Machine learning device, stacking molding system, machine learning method, welding condition adjustment method and storage medium

Abstract

A machine learning device that performs machine learning on welding conditions when shaping a stacked object by depositing a filler material and stacking welds. The machine learning device has a learning processing unit that performs learning processing for generating a learned model. The learned model uses two shape data of a weld or the difference between two shape data as input data and uses the difference between welding conditions corresponding to the difference between the two shape data as output data.

Description

Technical Field

[0001] The present invention relates to a machine learning device, a stacking modeling system, a machine learning method for welding conditions, a method for adjusting welding conditions, and a program. More specifically, it relates to a technique for adjusting welding conditions when shaping a stacked modeled object by stacking weld seams. Background Art

[0002] Traditionally, stacked objects have been shaped by stacking welded seams. To improve the accuracy of stacked objects, various welding conditions must be considered and controlled. These welding conditions offer numerous combinations, making manual selection of appropriate welding conditions extremely complex and tedious.

[0003] Regarding this situation, for example, Patent Document 1 discloses a learning device for automatically determining optimal welding conditions in a welding device without instruction from a skilled operator. In this case, the learning information includes information such as the appearance of the weld, the height and width of the weld, and the amount of weld penetration.

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Laid-Open No. 2017-30014 Summary of the Invention

[0007] Problems to be solved by the invention

[0008] As described above, when adjusting welding conditions during stacking, numerous combinations of conditions must be considered to understand the changing trends of weld seam shape (width, height, etc.), making it difficult to determine the appropriate combination. For example, creating a database that specifies the combinations of conditions is conceivable, but this approach is burdensome. Furthermore, when creating a database, the mechanical variations of the power supply and robot used for stacking cannot be ignored. Adjusting welding conditions based on the inherent influence of these devices makes extracting welding conditions even more complex and cumbersome. Patent Document 1, mentioned above, does not consider these mechanical variations of the power supply and robot, leaving room for improvement in this regard.

[0009] In view of the above-mentioned problems, an object of the present invention is to improve the accuracy of adjusting welding conditions when forming a laminated object.

[0010] Means for solving problems

[0011] In order to solve the above-mentioned problems, the present invention has the following configuration.

[0012] (1) A machine learning device for performing machine learning of welding conditions when shaping a laminated object by depositing a filler material and laminating a weld seam, characterized in that:

[0013] The machine learning device includes a learning processing unit that performs learning processing for generating a learned model. The learned model takes two shape data of a weld or the difference between two shape data as input data, and takes the difference between welding conditions corresponding to the difference between the two shape data as output data.

[0014] Furthermore, another embodiment of the present invention has the following configuration.

[0015] (2) A laminated molding system for molding a laminated molded object by depositing a filler material and laminating weld seams, characterized in that:

[0016] The stacking molding system has:

[0017] a production unit that produces shape data of a weld seam as first shape data based on the design data of the stacked shaped object;

[0018] a determination unit configured to determine welding conditions when forming the first shape data;

[0019] an acquiring unit that acquires, as second shape data, shape data of a weld formed using the welding conditions determined by the determining unit;

[0020] a deriving unit that inputs the first shape data and the second shape data, or the difference between the first shape data and the second shape data, into a learned model generated by performing a learning process that takes two shape data of a weld seam or a difference between two shape data as input data and takes a difference between welding conditions corresponding to the difference between the two shape data as output data, thereby deriving a difference between the welding conditions corresponding to the first shape data and the welding conditions corresponding to the second shape data; and

[0021] An adjusting unit adjusts the welding condition determined by the determining unit using the difference derived by the deriving unit.

[0022] Furthermore, another embodiment of the present invention has the following configuration.

[0023] (3) A machine learning method for determining welding conditions when a laminated object is formed by depositing a filler material and laminating a weld seam, characterized in that:

[0024] The machine learning method includes a learning processing step in which a learning process is performed to generate a learned model. The learned model takes two shape data of a weld or a difference between two shape data as input data and takes a difference between welding conditions corresponding to the difference between the two shape data as output data.

[0025] Furthermore, another embodiment of the present invention has the following configuration.

[0026] (4) A method for adjusting welding conditions in a stacking molding system for molding a stacked molded object by depositing a filler material and laminating a weld seam, characterized in that:

[0027] The method for adjusting the welding conditions includes:

[0028] a manufacturing step of manufacturing shape data of a weld seam as first shape data based on the design data of the stacked shaped object;

[0029] a determining step of determining welding conditions when forming the first shape data;

[0030] an acquiring step of acquiring, as second shape data, shape data of a weld formed using the welding conditions determined in the determining step;

[0031] a derivation step of inputting the first shape data and the second shape data, or the difference between the first shape data and the second shape data, into a learned model generated by performing a learning process that uses two shape data of a weld seam or a difference between two shape data as input data and uses a difference between welding conditions corresponding to the difference between the two shape data as output data, thereby deriving a difference between the welding conditions corresponding to the first shape data and the welding conditions corresponding to the second shape data; and

[0032] An adjustment step of adjusting the welding conditions determined in the determination step using the difference derived in the deriving step.

