup analysis and image comparison methods in practice, defect grade assessment requires multiple elements, including both structured data (such as rating tables and quantitative descriptions) and unstructured data (such as standard defect grade atlases). Therefore, existing assessment methods based solely on "image classification models" fall far short of industry accuracy requirements. Currently, there are no references available domestically or internationally for TOFD defect grade assessment.

[0004] In summary, the current defect grade assessment work mainly uses manual rating methods, which have problems such as low defect grade evaluation efficiency, high labor intensity, high personnel qualification requirements, high subjectivity in the assessment process, and poor consistency of results. The TOFD defect grade assessment method based on artificial intelligence is currently less studied and is not yet mature. The main reason is the lack of visual analysis of national standards and specifications related to weld defect quality. The assessment is based solely on the size of non-destructive testing defects. The assessment process and results are bound to be untrustworthy and unreliable and difficult to apply in enterprises. Therefore, the intelligent defect grade assessment method needs to solve two problems: one is to establish a quantifiable defect grade evaluation index system, and the other is to realize the intelligent matching decision problem of grade mode.

[0005] To address the above problems, we have developed a new similarity weld defect grade evaluation method based on multi-source information fusion. Summary of the Invention

[0006] (1) Technical problems solved

[0007] In response to the shortcomings of the existing technology, the present invention provides a similarity weld defect grade evaluation method based on multi-source information fusion, which solves the problems in the existing technology that the current defect grade assessment work mainly adopts the manual rating method, which has the disadvantages of low defect grade evaluation efficiency, high labor intensity, high personnel qualification requirements, high subjectivity in the assessment process and poor consistency of results, as well as the fact that the TOFD defect grade assessment method based on artificial intelligence is currently less studied and is not yet mature.

[0008] (2) Technical solution

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions: A similarity weld defect grade evaluation method based on multi-source information fusion includes the following specific steps:

[0010] Step 1: The established weld defect grade evaluation index system is used to quantitatively describe the standard reference map, and then the weld quality evaluation method process is designed in combination with the actual weld quality evaluation needs of the enterprise;

[0011] Step 2: Perform TOFD inspection on the weld to obtain a TOFD map, and pre-process the weld map;

[0012] Step 3: Detect the weld defects in the atlas and classify and identify the defects;

[0013] Step 4: Calculate the geometric dimensions of the defect, assess the grade according to the defect standard, calculate the final grade of the defect and generate a report.

[0014] Furthermore, a method for measuring the similarity of defects in images is provided:

[0015] Suppose there are two vectors and , the Mahalanobis distance is selected as the measure between the texture features of weld defects and the features of defects in the standard reference defect map:

[0016] (1)

[0017] Where: Ma is the Mahalanobis distance; is the covariance matrix of and .

[0018] Furthermore, an algorithm for calculating the geometric size of a defect is provided, comprising:

[0019] The sizes of each defect are: The indicators that need to be calculated are the characteristic vector composed of l and h of a single defect (non-point defect) and the texture feature indicators: ASM (energy), ENT (entropy), COR (correlation), IDM (homogeneity), and CON (contrast).

[0020] Furthermore, an algorithm for defect level evaluation is provided, including:

[0021] At the same time, the wavelet coefficient modulus mean and standard deviation of the output image of the Gabor kernel 0°, 45°, 90° and 135° are used as feature vectors for calculation;

[0022] Among them, Mahalanobis distance is used as a measure of the texture features between the atlas to be evaluated and the standard reference atlas;

[0023] The Mahalanobis distance Gabor kernel mean and standard deviation are used as a measure between the Gabor kernel features of the evaluated map and the standard reference map.

[0024] (3) Beneficial effects

[0025] The present invention provides a similarity weld defect grade evaluation method based on multi-source information fusion, which has the following beneficial effects:

[0026] This multi-source information fusion-based similarity weld defect grading method addresses the current issues with defect grading using ultrasonic time-of-flight diffraction (TOFD) image data, which is primarily manual, subjective, inefficient, and lacks standard defect grading atlases. By analyzing the characteristics of TOFD defect inspection data and establishing a multi-source information fusion evaluation system based on inspection standards and expert domain knowledge, this method enables the construction of a defect standard database. By identifying and judging defect types, and building upon the established defect grade index system and its quantitative indicators, a defect grade assessment method based on the Gabor-gray level co-occurrence matrix feature fusion similarity metric is developed using the Mahalanobis distance algorithm. The proposed method is demonstrated and validated using actual TOFD weld defect data from an enterprise. The results show that the proposed Gabor-gray level co-occurrence matrix fusion similarity metric achieves an accuracy of over 93.33% for defect grade assessment. Furthermore, the proposed method expands existing image defect grade assessment methods and can be applied to other image grade determination fields, demonstrating its strong versatility. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart for evaluating weld defect levels according to the present invention;

