JP7786178B2 - Welding procedure judgment method and welding procedure judgment device - Google Patents

Description

Article 30, paragraph 2 of the Patent Act applied. December 15, 2020: Presented at the 3rd (295th overall) Chemical Mechanical Welding Research Committee meeting held at the Welding Hall. March 10, 2021: Posted on the website as a lecture summary for the 2021 Spring National Conference of the Japan Welding Society. April 14, 2021: Presented at the 2021 Spring National Conference of the Japan Welding Society held online.

This disclosure relates to a welding construction assessment method and a welding construction assessment device.

Conventionally, methods exist for determining and evaluating the welding quality or operator skill in arc welding such as TIG welding or MAG welding, and applying this to adaptive control in automatic welding machines, or quantifying skill in manual welding. Patent Document 1 discloses technology related to a welding control method that calculates the amount of change in the root gap by extracting the left and right endpoints of the molten pool from images of the molten pool and its surrounding area captured by a visual sensor during welding, and applies this amount of change to the oscillation width of the welding torch or the welding speed. Meanwhile, Patent Document 2 discloses technology related to a manual welding work analysis device that evaluates the operator's skill based on feature values extracted from images of the molten pool and its surrounding area captured by a surveillance camera during welding.

Japanese Patent Application Laid-Open No. 2006-281282 Japanese Patent Application Laid-Open No. 2006-281270

In methods such as those disclosed in Patent Documents 1 and 2, the welding machine is adaptively controlled and the operator's skill is evaluated based on feature values extracted from images of the molten pool and its vicinity captured during welding. However, welding phenomena are governed by a complex interaction of countless factors with varying degrees of influence. In the technology disclosed in Patent Document 1, the change in root gap is used as the criterion for determining the quality of the weld. Because the root gap is expressed as a dimensional value, if some factor changes between welding operations, the change in the root gap corresponding to that shape may change, even if the molten pool shape is the same. Therefore, the change in root gap may not be appropriately reflected in the oscillation width of the welding torch of the automatic welding machine. On the other hand, in the technology disclosed in Patent Document 2, the monitoring camera is integrally attached to the manual welding torch. This means that the captured images are significantly affected by the operator's behavior, making it difficult to standardize the criteria for determining the quality of the skill and reflect them in subsequent judgments and evaluations.

The present disclosure therefore aims to provide a welding construction assessment method and welding construction assessment device that improve the accuracy of judging welding quality or evaluating the skills of a welding worker.

A welding performance evaluation method according to one aspect of the present disclosure includes the steps of: photographing a welded portion including a molten pool during welding; identifying primary feature values related to the shape of at least one of the arc and the molten pool based on the welding image obtained by photographing the weld; identifying secondary feature values as dimensionless quantities from a calculation formula using the primary feature values as variables; judging the welding quality or evaluating the skill of the welding worker based on the secondary feature values; and extracting an arc region corresponding to the arc or a molten pool region corresponding to the molten pool from the welding image as a mask image, wherein the primary feature values are identified from a plurality of feature points, a circumscribing rectangle of the arc region, or a circumscribing rectangle of the molten pool region, extracted based on the shape of the arc region or the molten pool region; and the molten pool region is extracted using machine learning that predicts the molten pool endpoints by learning and determining the correct molten pool endpoint positions from brightness values on a plurality of search lines of the molten pool extending from the center of gravity of the arc region to the edges of the welding image.

In the above-described welding construction evaluation method, one of the plurality of feature points may correspond to a tip point of a wire forming an arc, which is extracted as a depression in the mask image. One of the plurality of feature points may correspond to the center of gravity of the arc region. A secondary feature value may be set for each welding construction object, for each item of interest in judging welding quality or in evaluating skills, or for each of a plurality of welding images obtained by photographing a weld from different photographing positions and photographing directions. The primary feature value may be a line passing through two feature points, and the secondary feature value may be the slope of the line. The primary feature value may be a line segment connecting two feature points, and the secondary feature value may be the ratio between the length of one line segment formed by different combinations of two feature points and the length of the other line segment. The primary feature value may be the length of each side of a circumscribing rectangle of the arc region, and the secondary feature value may be the ratio between the area of the circumscribing rectangle based on a welding image photographed from one photographing position and photographing direction and the area of the circumscribing rectangle based on a welding image photographed from another photographing position and photographing direction. Furthermore, the welding quality or skill may be judged or evaluated based on standard information relating to the pass/fail of the judgment of welding quality or evaluation of skill, which is predefined based on the numerical value of the secondary feature.

In addition, a welding construction judgment device according to another aspect of the present disclosure includes a photographing device that photographs a welded portion including a molten pool during welding; a memory unit that stores calculation formulas for secondary feature quantities as dimensionless quantities with primary feature quantities related to the shape of at least one of the arc and the molten pool as variables; a control and calculation unit that identifies the primary feature quantities based on welding images acquired from the photographing device, identifies the secondary feature quantities based on the primary feature quantities and the calculation formula acquired from the memory unit, and judges the welding quality or evaluates the skill of the welding worker based on the secondary feature quantities ; and an input unit that inputs welding objects and items to be focused on in judging the welding quality or items to be focused on in evaluating the skills, wherein there are multiple photographing devices with different photographing positions and photographing directions, and the control and calculation unit selects from the multiple photographing devices a photographing device that acquires welding images to be used to identify the primary feature quantities for each welding object or each item to be focused on in judging the welding quality or each item to be focused on in evaluating the skills input via the input unit .

This disclosure provides a welding construction assessment method and welding construction assessment device that improve the accuracy of judging welding quality or evaluating the skills of a welding worker.

1 is a block diagram showing a schematic configuration of a welding construction determination device according to an embodiment; FIG. 2 is a perspective view showing the installation position and shooting direction of a first camera. FIG. 10 is a perspective view showing the installation position and shooting direction of a second camera. FIG. 10 is a perspective view showing the installation position and shooting direction of a third camera. FIG. 10 is a perspective view showing the installation position and shooting direction of a fourth camera. 10 is a flowchart showing a welding execution determination process according to one embodiment. 10 is a flowchart illustrating a primary feature amount specifying step. FIG. 10 is a diagram showing a welding image acquired in a welding image input step. FIG. 10 is a diagram showing a mask image displaying each heat quantity region of the arc region and the molten pool region. FIG. 5C is a diagram showing an image in which a circumscribing rectangle for each heat quantity region is added to the mask image of FIG. 5B. FIG. 10 is a diagram showing an image illustrating an example of feature points related to a molten pool region. FIG. 10 is a diagram in which a plurality of feature points are plotted on a welding image captured by a first camera. 6B is a graph for explaining extraction conditions for each feature point shown in FIG. 6A. FIG. 10 is a diagram in which a plurality of feature points are plotted on a welding image captured by a second camera. 7B is a graph for explaining extraction conditions for each feature point shown in FIG. 7A. FIG. 10 is a diagram in which a plurality of feature points are plotted on a welding image captured by a third camera. 8B is a graph for explaining extraction conditions for each feature point shown in FIG. 8A. FIG. 10 is a diagram in which a plurality of feature points are plotted on a welding image captured by a fourth camera. 9B is a graph for explaining extraction conditions for each feature point shown in FIG. 9A. FIG. 10 is a diagram showing a welding image in which primary feature amounts related to secondary feature amount I and the like are displayed. 10 is a graph showing a secondary feature value I relative to heat input. FIG. 10 is a diagram showing a welding image in which primary feature amounts and the like related to secondary feature amount II are displayed. 10 is a graph showing secondary feature quantity II with respect to heat input. FIG. 10 is a diagram showing a welding image in which primary feature amounts and the like related to secondary feature amount III are displayed. 10 is a graph showing secondary feature quantity III with respect to heat input. FIG. 10 is a diagram showing a welding image in which primary feature amounts and the like related to secondary feature amount IV are displayed. 10 is a graph showing a secondary feature value IV relative to the welding speed under a first standard condition. 10 is a graph showing secondary feature value IV versus welding speed under second standard conditions. FIG. 10 is a diagram showing a welding image in which primary feature amounts related to secondary feature amount V and the like are displayed. FIG. 10 is a diagram showing a welding image in which primary feature amounts and the like related to secondary feature amounts VI are displayed. FIG. 10 is a diagram showing a welding image in which primary feature amounts and the like related to secondary feature amount VII are displayed. FIG. 10 is a diagram showing a welding image in which primary feature amounts and the like related to secondary feature amount VII are displayed. FIG. 10 is a diagram in which a plurality of feature points relating to secondary feature amount XII are plotted on a welding image. 17B is a graph for explaining extraction conditions for each feature point shown in FIG. 17A. 10 is a table showing determination and evaluation results based on secondary feature amounts V to VII. 10 is a table showing determination and evaluation results based on secondary feature XII.

Below, several exemplary embodiments will be described with reference to the drawings. The dimensions, materials, and other specific numerical values shown in each embodiment are merely examples and, unless otherwise specified, do not limit the present disclosure. Furthermore, elements having substantially the same functions and configurations will be assigned the same reference numerals to avoid redundant explanation, and elements not directly related to the present disclosure will not be illustrated.

Figure 1 is a block diagram showing the general configuration of a welding construction judgment device 1 according to one embodiment. The welding construction judgment device 1 judges the welding quality or evaluates the skill of the welding worker based on images obtained by photographing the molten pool during welding.

The welding to be judged or evaluated by the welding execution judgment device 1 may be welding performed by an automatic welding machine or manual welding performed by a worker. Here, manual welding includes so-called semi-automatic welding. Examples of welding methods that can be judged or evaluated include arc welding, such as MAG welding, MIG welding, and TIG welding. In this embodiment, the welding execution judgment device 1 judges the quality or evaluates the skill of welding performed using a welding machine 80, which is an automatic or semi-automatic welding machine employing MAG welding, for example. The welding machine 80 includes a welding torch 81 and a wire feeder (not shown) that automatically feeds wire 82 to the welding torch 81 (see Figure 2A, etc.). The wire 82 is, for example, a solid wire, and is fed from the tip of the welding torch 81 toward the welding object 90 (see Figure 2A, etc.). The welding progresses by melting the wire 82 and the welding object 90 using an arc generated between the wire 82 and the welding object 90, which serves as the welded member (base material). The welding torch 81 supplies shielding gas toward the weld on the welding object 90. The shielding gas is an active gas such as carbon dioxide, and shields the area around the arc and molten pool from the atmosphere.