[0033] Furthermore, another embodiment of the present invention has the following configuration.

[0034] (5) A program, wherein:

[0035] The program is used to cause a computer to execute a learning processing step, in which a learning process is performed to generate a learned model. The learned model uses two shape data of a weld seam when shaping a stacked object or the difference between the two shape data as input data, and uses the difference between welding conditions corresponding to the difference between the two shape data as output data.

[0036] Furthermore, another embodiment of the present invention has the following configuration.

[0037] (6) A program, wherein:

[0038] The program is used to cause a computer to execute the following steps:

[0039] a manufacturing step of manufacturing shape data of the weld seam as first shape data based on design data of a laminated object to be shaped by depositing a filler material and laminating the weld seam;

[0040] a determining step of determining welding conditions when forming the first shape data;

[0041] an acquiring step of acquiring, as second shape data, shape data of a weld formed using the welding conditions determined in the determining step;

[0042] a derivation step of inputting the first shape data and the second shape data, or the difference between the first shape data and the second shape data, into a learned model generated by performing a learning process that uses two shape data of a weld seam or a difference between two shape data as input data and uses a difference between welding conditions corresponding to the difference between the two shape data as output data, thereby deriving a difference between the welding conditions corresponding to the first shape data and the welding conditions corresponding to the second shape data; and

[0043] An adjustment step of adjusting the welding conditions determined in the determination step using the difference derived in the deriving step.

[0044] Effects of the Invention

[0045] According to the present invention, it is possible to improve the accuracy of adjusting welding conditions when forming a laminated object. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a schematic structural diagram showing a structural example of a stacking molding system according to one embodiment of the present invention.

[0047] Figure 2A This is a conceptual diagram for explaining the relationship between the feed speed and the control value of the power supply.

[0048] Figure 2B This is a conceptual diagram for explaining the relationship between the input and output of the target position when forming a weld in a stacking modeling system.

[0049] Figure 3 This is a conceptual diagram for explaining the shape data of a weld.

[0050] Figure 4 This is a schematic diagram for explaining the concept of a learning process according to one embodiment of the present invention.

[0051] Figure 5 This is a flowchart of the shaping process according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following describes a method for implementing the present invention with reference to the accompanying drawings and other figures. It should be noted that the embodiment described below is intended to illustrate one embodiment of the present invention and is not intended to limit the present invention. Furthermore, not all structures described in each embodiment are necessarily required to solve the problems of the present invention. In the accompanying drawings, identical components are designated by identical reference numerals to indicate corresponding relationships.

[0053] <First embodiment>

[0054] Hereinafter, a first embodiment of the present invention will be described.

[0055] [System Structure]

[0056] Hereinafter, one embodiment of the present invention will be described in detail with reference to the accompanying drawings. Figure 1 This is a schematic structural diagram of a stacking molding system to which the present invention can be applied.

[0057] The laminated molding system 1 of the present embodiment is configured to include a laminated molding apparatus 100 and an information processing apparatus 200 that integrally controls the laminated molding apparatus 100 .

[0058] The laminated molding apparatus 100 includes a welding robot 104 , a filler material supply unit 105 that supplies a filler material (welding wire) M to the welding torch 102 , a robot controller 106 that controls the welding robot 104 , and a power source 107 .

[0059] Welding robot 104 is a multi-jointed robot that supports filler material M so that it can continuously supply it to welding torch 102, which is mounted on a distal shaft. Welding torch 102 holds filler material M protruding from its distal end. The position and posture of welding torch 102 can be freely set three-dimensionally within the range of freedom of the robotic arm that constitutes welding robot 104.

[0060] The welding torch 102 includes a shielded nozzle (not shown) through which shielding gas is supplied. The shielding gas cuts off the atmosphere, preventing oxidation and nitridation of the molten metal during welding, thereby suppressing welding defects. The arc welding method used in this embodiment may be any of consumable electrode methods such as sheathed arc welding or carbon dioxide gas arc welding, or non-consumable electrode methods such as TIG welding or plasma arc welding, and the method may be appropriately selected depending on the laminated object W being formed.

