KR102059272B1 - Device for detecting welding defects and method thereof - Google Patents

Abstract

According to one embodiment, provided is a method for detecting a welding defect comprising: an operation of obtaining an image for an object to be welded; an operation of extracting a region of interest including a welding part from the image; an operation of setting a section from a center of the welding part; an operation of generating a first histogram composed of a first image value and a second histogram composed of a second image value with respect to the section; an operation of coupling the first histogram and the second histogram to generate an overall histogram; and an operation of determining whether or not the welding is defective on the basis of the overall histogram. Therefore, the present invention provides a technique for detecting welding defects which cannot be identified by the naked eye.

Description

Welding defect detection device and its method {DEVICE FOR DETECTING WELDING DEFECTS AND METHOD THEREOF}

This embodiment relates to a welding failure detection technique.

In the welding technique of joining metal materials separated from each other, weldability which determines the good or bad of the welding result is very important. This is because welding is for the purpose of firm coupling, and welding methods, welding materials, and construction methods depend on weldability. Above all, the welded object must be firmly coupled to prevent malfunction due to poor welding. For example, the automobile sensor is coupled by spot welding. If the welding is poor, the automobile sensor is detached by vibration generated during driving, and a malfunction of the vehicle driving occurs. Therefore, it is required in advance to find and correct welding defects during manufacturing.

Weld defects may be found visually, but sometimes they are not. In the above example, it is difficult to visually identify whether spot welding of the automotive sensor is defective. Moreover, as miniaturization and integration of electronic devices is performed, there will be more cases where local and fine welding is performed that cannot be visually identified. If it is difficult to identify with the naked eye, the weld failure can only be determined through a direct physical test. However, this method is limited in that time and money are excessive and inefficient. In this regard, techniques have been developed to detect weld defects when visual identification and physical testing are difficult.

Against this background, it is an object of this embodiment to provide a technique for detecting weld defects that cannot be visually identified.

In order to achieve the above object, an embodiment is a method for detecting a defect of welding, the method comprising the steps of: obtaining an image for the welded object; Extracting a region of interest including a welded portion from the image; Setting a section from the center of the welding site; Generating, for the interval, a first histogram composed of first image values and a second histogram composed of second image values; Combining the first histogram and the second histogram to generate a full histogram; And it provides a welding failure detection method comprising the operation of determining whether the welding failure based on the entire histogram.

In the welding failure detection method, the first image value and the second image value may be the number of pixels corresponding to an arbitrary brightness value range.

In the welding failure detection method, a plurality of sample histograms, and the operation of obtaining information about the welding result indicating whether the plurality of sample histograms correspond to the normal welding or welding failure, and determining whether the welding failure The act of learning may include learning the weld results for each of the plurality of sample histograms; And determining based on the learning.

In the welding failure detection method, the operation of acquiring the information may include obtaining a histogram to which the plurality of sample histograms are combined and information about a welding result indicating whether the combined histogram corresponds to a normal welding or a welding failure. The determining of the welding failure may include: learning the welding result with respect to the combined sample histogram; And determining based on the learning.

In the welding failure detection method, the learning operation may be learned using a support vector machine (SVM) model.

In the welding failure detection method, the welding method may be spot welding, and the welding portion may be spot.

In the welding failure detection method, the spot is circular, and setting the section comprises: setting the spot circular; Setting a center of the circle; Setting a first radius and a second radius from the center of the circle; And determining the interval between the first radius and the second radius.

In the welding failure detection method, the operation of recognizing the circle of the spot and the operation of setting the center of the circle may be set using a Gaussian mixture (GM) model.

Another embodiment is an apparatus for detecting a defect in welding, the apparatus comprising: an image acquisition unit for obtaining an image for a welded object; Extracting a region of interest including a welded portion from the image, setting a section from the center of the welded portion, and for the section, a first histogram composed of first image values and a second histogram composed of second image values An image processor configured to generate the entire histogram by combining the first histogram and the second histogram; And it provides a welding failure detection device comprising a determination unit for determining whether or not welding based on the entire histogram.

In the welding failure detection apparatus, the determination unit obtains information on a plurality of sample histograms and welding results indicating whether the plurality of sample histograms correspond to a normal welding state or a welding failure state, and each of the plurality of sample histograms is obtained. The welding result for the learning, and based on the learning can determine whether the welding failure.