[0028] Figure 2 A bar chart showing the distribution of defect type samples in an example of the present invention;

[0029] Figure 3 This is a detection image of a stripe defect in an example of the present invention. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0031] 1. Example

[0032] like Figure 1 - Figure 3 As shown, the embodiment of the present invention provides a similarity weld defect level evaluation method based on multi-source information fusion, which includes the following specific steps:

[0033] Step 1: The established weld defect grade evaluation index system is used to quantitatively describe the standard reference map, and then the weld quality evaluation method process is designed in combination with the actual weld quality evaluation needs of the enterprise;

[0034] Among them, TOFD testing, a conventional method for detecting defects, utilizes diffraction energy from the corners and endpoints of internal defects within the test specimen. According to the NB / T47013.10-2015 standard "Nondestructive Testing of Pressure Equipment - Part 10: Time-of-Flight Diffraction Ultrasonic Testing" and the BS7706 standard, TOFD defect classification is based on the echo characteristics and phase state of the emitter, characterizing the texture, signal-to-noise ratio, and morphology of the defect. Defects such as upper surface openings, lower surface openings, buried cracks, and lack of fusion are classified as Level III and are independent of the test thickness. Other defect types are strongly correlated with the test thickness. The number of different defect types in the constructed standard reference map is shown in Table 1. Defects such as upper surface openings, lower surface openings, buried cracks, and lack of fusion are applicable to all test thicknesses. The test thickness for other defect types is based on the engineering thickness tested in the standard map.

[0035] Furthermore, based on the acceptance criteria for TOFD inspection quality, the number, size, and texture characteristics of defects were modeled from the perspectives of geometry and texture. Factors affecting the weld defect grade, such as geometric size, texture characteristics, and defect category, were integrated with geometric characteristic indicators such as detection thickness and maximum defect length, texture characteristic indicators such as energy and entropy, as well as indicators such as defect location and defect type in weld defect rating. A weld defect grade evaluation index system was established, resulting in a set of indicator feature vectors.

[0036] The mapping relationship between the characteristics of weld defects in actual production and the characteristics of defects in the standard reference atlas is essentially a measure of the similarity between the two. Taking strip defects as an example, the feature vector of the strip defect in actual detection is measured based on the similarity with the feature vector of the standard reference atlas to obtain its grade.

[0037] There are many methods to measure the similarity of defects in images, such as KL divergence, Euclidean distance, Minkowski distance, cosine similarity, Pearson correlation coefficient, Mahalanobis distance, etc.

[0038] Here is a method to measure the similarity of defects in images:

[0039] Suppose there are two vectors and , then several commonly used calculation formulas are used. According to this paper, a defect quality grade evaluation index system is established. Since the Mahalanobis distance does not consider the dimension of the two variables, the Mahalanobis distance is selected as the measurement method between the texture characteristics of the weld defect and the characteristics of the defect in the standard reference defect map:

[0040] (1)

[0041] Where: Ma is the Mahalanobis distance; is the covariance matrix of and .

[0042] Step 2: Perform TOFD inspection on the weld to obtain a TOFD map, and pre-process the weld map;

[0043] Step 3: Detect the weld defects in the atlas and classify and identify the defects;

[0044] Step 4: Calculate the geometric dimensions of the defect, perform a grade assessment based on the defect standard, calculate the final grade of the defect and generate a report;

[0045] An algorithm for calculating the geometric size of a defect is provided, comprising:

[0046] The size of each defect is: The indicators to be calculated are the feature vector composed of l and h of a single defect (non-point defect) and the feature vector composed of texture feature indicators: ASM, ENT, COR, IDM, and CON;

[0047] Furthermore, an algorithm for defect level evaluation is provided, including:

[0048] At the same time, the wavelet coefficient modulus mean and standard deviation of the output image of the Gabor kernel 0°, 45°, 90° and 135° are used as feature vectors for calculation;

[0049] Among them, Mahalanobis distance is used as a measure of the texture features between the atlas to be evaluated and the standard reference atlas;

[0050] The Mahalanobis distance Gabor kernel mean and standard deviation are used as a measure between the Gabor kernel features of the evaluated map and the standard reference map.

[0051] In summary, after extensive research on the quality evaluation methods of titanium alloy welds in actual production, the overall process of welds from initial inspection to quality evaluation is mainly as follows: TOFD inspection of welds to obtain TOFD maps - weld map preprocessing - weld defect detection - defect classification - grade assessment according to defect standards - report generation.