[0061] A shape sensor 101 is provided near the welding torch 102, capable of moving in response to the movement of the welding torch 102. The shape sensor 101 detects the shape of the stacked object W formed on the base 103. In this embodiment, the shape sensor 101 can detect the height, position, width, and other information of the weld seam 108 (also referred to simply as the "weld seam") that constitutes the stacked object W. The information detected by the shape sensor 101 is transmitted to the information processing device 200. The structure of the shape sensor 101 is not particularly limited; it can be a structure that detects shape through contact (a contact sensor) or a structure that detects shape using a laser or other means (a non-contact sensor). The means for deriving the shape of the formed weld seam is not limited to the shape sensor 101 provided near the welding torch 102. For example, a structure that indirectly derives the shape of the formed weld seam is also possible. For example, a database (database) that predefines the profile of the welding current or the feed speed of the filler material M, as well as trends in the weld seam height, can be used to derive the height of the formed weld seam based on the welding conditions during the formation process.

[0062] In welding robot 104, when the arc welding method is consumable electrode, a contact piece is placed inside the shielded nozzle, and filler material M, which is supplied with melting current, is held by the contact piece. A welding torch 102 holds the filler material M and generates an arc from the tip of the filler material M in a shielding gas atmosphere. The filler material M is fed from a filler material supply unit 105 to the welding torch 102 via a draw mechanism (not shown) mounted on a robotic arm, etc. The welding torch 102 then moves, causing the continuously fed filler material M to melt and solidify. At this point, a linear weld seam 108, representing the molten and solidified filler material M, is formed on the base 103. By stacking the weld seams 108, a laminated object W is formed.

[0063] It should be noted that the heat source for melting the filler material M is not limited to the aforementioned electric arc. For example, other heat sources such as a combined arc and laser heating method, a plasma heating method, an electron beam heating method, or a laser heating method may also be employed. When heating with an electron beam or laser, the amount of heat can be more finely controlled, the weld seam 108 can be more appropriately maintained, and the quality of the laminated object W can be further improved.

[0064] Based on instructions from the information processing device 200, the robot controller 106 drives the welding robot 104 through a prescribed driver program to form a stacked object W on the base 103. In other words, the welding robot 104 moves the welding torch 102 while melting the filler material M through an arc according to instructions from the robot controller 106. The power supply 107 is a welding power source that supplies the power required for welding to the robot controller 106. The power supply 107 can operate in multiple control modes and can switch the power (current or voltage, etc.) when supplying power to the robot controller 106 according to the control mode. The filler material supply unit 105 controls the supply of filler material M to the welding torch 102 and the feed speed based on instructions from the information processing device 200.

[0065] The information processing device 200 may be, for example, an information processing device such as a PC (Personal Computer). Figure 1 Each of the functions shown can be implemented by a control unit (not shown) reading out and executing a program for the functions of the present embodiment stored in a storage unit (not shown). The storage unit may include a RAM (Random Access Memory) as a volatile storage area, a ROM (Read Only Memory) as a non-volatile storage area, or an HDD (Hard Disk Drive). In addition, a CPU (Central Processing Unit), a GPU (Graphical Processing Unit), or a GPGPU (General-Purpose Computing on Graphics Processing Units) may be used as the control unit.

[0066] The information processing device 200 includes a shaping control unit 201, a power supply control unit 202, a feed control unit 203, a database management unit 204, a shape data acquisition unit 205, a learning data management unit 206, a learning processing unit 207, and a welding condition derivation unit 208. The shaping control unit 201 generates control signals for the robot controller 106 during shaping based on the design data (e.g., CAD / CAM data) of the stacked object W to be shaped. These control signals include the movement trajectory of the welding torch 102 formed by the welding robot 104, welding conditions during the formation of the weld seam 108, and the feed rate of the filler material M by the filler material supply unit 105. The movement trajectory of the welding torch 102 is not limited to the trajectory of the welding torch 102 during the formation of the weld seam 108 on the base 103, and also includes, for example, the movement trajectory of the welding torch 102 relative to the starting position for forming the weld seam 108.

[0067] The power supply control unit 202 controls the power supply (control mode) from the power supply 107 to the robot controller 106. Depending on the control mode, the current or voltage values, current waveform (pulse), and other characteristics may differ when forming welds of the same shape. Furthermore, the power supply control unit 202 timely obtains information on the current or voltage supplied from the power supply 107 to the robot controller 106.

[0068] The feed control unit 203 controls the feed speed or feed timing of the filler material M fed by the filler material supply unit 105. This feed control of the filler material M includes not only withdrawal (forward feed) but also return (reverse feed). The database management unit 204 manages the database (DB) of this embodiment. The DB of this embodiment will be described in detail later. The shape data acquisition unit 205 acquires the shape data of the weld seam 108 formed on the base 103, detected by the shape sensor 101.