As described above, according to this embodiment, it is possible to easily determine whether or not welding defects are difficult to visually identify. Through this, it is possible to prevent a malfunction caused by detachment of the welded object by poor welding.

In addition, according to the present embodiment, by learning a histogram indicating a welding state desired by the user, it is possible to determine whether the welded portion matches the welding state desired by the user. Through this, it is possible to provide a user-specific welding failure determination function for determining whether the welding meets the user setting.

1 is a block diagram of a welding failure detection apparatus according to an embodiment.
2 is an exemplary diagram illustrating an image and a region of interest about a welded object according to an exemplary embodiment.
3 is an exemplary diagram illustrating a region of interest in which a center of a welding portion is set according to an exemplary embodiment.
4 is a diagram illustrating setting a section at a welding site according to an exemplary embodiment.
5 is an exemplary diagram illustrating a histogram according to a welded section according to an exemplary embodiment.
6 is an exemplary diagram illustrating a total histogram generated by combining a plurality of histograms according to an exemplary embodiment.
7 is a flowchart illustrating an operation of a welding failure detection apparatus configuration, according to an exemplary embodiment.
8 is a flowchart illustrating an operation of a welding failure detection apparatus according to an exemplary embodiment.

Hereinafter, some embodiments of the present invention will be described in detail through exemplary drawings. In adding reference numerals to the components of each drawing, it should be noted that the same reference numerals are assigned to the same components as much as possible even though they are shown in different drawings. In addition, in describing the present invention, when it is determined that the detailed description of the related well-known configuration or function may obscure the gist of the present invention, the detailed description thereof will be omitted.

In addition, in describing the component of this invention, terms, such as 1st, 2nd, A, B, (a), (b), can be used. These terms are only for distinguishing the components from other components, and the nature, order or order of the components are not limited by the terms. If a component is described as being "connected", "coupled" or "connected" to another component, that component may be directly connected to or connected to that other component, but there may be another configuration between each component. It is to be understood that the elements may be "connected", "coupled" or "connected".

1 is a block diagram of a welding failure detection apparatus according to an embodiment.

Referring to FIG. 1, the welding failure detection apparatus 100 may include an image acquirer 110, an image processor 120, a determiner 130, a storage 140, and a display 150.

The welding failure detection method performed by the welding failure detection apparatus 100 may be mainly used to detect whether a resistance welding including spot welding is defective, but is not limited thereto. It can also be used in arc welding, gas welding or special welding, including electron beam, laser, ultrasonic and robot welding. Hereinafter, the spot welding will be described.

The image acquirer 110 may acquire an image of the welded object 210 or 220. The welded objects 210 and 220 are objects to be welded, and may be mainly metal objects. Images for the welded objects 210 and 220 may be image images and static images displayed on the screen. The image acquisition unit 110 may receive an image of the image to be welded 210 or 220 by a separate photographing apparatus, or may generate an image of the object to be welded 210 or 220 including a camera.

Images for the welded objects 210 and 220 may include image data. The image data may include information used to represent an image of the welded object 210 and 220 in a histogram. For example, the image data may be a brightness value, a saturation value, and / or a color value of a pixel.

2A is a conceptual diagram illustrating an image of a welded object according to an embodiment.

As in the example of FIG. 2A, an image for the welded object 210 and 220 may be obtained as a static image image. The first welded object 210 may be welded to the second welded object 220 through the welding portion 201. For example, the first to-be-welded material 210 may be an automobile sensor, and the second to-be-welded material 220 may be an automobile structure having a plate, and the welding portion 201 may be a spot formed by spot welding. The image acquirer 110 may acquire the image illustrated in FIG. 2A.

Referring back to FIG. 1, the image processor 120 may extract the ROI 230 including the welding portion 201 from the images of the welded objects 210 and 220. The ROI 230 is a portion corresponding to a region including the welded portion 201 and its vicinity of the region of the image to be welded, and may be a unit processed by the image processor 120. The image processor 120 may set and extract the ROI 230 from the images of the welded objects 210 and 220.

2B and 2C are exemplary views illustrating a region of interest according to an embodiment.

In FIG. 2B, the image processor 120 may set the ROI 230 including the welding portion 201 from the images of the welded objects 210 and 220. In the above-described example, when the automobile sensor is coupled to the plate by spot welding, the image processor 120 may move the spot, that is, the welding portion 201 and its vicinity to the region of interest 230, which is the portion of the automobile sensor welded to the plate. Can be set. As illustrated in FIG. 2C, the image processor 120 may separate the set ROI 230 from the images of the welded objects 210 and 220.

