Learning device, defect determination device, learning method, defect determination method, welding control device, and welding device

Abstract

To provide a learning device, a defect determination device, a learning method, a defect determination method, a welding control device, and a welding device which accurately predict a defect size and can prevent occurrence of defects.SOLUTION: A learning device 140 includes: a data acquisition part 141 for acquiring information on a welding condition when a bead B is stacked, a dimension concerning a narrow part N forming a valley part in a surface shape of a laminated molding before the bead B is stacked, a positional relation between the narrow part N and a target position of the bead B, and a defect size of a non-welded defect C; and a learning part 142 for learning a relation among the welding condition, the dimension concerning the narrow part N, the positional relation, and the defect size to generate an estimation model, wherein the dimension concerning the narrow part N includes at least any one of a bottom width U of the valley part, an opening width W indicating an interval between apexes of both sides of the valley part forming the valley part, and a valley depth H from the apex to the bottom of the valley part.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a learning device, a defect determination device, a learning method, a defect determination method, a welding control device, and a welding device. [Background technology]

[0002] In arc welding, a technique is known for detecting defects that occur in a welded structure and judging whether the welding is performed properly. For example, Patent Document 1 discloses a technique for judging whether the welding condition is good or bad by capturing an image of the molten pool state through a filter with a visual sensor. This document also describes that the welding current, voltage, and wire feed speed are measured at the same time as capturing the image with the visual sensor. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2008-110388 A Summary of the Invention [Problem to be solved by the invention]

[0004] Incidentally, there is an inspection means using ultrasonic testing, for example, as a means for non-contact inspection of the quality of a molded object formed by laminating beads. However, in this ultrasonic testing, since a probe must be brought into contact with the surface of the molded object, it is difficult to apply the method unless the surface of the molded object has been cut, and there is also a risk that reflected ultrasonic waves cannot be smoothly detected for complex molded objects.

[0005] In addition, a non-contact inspection method using an X-ray CT scanner is also possible, but this is not necessarily realistic because the size of the object to be inspected is restricted by the device and the device itself is expensive.

[0006] SUMMARY OF THE PRESENT EMBODIMENTS An object of the present invention is to provide a learning device, a defect determination device, a learning method, a defect determination method, a welding control device, and a welding device that are capable of accurately predicting defect size and preventing the occurrence of defects. [Means for solving the problem]

[0007] The present invention comprises the following configurations. (1) A learning device that learns the defect size of a non-weld defect that occurs inside an additive manufacturing body in which a plurality of beads are layered on a base material, and generates an estimation model that outputs the defect size according to input information, an information acquisition unit that acquires information regarding the welding conditions when the beads are stacked, the dimensions of a narrow portion that forms a valley in the surface shape of the layered product before the beads are stacked, the positional relationship between the narrow portion and a target position of the bead, and the defect size of the unwelded defect; a learning unit that learns the welding conditions, the dimensions of the narrow portion, and the positional relationship, and the relationship with the defect size, to generate the estimation model; The dimensions of the narrow portion include at least one of a bottom width of the valley portion, an opening width representing a distance between apexes of both sides of the valley portion forming the valley portion, and a valley depth from the apex to the bottom of the valley portion. Learning device. (2) A learning device according to (1), a determination unit that inputs information on a welding plan including at least dimensional information and the positional relationship related to the narrow portion into the estimation model, and compares an estimated value of the defect size of the non-welding defect output from the estimation model with a reference value that is a predetermined allowable limit; Equipped with Defect determination device. (3) A learning method for learning a defect size of a non-weld defect occurring inside an additive manufacturing body in which a plurality of beads are layered on a base material, and generating an estimation model that outputs the defect size according to input information, the method comprising the steps of: acquiring information on the welding conditions when the beads are stacked, the dimensions of a narrow portion that forms a valley in the surface shape of the layered product before the beads are stacked, the positional relationship between the narrow portion and a target position of the bead, and the defect size of the unwelded defect; generating the estimation model by learning the welding conditions, the dimensions of the narrow portion, and the positional relationship, and the relationship between the defect size; having The dimensions of the narrow portion include at least one of a bottom width of the valley portion, an opening width representing a distance between apexes of both sides of the valley portion forming the valley portion, and a valley depth from the apex to the bottom of the valley portion. How to learn. (4) Inputting welding plan information including at least dimensional information and the positional relationship regarding the narrow portion into the estimation model generated by the learning method described in (3), comparing an estimated value of the defect size of the non-weld defect output from the estimation model with a reference value that is a predetermined allowable limit, and determining that the non-weld defect will occur if the estimated value exceeds the reference value. Defect determination method. (5) A defect determination device according to (2), A welding control device comprising a control unit that performs arc welding in accordance with the result output by the defect determination device. (6) The welding control device according to (5), A welding robot that performs arc welding; Equipped with Welding equipment. Effect of the Invention