The welding object 90 may be the product itself or a component that constitutes part of the product. Alternatively, the welding object 90 may be, for example, a test piece used in the skill certification test of the Japanese Industrial Standards (JIS). The shape of the welding object 90 is, for example, a flat plate that is joined together by butt welding. However, the final shape of the component produced by welding is not limited to a flat plate shape, and may also be tubular.

The welding construction assessment device 1 includes a photography device 10, a control and calculation unit 11, a memory unit 12, an input unit 13, and an output unit 14.

The imaging device 10 captures an image of the welded portion, including the molten pool, formed on the welding object 90 while the welding object 90 is being welded by the welding machine 80. The imaging device 10 is, for example, a CCD camera. There are multiple imaging devices 10 so that they can capture images of the welded portion from different viewpoints. In this embodiment, as an example, a total of four imaging devices 10 are used: a first camera 10a, a second camera 10b, a third camera 10c, and a fourth camera 10d. Hereinafter, the images captured by the imaging devices 10 will be referred to as "welding images."

2A to 2D are perspective views illustrating the installation positions and imaging directions of each imaging device 10 relative to a molten pool in a welding object 90. 2A to 2D show a plate-shaped welding object 90, a welding torch 81 of a welding machine 80, and a wire 82 supplied from the welding torch 81 to a weld in the welding object 90. 2A to 2D also show the welding direction _DW with an outline arrow.

Here, the XY coordinate axes to be referred to in the following drawings are set. In this embodiment, the XY plane is assumed to be a plane parallel to the main plane of the plate-shaped welding object 90. The X axis is along the extension direction of a linear route provided on the welding object 90. In other words, the welding direction _DW in this embodiment is a direction generally along the X axis. The Y axis is perpendicular to the X axis.

Fig. 2A is a diagram showing the installation position and shooting direction of first camera 10a. The shooting direction of first camera 10a is a direction in which the horizontal angle θ _H = 45° and the vertical angle θ _V = 45° from the left side of the welding direction _DW , with the welded portion as the reference. Note that in Fig. 2A, a dashed line is displayed at the position on the left side of the welding direction _DW where the horizontal angle θ _H = 0°. Then, first camera 10a is installed at a position where the shooting distance L _S = 250 mm is achieved while satisfying the shooting direction conditions.

2B is a diagram showing the installation position and shooting direction of second camera 10b. The shooting direction of second camera 10b is a direction at a vertical angle θ _V = 45° on the left side of the welding direction _DW with respect to the welded portion. Second camera 10b is installed at a position that satisfies the shooting direction condition and has a shooting distance L _S = 250 mm.

2C is a diagram showing the installation position and shooting direction of the third camera 10c. The shooting direction of the third camera 10c is a direction in front of the welding direction _DW with respect to the welded portion as the reference, with a vertical angle _θV = 45°. The third camera 10c is installed at a position that satisfies the shooting direction condition and has a shooting distance _Ls = 250 mm.

2D is a diagram showing the installation position and shooting direction of the fourth camera 10d. The shooting direction of the fourth camera 10d is a direction in front of the welding direction _DW with respect to the welded portion as the reference, with a vertical angle _θV = 15°. The fourth camera 10d is installed at a position that satisfies the shooting direction condition and has a shooting distance _Ls = 250 mm.

Note that when the welding construction judgment device 1 judges or evaluates a welding construction object 90, evaluation object, or standard conditions are limited to specific ones, not all four image capture devices 10 will necessarily be used. Therefore, image capture devices 10 that are not anticipated for use do not need to be included in the welding construction judgment device 1. On the other hand, even image capture devices 10 that are not anticipated for use may be included in the welding construction judgment device 1 as spares, for example, in case a malfunction occurs in one of the other image capture devices 10. In this case, feature amounts based on the welding image that would be obtained from the malfunctioning image capture device 10 can be supplemented with feature amounts based on the welding image obtained from the spare image capture device 10.

The control and calculation unit 11 is, for example, a CPU (Central Processing Unit) and executes a program related to the welding performance evaluation process, which is a welding performance evaluation method according to this embodiment. The control and calculation unit 11 is electrically connected to all of the imaging devices 10, the memory unit 12, and the input unit 13. Furthermore, when the welding performance evaluation process is executed to evaluate the welding quality of an automatic welding machine, the control and calculation unit 11 may be electrically connected to the welding machine 80 in order to modify the welding performance conditions of the automatic welding machine in real time based on the evaluation results. Alternatively, when the welding performance evaluation process is executed to evaluate the skill of a welding worker, the control and calculation unit 11 may be electrically connected to the output unit 14 in order to display the evaluation results to the welding worker or an operator operating the welding performance evaluation device 1.

The memory unit 12 is, for example, a semiconductor memory or a magnetic storage device such as a hard disk drive (HDD), and stores (preserves) programs, databases, etc. The control and calculation unit 11 can read programs and data stored in the memory unit 12 as needed. The control and calculation unit 11 can also write new data to the memory unit 12 as needed.

The memory unit 12 also has a program storage unit 15 and a database storage unit 16. The program storage unit 15 stores at least a control program related to the series of controls in the welding construction assessment process and an image processing program, including a machine learning program, executed in the primary feature identification process S104 (see FIG. 3). The database storage unit 16 stores at least various databases queried in the welding construction assessment process. The databases that can be stored in the program storage unit 15 are a standard condition database queried in the standard condition acquisition process S101 (see FIG. 3) and a reference information database that stores reference information queried in the reference information acquisition process S106 (see FIG. 3). Details of the standard condition database and the reference information database will be described later in conjunction with the explanation of the standard condition acquisition process S101 or the reference information acquisition process S106. Furthermore, the memory unit 12 saves welding images captured by the imaging device 10.

The input unit 13 is, for example, an operation input device such as a keyboard or touch panel, a receiving device that receives information from an external device, or an input port that reads information from a portable memory device such as a memory card. The operator can input various information directly from the input unit 13 as an operation input device. Alternatively, the control and calculation unit 11 can receive various information from an external server or the like via a communication network such as a LAN or the Internet from the input unit 13 as a receiving device. Furthermore, the control and calculation unit 11 can transfer various information from a memory device via the input unit 13 as an input port. The information that can be input by the input unit 13 is the initial conditions set in the initial condition setting step S100 (see Figure 3). Details of the initial conditions will be described later in conjunction with the explanation of the initial condition setting step S100.

The output unit 14 may be a display unit such as a monitor, a transmission device that transmits information externally, or an output port that writes information to a memory device. The control and calculation unit 11 can display the judgment and evaluation results obtained in the welding construction judgment process on the output unit 14 as a display unit, allowing the contractor or operator to directly recognize them. Alternatively, the control and calculation unit 11 can transmit the judgment and evaluation results from the output unit 14 as a transmission device to an external server or the like via a communications network. Furthermore, the control and calculation unit 11 can transfer the judgment and evaluation results from the output unit 14 as an output port to a memory device.

Next, we will explain the welding construction judgment method and the operation of the welding construction judgment device 1 according to this embodiment.

Figure 3 is a flowchart showing the welding construction judgment process as part of the welding construction judgment method according to this embodiment. The welding construction judgment process includes a series of steps according to the flowchart shown in Figure 3 and is executed by the control and calculation unit 11 as a control program for the welding construction judgment device 1.

When the control and calculation unit 11 starts executing the welding construction judgment process, it first executes the initial condition setting process S100. In the initial condition setting process S100, the control and calculation unit 11 sets various initial conditions based on information input by the operator via the input unit 13. The initial conditions include, for example, the welding construction target 90 and the judgment/evaluation target.

The welding object 90 is set based on the type or specific shape of the welded joint. For example, when judging the welding quality of the welding machine 80 as an automatic welding machine, the product itself may be selected as the welding object 90. Alternatively, when evaluating the skill of the operator operating the welding machine 80 as a semi-automatic welding machine, various test pieces may be selected as the welding object 90.

In the following examples of this embodiment, the welding object 90 is assumed to be set in accordance with the dimensions of the test piece used in the JIS Proficiency Test. Specifically, as shown in Figures 2A to 2D, the welding object 90 is a pair of plate materials that have been pre-processed to have a V-shaped groove and are butt-welded together, with the grooved ends serving as the joining surfaces. For example, three types of plate materials with different thicknesses are used as the welding object 90: thin plate, medium plate, and thick plate. The thickness of the thin plate is 3.2 mm. Welding between the thin plates is performed under conditions of a 90° groove angle and a gap (root spacing) of approximately 2 mm. The thickness of the medium plate is 9.0 mm. Welding between the medium plates is performed under conditions of a 70° groove angle and a 5 mm gap. The thickness of the thick plate is 19.0 mm. Welding between thick plates is performed under conditions where the groove angle is 70° and the gap is approximately 5 mm.

The subject of judgment and evaluation is set as a specific item of focus regarding welding quality or skill. For example, the subject of judgment and evaluation regarding welding quality (welded joint quality) is the formation of back waves.

Next, the control and calculation unit 11 executes the standard condition acquisition step S101. In the standard condition acquisition step S101, the control and calculation unit 11 acquires and sets standard conditions corresponding to the initial conditions set in the initial condition setting step S100 by consulting the standard condition database. The standard conditions are, for example, the welding current, welding voltage, welding speed, and angle of the welding torch 81 for each initial condition, such as the type of weld joint or the plate thickness or material of the welding object 90, or a calculation formula for secondary feature quantities corresponding to the object to be judged or evaluated. Note that the secondary feature quantities in this embodiment will be described later in conjunction with the secondary feature quantity identification step S105. The standard condition database is pre-stored in the database storage unit 16 included in the memory unit 12. The calculation formula for secondary feature quantities stored in the standard condition database may be a formula identified in the secondary feature quantity identification step S105 during the welding operation judgment step during a previous welding operation and continuously saved.

Next, the control and calculation unit 11 starts the welding operation (start welding operation S102). If the welding operation is performed by the welding machine 80 as an automatic welding machine, the control and calculation unit 11 may start the operation of the welding machine 80. On the other hand, if the welding operation is performed by a worker as manual welding, the control and calculation unit 11 may notify the worker that the welding operation is ready to start, for example, by displaying on the output unit 14 as a display unit.