Referring back to FIG. 1, the image processor 120 may generate a histogram for the welded portion 201. In detail, the image processor 120 may generate specific histograms by setting specific sections of the welding portion 201, obtaining image values of the sections, and matching the image values to the sections. The histogram may be expressed by using each of the sections and the image value as one axis.

In order to set a section at the welded portion 201, the image processor 120 sets the center 310 of the welded portion 201. The image processor 120 may use a gaussian mixture (GM) algorithm to set the center 310 of the welding portion 201.

3 is an exemplary diagram illustrating a region of interest in which the center 310 of the welding site 201 is set, according to an exemplary embodiment. In FIG. 3, by means of a Gaussian mixing algorithm, each region of interest 230 of FIGS. 2B and 2C is transformed and the image processor 120 converts the center 310 of the welded portion 201 in each region of interest 230. You can find it.

Referring back to FIG. 1, the image processor 120 may set a section from the center 310 of the welded portion 201 to the entire welded portion 201. The sections may be divided according to the length of the radius. The section may include a region of the welding portion 201 divided according to the radius length.

4A is a conceptual diagram illustrating setting a section in a welding portion 201 according to an embodiment, and FIG. 4B is an exemplary diagram according to the same. In FIG. 4B, setting the section in the welded portion 201 shown in FIG. 4A is shown in the actual photographed image image, which is the same as in FIG. 4A. Hereinafter, a description will be given with reference to FIG. 4A.

In FIG. 4A, the image processor 120 may set the first radius r ₁ and the second radius r ₂ from the center 310 of the welding portion 201. The image processor 120 may calculate the first radius r _{1 by} specifying the first circle 401 and measuring a distance from the center 310 of the welding portion 201 to the first circle 401. . The image processor 120 may calculate the second radius r _{2 by} specifying the second circle 402 and measuring the distance from the center 310 of the welding portion 201 to the second circle 402. .

The image processor 120 may include a first section between the center 310 of the welding portion 201 and the first radius r ₁ , and a second section between the first radius r ₁ and the second radius r ₂ . Can be defined as The first section is an area A1 corresponding to the center 310 of the welding portion 201 and the first radius r ₁ , and the second section is the first radius r ₁ and the second radius r. ₂ ) each of the regions A2 may be included.

The widths of the sections may be different or equal to each other, and the width may be preset by the user. The method for setting the section is not limited to the above-described example. For example, instead of specifying a circle to obtain a radius, a section may be set by specifying a polygon to obtain a distance from the center 310 of the welding portion 201 to a vertex. This case may be appropriate when the welded portion 201 is angled rather than circular.

1, the image processor 120 may calculate an image value for the section. The image value may be calculated from image data of an image for the welded object 210 and 220. For example, the image value may include the number of pixels having a specific brightness. The image processor 120 may count the number of pixels having the specific brightness in each section. In detail, the image processor 120 may define the number of pixels corresponding to a certain brightness value range as the image value.

수식(1)

Formula (1)

수식(2)

Formula (2)

수식(3)

Formula (3)

According to equation (1), the image processing unit 120 determines the number of pixels having a brightness value I (d) of X _a or less, and according to equation (2), X _b or more X _c According to Equation (3), the number of pixels having the brightness value I (d) below is defined as the image value for each section by the number of pixels having the brightness value I (d) of X _d or more. can do. In the above-described example, the image processing unit 120 is formed in the first section A1 and the first radius r ₁ , which are areas corresponding to the first radius r ₁ , from the center 310 of the welding portion 201. The number of pixels having a brightness value I (d) equal to or less than X _a is calculated for each of the second intervals A2 corresponding to the second radius r ₂ , according to Equation (1), and the number of the pixels. May be defined as an image value for each section.

In addition, the image processor 120 may calculate an image value by applying a different brightness value range for each section. In the above-described example, the image processing unit 120 has _a number of pixels having a brightness value I (d) of X _a or less in the first section A1, and X _b or more and X _c or less in the second section A2. The number of pixels having a brightness value of I (d) may be defined as an image value, respectively.