[0008] According to the present invention, it is possible to predict the defect size with high accuracy and prevent the occurrence of defects. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is an overall configuration diagram of a welding system. [Diagram 2] FIG. 2 is a schematic diagram showing a welding torch, a shape detection unit, and a formation trajectory of a bead. [Diagram 3] FIG. 3 is a schematic functional block diagram of the defect determination device. [Figure 4] FIG. 4 is an explanatory diagram showing how a bead is formed in a narrow portion by a welding torch. [Diagram 5] FIG. 5 is an explanatory diagram for explaining a feature amount in a cross section perpendicular to the bead forming direction of an existing bead. [Figure 6] FIG. 6 is an explanatory diagram showing a state in which the interval between the existing beads shown in FIG. 5 is changed. [Figure 7] FIG. 7 is an explanatory diagram showing a state in which the interval between the existing beads shown in FIG. 5 is changed. [Figure 8] FIG. 8 is a cross-sectional view of a laminated specimen in which a defect has been formed. [Figure 9] FIG. 9 is a cross-sectional view of a laminated specimen in which a defect has been formed. [Figure 10] FIG. 10 is a flowchart showing a procedure for determining welding conditions. [Figure 11] FIG. 11 is a graph showing the relationship between the actual measured value and the predicted value of the defect size. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, configuration examples of the present invention will be described in detail with reference to the drawings. Here, additive manufacturing in which beads are stacked to form an additively shaped object will be described as an example, but the present invention can also be applied to general welding such as fillet welding and butt welding.

[0011] <Welding system configuration> FIG. 1 is an overall configuration diagram of a welding system. Welding system 100 includes welding device 110 and welding control device 120. Welding control device 120 includes control unit 11 and a defect determination device .

[0012] (Welding equipment) First, the configuration of the welding device 110 will be described. Welding device 110 includes welding robot 13, robot driving unit 15, filler metal supply unit 17, welding power supply unit 19, and shape detection unit 21. Welding robot 13, robot driving unit 15, filler metal supply unit 17, welding power supply unit 19, and shape detection unit 21 are each connected to control unit 11 of welding control device 120.

[0013] Welding robot 13 is an articulated robot, and a welding torch 27 is attached to the tip shaft of the robot. Robot driver 15 outputs a command to drive welding robot 13, and arbitrarily sets the position and posture of welding torch 27 three-dimensionally within the range of the degree of freedom of the robot arm. In addition, a filler material (welding wire) M that is continuously supplied is supported at the tip of welding torch 27.

[0014] The welding torch 27 has a shield nozzle (not shown) and is a torch for gas metal arc welding to which shielding gas is supplied from the shield nozzle. The arc welding method may be either a consumable electrode type such as shielded metal arc welding or carbon dioxide gas arc welding, or a non-consumable electrode type such as TIG welding or plasma arc welding, and is appropriately selected depending on the object (structure) to be produced. For example, in the case of a consumable electrode type, a contact tip is disposed inside the shield nozzle, and a filler material M to which a melting current is supplied is held by the contact tip. The welding torch 27 holds the filler material M and generates an arc from the tip of the filler material M in a shielding gas atmosphere.

[0015] The filler metal supply unit 17 includes a reel 17a around which the filler metal M is wound. The filler metal M is sent from the filler metal supply unit 17 to a payout mechanism (not shown) attached to a robot arm or the like, and is fed to the welding torch 27 while being fed forward and backward by the payout mechanism as necessary.

[0016] Any commercially available welding wire can be used as the filler metal M. For example, welding wires specified in MAG and MIG welding solid wires for mild steel, high tensile steel and low temperature steel (JIS Z 3312), arc welding flux-cored wires for mild steel, high tensile steel and low temperature steel (JIS Z 3313), etc. can be used. Furthermore, filler metals M such as aluminum, aluminum alloys, nickel, nickel-based alloys, etc. can be used depending on the required characteristics.

[0017] Welding power supply unit 19 supplies welding current and welding voltage to welding torch 27 for generating an arc from the tip of the torch.

[0018] Shape detection unit 21 is provided on or near the tip shaft of welding robot 13, and the measurement area is the vicinity of the tip of welding torch 27. Shape detection unit 21 may be another detection means provided at a position separate from welding torch 27.

[0019] Shape detection unit 21 of this configuration is moved together with welding torch 27 by the driving of welding robot 13, and measures the shape of bead B and the portion that will be the base when forming bead B. For example, a laser sensor that acquires reflected light of an irradiated laser light as height data can be used as this shape detection unit 21. Also, other detection means such as a camera for three-dimensional shape measurement may be used as shape detection unit 21.