Next, the control and calculation unit 11 executes the welded portion photographing process S103. In the welded portion photographing process S103, the control and calculation unit 11 causes each of the photographing devices 10, the first camera 10a, the second camera 10b, the third camera 10c, and the fourth camera 10d, to photograph the welded portion during welding. As described above, the welded portion photographed by the photographing devices 10 includes the molten pool formed in the welding object 90. The control and calculation unit 11 stores the photographed welding images in the memory unit 12.

Next, the control and calculation unit 11 executes the primary feature quantity identification process S104. Figure 4 is a flowchart showing the primary feature quantity identification process S104. The primary feature quantity identification process S104 is executed by the control and calculation unit 11 as an image processing program including a series of steps in accordance with the flowchart shown in Figure 4. Then, in the primary feature quantity identification process S104, the control and calculation unit 11 finally identifies the primary feature quantity based on the welding image.

5A to 5D are diagrams showing, as an example, images acquired or extracted in each step included in the primary feature identification step S104 based on the welding image _IP captured by the second camera 10b. In FIGS. 5B to 5D, the area corresponding to the arc in the welding image _IP shown in FIG. 5A is referred to as the "arc area _AC ," and the area corresponding to the molten pool (molten metal) is referred to as the "molten pool area _MP ." Note that areas other than the arc area _AC and the molten pool area _MP are background, and are shown in black in the figures.

Here, primary feature amounts refer to a function for directly recognizing the welded portion, including the molten pool, i.e., amounts representing the shape of the heat source area, such as the molten pool, obtained from each imaging device 10 that images the welded portion. Primary feature amounts are defined in advance for each secondary feature amount, described below. For example, a primary feature amount may be the position of a single feature point, or the length of a line segment connecting two feature points. Alternatively, a primary feature amount may be the area of a region consisting of multiple line segments or curves.

When the control and calculation unit 11 starts executing the primary feature amount specifying step S104, it first executes the welding image input step S200. In the welding image input step S200, the control and calculation unit 11 acquires from the storage unit 12 welding images _IP necessary for specifying desired primary feature amounts from among the welding images _IP captured by each imaging device 10 in the welded part imaging step S103.

5A is a diagram showing a welding image _IP acquired in the welding image input process S200. In the example shown in Fig. 5A, the welding image _IP is expressed in grayscale, but in reality it is a color image. In the welding image _IP , the arc H _A is expressed in white or a color close to white, the molten pool H _M is expressed in roughly gray, and the background is expressed in black.

Next, the control and calculation unit 11 executes a filtering process step S201. In the filtering process S201, the control and calculation unit 11 processes the welding image _IP obtained in the welding image input process S200 using, for example, a Gaussian filter to suppress fluctuations in the heat quantity region that are specific to the welding image _IP .

Next, the control and calculation unit 11 executes an arc region extraction step S202. In the arc region extraction step S202, the control and calculation unit 11 extracts a specific color component, for example, a blue component, from the filtered welding image _IP , thereby extracting it as an arc region _AC . At this time, the control and calculation unit 11 may finally extract the arc region _AC by binarizing the image of the blue component using a preset brightness threshold value.

Next, the control and calculation unit 11 executes a weld pool region extraction step S203. In this step S203, the control and calculation unit 11 extracts the weld pool region _MP by executing a machine learning program employing a learning model that predicts the endpoints of the weld pool from the waveform on the search line. In this case, the machine learning program may employ deep learning, which uses mean square error (MSE) as an evaluation function.

The control and calculation unit 11 executes a machine learning program to learn and determine, for example, the correct weld pool endpoint positions (1,717 images x 128 search lines = approximately 220,000 data points). Specifically, the control and calculation unit 11 calculates the center of gravity of the arc region _AC extracted in the arc region extraction step S202 and generates 128 search lines of the weld pool from the center of gravity of the arc region _AC to the edge of the image. Next, the control and calculation unit 11 calculates the weld pool endpoints on each search line according to the seven pattern classifications based on the waveform shapes of the brightness values on the search lines. The control and calculation unit 11 then connects the calculated weld pool endpoints between adjacent search line points and extracts them as the weld pool region _MP .

The seven patterns mentioned above for calculating the endpoints of the weld pool are as follows: Pattern 1 is a waveform shape in which an outline appears after a series of areas of only the weld pool. Pattern 2 is a waveform shape in which yellow areas appear two or more times. Pattern 3 is a waveform shape in which a yellow area with high heat is created around the arc. Pattern 4 is a waveform shape in which the edges of the arc and weld pool are close to each other. Pattern 5 is a waveform shape in which the shielding gas enters the background once and then re-emerges. Pattern 6 is a waveform shape in which the shielding gas spreads outside the weld pool. Pattern 7 is a waveform shape in which the brightness value of the weld pool outline is close to that of the background.

5B is a diagram showing an image displaying the heat quantity regions of the arc region _AC extracted in the arc region extraction step S202 and the weld pool region _MP extracted in the weld pool region extraction step S203. The arc region _AC and the weld pool region _MP are represented as a mask image I _M as shown in FIG.

Next, the control and calculation unit 11 executes a circumscribing rectangle extraction step S204. In the circumscribing rectangle extraction step S204, the control and calculation unit 11 extracts a circumscribing rectangle _RA of the arc region _AC from the mask of the arc region _AC extracted in the arc region extraction step S202. The control and calculation unit 11 also extracts a circumscribing rectangle _RB of the molten pool region _MP from the mask of the molten pool region _MP extracted in the molten pool region extraction step S203.

5C is a diagram showing an image in which a circumscribing rectangle _RA of the arc region _AC and a circumscribing rectangle _RB of the molten pool region _MP are added to the mask image I _M of FIG. The circumscribing rectangle _RB is extracted as a rectangle with the lower left corner of the image as the origin, the left side passing through the minimum value _XM1 of the mask in the X direction, the right side passing through the maximum value _XM2 of the mask in the X direction, the bottom side passing through the minimum value _YM1 of the mask in the Y direction, and the top side passing through the maximum value _YM2 of the mask in the Y direction. The mask for the circumscribing rectangle _RB is the mask for the molten pool region _MP . The circumscribing rectangle _RA is extracted using the same method as the extraction method for the circumscribing rectangle _RB .

Next, the control and calculation unit 11 executes a feature point extraction step S205. In the feature point extraction step S205, the control and calculation unit 11 extracts feature points to be used in subsequent calculation of primary feature amounts based on the circumscribing rectangles R 1 _A and R 1 _B extracted in the circumscribing rectangle extraction step S204.

Fig. 5D is a diagram showing an image showing an example of feature points related to the molten pool region _MP extracted based on the circumscribing rectangle _RB . The four feature points shown in Fig. 5D, i.e., feature points _P1A , _P1B , _P2A , and _P2B , correspond to the feature points shown in Fig. 7B, which will be described below.

The feature points extracted in the feature point extraction step S205 are based on the welding image _IP . Therefore, various conditions for extracting feature points are considered depending on conditions that affect the welding image _IP , specifically, which image capture device 10 captured the welding image _IP and the type of welding object 90. Below, feature points and their extraction conditions for each image capture device 10 and welding object 90 are exemplified.

Figures 6A and 6B show a first example of feature points and their extraction conditions extracted from a welding image captured by the first camera 10a when the welding target 90 is a middle plate.

6A is a diagram showing a welding image on which multiple feature points are plotted. The feature points in the first example are, for example, seven in total: feature point _P1A , feature point _P1B , feature point _P2 , feature point _P3A , feature point _P3B , feature point _P4A , and feature point _P4B . Feature point _P1A is the leading edge of the weld pool region. Feature point _P1B is the trailing edge of the weld pool region. Feature point _P2 is the midpoint between the leading and trailing edges of the weld pool region. Feature point _P3A is the left edge of the weld pool region. Feature point _P3B is the right edge of the weld pool region. Feature point _P4A is the top edge of the weld pool region. Feature point _P4B is the bottom edge of the weld pool region. Hereinafter, the expressions "front/back,""left/right," and "up/down" used to indicate each characteristic point are intended to simply distinguish the positional relationship between multiple characteristic points, and are not based on, for example, a strict direction relative to the welding direction _DW or a strict shape of the molten pool region.

Fig. 6B is a graph illustrating the extraction conditions for each feature point extracted based on the mask of the weld pool region. The masks of the arc region and weld pool region in Fig. 6B correspond to the arc region _AC or the weld pool region _MP shown in Fig. 5C and other figures, and are extracted based on the welding image shown in Fig. 6A. The circumscribing rectangle of the weld pool region corresponds to the circumscribing rectangle _RB of the weld pool region _MP shown in Fig. 5C and other figures. Regarding the circumscribing rectangle of the weld pool region, the minimum value in the X direction is denoted as _XM1 , the maximum value in the X direction as _XM2 , the minimum value in the Y direction as _YM1 , and the maximum value in the Y direction as _YM2 .

6B , the extraction conditions for feature points in the first example are as follows: Feature point _P1A is extracted as the point closest to the coordinates ( _XM1 , _YM1 ) of the two intersections between a line passing from the lower left coordinate ( _XM1 , _YM1 ) to the upper right coordinate ( _XM2 , _YM2 ) of the circumscribing rectangle of the weld pool region and the outline of the mask of the weld pool region. Feature point _P1B is similarly extracted as the point closest to the coordinates ( _XM2 , _YM2 ) of the two intersections between a line passing from the coordinates ( _XM1 , _YM1 ) to the coordinates ( _XM2 , _YM2 ) of the circumscribing rectangle of the weld pool region and the outline of the mask of the weld pool region. Feature point _P2 is extracted as the midpoint between feature points _P1A and _P1B . Feature point _P3A is extracted as the intersection of a half ray extending from feature point _P2 in the negative direction of the X-axis and the contour of the mask of the weld pool region. Feature point _P3B is extracted as the intersection of a half ray extending from feature point _P2 in the positive direction of the X-axis and the contour of the mask of the weld pool region. Feature point _P4A is extracted as the intersection of a half ray extending from feature point _P2 in the positive direction of the Y-axis and the contour of the mask of the weld pool region. Feature point _P4B is extracted as the intersection of a half ray extending from feature point _P2 in the negative direction of the Y-axis and the contour of the mask of the weld pool region. Note that Figures 6A and 6B illustrate line segments connecting two feature points that are referenced when extracting each feature point.