The image processor 120 may use all data that can be obtained from the image data in addition to the brightness value to calculate an image value for the section. For example, the image processor 120 may define the number of pixels having any range of saturation or color as an image value based on the saturation value or the color value.

The image processor 120 may calculate an image value using various kinds of image data, for example, brightness and saturation. In the above example, the number of pixels having brightness less than or equal to X _a and chroma less than or equal to Y _a in the first section A1 may be defined as an image value.

The image processor 120 may generate a histogram using the section as one axis and the image value as another axis. The image processor 120 may match an image value corresponding to each section.

5 is an exemplary diagram illustrating a histogram according to a welded section according to an exemplary embodiment.

FIG. 5A shows a histogram in which the number of pixels having brightness values corresponding to Equation (1) is matched for each section. The image processor 120 may separate and set the welding portion 201 of FIG. 4B into 10 sections, calculate the number of pixels having a brightness value of X _a or less for each section, and match the corresponding section. The first section A1 is from the center 310, r ₀ of the welding portion 201 to the first radius r1 in the histogram, and the second section A2 is the second radius at the first radius r ₁ ( up to r ₂ ), the third section corresponds to a first radius r ₃ to a second radius r ₄ . The fourth to tenth sections may be defined as well.

The image processor 120 may generate a plurality of histograms for the same section. The image processor 120 may define a first image value in a section set in the welding portion 201 and then generate a first histogram having the section and the first image value as an axis. The image processor 120 may define a second image value in the same section, and then generate a second histogram having the section and the second image value as an axis. FIG. 5B shows an example of a second histogram in which the number of pixels having brightness values corresponding to Equation (2) is matched for the same section as that shown in FIG. 5A. If the number of pixels having a brightness value corresponding to Equation (1) is a first image value, the number of pixels having a brightness value corresponding to Equation (2) may be a second image value.

FIG. 5C illustrates a third histogram in which the number of pixels having brightness values corresponding to Equation (3) is matched for each section in the same section as illustrated in FIG. 5A. The number of pixels having a brightness value corresponding to Equation (3) may be a third image value.

As described above, the image processor 120 may calculate various types of image values for the same section and generate a plurality of histograms according to the calculated image values of each type. In FIG. 5, the image processor 120 may calculate a first image value for each section and generate a first histogram for ten sections according to a type according to Equation (1). The image processor 120 may calculate a second image value for each section and generate a second histogram for the ten sections according to the type according to Equation (2). The image processor 120 may calculate a third image value for each section and generate a third histogram for the ten sections according to the type according to Equation (3).

Referring back to FIG. 1, the image processor 120 may combine a plurality of histograms to generate a single whole histogram. The entire histogram may be used by the determination unit 130 to determine whether the welding is bad.

 6 is an exemplary diagram illustrating a total histogram generated by combining a plurality of histograms according to an exemplary embodiment. Referring to FIG. 6, the entire histogram is illustrated by combining the first histogram of FIG. 5A, the second histogram of FIG. 5B, and the third histogram of FIG. 5C. The entire histogram may have a form in which the first to third histograms are arranged in sequence, but the coupling method is not limited thereto.

1, the image processor 120 may convert the histogram into a vector. The transformed vector may include a row vector or a column vector. The transformed vector may represent an image value corresponding to sections of the histogram. For example, in the case of the first histogram of FIG. 5A, the transformed vector is' (first image value of the first interval, first image value of the second interval, ..., first image value of the tenth interval, and the like. ) '

The image processor 120 may convert the entire histogram generated from the plurality of histograms into a vector. For example, in the histogram of FIG. 6, the transformed vector is' (first image value of the first interval, first image value of the second interval, ..., first image value of the tenth interval, Second image value in one section, second image value in a second section, ..., second image value in a tenth section, third image value in a first section, third image value in a second section, .. , The third image value of the tenth section).

The determination unit 130 may determine whether welding is defective based on the histogram. In detail, the determination unit 130 may use machine learning including deep learning and a support vector machine (SVM) model for learning.

The determination unit 130 may further include a learning unit 132. Here, the learner 132 may learn through the SVM model. The learner 132 may obtain a sample histogram including a plurality of histograms for various welding sites 201 and welding result information about the sample histogram. For example, the learner 132 may acquire information, 'welding normal', which is a welding result of the first sample histogram and the first sample histogram. In addition, the learning unit 132 may include information 'welding normal', which is a welding result of the second sample histogram and the second sample histogram, and 'welding failure', which is a welding result of the third sample histogram and the third sample histogram. Information can also be obtained. The learner 132 learns to determine that the first sample histogram is 'welding normal', the second sample histogram is 'welding normal', and the third sample histogram is 'bad welding'.