[0020] According to the welding device 110 having the above-mentioned configuration, a modeling program corresponding to a model to be produced is transmitted from the control unit 11 to the robot driving unit 15. The modeling program is composed of a large number of command codes, and is created based on an appropriate algorithm according to various conditions such as the shape data (CAD data, etc.) of the model, the material, and the amount of heat input of the model.

[0021] The robot driving unit 15 executes the received molding program, drives the welding robot 13, the filler material supply unit 17, the welding power supply unit 19, etc., and forms the bead B according to the molding program. That is, the robot driving unit 15 drives the welding robot 13 to move the welding torch 27 along the trajectory of the welding torch 27 (bead formation trajectory) set in the molding program. At the same time, the filler material supply unit 17 and the welding power supply unit 19 are driven according to the set welding conditions, and the filler material M at the tip of the welding torch 27 is melted and solidified by the arc. As a result, the bead B is formed along the trajectory of the welding torch 27 on the base plate P, which is the base material. The beads B are formed adjacent to each other to form a bead layer consisting of a plurality of beads B. The next bead layer is layered on this bead layer, and a desired three-dimensional object WK is formed.

[0022] Although not shown, welding control device 120 is configured with a computer device including a processor such as a CPU, memories such as a ROM and a RAM, and storage units such as a HD (hard disk drive) and an SSD (solid state drive). Each of the components of welding control device 120 described above operates according to commands from the CPU to perform each function. Welding control device 120 may be configured to be disposed separately from welding device 110 and connected to welding device 110 from a remote location via a communication means such as a network.

[0023] The control unit 11 constituting the welding control device 120 has a function of collectively controlling the robot driving unit 15, the filler metal supply unit 17, the welding power source unit 19, and the shape detection unit 21 shown in Fig. 1. The control unit 11 executes a driving program prepared in advance or a driving program created under desired conditions to drive each unit such as the welding robot 13. As a result, the welding torch 27 is moved according to the driving program, and multiple layers of beads B are stacked on the base plate P based on the created welding plan, thereby forming a multi-layered object WK.

[0024] FIG. 2 is a schematic diagram showing welding torch 27, shape detection unit 21, and the bead forming trajectory of bead B. As shown in FIG. Bead B is formed sequentially by moving welding torch 27 along a bead formation trajectory created in advance on base plate P. At the same time as moving welding torch 27, shape detection unit 21 measures the surface shapes of existing bead B and bead formation surface G. Then, shape detection unit 21 outputs the surface shapes of bead B and bead formation surface G (collectively referred to as a shape profile) to welding control device 120.

[0025] It is preferable to measure the shape profile at the same time as the bead B is formed. In that case, the shape detection unit 21 may be disposed behind the movement direction of the welding torch 27. This allows the shape of the formed bead B to be efficiently measured along the movement path while the bead B is being formed by the movement of the welding torch 27, thereby shortening the takt time. This shape profile measurement may be performed at a timing other than the formation of the bead B, or may be performed at a desired timing according to various conditions. Hereinafter, a bead to be formed in the future is also referred to as a "new bead", and a bead that has already been formed is also referred to as an "existing bead".

[0026] The shaped object WK shown here has a frame-shaped wall portion Aw formed by beads B, and a filling portion Af that fills the area surrounded by the wall portion Aw with beads B. This filling portion Af is formed after the wall portion Aw is formed. That is, after the wall portion Aw is formed, beads B that will become the filling portion Af are formed inside the wall portion Aw along bead forming trajectories F1 to F3 shown by dotted lines. Then, beads B are formed along a bead forming trajectory F4. The order of forming the beads B in the filling portion Af is arbitrary.

[0027] FIG. 3 is a schematic functional block diagram of the defect determination device 130. As shown in FIG. Defect determination device 130 constituting welding control device 120 includes a learning device 140, a determination unit 151, and a welding plan correction unit 152. Learning device 140 has a data acquisition unit (information acquisition unit) 141, a learning unit 142, and an estimation model unit (estimation model) 143. Determination unit 151 and welding plan correction unit 152 are each connected to estimation model unit 143 of learning device 140.

[0028] The learning device 140 is a device that learns the defect size of an unwelded defect occurring inside an additive manufacturing body in which a plurality of beads B are layered on a base plate P, and generates an estimation model that outputs a defect size according to the input information. The data acquisition unit 141 acquires information on the welding conditions when stacking the beads B, the dimensions of a narrow portion that forms a valley in the surface shape of the additive manufacturing body before stacking the beads B, the positional relationship between the narrow portion and the target position of the bead B, and the defect size of the unwelded defect. Note that the positional relationship between the narrow portion and the target position of the bead B may be, for example, a differential distance between a representative position such as the center of the narrow portion and the target position of the bead B to be formed next, but is not limited to this. The learning unit 142 learns the relationship between the welding conditions, the dimensions and positional relationship of the narrow portion, and the defect size, and generates an estimation model. The estimation model unit 143 registers the estimation model generated by the learning unit 142.