Figures 7A and 7B show a second example of feature points and their extraction conditions extracted from a welding image captured by the second camera 10b when the welding object 90 is an intermediate plate.

7A is a diagram showing a welding image on which multiple feature points are plotted. The feature points in the second example are, for example, five in total: feature point _P1A , feature point _P1B , feature point _P2A , feature point _P2B , and feature point _P3 . Feature point _P1A is the leading end point of the weld pool region. Feature point _P1B is the trailing end point of the weld pool region. Feature point _P2A is the left end point of the weld pool region. Feature point _P2B is the right end point of the weld pool region. Feature point _P3 is the leading end point of the wire 82.

7B is a graph illustrating the extraction conditions for each feature point extracted based on the masks of the arc region and the weld pool region. The drawing conditions in FIG. 7B are the same as those in FIG. 6B. The extraction conditions for feature points in the second example are as follows: Feature point _P1A is extracted as the point where Y is the median value from the point group where X = _XM1 . Feature point _P1B is extracted as the point where Y is the median value from the point group where X = _XM2 . Feature point _P2A is extracted as the point where X is the median value from the point group where Y = _YM2 . Feature point _P2B is extracted as the point where X is the median value from the point group where Y = _YM1 . Feature point _P3 is extracted as the bottom edge of the depression in the mask at the top of the arc region.

Here, the control and calculation unit 11 may identify the depression of the mask to be referenced when extracting the feature point _P3 by, for example, analyzing pixel values along the X direction of the upper half of the mask of the arc region, where the pixel values on the mask are 255 and the pixel values in the background are 0 (zero). In this case, the control and calculation unit 11 extracts, as a depression candidate, a position obtained by averaging the X coordinates from the coordinates where the pixel values on one line are 0 (zero). Next, the control and calculation unit 11 narrows down the depression candidates based on the conditions that the depression must consist of three or more consecutive lines in the Y direction and that the bottom end of the consecutive lines in the Y direction is used as the end of the depression. Then, from the final depression candidates, the control and calculation unit 11 selects the one whose X coordinate is closest to the center of gravity of the arc region as the depression of the mask to be referenced when extracting the feature point _P3 .

Figures 8A and 8B show a third example of feature points and their extraction conditions extracted from a welding image captured by the third camera 10c when the welding target 90 is an intermediate plate.

8A is a diagram showing a welding image on which multiple feature points are plotted. The feature points in the third example are, for example, five in total: feature point _P1A , feature point _P1B , feature point _P2A , feature point _P2B , and feature point _P3 . Feature point _P1A is the upper end point of the weld pool region. Feature point _P1B is the lower end point of the weld pool region. Feature point _P2A is the left end point of the weld pool region. Feature point _P2B is the right end point of the weld pool region. Feature point _P3 is the tip point of the wire 82.

8B is a graph illustrating the extraction conditions for each feature point extracted based on the masks of the arc region and the molten pool region. The drawing conditions in FIG. 8B are the same as those in FIG. 6B. The extraction conditions for feature points in the third example are as follows: Feature point _P1A is extracted as the point where X is the median value from the point cloud where Y = _YM2 . Feature point _P1B is extracted as the point where X is the median value from the point cloud where Y = _YM1 . Feature point _P2A is extracted as the point where Y is the median value from the point cloud where X = _XM1 . Feature point _P2B is extracted as the point where Y is the median value from the point cloud where X = _XM2 . Feature point _P3 is extracted as the bottom edge of the depression in the mask at the top of the arc region, as in the second example above.

Figures 9A and 9B show a fourth example of feature points and their extraction conditions extracted from a welding image captured by the fourth camera 10d when the welding target 90 is an intermediate plate.

9A is a diagram showing a welding image on which multiple feature points are plotted. The feature points in the fourth example are, for example, five in total: feature point _P1 , feature point _P2A , feature point _P2B , feature point _P3 , and feature point _P4 . Feature point _P1 is the front end point of the weld pool region. Feature point _P2A is the left end point of the weld pool region. Feature point _P2B is the right end point of the weld pool region. Feature point _P3 is the tip point of the wire 82. Feature point _P4 is the bottom end point of the arc region.

FIG. 9B is a graph illustrating the extraction conditions for each feature point extracted based on the masks of the arc region and the molten pool region. The drawing conditions in FIG. 9B are the same as those in FIG. 6B. However, in FIG. 9B, the minimum value in the Y direction of the arc region is denoted as Y _A1 . The extraction conditions for feature points in the fourth example are as follows: Feature point _P1 is extracted as the point where X is the median value from the point cloud where Y = Y _M1 . Feature point _P2A is extracted as the point where Y is the median value from the point cloud where X = X _M1 . Feature point _P2B is extracted as the point where Y is the median value from the point cloud where X = X _M2 . Feature point _P3 is extracted as the bottom edge of the depression in the mask at the top of the arc region, as in the second example above. Feature point _P4 is extracted as the point where X is the median value from the point cloud where Y = Y _A1 .

Up to this point, an example of feature points and their extraction conditions has been described for the case where the welding object 90 is a medium-sized plate. However, in the feature point extraction process S205, feature points are similarly extracted even when the welding object 90 is, for example, a thick plate, depending on the object to be judged and evaluated. Feature points and their extraction conditions for the case where the welding object 90 is a thick plate will also be exemplified in the explanation of secondary feature quantity XII using Figures 17A and 17B below.

Then, following the feature point extraction step S205, the control and calculation unit 11 executes a primary feature value calculation step S206. In the primary feature value calculation step S206, the control and calculation unit 11 calculates primary feature values using the circumscribing rectangle _RA of the arc region _AC or the circumscribing rectangle _RB of the molten pool region _MP extracted in the circumscribing rectangle extraction step S204, or the feature points extracted in the feature point extraction step S205. Here, the primary feature values are defined from secondary feature values selected based on the judgment/evaluation targets set in the initial condition setting step S100. Therefore, specific primary feature values that can be calculated in the primary feature value calculation step S206 will be exemplified for each secondary feature value in the following description of the secondary feature values identified in the secondary feature value identification step S105. After completing the primary feature value calculation step S206, the control and calculation unit 11 ends the primary feature value identification step S104 in the welding construction judgment step.

Next, following the primary feature quantity identification process S104, the control and calculation unit 11 executes the secondary feature quantity identification process S105. A secondary feature quantity is a quantity obtained from a calculation formula (secondary feature quantity calculation formula) that uses the primary feature quantity as a variable. A secondary feature quantity may be a ratio between primary feature quantities, or a dimensionless quantity such as the slope of a line segment connecting two feature points. In the secondary feature quantity identification process S105, the control and calculation unit 11 calculates the secondary feature quantity using the secondary feature quantity calculation formula acquired in the standard condition acquisition process S101 according to the judgment/evaluation object. Below, specific secondary feature quantities selected and identified for each welding construction object 90 and judgment/evaluation object are illustrated.

Figures 10A and 10B are diagrams relating to secondary feature value I when the welding target 90 is a middle plate and the object of judgment/evaluation is the presence or absence of incomplete fusion related to the stability of the movement of the molten pool.

FIG. 10A is a diagram showing the primary feature values and the line segments related to the calculation formula for the secondary feature value I on the welding image used to identify the primary feature values. The primary feature values used to identify the secondary feature value I are the positions of feature points _P1A and _P1B obtained based on the welding image captured by the second camera 10b. The welding image shown in FIG. 10A corresponds to the welding image shown in FIG. 7A. That is, feature point _P1A is the leading end point of the weld pool region. Feature point _P1B is the trailing end point of the weld pool region. In this case, the calculation formula for the secondary feature value I is the equation for a straight line passing through feature points _P1A and _P1B . The finally calculated secondary feature value I is the slope of the line passing through feature points _P1A and _P1B , which is a dimensionless quantity.

Figures 11A and 11B are diagrams relating to secondary feature value II when the welding target 90 is a middle plate and the object of judgment/evaluation is the presence or absence of incomplete fusion related to the followability of the molten pool to the wire 82.

FIG. 11A is a diagram showing the primary feature values and the line segments related to the calculation formula for secondary feature value II on the welding image used to identify the primary feature values. The primary feature values used to identify secondary feature value II are the positions of feature point _P3 and feature point _P1B obtained based on the welding image captured by the third camera 10c. The welding image shown in FIG. 11A corresponds to the welding image shown in FIG. 8A. That is, feature point _P3 is the tip point of the wire 82. Feature point _P1B is the front end point of the weld pool region. In this case, the calculation formula for secondary feature value II is the equation for a straight line passing through feature point _P3 and feature point _P1B . The finally calculated secondary feature value II is the slope of the line passing through feature point _P3 and feature point _P1B , which is a dimensionless quantity.

Figures 12A and 12B are diagrams relating to secondary feature value III when the welding object 90 is a middle plate and the object of judgment/evaluation is heat input.

FIG. 12A is a diagram showing the primary feature values and the circumscribing rectangle related to the calculation formula for the secondary feature value III on the welding image used to identify the primary feature values. The primary feature value used to identify the secondary feature value III is the length of each side of the circumscribing rectangle _RB of the weld pool region obtained based on the welding image captured by the second camera 10b. In other words, when identifying the secondary feature value III, it is sufficient to refer to the circumscribing rectangle _RB extracted in the circumscribing rectangle extraction step S204, and no feature points are required. Note that the welding image shown in FIG. 12A also corresponds to the welding image shown in FIG. 7A. In this case, the calculation formula for the secondary feature value III is a formula for calculating the area of the circumscribing rectangle _RB from the length of each side of the circumscribing rectangle _RB . The finally calculated secondary feature value III is the area of the circumscribing rectangle _RB .

Figures 13A, 13B, and 13C are diagrams relating to secondary feature value IV when the welding object 90 is a middle plate and the object of judgment/evaluation is the welding speed.