If the learner 132 makes another determination, the determination may be corrected through a correction value. For example, when the learner 132 determines that the welding is normal, the learner 132 includes a correction for the third sample histogram, which includes information of 'welding failure'. Receive the value and adjust the third sample histogram to make a 'welding failure' decision.

The learner 132 may learn a histogram in which a plurality of sample histograms are combined. For example, the learner 132 may receive the histogram to which the first to third sample histograms are combined and the welding result information of the combined histograms, and may learn whether the combined histograms are normal or defective welds.

After the learning unit 132 finishes the learning, the determination unit 130 may determine whether the single histogram or the entire histogram received from the image processing unit 120 is 'bad welding' or 'welding normal'. In the above-described example, the determination unit 130 may receive a histogram for the welding portion 201 from the image processing unit 120. The determination unit 130 may determine that a histogram equal to or similar to the first sample histogram is' welded normal ', a histogram equal to or similar to the second sample histogram is' welded normal', and a histogram equal to or similar to the third sample histogram is' Each of them judges that the welding is bad.

As another determination method, the determination unit 130 may determine whether the welding is bad by comparing the histogram generated by the image processing unit 120 with the sample histogram. The sample histogram may include a histogram for the weld site 201 that appears when the weld is normal. The sample histogram may be stored in advance in the storage 140, and may be read out from the storage 140 by the determination unit 130 and used. When the histogram generated by the image processor 120 is the same as or similar to the sample histogram, the determination unit 130 may determine that the welding is normal, and if the histogram is not the same or similar, determine that the welding is bad.

The determination unit 130 may determine in the same manner by using a vector converted from the histogram instead of the histogram.

Each histogram represents the characteristics of the welded portion 201 represented in the image image. The more the histogram, the more the characteristics of the welded portion 201 can be defined in detail. Also, if the segment is subdivided and stretched, a single histogram may be more specific in characterizing the weld site 201. For example, rather than calculating image values according to Equation (1) for 10 sections, calculating for 50 sections may more clearly indicate the characteristics of the welded portion 201.

)개 구간에 대한 이미지 값이 정확해야 하기 때문에 '용접 불량'이란 판단의 빈도는 매우 높을 것이다. Defining the characteristics of the welded portion 201 in detail through the histogram may increase the rigor and accuracy of the determination of weld failure in the determination unit 130. This is because whether the histogram is large or the histogram interval is subdivided, the requirements to be satisfied for the 'welding normal' result in more and more accurate and accurate judgments. For example, when the determination unit 130 determines only the first histogram, the determination of 'bad welding' will hardly be made since the image values for the 10 sections need to be correct. However, when the determination unit 130 determines with the entire histogram combining the first to third histograms, 30 (

Since the image values for the) sections must be accurate, the frequency of determination of 'bad welding' will be very high.

The storage 140 may store an image of the welded object 210 and 220 photographed from the image acquisition unit 110 and / or a pattern of the sample histogram.

The display unit 150 may receive determination result information on whether welding is defective from the determination unit 130, and display the result to the outside.

7 is a flowchart illustrating an operation of a welding failure detection apparatus configuration, according to an exemplary embodiment. Referring to FIG. 7, the image acquirer 110 may acquire an image of the welded object 210 or 220 (S702). The image processor 120 may receive an image of the welded object 210 and 220 from the image acquirer 110 (S704).

The image processor 120 may extract the ROI 230 from the images of the welded objects 210 and 220 (S706).

The image processor 120 may set a section at the welding portion 201 included in the ROI 230 (S708). In order to set a section, the image processor 120 may set a center of the welding portion 201 and a plurality of circles surrounding the center. The image processor 120 may obtain a plurality of radii from the center to the plurality of circles, and then set the intervals of the plurality of radii as intervals. The section may be an image area between the plurality of circles.

The image processor 120 may calculate an image value for each section (S710). The image value is generated from image data such as brightness, saturation, and color, and may include the number of pixels corresponding to an arbitrary brightness value range.

The image processor 120 may generate a histogram using each axis as one axis and the image value for each period as another axis (S712).