[0029] The determination unit 151 inputs information on the welding plan, including dimensional information and positional relationship regarding the narrow portion, to the estimation model registered in the estimation model unit 143. Then, the determination unit 151 compares the estimated value of the defect size of the non-welding defect output from the estimation model unit 143 with a reference value that is a predetermined allowable limit.

[0030] When the determination unit 151 determines that the estimated value of the defect size exceeds the reference value, the welding plan modification unit 152 modifies at least one of the welding conditions and the positional relationship to create a modified welding plan. Then, the modified welding plan is transmitted to the estimation model unit 143.

[0031] <Generating learning data> Next, the process of generating training data will be described. (Preparing training data) FIG. 4 is an explanatory diagram showing how a bead B is formed in a narrow portion N by a welding torch 27. As shown in FIG. First, learning data to be acquired by the data acquisition unit 141 of the learning device 140 is prepared. Specifically, as shown in FIG. 4, the following learning data are prepared: welding conditions for the bead B filling the narrow portion N, feature quantities of the shape of the narrow portion N of the bead B in the previous layer, and the horizontal distance δ between the welding torch 27 and the narrow portion N. Examples of the welding conditions for the bead B filling the narrow portion N include the welding voltage and welding current set when stacking the bead B, the welding speed and the feed speed of the filler metal M, the torch angle α which is the inclination angle of the welding torch 27, the volume of the bead B, and the cross-sectional area of ​​the cross section perpendicular to the longitudinal direction of the bead B. The torch angle α here is the inclination angle inclined in a direction perpendicular to the longitudinal direction in the cross section perpendicular to the longitudinal direction of the bead B, but may be the inclination angle inclined in the longitudinal direction (torch advance angle, torch retreat angle) in the cross section parallel to the longitudinal direction of the bead B.

[0032] FIG. 5 is an explanatory diagram showing examples of feature amounts W, U, and H in a cross section perpendicular to the bead formation direction of an existing bead B. The narrow portion N of the bead B in the previous layer refers to a portion formed in a valley shape between the base plate or existing beads in the lower layer, and at least three feature amounts, namely, the bead spacing W, the bottom spacing U, and the average depth H, are extracted as feature amounts of the shape of the narrow portion N. In other words, when a pair of adjacent existing beads B are formed on the base surface FL representing the surface of the base plate or existing bead in the lower layer, in addition to the bottom spacing U and bead spacing W described above, the average depth H from the base surface FL to the top Pt of each bead B is used as a feature amount. The average depth H corresponds to the valley depth to the bottom of the valley formed by the pair of existing beads B in the stacking direction.

[0033] 6 and 7 are explanatory diagrams showing a state in which the intervals of the existing beads B shown in FIG. 5 are changed. As shown in Fig. 6, when the existing beads B approach each other to the point where they touch each other, the bottom spacing U is 0, and the average depth H is the valley depth between the beads, which is represented by the triangle indicated by the dotted line. Furthermore, as shown in Fig. 7, when the existing beads B overlap each other, the bottom spacing U is 0, and the average depth H is shallower than in the cases shown in Fig. 5 and Fig. 6. In this way, by including a combination of the bottom spacing U, bead spacing W, and average depth H as the feature amount, the shape of the valley can be easily specified.

[0034] These feature quantities W, U, and H may be calculated by applying a model function that simulates the shape of the bead to a shape profile obtained by actual measurement. The sensor that measures the shape of the narrow portion N is preferably a non-contact type like the shape detection unit 21 of this example, and more preferably, is attached near the welding torch 27 and measures while scanning the surface of the bead B.

[0035] The horizontal distance δ between the welding torch 27 and the narrow portion N is calculated as the difference distance between the representative position of the narrow portion N and the center position where the bead B is to be deposited next. The representative position of the narrow portion N may be, for example, the position where the depth is minimum or the midpoint of the portion evaluated by the feature amount U.

[0036] 8 and 9 are cross-sectional views of a laminated specimen in which defect C is formed. The data on the defect size can be measured by directly observing the cut surface after cutting the bead stack specimen shown in Figures 8 and 9. The data on the defect size can also be data obtained by observing the bead stack specimen with a CT device to determine the defect size.

[0037] Here, when a bead is formed, a recess is likely to be formed at the base of the side edge of the bead, and foreign matter is likely to accumulate at the base of the side edge. For this reason, as shown in Fig. 8, when a bead B2 is formed adjacent to a side edge of a bead B1 so as to overlap the other bead B2, a small gap may be generated due to insufficient contact of the arc with the base of the side edge of the bead B1, resulting in a defect (unwelded defect) C. Also, as shown in Fig. 9, when a pair of beads B1 and B2 are formed with a gap, even if a bead B3 is formed to fill the valley between the beads B1 and B2, an unwelded defect C may be generated in the recess at the base of the side edge of the existing beads B1 and B2 because the arc does not contact the bottom sufficiently.