FIG. 13A shows the primary feature values and the line segments related to the calculation formula for the secondary feature value IV on the welding image used to identify the primary feature values. The primary feature values used to identify the secondary feature value IV are calculated from four feature points obtained based on the welding image captured by the second camera 10b. Specifically, the primary feature values are the length of the line segment connecting feature points _{P1A and P1B ,} which corresponds to the length of the weld pool, and the length of the line segment connecting feature points _P2A and _P2B , which corresponds to the width of the weld pool. The welding image shown in FIG. 13A corresponds to the welding image shown in FIG. 7A. That is, feature point _P1A is the leading end point of the weld pool region. Feature point _P1B is the trailing end point of the weld pool region. Feature point _P2A is the left end point of the weld pool region. Feature point _P2B is the right end point of the weld pool region. In this case, the calculation formula for the secondary feature value IV is a formula for calculating the ratio between the length of the line segment connecting the feature points _P1A and _P1B and the length of the line segment connecting the feature points _P2A and _P2B . The finally calculated secondary feature value IV is expressed as the ratio between the length of the line segment connecting the feature points _P1A and _P1B and the length of the line segment connecting the feature points _P2A and _P2B , which is expressed as the ratio between the length and width of the molten pool (aspect ratio), and is a dimensionless quantity.

So far, we have explained examples of secondary feature values when the welding object 90 is a medium-sized plate. Next, we will explain examples of secondary feature values when the welding object 90 is a thin plate or a thick plate.

Figures 14, 15, 16A, and 16B are diagrams relating to secondary feature values related to whether the welding object 90 is a thin plate and the object of judgment/evaluation is, for example, the backside swell formation property, and whether the welding object is classified as "burn-through," "sound weld," or "insufficient backside swell." Figure 14 is a diagram relating to secondary feature value V. Figure 15 is a diagram relating to secondary feature value VI. Figures 16A and 16B are diagrams relating to secondary feature value VII.

Figure 14 shows the primary feature values and lengths related to the calculation formula for the secondary feature value V on the welding image used to identify the primary feature values. The primary feature values used to identify the secondary feature value V are calculated from three feature points obtained based on the welding image captured by the fourth camera 10d. Specifically, the primary feature values are a first length L1 between the tip point of the wire 82 and the rear end point of the weld pool region, which corresponds to the length of the upper groove portion of the weld pool, and a second length L2 between the tip point of the wire 82 and the front end point of the weld pool region, which corresponds to the length of the lower groove portion of the weld pool. In other words, the three feature points here correspond to the front end point of the weld pool region, the rear end point of the weld pool region, and the tip point of the wire 82. In this case, the calculation formula for the secondary feature value V is a formula for calculating the ratio expressed as (first length L1 / second length L2). The final calculated secondary characteristic value V is expressed as (first length L1/second length L2) (length of upper groove of molten pool/length of lower groove of molten pool), and is a dimensionless quantity.

Figure 15 shows the primary feature values and the lengths related to the calculation formula for the secondary feature value VI on the welding image used to identify the primary feature values. The primary feature values used to identify the secondary feature value VI are calculated from four feature points obtained based on the welding image captured by the fourth camera 10d. Specifically, the primary feature values are the third length L3 between the left end point of the weld pool region and the right end point of the weld pool region, which corresponds to the width of the weld pool, and the fourth length L4 between the tip point of the wire 82 and the front end point of the weld pool region, which corresponds to the length of the lower groove of the weld pool. In other words, the four feature points here correspond to the left end point of the weld pool region, the right end point of the weld pool region, the front end point of the weld pool region, and the tip point of the wire 82. In this case, the calculation formula for the secondary feature value VI is a formula for calculating the ratio expressed as (third length L3 / fourth length L4). The final calculated secondary characteristic value VI is expressed as (third length L3/fourth length L4) (width of molten pool/length of bottom of groove of molten pool), and is a dimensionless quantity.

16A and 16B are diagrams showing the primary feature values and the circumscribing rectangle associated with the calculation formula for the secondary feature value VII on the welding image used to identify the primary feature values. The primary feature value used to identify the secondary feature value VII is the length of each side of the circumscribing rectangle R _A1 of the arc region obtained based on the welding image captured by the second camera 10b, as shown in FIG. 16A . The primary feature value used to identify the secondary feature value VII is the length of each side of the circumscribing rectangle R _A2 of the arc region obtained based on the welding image captured by the fourth camera 10d, as shown in FIG. 16B . In other words, the welding image used to identify the circumscribing rectangle R _A1 can be considered to have been captured from above, and the welding image used to identify the circumscribing rectangle R _A2 can be considered to have been captured from the front. Furthermore, when identifying the secondary feature value VII, it is sufficient to refer to the circumscribing rectangle R _A extracted in the circumscribing rectangle extraction step S204; feature points are not required. In this case, the calculation formula for secondary feature VII is a formula for calculating a ratio expressed as (area of circumscribing rectangle R _A1 derived from the length of each side of circumscribing rectangle R _A1 / area of circumscribing rectangle R _B1 derived from the length of each side of circumscribing rectangle R _B1 ). The finally calculated secondary feature VII is (area of circumscribing rectangle R _A1 / area of circumscribing rectangle R _B1 ), which is a dimensionless quantity.

In addition, when the welding object 90 is a thin plate and the object to be judged/evaluated is classified as "burn-through," "sound weld," or "insufficient weld," the secondary feature may be identified based on the welding image captured by the first camera 10a as follows:

For example, the primary feature used to identify secondary feature VIII as another secondary feature is calculated from the following four feature points obtained based on the welding image captured by the first camera 10a. That is, the primary feature is the length between the left end point of the weld pool area and the right end point of the weld pool area, which corresponds to the left-to-right width of the weld pool, and the length between the top end point of the weld pool area and the bottom end point of the weld pool area, which corresponds to the top-to-bottom width of the weld pool. In this case, the calculation formula for secondary feature VIII is a formula for calculating the ratio expressed as (top-to-bottom width of the weld pool / left-to-right width of the weld pool). The finally calculated secondary feature VIII is then (top-to-bottom width of the weld pool / left-to-right width of the weld pool), which is a dimensionless quantity.

Furthermore, the primary feature used to identify secondary feature IX as another secondary feature is calculated from the following four feature points obtained based on the welding image captured by the first camera 10a. That is, the primary feature is the length between the upper end point of the weld pool area and the left end point of the weld pool area, which corresponds to the upper left width of the weld pool, and the length between the upper end point of the weld pool area and the lower end point of the weld pool area, which corresponds to the vertical width of the weld pool. In this case, the calculation formula for secondary feature IX is a formula for calculating the ratio expressed as (vertical width of the weld pool / upper left width of the weld pool). The finally calculated secondary feature IX is (vertical width of the weld pool / upper left width of the weld pool), which is a dimensionless quantity.

Furthermore, the primary feature used to identify secondary feature X as another secondary feature is calculated from the following four feature points obtained based on the welding image captured by the first camera 10a. That is, the primary feature is the length between the left end point of the weld pool area and the right end point of the weld pool area, which corresponds to the left-right width of the weld pool, and the length between the right end point of the weld pool area and the bottom end point of the weld pool area, which corresponds to the bottom-right width of the weld pool. In this case, the calculation formula for secondary feature X is a formula for calculating the ratio expressed as (bottom-right width of the weld pool / left-right width of the weld pool). The finally calculated secondary feature X is (bottom-right width of the weld pool / left-right width of the weld pool), which is a dimensionless quantity.

Furthermore, the primary feature used to identify secondary feature XI as another secondary feature is calculated from the following four feature points obtained based on the welding image captured by the first camera 10a. That is, the primary feature is the length between the left end point of the weld pool area and the top end point of the weld pool area, which corresponds to the upper left width of the weld pool, and the length between the right end point of the weld pool area and the bottom end point of the weld pool area, which corresponds to the lower right width of the weld pool. In this case, the calculation formula for secondary feature XI is a formula for calculating the ratio expressed as (lower right width of weld pool / upper left width of weld pool). The finally calculated secondary feature XI is (lower right width of weld pool / upper left width of weld pool), which is a dimensionless quantity.

Figures 17A and 17B are diagrams showing a fifth example of feature points and their extraction conditions extracted from a welding image captured by the third camera 10c when the welding object 90 is a thick plate. When the welding object 90 is a thick plate and the object to be judged/evaluated is the presence or absence of incomplete fusion in a welding operation with classification, secondary feature XII can be identified based on the feature points and their extraction conditions shown in Figures 17A and 17B.

17A is a diagram showing a welding image on which multiple feature points are plotted. The feature points in the fifth example are, for example, four in total: feature point _P1B , feature point _P2A , feature point _P2B , and feature point _P3 . Feature point _P1B is the bottom end point of the weld pool region. Feature point _P2A is the left end point of the weld pool region. Feature point _P2B is the right end point of the weld pool region. Feature point _P3 is the center of gravity of the arc region.

Figure 17B is a graph illustrating the extraction conditions for each feature point extracted based on the masks for the arc region and the weld pool region. The masks for the arc region and the weld pool region in Figure 17B are extracted based on the welding image shown in Figure 17A. Regarding the circumscribing rectangle of the arc region, the minimum value in the X direction is denoted as _XA1 , the maximum value in the X direction is denoted as _XA2 , the minimum value in the Y direction is denoted as _YA1 , and the maximum value in the Y direction is denoted as _YA2 . The feature points in the fifth example can basically be extracted by applying the feature point extraction conditions for the third example described using Figure 8B. Meanwhile, feature point _P3 is extracted as the center of gravity of the circumscribing rectangle of the arc region.

In this case, the primary feature quantities used to determine the secondary feature quantity XII are the fifth length L5 between feature points _P3 and _P1B and the sixth length L6 between feature points _P2A and _P2B . The fifth length L5 corresponds to the length between the center of gravity of the arc region and the leading end point of the weld pool region. The sixth length L6 corresponds to the width of the weld pool. The formula for calculating the secondary feature quantity XII is a formula for calculating the ratio expressed as (fifth length L5/sixth length L6). The finally calculated secondary feature quantity XII is (the length between the center of gravity of the arc region and the leading end point of the weld pool region/the width of the weld pool), which is expressed as (fifth length L5/sixth length L6), and is a dimensionless quantity.

Next, following the secondary feature identification step S105, the control and calculation unit 11 executes the reference information acquisition step S106. In the reference information acquisition step S106, the control and calculation unit 11 queries the database storage unit 16 to acquire reference information that will serve as the basis for judgment and evaluation in the subsequent judgment and evaluation step S107. Reference information refers to information selected for each secondary feature, such as welding condition numerical values, quality judgment based on quantification, or relational equations that lead to the quantification of skills. The reference information is continuously stored in various reference information databases in the database storage unit 16 that are categorized by secondary feature. Below, specific reference information and the databases that the control and calculation unit 11 queries when acquiring the reference information are exemplified for each secondary feature.