The image processor 120 may process and convert the histogram (S714). The image processor 120 may combine a plurality of histograms to generate a single histogram or convert the histogram into a vector.

The image processor 120 may transmit histogram information including the information on the pattern of the histogram or the information on the vector whose histogram is converted (S716).

The determination unit 130 may determine whether welding is defective by using the histogram information (S718). The determination unit 130 may perform the determination by learning or comparing patterns according to the SVM model.

The determination unit 130 may transmit the determination result information to the display unit 150 (S720).

The display unit 150 may inform the user of welding failure based on the determination result information (S722).

8 is a flowchart illustrating an operation of a welding failure detection apparatus according to an exemplary embodiment. Referring to FIG. 8, the welding failure detection apparatus 100 may acquire an image of the welded object 210 and 220 (S802) and extract an ROI 230 including a welded portion 201 (S804). ). The welding failure detection apparatus 100 may set a section at the welding site 201 (S806) and generate a histogram having the section as one axis (S808). The welding failure detection apparatus 100 may determine whether welding is defective based on the histogram (S810).

The terms "comprise", "comprise" or "having" described above mean that a corresponding component may be included unless specifically stated otherwise, and thus does not exclude other components. It should be construed that it may further include other components. All terms, including technical and scientific terms, have the same meanings as commonly understood by one of ordinary skill in the art unless otherwise defined. Terms commonly used, such as terms defined in a dictionary, should be interpreted to coincide with the contextual meaning of the related art, and shall not be construed in an ideal or excessively formal sense unless explicitly defined in the present invention.

The above description is merely illustrative of the technical idea of the present invention, and those skilled in the art to which the present invention pertains may make various modifications and changes without departing from the essential characteristics of the present invention. Therefore, the embodiments disclosed in the present invention are not intended to limit the technical idea of the present invention but to describe the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. The protection scope of the present invention should be interpreted by the following claims, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of the present invention.

Claims (10)

In the method of detecting the defect of welding,
Obtaining an image for the object to be welded;
Extracting a region of interest including a welded portion from the image;
Setting a section from the center of the welding site;
Generating a first histogram composed of first image values and a second histogram composed of second image values for the interval;
Combining the first histogram and the second histogram to generate a full histogram; And
Weld failure detection method comprising the step of determining whether the weld failure based on the entire histogram. The method of claim 1,
And the first image value and the second image value are the number of pixels corresponding to an arbitrary range of brightness values. The method of claim 1,
Acquiring a plurality of sample histograms and information on a welding result indicating whether the plurality of sample histograms correspond to a weld normal or a weld failure;
The operation of determining whether the welding is poor,
Learning the weld results for each of the plurality of sample histograms; And
Weld failure detection method comprising the operation of determining based on the learning. The method of claim 3,
The acquiring of the information may include obtaining a histogram to which the plurality of sample histograms are combined, and a welding result indicating whether the combined histogram corresponds to a normal welding or a bad welding,
The operation of determining whether the welding is poor,
Learning the weld results for the combined sample histogram; And
Weld failure detection method comprising the operation of determining based on the learning. The method according to claim 3 or 4,
The learning operation is a welding failure detection method for learning using a support vector machine (SVM) model. The method of claim 1,
The method of welding is spot welding,
The welding failure detection method of the welding site is a spot. The method of claim 6,
The spot is circular,
The operation of setting the section,
Setting a circle of the spot;
Setting a center of the circle;
Setting a first radius and a second radius from the center of the circle; and
Weld failure detection method comprising the step of determining the interval between the first radius and the second radius. The method of claim 7, wherein
The operation of recognizing the circle of the spot and the operation of setting the center of the circle, the welding failure detection method is set using a gaussian mixture (GM) model. In the apparatus for detecting a defect of welding,
An image obtaining unit obtaining an image of the object to be welded;
Extracting a region of interest including a welded portion from the image, setting a section from the center of the welded portion, and for the section, a first histogram composed of first image values and a second histogram composed of second image values An image processor configured to generate the entire histogram by combining the first histogram and the second histogram; And
Determination unit for determining whether the weld failure based on the entire histogram
Welding failure detection device comprising a. The method of claim 9,
The determination unit acquires information on a plurality of sample histograms and a welding result indicating whether the plurality of sample histograms correspond to a normal welding state or a welding defect, and learn the welding results for each of the plurality of sample histograms. And determining a welding failure based on the learning.

Images