[0038] The defect size is a size including the cross-sectional area or length in a cross section perpendicular to the longitudinal direction of the bead B, and includes indicators such as the diameter when the shape of the defect C is approximated by a perfect circle, the cross-sectional area of ​​the approximated circle, or the area of ​​the observed defect C, and the long axis length and short axis length when the shape of the defect is approximated by an ellipse.

[0039] (Generating an Estimation Model) When the prepared learning data is input to data acquisition unit 141 of learning device 140, learning unit 142 generates an estimation model consisting of a relationship between the learning data and defect size based on the learning data. The estimation model generated by learning unit 142 is transmitted to estimation model unit 143 and registered in this estimation model unit 143. Examples of means for generating the estimation model in learning unit 142 include well-known means such as decision tree, linear regression, random forest, support vector machine, Gaussian process regression, and neural network. At this time, in addition to the defect size, the probability of occurrence of defects of a specific size, the occurrence density of defects occurring per specific region (area or volume), and the like may also be learned.

[0040] <Determining welding conditions> Next, the process of determining the welding conditions in welding control device 120 will be described. FIG. 10 is a flowchart showing a procedure for determining welding conditions.

[0041] By moving welding torch 27 along a bead formation track created in advance, the surface shapes of existing bead B and surface G on which the bead is to be formed are measured by shape detection unit 21 arranged in parallel with welding torch 27. In this way, the shape of narrow portion N on surface G on which the bead is to be formed is measured (S1).

[0042] Based on the measured shape of the narrow portion N on the bead formation planned surface G, the feature quantities W, U, and H related to the narrow portion N are calculated (S2). The feature quantities W, U, and H may be calculated by fitting a model function that simulates the narrow portion N prepared in advance. Note that the shape profile obtained by measurement may be subjected to a smoothing process or the like before fitting with the model function.

[0043] The calculated feature amounts W, U, and H, the welding conditions for the bead B to be formed on the bead formation surface G, and information regarding the target position of the welding torch 27 are input to the estimation model unit 143 to obtain an estimate of the defect size (S3).

[0044] The determination unit 151 compares the estimated value of the defect size obtained by the estimation model unit 143 with a preset allowable value, and determines whether the estimated value is equal to or smaller than the allowable value (S4). This allowable value is a value of an allowable limit size, which is an allowable size of a defect. Note that the allowable limit size may be set for each object to be modeled or each material to be used for modeling.

[0045] When the determination by determination unit 151 indicates that the estimated value of the defect size exceeds the allowable value (S4: No), welding plan correction unit 152 searches for a condition that suppresses the defect size to the allowable limit size or less (S5). For example, welding plan correction unit 152 corrects a welding plan such as the welding conditions of bead B to be formed on planned bead formation surface G and the target position of welding torch 27, and transmits the corrected welding plan to estimation model unit 143. As a result, an estimate of the defect size is obtained again (S3), and determination unit 151 compares the re-obtained estimate of the defect size with the allowable value and makes a determination. Then, the correction of the welding plan by welding plan correction unit 152, the estimation of the defect size by estimation model unit 143, and the determination by determination unit 151 are repeated.

[0046] When the estimated value of the defect size is determined by determination unit 151 to be equal to or smaller than the allowable value (S4: Yes), the welding conditions for bead B to be formed on planned bead formation surface G, the target position of welding torch 27, and the like, which were input to estimation model unit 143 to obtain the estimated value, are determined as a welding plan (S6). After that, bead B is formed on planned bead formation surface G according to this welding plan.

[0047] According to the present welding system 100 described above, the relationship between the welding conditions, the dimensions and positional relationship of the narrow portion, and the defect size is learned to generate an estimation model. At this time, as the dimensions of the narrow portion N, at least one of the bottom interval (bottom width) U of the valley portion of the narrow portion N serving as the base, the opening width (bead interval) W representing the interval between the apexes on both sides of the valley portion forming the valley portion, and the valley depth H from the apex to the bottom of the valley portion is used. In this way, by performing machine learning based on the shape of the narrow portion N serving as the base and the conditions of the bead B to be layered thereon, the defect size can be predicted with high accuracy even with a relatively small amount of data.

[0048] Furthermore, the defect size can be predicted with high accuracy by generating an estimation model using welding conditions including at least one of the feed speed of the filler metal M, the welding speed, the welding current, the welding voltage, the torch angle α of the welding torch 27, the volume of the bead B, and the cross-sectional area of ​​the cross section perpendicular to the longitudinal direction of the bead B. In particular, by learning an index related to the volume or heat input of the bead B, the relationship as to whether or not the narrow portion N of the previous layer can be filled can be incorporated into the machine learning.