When the secondary feature value according to the object of judgment/evaluation is secondary feature value I, the reference information is a threshold value for the presence or absence of poor fusion for secondary feature value I. Similarly, when the secondary feature value according to the object of judgment/evaluation is secondary feature value II, the reference information is a threshold value for the presence or absence of poor fusion for secondary feature value II. The reference information regarding secondary feature value I or secondary feature value II is stored in a first reference information database related to welding quality judgment or skill quantification.

Furthermore, when the secondary feature quantity according to the object of judgment/evaluation is secondary feature quantity III, the reference information is the correlation between secondary feature quantity III and heat input. On the other hand, when the secondary feature quantity according to the object of judgment/evaluation is secondary feature quantity IV, the reference information is the correlation between secondary feature quantity IV and welding speed. The reference information regarding secondary feature quantity III or secondary feature quantity IV is each stored in a second reference information database related to the quantification of welding conditions.

Furthermore, when the secondary feature quantity corresponding to the object of judgment/evaluation is secondary feature quantity V, the reference information is the optimal value that results in an appropriate weld pool shape for secondary feature quantity V. The weld pool shape corresponds to the properties of back-beam formation. Similarly, when the secondary feature quantity corresponding to the object of judgment/evaluation is any of secondary feature quantities VI to XI, the reference information is the optimal value that results in an appropriate weld pool shape for that secondary feature quantity. The reference information for each of secondary feature quantities V to XI is stored in a third reference information database related to back-beam formation.

Furthermore, when the secondary feature corresponding to the object of judgment/evaluation is secondary feature XII, the reference information is a threshold for determining whether or not there is poor fusion for each number of passes for secondary feature XII. The reference information for secondary feature XII is stored in the first reference information database, similar to the reference information for secondary feature I or secondary feature II.

Next, the control and calculation unit 11 executes the judgment and evaluation step S107. In the judgment and evaluation step S107, the control and calculation unit 11 judges or evaluates the welding quality, skill, etc. in real time from the secondary feature quantities quantified in the secondary feature quantity identification step S105, using the reference information acquired in the reference information acquisition step S106 as a reference. The control and calculation unit 11 may derive a judgment and evaluation result based on any one of the multiple secondary feature quantities exemplified above, or may derive multiple judgment and evaluation results based on multiple secondary feature quantities. In this embodiment, the welding to be judged or evaluated is welding performed by an automatic welding machine, or manual welding performed by a worker, including semi-automatic welding. Therefore, examples of judgment and evaluation for each secondary feature quantity will be described below in relation to the welding to be judged or evaluated and the object of evaluation.

In determining or evaluating based on secondary feature I, in welding using an automatic welding machine, the presence or absence of poor fusion in terms of weld quality can be determined when the welding object 90 is a medium-sized plate. On the other hand, in manual welding, the presence or absence of poor fusion in terms of skill can be evaluated when the welding object 90 is a medium-sized plate.

Figure 10B is a graph showing the secondary feature value I versus heat input (J/cm). In the graph, plots indicated by solid circles correspond to the secondary feature value I when it is determined that there is no incomplete fusion in the weld, i.e., that there is no defect. On the other hand, plots indicated by open circles correspond to the secondary feature value I when it is determined that there is incomplete fusion in the weld.

As for the relationship between secondary feature value I and the object of judgment/evaluation, when secondary feature value I is small, the shape of the molten pool is close to a straight line and stable. On the other hand, when secondary feature value I is large, the leading edge point of the molten pool region and the trailing edge point of the molten pool region are misaligned in the Y direction, resulting in a meandering and unstable molten pool. Therefore, when the value of secondary feature value I is equal to or greater than a threshold value, which is reference information related to secondary feature value I, the control and calculation unit 11 judges and evaluates that there is a high possibility of insufficient fusion. In general welding operations, when the heat input is insufficient, welding is performed under low heat input conditions, and the molten pool is likely to have a meandering shape. In contrast, judgment or evaluation based on secondary feature value I allows the presence or absence of insufficient fusion to be judged or evaluated regardless of the amount of heat input.

For judgments or evaluations based on secondary feature II, the presence or absence of poor fusion can be judged or evaluated in both welding using an automatic welding machine and manual welding, in the same way as when based on secondary feature I.

Figure 11B is a graph showing secondary feature value II versus heat input (J/cm). In the graph, plots indicated by solid circles correspond to secondary feature value II when it is determined that there is no incomplete fusion in the weld, i.e., that there is no defect. On the other hand, plots indicated by open circles correspond to secondary feature value II when it is determined that there is incomplete fusion in the weld.

The relationship between secondary feature quantity II and the object of judgment/evaluation is that when secondary feature quantity II is small, the movement of the molten pool does not follow the movement of the wire 82, resulting in an unstable shape of the molten pool. On the other hand, when secondary feature quantity II is large, the front end point of the molten pool region is located directly below the wire 82, resulting in a stable shape of the molten pool. Therefore, when the value of secondary feature quantity II is equal to or greater than a threshold value, which is reference information related to secondary feature quantity II, the control and calculation unit 11 judges and evaluates that there is a high possibility of incomplete fusion. In general welding operations, if the rod operation is inappropriate, welding will be performed under low heat input conditions, making it difficult for the movement of the molten pool to follow the movement of the wire 82. In contrast, judgment or evaluation based on secondary feature quantity II allows the presence or absence of incomplete fusion to be judged or evaluated regardless of the amount of heat input.

In determining or evaluating based on secondary feature III, when welding is performed using an automatic welding machine, the impact on the material regarding welding quality when the welding target 90 is a medium-sized plate can be determined. On the other hand, when manual welding is performed when the welding target 90 is a medium-sized plate, the error from the allowable range of heat input regarding skill can be evaluated.

Figure 12B is a graph showing secondary feature III versus heat input (J/cm). In the graph, the predefined allowable range of heat input is shown as the area surrounded by a dashed line.

The relationship between secondary feature quantity III and the object of judgment/evaluation is that when secondary feature quantity III is small, the heat input is insufficient, whereas when secondary feature quantity III is large, the heat input is excessive. Therefore, the control and calculation unit 11 estimates the heat input by applying secondary feature quantity III calculated based on welding images captured during the welding work to the correlation, which is reference information related to secondary feature quantity III, and then determines and evaluates whether the estimated heat input is within the allowable range.

Here, the secondary feature quantity III is based on the welding image captured by the second camera 10b. The second camera 10b may be mounted on the welding worker himself. However, the secondary feature quantity III is the area of the circumscribing rectangle _RB related to the dimensions of the molten pool, and is not a dimensionless quantity. Therefore, for example, if the second camera 10b is mounted on the welding worker himself, when the distance between the welding worker and the molten zone changes, the dimensions of the molten pool at the angle of view of the second camera 10b also change, making it impossible to accurately estimate the heat input. Therefore, the control and calculation unit 11 may use a separately installed IMU (inertial measurement unit) sensor to measure the distance between the welding worker and the weld zone during welding, and convert the area of the circumscribing rectangle _RB into a value that can be determined and evaluated based on the distance measured by the IMU sensor.

In determining or evaluating based on the secondary feature IV, when welding is performed using an automatic welding machine, the impact on the material regarding welding quality when the welding target 90 is a medium-sized plate can be determined. On the other hand, when manual welding is performed when the welding target 90 is a medium-sized plate, the error from the allowable range of welding speed regarding skill can be evaluated.

Figures 13B and 13C are graphs showing the secondary feature value IV versus welding speed (cm/min). The secondary feature value IV in Figure 13B was obtained under first standard conditions where the welding current was 160 A and the welding voltage was 21.9 V. The secondary feature value IV in Figure 13C was obtained under second standard conditions where the welding current was 230 A and the welding voltage was 26.0 V. In each graph, the predetermined allowable range of welding speed is shown as the area surrounded by a dashed line.

The relationship between the secondary feature value IV and the object of judgment/evaluation is that under standard conditions of the same current and voltage, when the secondary feature value IV is small, the welding speed is fast, whereas when the secondary feature value IV is large, the welding speed is slow. Therefore, the control and calculation unit 11 estimates the welding speed by applying the secondary feature value IV calculated based on the welding image captured during the welding operation to the correlation, which is reference information related to the secondary feature value IV, and determines/evaluates whether the estimated welding speed is within the allowable range. Under the same current and voltage conditions, when the welding speed is fast, the length of the molten pool tends to be long, and therefore the secondary feature value IV becomes small. Therefore, in judgments or evaluations based on the secondary feature value IV, the welding speed can be estimated based on the welding image.

Figure 18 is a table showing the results of the evaluation based on secondary feature quantities V through VII when the welding target 90 is a thin plate and the evaluation target is the back-beam formation property. The numerical values for each secondary feature quantity in the table are obtained from 11 welding runs, each assigned a evaluation number. In other words, a total of three secondary feature quantities, secondary feature quantities V through VII, are identified for each welding run. The table also shows the welding current and welding speed for each welding run. Here, the evaluation target is visually classified into three evaluation patterns: "burn-through," "sound weld," or "insufficient back-beam" after welding. According to this classification method, if the evaluation target is classified as either "burn-through" or "insufficient back-beam," it is not a sound weld.

According to the judgment/evaluation results shown in Figure 18, the magnitude of each value of secondary feature quantity V to secondary feature quantity VII is thought to be related to the tendency of the judgment/evaluation target to be good or bad. Therefore, the control/calculation unit 11 can refer to the tendency of these secondary feature quantities relative to the judgment/evaluation target and perform judgment or evaluation based on secondary feature quantities V to secondary feature quantity VII as follows:

In determining or evaluating based on secondary feature quantities V to VII, when welding is performed using an automatic welding machine, it is possible to determine which of the three evaluation patterns above applies to the welding quality when the welding object 90 is a thin plate. On the other hand, when manual welding is performed, it is possible to evaluate which of the three evaluation patterns above applies to the skill when the welding object 90 is a thin plate.

The relationship between the secondary feature value V and the object of judgment/evaluation is that when the secondary feature value V is small, the length of the lower part of the groove in the molten pool, corresponding to the second length L2 in Figure 14, becomes long, making the state after welding more likely to be burn-through. On the other hand, when the secondary feature value V is large, the length of the lower part of the groove in the molten pool becomes short, making it more likely that the back ribs formed after welding will be insufficient. Therefore, the control and calculation unit 11 judges and evaluates that the welding is sounder the closer the value of the secondary feature value V is to the optimal value, which is the reference information for the secondary feature value V. The optimal value here is, for example, 1.3.