[0069] The learning data management unit 206 generates and manages learning data used in the learning process performed by the learning processing unit 207. The learning processing unit 207 performs learning processing using the learning data managed by the learning data management unit 206. The learning data and learning process of this embodiment will be described in detail later. In addition, the learning processing unit 207 manages the learned model obtained as a result of the learning process. As described above, the power supply 107 of this embodiment can operate in multiple control modes. In conjunction with this, the learning processing unit 207 of this embodiment performs learning corresponding to each of the multiple control modes of the power supply 107 and generates a learned model.

[0070] The welding condition derivation unit 208 derives the adjustment amount for the welding condition of the formation control unit 201 using the learned model generated by the learning processing unit 207, and notifies the formation control unit 201. The method of deriving the adjustment amount in this embodiment will be described later.

[0071] In this embodiment, if Figure 1 As shown in FIG, the structure of forming a stacked object W by moving the welding torch 102 on the cylindrical base 103 to form a welding seam 108 is described as an example. Figure 1 In this embodiment, a structure is shown in which base 103 forms a stacked object W on a cylindrical plane, but the present invention is not limited to this. For example, a structure in which base 103 is cylindrical and weld seams 108 are formed on the outer periphery of its side surface may also be employed. Furthermore, the coordinate system in the design data of this embodiment corresponds to the coordinate system on base 103 that forms stacked object W. The three axes of the coordinate system (X-axis, Y-axis, and Z-axis) are defined so that a three-dimensional position can be defined with an arbitrary position as the origin.

[0072] The stacked molding system 1 configured as described above melts the filler material M while moving the welding torch 102 according to the trajectory of movement specified by the set design data, driven by the welding robot 104. The melted filler material M is then supplied onto the base 103. This forms a stacked molding W having a plurality of linear weld seams 108 stacked and arranged on the upper surface of the base 103.

[0073] [Relationship between factors during modeling]

[0074] When shaping a laminated object W, the control parameters used during shaping need to be adjusted depending on factors such as the operating state of the power supply 107, the inherent characteristics of the device, and the structure of the laminated object W. More specifically, the shape of the weld can vary depending on various control parameters used during welding. The following describes examples of control parameters that affect weld shape.

[0075] Examples of control parameters that affect the shape of the weld include the feed rate of the filler material M, welding speed, weld volume, target position, oscillation amplitude or vibration frequency, and heat input. The feed rate of the filler material M will be described as an example. Figure 2AThis graph shows the relationship between the feed speed of the filler material M and the current (or voltage) supplied by the power supply 107. The horizontal axis represents the feed speed of the filler material M, and the vertical axis represents the controlled value of the current (or voltage) supplied by the power supply 107. As the feed speed increases, the current (or voltage) supplied by the power supply 107 increases, but this increase is not constant. The trend of this change can vary depending on the control mode of the power supply 107. Therefore, due to differences in this trend of change, the shape of the resulting weld can vary even with the same control parameters.

[0076] As another example, the target position when forming a weld will be described. Figure 2B This is a diagram showing the relationship between the target position (input) of the weld on the base 103 determined based on the design data and the target position (output) of the weld obtained as a result of the formation, with the horizontal axis representing the input and the vertical axis representing the output. Figure 2B In the figure, the dotted line represents the ideal relationship between input and output, and the input value (i.e., the design value) and the output value are the same. However, in reality, due to various factors such as the performance or specifications of the equipment, the input value and the output value may not necessarily be the same. For example, Figure 2B The solid lines represent an example of the relationship between actual input and output values. As shown by these lines, the designed values and the output results may differ. Therefore, even with the same control parameters, the weld shape may vary depending on the difference (offset) in the target position.

[0077] Figure 3 This is a conceptual diagram for explaining the shape data of a weld seam. Figure 3 FIG. 1 shows a cross section of a state where a weld seam 108 is formed on a base 103 and viewed from the direction of travel of the welding torch 102 during formation. Figure 3 As shown, as the shape data of the weld seam 108 , information such as the height h, the width w, the angle α at the base, and surface irregularities can be used.

[0078] [database]

[0079] In this embodiment, a database is used that represents the relationship between welding conditions and the shape information of the weld formed under those conditions. This database is managed by the DB management unit 204 and is predefined. As described above, the power supply 107 of this embodiment can operate in multiple control modes. Accordingly, multiple databases corresponding to the multiple control modes are defined and managed.

[0080] In the database of this embodiment, predetermined control parameters as welding conditions are associated with information on the shape of the weld formed when welding is performed using these control parameters. Items of welding conditions include the welding amount of filler material M, target position, oscillation conditions, heat input, number of stacking paths, base material temperature, and time between paths, as described above. Items of information on the weld shape include Figure 3 The height, width, root angle, surface unevenness, etc. of the weld shown. It should be noted that the items of various information specified in the database are not limited to the above, and can be increased or decreased as needed.

