JPH03209582A - Method for estimating type of welding defect - Google Patents

Abstract

PURPOSE:To make correspondence to estimation for the type or cause of a defective part by a skillful inspector by processing the data of a defect feature amount concerning welding defect, collating the data with respective defect estimation rules, applying the degree of conviction corresponding to the processing data to satisfy the defect estimation rules and estimating the type of the welding defect according to this degree of conviction in the manner of probability. CONSTITUTION:An experiences and skillful inspector 2, etc., equipped with the type rules based on a knowledge base according to the quantitative data concerning the welding defect feature amount, etc., or quantitatively uncertain information based on user input information applies the degree of conviction such as a weight coefficient and therefore, the respective rules, which can be expressed by the semantic network of the knowledge rule, are made parallel. Then, the data are collated with the rules by using the defect feature amount of a picture and the expert system of user input to a welding defect part 6. The degree of conviction is applied according to the contents of one satisfactory rule at least. Thus, the skillful inspector 2 can make correspondence to the estimation for the type or cause of the defective part.

Description

[Detailed Description of the Invention] <Industrial Application Field> The disclosed technology is used to deal with welding defects in welded parts such as welded joints of welded structures such as power plant equipment, and to eliminate defects such as blowholes and poor fusion. It belongs to the technical field of estimating the type of image using a computer or the like from film images obtained by radiological examinations such as X-rays or images from image intensifiers.

<Summary of the gist> The invention of this application irradiates a welded part such as a welded joint part of piping of power plant equipment with radiation such as X-rays, obtains a transmitted image of the welded part, and obtains a transmitted image of the welded part. The welding defects that appear in the image are measured with a specified instrument, and the defect characteristics such as size, position, and inclination are converted into quantitative data, and the data is processed in a specified manner to determine the type of welding defect such as blowholes and poor fusion. This invention relates to a method for estimating from film images, and in particular,
The relationship between the defect features is constructed as a knowledge base of data for each type of welding defect, a predetermined number of rules are created, and the defect features from the film image of the welding site where welding defects are to be detected are set as data. Processed data is processed and compared as a defect estimation rule, and as a result of selective matching, a confidence level is directly assigned to the processed data that satisfies the predetermined defect estimation rule.
Alternatively, the welder or inspector compares the user input information they have regarding the welding with comprehensively set estimation rules, mutually determines the confidence level, and assigns the final confidence level. This invention relates to a method for estimating the type of welding defect.

<Prior Art> As is well known, mechanical devices and plant equipment are made up of complex mechanical parts, and of course mechanical assembly parts,
Quite often they consist of many welded joints.

Therefore, not only to maintain the functions of such machinery and plant equipment, but also to ensure the safety of the plant itself and its surroundings, maintenance of the joints and joints of these complex parts, It goes without saying that management of inspection and maintenance is extremely important.

In particular, for large-scale facilities such as power generation facilities, which have a strong public nature and whose social existence value is extremely important, construction, maintenance, inspection and servicing, whether regular or irregular, are required. Therefore, inspection management at the time of construction and maintenance is required based on the Electricity Business Law and regulations, especially in areas where internal defects cannot be confirmed with the naked eye or with simple inspection equipment. Radiographic tests such as X-rays using non-destructive inspection methods are prescribed by JIS and JIS standards, and there are also inspection and judgment standards for welding defects detected by X-ray films etc. obtained by such radiographic tests. It is strictly defined.

However, it requires an extremely high level of skill for inspectors to visually judge the welding defect image shown on such XI film based on their experience, and the results are extremely important. There were complaints that there were few misidentifications and that reliable judgment results could not be obtained.

However, the criteria for visual inspection and judgment by experts are slightly different, and due to factors such as frequency and time, fatigue may overlap, making it difficult to make stable, quantitative, and objective judgments. The problem was that it was difficult to do so.

In particular, unless the type of welding defect is determined in advance through non-destructive judgment, it is impossible to take specific countermeasures, which can seriously affect the operation and maintenance of equipment, and However, there were some inconveniences that could raise questions about safety.

Therefore, estimating the type of welding defect using X-ray film or the like is an extremely important problem for power generation projects and the like.

To deal with this point, there is automation of image processing of X-ray film through computer systems and so-called expert systemization by skilled inspectors. For example, Mr. Koshimizu et al. of Chukyo University and Mr. Inoue et al. by 3ayes
Inspection systems using the rules have been developed and proposed.

<Problems to be Solved by the Invention> However, the inspection and determination systems developed by Mr. Koshimizu and Mr. Inoue have had the following problems.