Furthermore, in terms of the relationship between the secondary feature value VI and the object of judgment/evaluation, when the secondary feature value VI is small, the width of the molten pool corresponding to the third length L3 in Figure 15 becomes narrower, making the state after welding more likely to result in burn-through. On the other hand, when the secondary feature value VI is large, the width of the molten pool becomes wider, making it more likely that the backside ribs formed after welding will be insufficient. Therefore, the control and calculation unit 11 judges and evaluates that the welding is sounder the closer the value of the secondary feature value VI is to the optimal value, which is the reference information for the secondary feature value VI. The optimal value here is, for example, 2.5.

Furthermore, with regard to the relationship between the secondary feature quantity VII and the object of judgment/evaluation, when the secondary feature quantity VII is small, the area of the circumscribing rectangle R _A2 in FIG. 16B based on the welding image captured from the front becomes large, and the state after welding is likely to be burn-through. On the other hand, when the secondary feature quantity VII is large, the area of the circumscribing rectangle R _A2 becomes small, and the back ribs formed after welding are likely to be insufficient. Therefore, the control/calculation unit 11 judges/evaluates that the welding is sound when the value of the secondary feature quantity VI is closer to the optimal value, which is the reference information for the secondary feature quantity VI.

The above-described judgment or evaluation based on secondary feature quantities V to VII can also be applied to judgment or evaluation based on secondary feature quantities VIII to XI, which are secondary feature quantities identified based on the welding image captured by the first camera 10a. In this case, too, in welding using an automatic welding machine, it can be determined whether the welding quality of the welding object 90, which is a thin plate, falls into one of the above three judgment patterns. On the other hand, in manual welding, it can be evaluated whether the skill of the welding object 90, which is a thin plate, falls into one of the above three judgment patterns.

The relationship between secondary feature value VIII and the object of judgment/evaluation is that when secondary feature value VIII is small, the left and right width of the weld pool is narrow, making the state after welding more likely to result in burn-through. On the other hand, when secondary feature value VIII is large, the left and right width of the weld pool is wide, making the backside ribs formed after welding more likely to be insufficient. Therefore, the control and calculation unit 11 judges and evaluates that the welding is sounder the closer the value of secondary feature value VIII is to the optimal value, which is the reference information for secondary feature value VIII. The optimal value here is, for example, 0.8.

Furthermore, in terms of the relationship between the secondary feature quantity IX and the object of judgment/evaluation, when the secondary feature quantity IX is small, the width of the upper left corner of the molten pool is narrow, making the state after welding more likely to result in burn-through. On the other hand, when the secondary feature quantity IX is large, the width of the upper left corner of the molten pool is wide, making it more likely that the backside ribs formed after welding will be insufficient. Therefore, the control and calculation unit 11 judges and evaluates that the welding is sounder the closer the value of the secondary feature quantity IX is to the optimal value, which is the reference information for the secondary feature quantity IX. The optimal value here is, for example, 0.9.

Furthermore, in terms of the relationship between the secondary feature quantity X and the object of judgment/evaluation, when the secondary feature quantity X is small, the left and right width of the molten pool becomes narrow, making the state after welding more likely to result in burn-through. On the other hand, when the secondary feature quantity X is large, the left and right width of the molten pool becomes wide, making it more likely that the backside ribs formed after welding will be insufficient. Therefore, the control calculation unit 11 judges and evaluates that the welding is sounder the closer the value of the secondary feature quantity X is to the optimal value, which is the reference information for the secondary feature quantity X. The optimal value here is, for example, 0.62.

Furthermore, in terms of the relationship between the secondary feature quantity XI and the object of judgment/evaluation, when the secondary feature quantity XI is small, the width at the top left of the molten pool is narrow, making the state after welding more likely to result in burn-through. On the other hand, when the secondary feature quantity XI is large, the width at the top left of the molten pool is wide, making it more likely that the backside ribs formed after welding will be insufficient. Therefore, the control and calculation unit 11 judges and evaluates that the welding is sounder the closer the value of the secondary feature quantity XI is to the optimal value, which is the reference information for the secondary feature quantity XI. The optimal value here is, for example, 1.2.

Figure 19 is a table showing the results of judgment and evaluation based on secondary feature XII when the welding object 90 is a thick plate. When welding thick plates, each layer is welded in multiple passes, a process known as allocation. Therefore, even under the same welding conditions, it is expected that insufficient fusion as a defect may or may not occur. Therefore, judgment or evaluation based on secondary feature XII focuses on left-right asymmetry in addition to welding phenomena. Figure 19 shows the presence or absence of defects and the value of secondary feature XII identified in the welding process for each pass. However, values for pass numbers from "1" to "3" are excluded from the table because there is no allocation for each pass. The presence or absence of defects is also determined visually for each pass.

According to the judgment/evaluation results shown in FIG. 19, the value of secondary feature XII in a pass determined to have a defect varies significantly from the value in a pass determined to have no defect. In other words, the magnitude of the value of secondary feature XII is thought to be related to the tendency of the judgment/evaluation object to be good or bad. Therefore, the control/calculation unit 11 can refer to the tendency of secondary feature XII for the judgment/evaluation object and perform judgment or evaluation based on secondary feature XII.

In determining or evaluating based on secondary feature XII, when welding is performed using an automatic welding machine, the presence or absence of poor fusion in terms of welding quality can be determined when the welding object 90 is a thick plate. On the other hand, when manual welding is performed, the presence or absence of poor fusion in terms of skill can be evaluated when the welding object 90 is a thick plate.

The relationship between secondary feature quantity XII and the object of judgment/evaluation is that when secondary feature quantity XII is small, the arc is downward and stable, and it is therefore assumed that heat has been appropriately input throughout the entire molten pool. On the other hand, when secondary feature quantity XII is large, the arc is deflecting significantly toward the groove wall, and it is therefore assumed that heat has not been input to the groove-shaped area created by the previous pass. Therefore, when the value of secondary feature quantity XII is equal to or greater than a threshold value, which is reference information regarding secondary feature quantity XII, the control and calculation unit 11 judges and evaluates that there is a high possibility of insufficient fusion occurring.

Next, the control and calculation unit 11 ends the welding operation (welding operation end S108), and then ends the execution of the welding operation determination process.

Next, we will explain the welding construction judgment method and the effects of the welding construction judgment device 1 according to this embodiment.

First, the welding performance evaluation method according to this embodiment includes a step of photographing a weld including a molten pool _HM during welding (welding image input step S200). The welding performance evaluation method also includes a step of identifying primary feature quantities related to the shape of at least one of the arc H _A or the molten pool _HM based on the welding image _IP obtained by photographing the weld (primary feature quantity identification step S104). The welding performance evaluation method also includes a step of identifying secondary feature quantities as dimensionless quantities from a calculation formula using the primary feature quantities as variables (secondary feature quantity identification step S105). The welding performance evaluation method also includes a step of evaluating the welding quality or the skill of the welding worker based on the secondary feature quantities (evaluation step S107).

On the other hand, the welding construction judgment device 1 according to this embodiment includes an imaging device 10 that captures an image of a welded portion including a molten pool during welding. The welding construction judgment device 1 also includes a memory unit 12 _that stores _a calculation formula for a secondary feature quantity as a dimensionless quantity in which a primary feature quantity related to the shape of at least one of the arc H A and the molten pool H M is a variable. The welding construction judgment device 1 also includes a control and calculation unit 11 that identifies the primary feature quantity based on a welding image _IP acquired from the imaging device 10 and identifies the secondary feature quantity based on the primary feature quantity and the calculation formula acquired from the memory unit 12. The control and calculation unit 11 further judges the welding quality or evaluates the skill of a welding worker based on the secondary feature quantity.

In the welding performance evaluation method and welding performance evaluation device 1 according to this embodiment, when judging weld quality or evaluating the skill of a welding worker, secondary features that use the primary features as variables are referenced rather than directly referencing primary features based on the welding image _IP . The secondary features are quantified as dimensionless quantities, and are therefore less susceptible to the influence of welding phenomena that may vary from welding performance to welding performance. Therefore, variations in the evaluation and assessment results for the same welding quality or skill are less likely to occur for each welding performance, improving the accuracy of the welding quality evaluation or skill evaluation.

In this way, this embodiment can provide a welding construction judgment method and welding construction judgment device 1 that improve the accuracy of judging welding quality or evaluating the skills of a welding worker.

When determining the welding quality of welding performed by an automatic welding machine such as welding machine 80, the welding quality is quantified using secondary features. Therefore, by directly reflecting the results of such welding quality determination in the automatic welding machine's operation conditions, the automatic welding machine can perform adaptive control in real time to correct deviations from the optimal value.

On the other hand, when evaluating a worker's manual welding skills, those skills are also quantified using secondary features. Therefore, the evaluation results for such skills can be more suitable as indicators for worker skill training or skill transfer.

Furthermore, the welding construction determination method according to this embodiment may include a step of extracting an arc region _AC corresponding to the arc _HA or a molten pool region _MP corresponding to the molten pool _HM from the welding image _IP as a mask image I _M. This step corresponds to the arc region extraction step S202 or the molten pool region extraction step S203. The primary feature amount may be identified from a plurality of feature points corresponding to the feature point _P1A in the above example, extracted based on the shape of the arc region _AC or the molten pool region _MP , or from the circumscribing rectangle _RA of the arc region _AC or the circumscribing rectangle _RB of the molten pool region _MP .

According to this welding construction judgment method, a plurality of feature points, a circumscribing rectangle _RA , and a circumscribing rectangle _RB are derived from the arc region _AC or the molten pool region _MP extracted as the mask image _IM . This can be advantageous in improving the accuracy of the position or shape and in facilitating the selection of desired feature points.

In the welding construction determination method according to this embodiment, the molten pool region _MP may be extracted by machine learning. The machine learning employed here may predict the molten pool endpoints by learning and determining correct molten pool endpoint positions from brightness values on multiple search lines of the molten pool extending from the center of gravity of the arc region _AC to the edges of the welding image _IP .

For example, if the weld pool region is extracted by focusing on the increase or decrease in brightness, it may not be possible to accurately extract the feature points due to the influence of the brightness of the fumes generated during welding or the darkness of positions far from the arc. In contrast, according to this embodiment, machine learning that statistically considers the characteristics of the data is applied when extracting the weld pool region _MP , thereby improving the accuracy of the extracted feature points.