[0049] Moreover, by comparing the estimated defect size of defect C with a reference value that is a predetermined tolerance limit, even if the occurrence of defect C is predicted, it is possible to determine whether the estimated defect C is a harmless defect or a harmful defect, thereby preventing unnecessary defect responses.

[0050] Then, by repeating the correction of the welding plan, the calculation of the estimated value, and the judgment of the estimated value, it is possible to extract welding conditions and a target position of welding torch 27 suitable for suppressing defect C.

[0051] This allows arc welding to be performed under welding conditions and with a target position for the welding torch that are suitable for suppressing defect C, making it possible to form a molded object while suppressing defects C, such as unwelded defects in narrow areas N.

[0052] In the learning device 140 having the above-described configuration, the learning unit 142 may register the variance of the defect size in addition to the defect size in the estimation model unit 143, and the estimation model unit 143 may output the defect size and its variance value corresponding to the input information in accordance with the input information.

[0053] Here, Fig. 11 is a graph showing the relationship between the actual measurement value and the predicted value of the defect size. In Fig. 11, the defect size is the diameter when the shape of the defect is approximated by a perfect circle. The predicted value of the defect size (predicted defect equivalent diameter) varies to a certain extent from the actual measurement value of the defect size (actual defect equivalent diameter), and this variation decreases when the depth of the narrow portion N is in the range of 2 mm to 4 mm. In other words, the defect size can be predicted with high accuracy, particularly when the depth of the narrow portion N is 2 mm to 4 mm.

[0054] The learning unit 142 registers the variance of the defect size in addition to the defect size in the estimation model unit 143, and the estimation model unit 143 is capable of outputting the defect size and its variance value, thereby making it possible to predict defect C taking into account the variation that occurs depending on the defect size.

[0055] As such, the present invention is not limited to the above-described embodiments, and the present invention also contemplates mutual combinations of the various components of the embodiments, as well as modifications and applications by those skilled in the art based on the descriptions in the specification and well-known techniques, and these are included in the scope of the protection sought.

[0056] As described above, the present specification discloses the following: (1) A learning device that learns the defect size of a non-weld defect that occurs inside an additive manufacturing body in which a plurality of beads are layered on a base material, and generates an estimation model that outputs the defect size according to input information, an information acquisition unit that acquires information regarding the welding conditions when the beads are stacked, the dimensions of a narrow portion that forms a valley in the surface shape of the layered product before the beads are stacked, the positional relationship between the narrow portion and a target position of the bead, and the defect size of the unwelded defect; a learning unit that learns the welding conditions, the dimensions of the narrow portion, and the positional relationship, and the relationship with the defect size, to generate the estimation model; A learning device, wherein the dimensions related to the narrow portion include at least one of the bottom width of the valley portion, the opening width representing the distance between the peaks on both sides of the valley portion that form the valley portion, and the valley depth from the peak to the bottom of the valley portion. According to this learning device, the relationship between the welding conditions, the dimensions and positional relationship of the narrow portion, and the defect size is learned to generate an estimation model. At this time, as the dimensions of the narrow portion, at least one of the bottom width of the valley of the narrow portion that serves as the base, the opening width representing the distance between the apexes on both sides of the valley that form the valley, and the valley depth from the apex to the bottom of the valley is used. In this way, by performing machine learning based on the shape of the narrow portion that serves as the base and the conditions of the bead to be layered on it, the defect size can be predicted with high accuracy even with a relatively small amount of data.

[0057] (2) The learning device described in (1), wherein the welding conditions include at least one of a welding wire feed speed, a welding speed, a welding current, a welding voltage, a torch angle of a welding torch, a volume of the bead, and a cross-sectional area of ​​a cross section perpendicular to the longitudinal direction of the bead. According to this learning device, the defect size can be predicted with high accuracy by generating an estimation model using welding conditions including at least one of the welding wire feed speed, welding speed, welding current, welding voltage, torch angle of the welding torch, bead volume, and cross-sectional area of ​​the cross section perpendicular to the longitudinal direction of the bead. In particular, by learning an index related to the bead volume or heat input, the relationship between whether or not the narrow part of the previous layer can be filled can be incorporated into the machine learning.

[0058] (3) The learning device according to (1) or (2), wherein the positional relationship includes a differential distance between a representative position of the narrow portion and a target position of the bead to be formed next. According to this learning device, a highly accurate estimation model can be generated by using a positional relationship including a differential distance between a representative position of a narrow portion and a target position of the bead to be formed next.

[0059] (4) The learning device according to any one of (1) to (3), wherein the defect size includes a cross-sectional area or length in a cross section perpendicular to the longitudinal direction of the bead. According to this learning device, a highly accurate estimation model of the defect size can be generated using the defect size including the cross-sectional area or length in a cross section perpendicular to the longitudinal direction of the bead.