In other words, the inspection system proposed by Mr. Koshimizu et al. is a deterministic method based on a logical flow, a so-called algorithmic processing method, which has the disadvantage that it is extremely inconvenient to add new rules to serve as judgment criteria and is burdened with logic. Because the method uses graphical data such as the size and complexity of the welding defect and the concentration and position of the area near the defect, it has the advantage of logical flow processing and quantitative data processing, but it requires experience. There is an inconvenience that the rich knowledge of experienced inspectors cannot be used as an expert system or cannot be added to it.

In the latter system of Mr. Inoue et al., 3 aye
In order to use the s-law, the collected data must be completely organized, for example, nine defect features are required,
As mentioned above, integration with expert systems is being attempted,
There were negative points that could not be added.

<Object of the invention> The object of the invention of this application is to solve the problems of the inference method for welding defects based on the above-mentioned prior art, and to solve the problems of the inference method for welding defects based on the above-mentioned prior art. It is possible to efficiently combine welding defect features obtained from fire images with knowledge information from an expert system, avoid errors caused by deterministic methods, and provide fuzzy probabilistic certainty. Moreover, it is easy to add and update new rules, can tolerate insufficient collected data, and can also take into account expert systems, making it an excellent product that benefits the field of use of management technology in the machinery equipment and equipment plant industries. The purpose of this paper is to provide a method for estimating the types of welding defects.

<Means/effects for solving the problem> In order to solve the above-mentioned problem, the structure of the invention of this application, which is summarized in the above-mentioned claims, is to solve the above-mentioned problem. In order to estimate the type of defect in order to deal with defects, we use on-site radiation inspection such as X-rays to capture images of welded parts on film.
The image obtained from Image Inch Fire is measured by a specified measuring device, the defect data is quantified into data, and the relationship between the data processed defect feature values is found for each type of defect. Build a knowledge base, parallelize a predetermined number of rules by type, and match the defect features obtained from the film image of the welding defect and user input information held by welding engineers and inspectors to the rules. Then, a predetermined degree of certainty is given to the results of data that satisfy predetermined rules, so that the defect type can be estimated as closely as possible to the processing procedures and judgment results of experienced inspectors. , it is easy to change or add to the estimation logic, and the amount of collected data does not necessarily have to be statistically sufficient, and the data to be handled does not have to be a dimensionless relative quantity. "Certainty" refers to types of welding defects that are highly reliable based on fuzzy theory.
This is a technical measure that allows estimation to be made based on the following.

<Principle of the Invention> Next, to briefly explain the principle of the invention of this application, conventional inspectors and judges can detect welding defects by macroscopic observation of developed film images taken by X-ray inspection equipment etc. on-site. There is no need to estimate the type or cause based on the determination of
In addition to converting the characteristics of the obtained film images and image images from the image inching fire into data quantitatively and creating a knowledge base, we also use the rich experiential knowledge base of inspectors and judges as their confidence level. By using all of these knowledge bases as predetermined rules, comparing the image of the welded object and selectively matching it with the set rules and accumulating the confidence level, the final defect in the welding can be determined. If the type is 1, then it is estimated that it is a type with a ``likelihood'' of 6.

<Example> Next, an example of the invention of this application will be described below based on the drawings.

As shown in FIG. 7, the method for estimating the type of welding defect according to the invention of this application is as follows. Confidence of knowledge-based rules (a), (b), and (c) based on defect characteristics, and user input information (d) that only at least one of the welder and inspector can know empirically , (e). Through the comprehensive certainty, it is possible to estimate the "certainty" 11 of defects such as poor fusion or slag entrainment of the welding defect. Regarding the above user input information) and (e), for example, welding conditions that only the welder can know, such as welding positions such as upward welding and downward welding, the magnitude of the welding current used, and the weld bead. slag,
The scale condition, groove condition, type of welding, etc. are assigned according to the degree of certainty based on the experience of the welder concerned, and predetermined rules are created as a knowledge base, as shown in Fig. 7. In addition, we conducted a parallel comparative study of each rule based on comprehensive comparison with the confidence level of the defect feature quantity based on the image measurement of the welding defect part by radiation described below) and the comparison data with the above user input information (g). This can be obtained by calculating the overall confidence level based on .

On the other hand, although the welding feature quantity requires a complicated procedure, in a sense it is possible to process the data by means such as measurements that are easy to specify, and the specific procedure for creating rules is necessary. is as follows.

In other words, as shown in Figure 1, the site 1 of the power plant piping equipment, etc.
, the inspector 2 uses a predetermined radiographic inspection device 3 such as X-ray
A radiographic inspection film for welding defects in the welded part of the piping, etc. is taken by the operator, and as shown in FIG. The presence of the welding defect 6 (including a group of defect candidates) between the base metals 7 and 7 is confirmed, and the welding defect 6 in the image film 4 is directly measured using a predetermined measuring device. , obtain input information regarding defect features as shown in Table 1 below.