[0081] [Learning Process]

[0082] In this embodiment, a deep learning method based on a neural network in machine learning is used as a learning method, and supervised learning is used as an example for explanation. It should be noted that the more specific method (algorithm) of deep learning is not particularly limited. For example, a well-known method such as a convolutional neural network (CNN) can be used. In addition, the type and number of layers that constitute the neural network are not particularly limited.

[0083] Figure 4 This is a schematic diagram used to illustrate the concept of the learning process in this embodiment. First, in this embodiment, as raw data, multiple pairs of shape data representing the shape of a weld and the welding conditions used to form the weld are used. It should be noted that the raw data can be data stored as a weld formation history. Using these pairs of data, the differences between the shape data and the welding conditions are calculated. For example, the difference between shape data A for weld shape A and shape data B for weld shape B, as well as the difference between welding condition A corresponding to weld shape A and welding condition B corresponding to weld shape B, are calculated. Then, multiple learning data sets are prepared, using the differences in weld shape as input data and the differences in welding conditions as teaching data. It should be noted that in this embodiment, while the differences in weld shape are used as an example of input data included in the learning data, the present invention is not limited to this. Two pairs of shape data used to derive the differences can also be used as input data.

[0084] In this embodiment, the learning process is performed using the above-mentioned learning data. When input data prepared as learning data (here, the difference in weld shape) is input to the learning model, the difference in welding conditions is output as output data for the input data. This output data corresponds to the adjustment amount of the welding conditions. Next, using this output data and the teaching data prepared as learning data (here, the difference in welding conditions), an error is derived using a loss function. Then, each parameter in the learning model is adjusted to reduce the error. For example, the error backpropagation method can be used to adjust the parameters. In this way, by repeatedly learning using multiple learning data, a learned model is generated. The learned model is updated each time the learning process is executed. Therefore, depending on the timing used, the parameters constituting the learned model are changed, and the output results for the input data are also different. It should be noted that even when two pairs of shape data are used as input data, basically the same operation is performed.

[0085] It should be noted that the learning process does not necessarily need to be performed by the information processing device 200. For example, the information processing device 200 can also be configured to provide learning data to a learning server (not shown) provided outside the information processing device 200, and perform learning processing on the server side. Moreover, it can also be configured as needed so that the server provides the information processing device 200 with a learned model. Such a learning server can be located on a network such as the Internet (not shown), for example, and the learning server is connected to the information processing device 200 in a communicative manner. That is, the information processing device 200 can also operate as a machine learning device, and the external device can also operate as a machine learning device. In either case, the information processing device 200 can obtain the learned model obtained in the learning process and use it when shaping the stacked shape object W.

[0086] [Processing Flow]

[0087] Figure 5 This is a flowchart of the welding condition adjustment process of this embodiment. This process is executed and controlled by the information processing device 200. For example, the processing unit such as the CPU or GPU of the information processing device 200 can read and execute the data from the storage unit (not shown) to achieve the desired result. Figure 1. In addition, before starting this processing flow, the above-mentioned learning process is performed to generate a learned model. In addition, the process for adjusting parameters in this processing flow can also be performed before the actual molding of the stacked object W is about to begin. Alternatively, it can also be performed during the molding of the stacked object W, when the power control mode is switched or when the layer of the formed weld is moved to the next layer. Here, the following case is explained: before molding the stacked object W, adjustment is performed at a position different from the position at which the stacked object W is molded on the base 103.

[0088] In S501, the information processing device 200 obtains design data for a stacked object W. This design data specifies the shape of the stacked object W and is created based on user instructions. The design data can be input from an external device (not shown) connected in a communicable manner, or created on the information processing device 200 using a predetermined application (not shown).

[0089] In S502, the information processing device 200 generates path data corresponding to each weld seam forming the laminated object W in the laminated molding apparatus 100 based on the design data acquired in S501. This path data includes information such as the movement trajectory of the welding torch 102 and shape data representing the shape of the weld seam. The generated shape data corresponds to the design value and can be stored and managed in a storage unit (not shown).

[0090] In S503, the information processing device 200 focuses on a control mode for which parameter adjustment processing has not been performed, among the plurality of control modes in which the power supply 107 can operate. The control modes to be processed here may include all control modes in which the power supply 107 can operate, or may be limited to one or more control modes used when forming the stacked object W using the design data obtained in S501.

[0091] In S504, the information processing device 200 acquires the learned model corresponding to the control mode focused on in S503. As described above, different learned models are generated depending on the control mode, and the corresponding learned model is acquired from these.