[0052] First, the index system established above is used to quantitatively describe the standard reference map. Then, combined with the actual weld quality evaluation needs of enterprises, the weld quality evaluation method process is designed as follows: Figure 1 shown.

[0053] 2. Example

[0054] In order to verify the effectiveness of the titanium alloy weld defect grade evaluation index system proposed in this paper, we first combined the above-mentioned method and took the data from the actual TOFD inspection process of the enterprise as an example to conduct a classification test on 163 TOFD inspection images.

[0055] There are seven types of samples, including upper surface opening, lower surface opening, crack, lack of fusion, point defect, line defect and strip defect. The sample quantity distribution is as follows: Figure 2 As shown, the quality grade is assessed; then the atlas is evaluated and compared with the conclusions given by actual reviewers to verify the accuracy of this method.

[0056] 2.1 Quantitative description of defect standards with reference to TOFD images

[0057] According to the evaluation index content in the weld defect grade evaluation index system, the standard reference atlas is quantitatively described, and the defect texture feature results are calculated according to the gray-level co-occurrence matrix.

[0058] 2.2 Weld defect grade evaluation

[0059] 1) Calculate the geometric dimensions of the defect

[0060] like Figure 3 The sizes of the defects shown are: The indicator to be calculated is the characteristic vector composed of l and h of a single defect (non-point defect). The calculation results are as follows:

[0061] 1# Defect:

[0062] (2)

[0063] 2) Defect level evaluation

[0064] For defect #1, the geometric vectors l and h need to be calculated as (2). The texture feature indexes are the feature vectors composed of ASM, ENT, COR, IDM, and CON. The calculation results are as shown in (3):

[0065] (3)

[0066] At the same time, the wavelet coefficient modulus mean and standard deviation of the output image of the Gabor kernel 0°, 45°, 90° and 135° for defect 1# are used as feature vectors for calculation:

[0067] Mean:

[0068]

[0069] Standard deviation:

[0070]

[0071] According to the above results, combined with the weld defect grade evaluation process, Table 2 selects the standard reference spectrum quantitative indicators for buried defects. From the geometric vector indicators, it can be seen that the defect grade is grade III;

[0072] The Mahalanobis distance is used as a measure of the texture features between the atlas to be evaluated and the standard reference atlas. The calculation results are shown in Table 2.

[0073] The Mahalanobis distance Gabor kernel mean and standard deviation are used as the measurement method between the Gabor kernel features of the evaluated map and the standard reference map. The calculation results are shown in Table 3.

[0074] 3) Comparison of weld defect grade evaluation experiments

[0075] To verify the consistency of this method with the evaluation results of actual reviewers, 97 TOFD images produced in actual enterprise production were selected and the defect level was assessed using this method. The results were compared with the evaluation results of three reviewers in actual production. The results are shown in Table 4.

[0076] Table 1 Standard reference spectrum example

[0077]

[0078] Table 2: Texture feature index measurement of actual strip defects and standard reference atlas

[0079]

[0080] Table 3 Texture feature index measurement of actual strip defects and standard reference atlas

[0081]

[0082] Table 4 Comparison of this application method and the evaluation results of the reviewers

[0083]

[0084]

[0085]

[0086]

[0087] To analyze the table above, based on actual conditions, we define consistency as follows: For a particular atlas, if the evaluation results of this method and the reviewers are the same, then the evaluation results of this method for this atlas are consistent. The consistency rate is defined as the ratio of the number of atlases with consistent evaluation results using this method to the total number of atlases evaluated using this method, as shown in Equation (4).

[0088] (4)

[0089] Where: is the consistency rate; The number of atlases with consistent results for the method assessment of this application; This is the number of all spectra evaluated using the method of this application.

[0090] Analysis of Table 4 shows that for the 97 images evaluated using our method, the consistency rate based on the mean and standard deviation of Gabor features was 76.28%, the consistency rate based on gray-level co-occurrence matrix features was 75.25%, and the consistency rate of our method was 82.47%. Specifically, for individual defects of relatively large, critical area defects such as cracks and unfused defects in images 1-38, the consistency rate of the evaluation results based on Gabor features was 94.73%, while the consistency rate based on gray-level co-occurrence matrix features was 63.15%, and the consistency rate of our method was 100%. This shows that Gabor features are highly capable of extracting the edge features of area defects. For individual point defects in images 39-77, the consistency rate between our method and the evaluation results based on gray-level co-occurrence matrix features was 82.05%, while the consistency rate based on Gabor features was 64.10%. Since individual point defects only require intuitive comparison based on geometric features, it is easy to conclude that they are consistent. For the strip defects in atlas numbers 78 to 88 and the linear defects in atlas numbers 89 to 97, the consistency rates of the assessment results based on Gabor features were 63.63% and 100%, respectively. The consistency rates of the strip defect assessment results based on gray-level co-occurrence matrix features were 72.72% and 66.66%, respectively. For the defect assessment results of the present application method, the consistency rates were 72.72% and 100%, respectively. It can be seen that for strip defects, the consideration of their abstract texture features is more important, so it is more difficult to draw a consistent conclusion. For linear defects, due to the need to comprehensively consider intuitive geometric features and more abstract texture features, and the greater influence of geometric features, there will be some deviation in the conclusion.