[0060] (5) the learning unit registers in the estimation model a variance of the defect size in addition to the defect size with respect to the welding conditions, the dimensions of the narrow portion, and the positional relationship; The learning device according to any one of (1) to (4), wherein the estimation model is capable of outputting, in response to input information, the defect size and its variance value corresponding to the information. According to this learning device, by outputting the variance value together with the defect size, it is possible to judge the reliability of the estimation, and based on that reliability, it is possible to determine a range of the estimated value that is practically reliable.

[0061] (6) A learning device according to any one of (1) to (5), a determination unit that inputs information on a welding plan including at least dimensional information and the positional relationship related to the narrow portion into the estimation model, and compares an estimated value of the defect size of the non-welding defect output from the estimation model with a reference value that is a predetermined allowable limit; A defect determination device comprising: This defect judgment device is equipped with a judgment unit that compares an estimated value of the defect size of an unwelded defect with a predetermined reference value that is the tolerance limit, so that even if a defect is predicted to occur, it can determine whether the estimated defect is a harmless defect or a harmful defect, thereby preventing unnecessary defect responses.

[0062] (7) A welding plan correction unit that corrects at least one of the welding conditions and the positional relationship to create a corrected welding plan when the determination unit determines that the reference value is exceeded, The defect determination device according to (6), wherein the welding plan correction unit repeats correction of the welding plan until an estimated value of the defect size according to the information of the corrected welding plan output by the estimation model becomes equal to or smaller than the reference value. According to this defect determination device, by repeatedly modifying the welding plan, calculating the estimated value, and determining the estimated value, it is possible to extract welding conditions and a target position for the welding torch suitable for suppressing unwelded defects.

[0063] (8) A learning method for learning a defect size of a non-weld defect occurring inside an additive manufacturing body in which a plurality of beads are layered on a base material, and generating an estimation model that outputs the defect size according to input information, the method comprising the steps of: acquiring information on the welding conditions when the beads are stacked, the dimensions of a narrow portion that forms a valley in the surface shape of the layered product before the beads are stacked, the positional relationship between the narrow portion and a target position of the bead, and the defect size of the unwelded defect; generating the estimation model by learning the welding conditions, the dimensions of the narrow portion, and the relationship between the positional relationship and the defect size; having A learning method in which the dimensions related to the narrow portion include at least one of a bottom width of the valley portion, an opening width representing the distance between the peaks on both sides of the valley portion that form the valley portion, and a valley depth from the peak to the bottom of the valley portion. According to this learning method, the relationship between the welding conditions, the dimensions and positional relationship of the narrow portion, and the defect size is learned to generate an estimation model. At this time, as the dimensions of the narrow portion, at least one of the bottom width of the valley of the narrow portion that serves as the base, the opening width representing the distance between the apexes on both sides of the valley that form the valley, and the valley depth from the apex to the bottom of the valley is used. In this way, by performing machine learning based on the shape of the narrow portion that serves as the base and the conditions of the bead to be layered on it, the defect size can be predicted with high accuracy even with a relatively small amount of data.

[0064] (9) A defect determination method comprising: inputting welding plan information including at least dimensional information and the positional relationship regarding the narrow portion into the estimation model generated by the learning method described in (8); comparing an estimated value of the defect size of the unweld defect output from the estimation model with a reference value which is a predetermined allowable limit; and determining that the unweld defect will occur if the estimated value exceeds the reference value. According to this defect determination method, by comparing an estimated value of the defect size of a non-weld defect with a predetermined reference value that is the allowable limit, even if a defect is predicted to occur, it is possible to determine whether the estimated defect is a harmless defect or a harmful defect, thereby preventing unnecessary defect responses.

[0065] (10) when it is determined that the reference value is exceeded, modifying at least one of the welding conditions and the positional relationship to create a modified welding plan; inputting the modified welding plan into the estimation model; The defect determination method according to (9), further comprising repeating the correction of the welding plan until the estimated value of the defect size according to the information of the corrected welding plan output by the estimation model becomes equal to or smaller than the reference value. According to this defect determination method, by repeatedly modifying the welding plan, calculating the estimated value, and determining the estimated value, it is possible to extract welding conditions and a target position of the welding torch suitable for suppressing unwelded defects.

[0066] (11) A defect determination device according to (6) or (7), A welding control device comprising a control unit that performs arc welding in accordance with the result output by the defect determination device. According to this welding control device, arc welding can be performed under welding conditions and with a target position for the welding torch that are suitable for suppressing non-welding defects.