[0092] In S505, the information processing device 200 refers to the DB corresponding to the control mode of interest in S503 and determines the welding conditions corresponding to the shape data generated in S502. As described above, the DB associates the welding conditions with the weld shape data, and the welding conditions can be determined by specifying the shape data.

[0093] In S506, the information processing device 200 executes a shaping operation by the welding robot 104 based on the welding conditions determined in S505. This shaping operation is not performed to shape a portion of the stacked object W, but to form a weld for parameter adjustment at a different location.

[0094] In S507, the information processing device 200 obtains the shape data of the weld seam formed in S506 via the shape sensor 101. As described above, the shape sensor 101 of this embodiment is set to move following the welding torch 102. It can also be configured to obtain the shape data in parallel with the formation of the weld seam, or it can be configured to obtain the shape data after the formation of the weld seam is completed. As the shape data obtained here, Figure 3 As shown, the height or width of the weld formed, the angle of the root, the surface unevenness, etc. are given.

[0095] In S508, the information processing apparatus 200 derives the difference between the shape data (measured value) acquired in S507 and the shape data (design value) created in S502. For example, when the shape data includes multiple items such as height and width, each difference is derived.

[0096] In S509, the information processing device 200 compares the difference derived in S508 with a predetermined threshold value to determine whether the difference is above the threshold value. The threshold value is set for each item of the shape data and is stored in a storage unit not shown. The threshold value used here may vary depending on the control mode, or a fixed value may be used. When the difference is above the threshold value (yes in S509), the processing of the information processing device 200 proceeds to S510. On the other hand, when the difference is smaller than the threshold value (no in S509), the processing of the information processing device 200 proceeds to S513. In addition, in the shape data, when multiple items are used for judgment, as a result of comparing each item with the threshold value, it can be judged as yes when all items are above the threshold value. In this case, a threshold value is set for each item.

[0097] In S510, the information processing device 200 inputs the difference derived in S508 as input data to the learned model obtained in S503, thereby obtaining the difference in welding conditions as output data. This difference corresponds to the adjustment amount for the welding conditions used in forming the immediately preceding weld. It should be noted that, as described above, when learning is performed using a pair of two shape data as input data, a pair of the shape data obtained in S507 (measured values) and the shape data created in S502 (designed values) may be input instead of the difference derived in S508.

[0098] In S511 , the information processing apparatus 200 corrects the welding conditions used in forming the immediately preceding weld by reflecting the adjustment amount acquired in S510 .

[0099] In S512, information processing device 200 performs weld formation again using the welding conditions corrected in S511. The process then returns to S507 and repeats the subsequent processing. Specifically, the processes from S507 to S512 are repeated until the difference between the design value based on the design data and the measurement result based on the actual welding result falls below a threshold. Therefore, by repeatedly accumulating the adjustment values obtained in S510 and reflecting them in the welding conditions, the difference gradually decreases (converges).

[0100] In S513, the information processing device 200 stores the welding conditions based on the current adjustment amount in a storage unit (not shown) in association with the control mode of the power source 107 in question. The stored welding conditions (or adjustment amount) are used when forming the laminated object W. The process then proceeds to S514.

[0101] In S514, the information processing device 200 determines whether the parameter adjustment process is complete for all control modes in which the power supply 107 can operate. If there are any unprocessed control modes (No in S514), the information processing device 200 returns to S503 and repeats the subsequent processes. On the other hand, if the process is complete for all control modes (Yes in S514), the present process flow ends.

[0102] The above flowchart illustrates an example of parameter adjustment based on the design data of a stacked object W. In this case, parameter adjustments corresponding to the layers or positional relationships can also be performed based on the number of welds stacked and the positional relationship with adjacent welds (adjacent welds) indicated in the design data. More specifically, information related to the number of layers or positions can be further used as shape data. Including such information allows for parameter adjustments that take into account factors such as the droop of welds or their fusion with adjacent welds depending on the weld formation location. Alternatively, parameter adjustments can be performed based on predefined shape data for parameter adjustment, rather than using the design data of the stacked object W.

[0103] As described above, according to this embodiment, the accuracy of adjusting welding conditions during the molding of laminated objects can be improved. In particular, by using a learned model to derive the relationship between the changing trends of welding conditions and the changing trends of weld bead shape, welding condition adjustments can be made independently of the system. Furthermore, there is no need to create a database that takes into account system variations; using only a universal database allows for application to various laminated molding systems.

[0104] <Other Implementation Methods>

[0105] In addition to the structure shown in the first embodiment, the stacking molding system 1 may also have a structure in which learning data for learning processing is generated. For example, when the stacking molding object W is molded, the shape of the weld is detected by the shape sensor 101 each time a weld is formed, and the welding conditions, shape data, and power supply control mode when the weld is formed are stored in correspondence. In addition, the stored data can be used, for example, Figure 4 At this time, the data used to generate the learning data may be specified by the user of the stacking modeling system 1 or may be extracted by filtering the stored data under arbitrary conditions.