[0091] Therefore, combining the advantages of the gray-level co-occurrence matrix in spatial distribution with the advantages of the Gabor kernel in local structural detail, the resulting joint texture feature can better describe the texture characteristics of TOFD images. Furthermore, although the similarity metric method yielded inconsistent conclusions for point defects, it can be seen that the proposed method provides a safer defect grade, especially for more hazardous area defects, where the proposed method achieved 100% accuracy. This demonstrates that the proposed method is safer for weld defect grade assessment and meets the requirements for weld quality assessment in actual production.

[0092] Based on the above analysis, the TOFD weld defect grade evaluation index system proposed in this paper fits the actual production of enterprises, realizes the quantitative assessment of product weld quality grade, and can meet the use needs of enterprises.

[0093] 3. Conclusion

[0094] This paper proposes a quantitative evaluation index system and evaluation method for weld defect levels based on multi-source information fusion. The effectiveness of this method is verified by using TOFD data examples. The main conclusions are as follows:

[0095] (1) Based on the structured data elements (assessment table and quantitative description) of the standard specifications, and based on the TOFD defect data in engineering, combined with the judgment of field experts, a typical defect grade case library was constructed, which supplemented the unstructured data elements in the TOFD weld defect grade assessment (such as the standard defect grade atlas).

[0096] (2) Based on TOFD standards and expert domain knowledge, an evaluation system and quantitative indicators of multi-source information fusion are constructed, and a defect grade evaluation algorithm based on multi-dimensional feature similarity measurement is adopted to achieve accurate defect assessment, which expands the existing defect grade evaluation ideas and methods. It can not only improve the current defect grade assessment efficiency, but also ensure the consistency of the assessment results and the process and standardization of the assessment process.

[0097] (3) The proposed method was verified and demonstrated using TOFD weld defect data. The results showed that the proposed method is safer for weld defect grade assessment and meets the weld quality assessment requirements of products in actual production. In particular, it has a high assessment accuracy for linear defects that are difficult to assess, providing a technical approach for the in-depth analysis and application of current TOFD inspection data. At the same time, the proposed method has a certain degree of universality and can be technically applied to the analysis of inspection data with image features such as X-ray, phased array, and terahertz; it can also be applied to other fields such as medical imaging.

[0098] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A similarity weld defect grade evaluation method based on multi-source information fusion is characterized by: The specific steps include: Step 1: The established weld defect grade evaluation index system is used to quantitatively describe the standard reference map, and then the weld quality evaluation method process is designed in combination with the actual weld quality evaluation needs of the enterprise; Step 2: Perform TOFD inspection on the weld to obtain a TOFD map, and pre-process the weld map; Step 3: Detect the weld defects in the atlas and classify and identify the defects; Step 4: Calculate the geometric dimensions of the defect, perform a grade assessment based on the defect standard, calculate the final grade of the defect and generate a report; Calculation of the defect geometry includes: The size of each defect is: The indicators to be calculated are the feature vector composed of l and h of the non-point single defect and the texture feature indicators: energy ASM, entropy ENT, correlation COR, homogeneity IDM, and contrast CON. Among them, l represents the length of the defect and h represents the height of the defect. The similarity of defects in the image is measured for grading. The similarity of defects in the image is measured by: Suppose there are two vectors and , the Mahalanobis distance is selected as the measure between the texture features of weld defects and the features of defects in the standard reference defect map: (1) Where: Ma is the Mahalanobis distance; is the covariance matrix of and ; The grading process includes: At the same time, the wavelet coefficient modulus mean and standard deviation of the output image of the Gabor kernel 0°, 45°, 90° and 135° are used as feature vectors for calculation; Among them, Mahalanobis distance is used as a measure of the texture features between the atlas to be evaluated and the standard reference atlas; The Mahalanobis distance Gabor kernel mean and standard deviation are used as a measure between the Gabor kernel features of the evaluated map and the standard reference map.

Images