(Leaving space below) Table 1 Furthermore, regarding the data in Table 1 above, items numbered 1 to 5 have already been defined in the previously mentioned conventional technology by Mr. Inoue et al. The nonlinearity data from No. 6 average density to No. 10 were newly provided in the invention of this application.

Regarding such data, as shown in FIG. 4, the welding center line 8, defect image 6, and their inclination φ, etc. for the bead 5 between the base materials 7 and 7 can be obtained from the image film 4. The welding defect trace data 9 based on actual measurement data from film images as shown in Fig. 3.4 may be corrected as model sample graphic data. Create it as shown in Table 1.

Therefore, FIG. 5 shows the trace amounts of welding defects in the data in Table 1 above.

Welding defect characteristics based on data directly measured by measuring instruments from film images as shown in Table 1 above!
! I! Regarding the amount, as is clear from experience, a specific relationship is observed between the predetermined welding defective trace amounts,
For example, as shown in Figure 6, the horizontal axis represents the defect length (LLT)
Also, if we take the flatness (FLT) on the vertical axis, there is a linear relationship of FLT = 1/3 LLT, and above and below there is a specific distribution relationship where the fused areas marked with ○ and the slag entrapment in the mouth are distributed. obtained, therefore the defect length (
A kind of knowledge-based identification rule is created by organizing the data in Table 1 regarding the relationship between 1F seasonality and flatness (FLT), and in the example shown in FIG. , 1/311-0,2<FLT <1/8 LLT +
If the data relationship is 0.3 and [[1<12, then the identification rule indicates that the confidence level that the welding defect is due to poor fusion or slag entrainment is necessarily 0.95. It is determined that an estimate of ``similarity'' can be obtained.

In this case, as a method of assigning the certainty factor, a corresponding probability for the total number of data may be used.

Of course, the above-mentioned embodiment is an example of the identification rule as a knowledge base of the welding defect feature amount for the six welding defect parts,
Such knowledge-based identification rules can also be obtained for other predetermined relationships of welding defect features, and a predetermined number of these rules can be set in parallel.

Identification rules as a knowledge base obtained in this way), the above-mentioned welding defect feature quantity 10, and the above-mentioned user input information (g), user input information (d), (e)
As shown in Fig. 7, if the respective rules are checked comprehensively and each of these degrees of confidence is considered and a particular degree of certainty can be adopted preferentially, then the welding The type and cause of the welding defect can be estimated with a certain probability of 1, or the type and cause of the welding defect can be estimated based on a predetermined certainty calculation method to obtain a comprehensive confidence level. Causes and [certainty] should be determined as composite degrees of certainty.

In addition to the above-mentioned embodiments, in addition to the above-mentioned embodiments, the reliability is given by the knowledge-based identification rule based on the processing of a small amount of defective data about the defective part using an image film taken by a radiographic inspection device for the welding defective part. Rules are created by data processing as shown in FIG. 10, and as shown in FIG.
Determine the trace amount of each defect as shown in the table, and compare it with the database of histograms for each defect feature obtained in advance from various data as shown in Figure 9 to find the standard deviation. For example, in the case of a blowhole, the welding defect is determined by determining whether it falls within the range of 1σ, 2σ, or 3σ.
If there is a standard deviation of 1σ between bl and b2, the concentration is bl and b2.
If the defect length is between C2 and the defect length is smaller than C2, the weighting coefficient is set to 1.0.0.6.0.1 as shown in Table 2 below, and as a result, the weighting coefficient of flatness is set to 1. .0.1 degree is 0.6, weld length is 0.1, and each weighting coefficient is set as the confidence level. (Therefore, since the weighting coefficients and confidence levels differ depending on the level of experience of the inspector, etc., it is desirable that the weighting coefficients and confidence levels be assigned by inspectors with an average of years of experience and skill.)

Then, a confidence level is calculated for each defective trace amount using a weighting coefficient, and the final confidence level is determined in the same manner as described above by combining the confidence level with the user input information.

Regarding the construction style of the knowledge base, it goes without saying that there are other methods in addition to the embodiments described above.

Regarding user input information, it is of course possible to take steps such as obtaining confidence in advance through interviews with welding personnel and welding inspectors.

Through matching against a parallel set of identification rules in such a knowledge base, the final confidence can be e.g. "If... then... then the confidence is...".
The type and cause of the welding defect are expressed using the rule method, and a fuzzy "certainty" estimate is given, avoiding deterministic determination of the type and cause.