[0106] In addition, in the present invention, the following processing can also be achieved: a program or application for implementing the functions of one or more of the above-mentioned embodiments is supplied to a system or device using a network or storage medium, and the program is read and executed by one or more processors in the computer of the system or device.

[0107] Alternatively, the present invention may be implemented by a circuit that realizes one or more functions (for example, an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array)).

[0108] Based on the above, this specification discloses the following matters.

[0109] (1) A machine learning device for performing machine learning of welding conditions when shaping a laminated object by depositing a filler material and laminating a weld seam, characterized in that:

[0110] The machine learning device includes a learning processing unit that performs learning processing for generating a learned model. The learned model takes two shape data of a weld or the difference between two shape data as input data, and takes the difference between welding conditions corresponding to the difference between the two shape data as output data.

[0111] This configuration improves the accuracy of welding condition adjustment when shaping a laminated object. In particular, it generates a learned model for deriving welding condition adjustment amounts corresponding to the changing trend of shape data, which is used when adjusting welding conditions.

[0112] (2) The machine learning device according to (1), wherein the shape data includes at least one of the height, width, root angle, and surface irregularities of the weld seam.

[0113] According to this configuration, the welding conditions can be adjusted by taking into account the height, width, angle, and surface shape of the weld as shape data.

[0114] (3) The machine learning device according to (1) or (2), characterized in that the welding conditions include at least any one of the feed speed of the filler material, the welding speed, the welding current or voltage, the target position on the base on which the stacked object is formed, the input heat during forming, and the swing control conditions.

[0115] According to this configuration, as welding conditions, the feed speed of the filler material, the target position on the base, the input heat amount during molding, and the swing control conditions can be adjusted.

[0116] (4) The machine learning device according to (3), wherein the welding conditions further include the number of stacking paths or the base material temperature.

[0117] This configuration allows for adjustments to the welding conditions, including the number of stacking paths and the base metal temperature. For example, learning can be performed to account for load accumulation, such as electrode friction and spatter deposition on the nozzle, associated with an increase in the number of stacking paths. Furthermore, learning can be performed to account for heat accumulation in the susceptor.

[0118] (5) The machine learning device according to any one of (1) to (4), characterized in that the learning processing unit generates a learned model according to each control mode of the power supply used when shaping the stacked shaped object.

[0119] According to this configuration, a learned model corresponding to the power supply control mode can be generated, and welding conditions can be adjusted with higher accuracy.

[0120] (6) The machine learning device according to (5), characterized in that at least any one of the voltage value, current value, or current pulse supplied in accordance with the welding conditions is different for each control mode of the power supply.

[0121] According to this configuration, a learned model can be generated that takes into account voltage values, current values, and pulses that differ in the power supply control mode.

[0122] (7) The machine learning device according to any one of (1) to (6), wherein the learning processing unit performs the learning process using a supervised learning method using a neural network.

[0123] According to this structure, machine learning corresponding to supervised learning using a neural network can be performed.

[0124] (8) A laminated molding system for molding a laminated molded object by depositing a filler material and laminating weld seams, characterized in that:

[0125] The stacking molding system has:

[0126] a production unit that produces shape data of a weld seam as first shape data based on the design data of the stacked shaped object;

[0127] a determination unit configured to determine welding conditions when forming the first shape data;

[0128] an acquiring unit that acquires, as second shape data, shape data of a weld formed using the welding conditions determined by the determining unit;

[0129] a deriving unit that inputs the first shape data and the second shape data, or the difference between the first shape data and the second shape data, into a learned model generated by performing a learning process that takes two shape data of a weld seam or a difference between two shape data as input data and takes a difference between welding conditions corresponding to the difference between the two shape data as output data, thereby deriving a difference between the welding conditions corresponding to the first shape data and the welding conditions corresponding to the second shape data; and

[0130] An adjusting unit adjusts the welding condition determined by the determining unit using the difference derived by the deriving unit.

[0131] This configuration improves the accuracy of welding condition adjustment during the molding of laminated objects. Furthermore, by using a learned model to derive the relationship between the changing trends in welding conditions and weld bead shape, welding condition adjustment can be performed independently of the system. Furthermore, there is no need to create a separate database that accounts for system variations; using a single, universal database allows for application to various laminated molding systems.

[0132] (9) The stacking molding system according to (8) is characterized in that the deriving unit derives the difference for adjusting the welding conditions determined by the determining unit when the difference between the first shape data and the second shape data is greater than a predetermined threshold.