Furthermore, in the welding construction determination method according to this embodiment, one of the plurality of feature points may correspond to the tip point of the wire 82 forming the arc H _A , which is extracted as a depression in the mask image I _M.

The feature point corresponding to the tip point of the wire 82 extracted as a depression in the mask image I _M is, for example, feature point _P3 illustrated in Figure 11 and other figures relating to secondary feature quantity II. Whether the movement of the molten pool H _M follows the movement of the wire 82 is related to whether the shape of the molten pool H _M is stable. This welding procedure evaluation method may be suitable, for example, when the welding object 90 is an intermediate plate and the object of evaluation is the presence or absence of incomplete fusion related to the ability of the molten pool to follow the wire 82.

In the welding construction determination method according to this embodiment, one of the plurality of characteristic points may correspond to the center of gravity of the arc region _AC .

The feature point corresponding to the center of gravity of the arc region A- _C is, for example, feature point _P3 illustrated in FIG. 17A and other figures for secondary feature quantity XII. Generally, when welding thick plates, each layer is divided into multiple passes. Since the movement of the center of gravity of the arc region A- _C is related to the arc swing amplitude, when the arc swings significantly toward the groove wall, it can be assumed that heat is not being input to the groove-shaped area formed by the previous pass. This welding performance evaluation method can be suitable, for example, when the welding target 90 is a thick plate and the evaluation target is the presence or absence of incomplete fusion.

In the welding construction judgment method according to the present embodiment, the secondary feature amount may be set for each welding construction object 90, or for each item focused on in judging welding quality or in evaluating skills, or for each of a plurality of welding images _IP obtained by photographing the weld from different photographing positions and photographing directions.

According to this welding construction judgment method, multiple secondary feature amounts can be set, such as secondary feature amounts I to XII exemplified above, depending on at least which welding construction object 90, judgment/evaluation object, or imaging device 10 that acquires welding image _IP is selected. The judgment/evaluation result can be derived based on any one of the multiple secondary feature amounts, or multiple judgment/evaluation results can be derived in a composite manner based on multiple secondary feature amounts. In particular, the reliability of the judgment/evaluation result can be improved by deriving multiple judgment/evaluation results in a composite manner based on multiple secondary feature amounts.

Furthermore, in the welding construction judgment method according to this embodiment, the primary feature value may be a straight line passing through two feature points, and the secondary feature value may be the slope of the line.

In this case, the primary feature and secondary feature correspond to, for example, the combination of secondary feature I shown in Figures 10A and 10B, or the combination of secondary feature II shown in Figures 11A and 11B. As explained above, this welding performance assessment method can be advantageous for determining or evaluating the presence or absence of poor fusion in relation to welding quality or skill when the welding performance object 90 is a middle plate.

Furthermore, in the welding construction judgment method according to this embodiment, the primary feature value may be a line segment connecting two feature points, and the secondary feature value may be the ratio between the length of one line segment formed by different combinations of two feature points and the length of another line segment.

In this case, the primary and secondary features correspond to, for example, the combination of secondary features IV shown in Figures 13A to 13C. As explained above, this welding performance assessment method can be useful when the welding target 90 is a medium plate, in welding performed by an automatic welding machine, for assessing the impact on the material regarding weld quality, or in manual welding, for evaluating the error from the allowable range of welding speeds related to the skill of the welding operator.

In this case, the primary feature and secondary feature correspond to, for example, the combination related to secondary feature V shown in FIG. 14 or the combination related to secondary feature VI shown in FIG. 15. As explained above, this welding procedure determination method can be advantageous for determining or evaluating whether the back-beam formation characteristics are classified as "burn-through," "sound weld," or "insufficient back-beam" when the welding procedure object 90 is a thin plate.

In this case, the primary feature amount and secondary feature amount correspond to, for example, the combination related to secondary feature amount XII shown in Figures 17A and 17B. As explained above, this welding performance evaluation method can be advantageous for determining or evaluating the presence or absence of poor fusion related to welding quality or skill when the welding performance object 90 is a thick plate.

Furthermore, in the welding construction determination method according to this embodiment, the primary feature amount may be the length of each side of the circumscribing rectangle R _A of the arc region A _C. In this case, the secondary feature amount may be the ratio between the area of the circumscribing rectangle R _A1 based on the welding image I _P captured from one shooting position and shooting direction and the area of the circumscribing rectangle R _A2 based on the welding image I _P captured from another shooting position and shooting direction.

In this case, the primary feature and secondary feature correspond to, for example, the combination of secondary feature VII associated with Figures 16A and 16B. As explained above, this welding procedure assessment method can be advantageous for determining or evaluating whether the back-beam formation characteristics are classified as "burn-through," "sound weld," or "insufficient back-beam" when the welding procedure target 90 is a thin plate.

Furthermore, in the welding construction judgment method according to this embodiment, welding quality or skill may be judged or evaluated based on predetermined reference information relating to the pass/fail of the welding quality judgment or skill evaluation, which is determined based on the numerical values of the secondary feature quantities.

This welding construction evaluation method allows for easy evaluation of welding quality or skill by comparing the secondary feature values identified in the welding construction with reference information that has been predefined based on the numerical values of the secondary feature values.

The welding construction judgment device according to this embodiment may also include an input unit 13 for inputting the welding construction object 90 and an item to be focused on in judging the welding quality or an item to be focused on in evaluating the skill. There may be a plurality of image capturing devices 10, each with a different image capturing position and image capturing direction. The control and calculation unit 11 may select an image capturing device 10 from the plurality of image capturing devices 10 to acquire the welding image _IP used to identify the primary feature. In this case, the image capturing device 10 may be selected for each welding construction object 90, or for each item to be focused on in judging the welding quality or for each item to be focused on in evaluating the skill, input via the input unit 13.

According to this welding construction judgment device 1, the control and calculation unit 11 acquires the welding image _IP from at least one of a plurality of image capture devices 10 each having a different image capture position and image capture direction. Therefore, the control and calculation unit 11 can use the welding image _IP captured by the appropriate image capture device 10 in accordance with the secondary feature amount to be adopted.

Several embodiments have been described, but modifications or variations of the embodiments can be made based on the above disclosure. All components of the above embodiments and all features described in the claims may be individually extracted and combined, as long as they are not mutually inconsistent.

1 Welding execution determination device 10 Photographing device 11 Control calculation section 12 Storage section 13 Input section A _C arc area H _A arc _HM molten pool _IM mask image I _P welding image L1 1st length L2 2nd length L3 3rd length L4 4th length L5 5th length L6 6th length M _P molten pool area P ₁ , P _1A , P _1B , P ₂ , P _2A , P _2B , P ₃ , P _3A , P _3B , P _4A , P _4B Feature points R _A circumscribed rectangle R _A1 circumscribed rectangle R _A2 circumscribed rectangle R _B circumscribed rectangle

Claims (9)

a step of photographing a weld including a molten pool during welding;
identifying a primary feature value relating to the shape of at least one of the arc and the molten pool based on a welding image obtained by photographing the weld;
specifying a secondary feature value as a dimensionless quantity from a calculation formula using the primary feature value as a variable;
a step of determining welding quality or evaluating the skill of a welding worker based on the secondary feature amount;
extracting an arc region corresponding to the arc or a molten pool region corresponding to the molten pool from the welding image as a mask image;
and
the primary feature amount is identified from a plurality of feature points, a circumscribing rectangle of the arc region, or a circumscribing rectangle of the weld pool region, which are extracted based on the shape of the arc region or the weld pool region;
A welding construction judgment method in which the molten pool region is extracted using machine learning that predicts the endpoints of the molten pool by learning and determining the correct molten pool endpoint positions from the brightness values on multiple search lines of the molten pool extending from the center of gravity of the arc region to the edge of the welding image. The welding construction judgment method according to claim 1 , wherein one of the plurality of feature points corresponds to a tip point of a wire forming the arc, the tip point being extracted as a depression in the mask image. The welding construction determination method according to claim 1 , wherein one of the plurality of characteristic points corresponds to a center of gravity of the arc region. 4. The welding construction judgment method according to claim 1, wherein the secondary feature amount is set for each welding construction object, for each item focused on in judging the welding quality or for each item focused on in evaluating the skill, or for each of the plurality of welding images obtained by photographing the welded part from different photographing positions and photographing directions. the primary feature is a straight line passing through the two feature points,
The welding construction judgment method according to claim 1 , wherein the secondary feature amount is a slope of the straight line. the primary feature is a line segment connecting two of the feature points,
4. The welding construction judgment method according to claim 1 , wherein the secondary feature amount is a ratio between a length of one of the line segments formed by different combinations of two of the feature points and a length of another of the line segments. the primary feature amount is the length of each side of the circumscribing rectangle of the arc region;
The welding construction judgment method according to any one of claims 1 to 3, wherein the secondary feature is a ratio between an area of the circumscribing rectangle based on the welding image photographed from one photographing position and photographing direction and an area of the circumscribing rectangle based on the welding image photographed from another photographing position and photographing direction. The welding execution judgment method according to any one of claims 1 to 7, wherein the welding quality or the skill is judged or evaluated based on reference information relating to whether the judgment of the welding quality or the evaluation of the skill is good or bad, the reference information being specified in advance based on the numerical values of the secondary feature amounts . an imaging device for imaging the welded portion including the molten pool during welding;
a storage unit that stores a calculation formula for a secondary feature value as a dimensionless quantity using a primary feature value relating to the shape of at least one of the arc and the molten pool as a variable;
a control and calculation unit that identifies the primary feature amount based on the welding image acquired from the imaging device, identifies the secondary feature amount based on the primary feature amount and the calculation formula acquired from the storage unit, and judges the welding quality or evaluates the skill of a welding worker based on the secondary feature amount;
an input unit for inputting a welding object and an item to be focused on in determining the welding quality or an item to be focused on in evaluating the skill;
Equipped with
The photographing device is provided in a plurality of positions and directions different from each other,
The control and calculation unit selects, from the plurality of photographing devices, the photographing device that acquires the welding image used to identify the primary feature for each of the welding work objects, or for each of the items that are focused on in determining the welding quality or for each of the items that are focused on in evaluating the skill, which are input via the input unit .