[0067] (12) The welding control device according to (11), A welding robot that performs arc welding; A welding device comprising: This welding device makes it possible to manufacture a structure while suppressing defects such as unwelded areas in narrow areas. [Explanation of symbols]

[0068] 11 Control section 13 Welding robot 27 Welding Torch 100 Welding System 110 Welding equipment 120 Welding control device 130 Defect Judgment Device 140 Learning Device 141 Data Acquisition Unit (Information Acquisition Unit) 142 Learning Department 143 Estimation Model Section (Estimation Model) 151 Judgment section 152 Welding Plan Modification Section B, B1, B2, B3 Bead C defect (unwelded defect) H Valley Depth M Filler metal (welding wire) N Narrow area P Base plate (base material) U bottom width W Opening width α Torch angle δ horizontal distance (differential distance)

Claims

1. A learning device that learns the defect size of unwelded defects occurring inside a laminated structure in which a plurality of beads are laminated in layers on a base material and generates an estimation model that outputs the defect size according to the input information, an information acquisition unit that acquires information regarding welding conditions when laminating the beads, dimensions regarding narrow portions that form valleys among surface shapes of the laminated structure before laminating the beads, a positional relationship between the narrow portions and target positions of the beads, and the defect size of the unwelded defects; a learning unit that learns the relationship between the welding conditions, the dimensions regarding the narrow portions, and the positional relationship and the defect size to generate the estimation model; wherein the dimensions regarding the narrow portions include at least any one of a bottom width of the valleys, an opening width representing an interval between tops on both sides of the valleys forming the valleys, and a valley depth from the tops to the bottom of the valleys; a learning device.

2. The welding conditions include at least any one of a feeding speed of a welding wire, a welding speed, a welding current, a welding voltage, a torch angle of a welding torch, a volume of the beads, and a cross-sectional area of a cross-section orthogonal to the longitudinal direction of the beads; The learning device according to claim 1.

3. The positional relationship includes a difference distance between a representative position of the narrow portion and a target position of the bead to be formed next; The learning device according to claim 1 or 2.

4. The defect size includes a cross-sectional area or a length in a cross-section orthogonal to the longitudinal direction of the beads; The learning device according to any one of claims 1 to 3.

5. In addition to the defect size with respect to the welding conditions, the dimensions regarding the narrow portions, and the positional relationship, the learning unit registers the variance of the defect size in the estimation model; The estimation model can output the defect size corresponding to the input information and its variance value according to the input information; The learning device according to any one of claims 1 to 4.

6. The learning device according to any one of claims 1 to 5, a determination unit that inputs information of a welding plan including at least dimension information regarding the narrow portions and the positional relationship to the estimation model, and compares an estimated value of the defect size of the unwelded defect output from the estimation model with a reference value that is a predetermined tolerance limit; comprising; a defect determination device.

7. further comprising a welding plan correction unit that corrects at least any one of the welding conditions and the positional relationship to create a corrected welding plan when the determination unit determines that the reference value is exceeded. Repeating the modification of the welding plan by the welding plan modification unit until the estimated value of the defect size according to the information of the modified welding plan output by the estimation model is equal to or less than the reference value. The defect determination device according to claim 6.

8. A learning method for generating an estimation model that learns the defect size of an unwelded defect occurring inside a laminated structure in which a plurality of beads are laminated in layers on a base material and outputs the defect size according to the input information, A step of acquiring information regarding welding conditions when laminating the beads, dimensions regarding a narrow portion that forms a valley portion among the surface shapes of the laminated structure before laminating the beads, a positional relationship between the narrow portion and the aiming position of the beads, and the defect size of the unwelded defect. A step of learning the relationship between the welding conditions, the dimensions regarding the narrow portion, and the positional relationship and the defect size to generate the estimation model. Having The dimensions regarding the narrow portion include at least any one of the bottom width of the valley portion, the opening width representing the distance between the tops on both sides of the valley portion forming the valley portion, and the valley depth from the top to the bottom of the valley portion. Learning method.

9. Inputting information of a welding plan including at least dimension information regarding the narrow portion and the positional relationship into the estimation model generated by the learning method according to claim 8, comparing the estimated value of the defect size of the unwelded defect output from the estimation model with a reference value that is a predetermined tolerance limit, and determining that the unwelded defect occurs when the estimated value exceeds the reference value. Defect determination method.

10. When it is determined that the value exceeds the reference value, modifying at least one of the welding conditions and the positional relationship to create a modified welding plan. Inputting the modified welding plan into the estimation model. Repeating the modification of the welding plan until the estimated value of the defect size according to the information of the modified welding plan output by the estimation model is equal to or less than the reference value. The defect determination method according to claim 9.

11. The defect determination device according to claim 6 or 7, A welding control device including a control unit that executes arc welding according to the result output by the defect determination device.

12. The welding control device according to claim 11, A welding robot that performs arc welding, Comprising Welding device.

Images