Of course, as mentioned above, if multiple confidence levels are obtained as a result of matching against multiple identification rules, a final overall confidence level is assigned based on a predetermined confidence level calculation method as described above. do.

This procedure is as shown in Figures 10, 11 and 12.

Note that the embodiment of the invention of this application is not limited to a case in which a radiographic image of a welded part is merely an image on a film; for example, it may be directly captured as a digital image through an image inverter and displayed on a CRT. It goes without saying that they may be equal.

<Effects of the Invention> As described above, according to the invention of this application, identification rules based on a knowledge base based on quantified data such as welding defect feature values and quantitatively uncertain information based on user input information can be , and parallelize each rule that can be expressed in a knowledge-based semantic network by adding confidence levels such as weighting coefficients by a skilled inspector, and then calculate the defect features of the image for the welding defect and the user. By making it possible to perform verification using an input-based expert system and assigning confidence based on the content of at least one rule that is satisfied, it is not possible to draw a definitive conclusion on the type or cause of the defect as described above. , it is possible to estimate "likelihood" without using conventional statistical methods, and avoid errors caused by deterministic methods.
It has the effect of being able to provide a level of certainty that is extremely close to the estimation of realistic defect types and causes by highly skilled inspectors, and compared to the conventional theoretical algorithms of Mr. Koshimizu and Mr. Inoue, etc. It eliminates the negative point that it is difficult to add new rules or update logic based on the 3ayes rule, it is extremely easy to add new rules or update, and it does not require complete replenishment of collected data. Some effects have the advantage of being able to be estimated based on the data that can be collected.

In addition to the size and complexity of graphical defect features, density and position in their vicinity, as shown in Table 1 above, which can be obtained from image information, 10 types of defect features and skilled inspectors can be found. By taking into account user input information such as There is also the flexibility of being able to change the confidence level based on statistical processing of feature amounts and deviation values depending on the data.

Furthermore, depending on the case, there is flexibility in that information other than information on defect feature quantities can also be incorporated.

[Brief explanation of drawings]

The drawings are detailed explanatory drawings of the invention of this application, and Figs. 1 to 7 are flowcharts of one embodiment. Fig. 1 is a schematic diagram of photographing a defective part of a welding defect in the field, and Fig. 2 is a flowchart of one embodiment. FIG. 3 is a schematic plan view of the image film; FIG. 4 is a plan view of the modeled image film; FIG. 5 is a perspective view of a table of defect feature data; Figure 6 is a correlation diagram of identification rules for predetermined defect feature quantities, Figure 7 is a schematic diagram of confidence imparting through comparison between image processing data such as defect feature quantities and user input information, and Figures 8 to 12.
The figure is a schematic flowchart of reliability imparting according to the embodiment, Fig. 8 is a schematic diagram of an image film, Fig. 9 is a graph of standard deviation and histogram, and Fig. 10 is a graph of defect characteristic wi amount with respect to standard deviation. FIG. 11 is a schematic diagram of the accumulation of weighting coefficients and confidence degrees, FIG. 11 is a schematic diagram of fusion through verification of confidence degrees, and FIG. 12 is a schematic diagram of giving comprehensive confidence degrees. 4...Image film 6...Welding area

Claims (4)

[Claims] (1) In a method of estimating the type of welding defect by processing the data using multiple defect feature values obtained by measuring the welding defect appearing in an image obtained by irradiating the welding site with radiation, A predetermined number of defect estimation rules are created as a knowledge base in advance, and the defect feature data of the welding defect is processed and compared with each defect estimation rule, and the confidence level corresponding to the processed data that satisfies the defect estimation rules is calculated. A method for estimating a welding defect, characterized in that the type of the welding defect is probabilistically estimated based on the certainty factor. (2) Claim 1 characterized in that the relationship between the defect feature quantities is quantitatively established as a rule.
Method for estimating welding defects described in section. (3) Claim 1, characterized in that the defect feature amount is compared with a frequency distribution database for each defect type based on known data, and a deviation range is determined and a confidence level is given. Described method for estimating welding defects. (4) Estimating a predetermined number of defects as a knowledge base for each type of welding defect by using multiple defect feature quantities as data by measuring welding defects that appear in images obtained by irradiating the welding site with radiation, and processing the data. A rule is created in advance, and the defect feature data of the welding defect is processed and compared with the defect estimation rule along with user input information regarding the welding that only at least one of the welder and the inspector has. A method for estimating a welding defect, comprising: assigning a degree of certainty corresponding to processed data that satisfies a defect estimation rule, and probabilistically estimating the type of the welding defect based on the degree of certainty.