[0133] According to this configuration, by repeatedly adjusting the welding conditions, control can be performed so as to obtain predetermined accuracy.

[0134] (10) The stacking molding system according to (8) or (9) is characterized in that the determination unit uses a database in which the shape of the weld seam and the welding conditions are pre-established to determine the welding conditions when forming the first shape data.

[0135] According to this configuration, a common database can be used to determine welding conditions that serve as a reference, and adjustments can be made based on the reference, thereby reducing the time and effort required to create a database of welding conditions for each device.

[0136] (11) The stacking molding system according to any one of (8) to (10), wherein the acquisition unit acquires shape data of the weld seam by measuring the weld seam with a sensor.

[0137] According to this configuration, the actual measured value of the weld bead shape can be acquired by the sensor and used for comparison with the designed value.

[0138] (12) A machine learning method for learning welding conditions when shaping a laminated object by depositing a filler material and laminating a weld seam, characterized in that:

[0139] The machine learning method includes a learning processing step in which a learning process is performed to generate a learned model. The learned model takes two shape data of a weld or a difference between two shape data as input data and takes a difference between welding conditions corresponding to the difference between the two shape data as output data.

[0140] This configuration improves the accuracy of welding condition adjustment when shaping a laminated object. In particular, it generates a learned model for deriving welding condition adjustment amounts corresponding to the changing trend of shape data, which is used when adjusting welding conditions.

[0141] (13) A method for adjusting welding conditions in a stacking molding system for molding a stacked molded object by depositing a filler material and laminating a weld seam, characterized in that:

[0142] The method for adjusting the welding conditions includes:

[0143] a manufacturing step of manufacturing shape data of a weld seam as first shape data based on the design data of the stacked shaped object;

[0144] a determining step of determining welding conditions when forming the first shape data;

[0145] an acquiring step of acquiring, as second shape data, shape data of a weld formed using the welding conditions determined in the determining step;

[0146] a derivation step of inputting the first shape data and the second shape data, or the difference between the first shape data and the second shape data, into a learned model generated by performing a learning process that uses two shape data of a weld seam or a difference between two shape data as input data and uses a difference between welding conditions corresponding to the difference between the two shape data as output data, thereby deriving a difference between the welding conditions corresponding to the first shape data and the welding conditions corresponding to the second shape data; and

[0147] An adjustment step of adjusting the welding conditions determined in the determination step using the difference derived in the deriving step.

[0148] This configuration improves the accuracy of welding condition adjustment during the molding of laminated objects. Furthermore, by using a learned model to derive the relationship between the changing trends in welding conditions and weld bead shape, welding condition adjustment can be performed independently of the system. Furthermore, there is no need to create a separate database that accounts for system variations; using a single, universal database allows for application to various laminated molding systems.

[0149] (14) A program, wherein:

[0150] The program is used to cause a computer to execute a learning processing step, in which a learning process is performed to generate a learned model. The learned model uses two shape data of a weld seam when shaping a stacked object or the difference between the two shape data as input data, and uses the difference between welding conditions corresponding to the difference between the two shape data as output data.

[0151] This configuration improves the accuracy of welding condition adjustment when shaping a laminated object. In particular, it generates a learned model for deriving welding condition adjustment amounts corresponding to the changing trend of shape data, which is used when adjusting welding conditions.

[0152] (15) A program, wherein:

[0153] The program is used to cause a computer to execute the following steps:

[0154] a manufacturing step of manufacturing shape data of the weld seam as first shape data based on design data of a laminated object to be shaped by depositing a filler material and laminating the weld seam;

[0155] a determining step of determining welding conditions when forming the first shape data;

[0156] an acquiring step of acquiring, as second shape data, shape data of a weld formed using the welding conditions determined in the determining step;

[0157] a derivation step of inputting the first shape data and the second shape data, or the difference between the first shape data and the second shape data, into a learned model generated by performing a learning process that uses two shape data of a weld seam or a difference between two shape data as input data and uses a difference between welding conditions corresponding to the difference between the two shape data as output data, thereby deriving a difference between the welding conditions corresponding to the first shape data and the welding conditions corresponding to the second shape data; and

[0158] An adjustment step of adjusting the welding conditions determined in the determination step using the difference derived in the deriving step.

[0159] This configuration improves the accuracy of welding condition adjustment during the molding of laminated objects. Furthermore, by using a learned model to derive the relationship between the changing trends in welding conditions and weld bead shape, welding condition adjustment can be performed independently of the system. Furthermore, there is no need to create a separate database that accounts for system variations; using a single, universal database allows for application to various laminated molding systems.

[0160] While various embodiments have been described above with reference to the accompanying drawings, the present invention is not limited to the aforementioned examples. It is apparent to those skilled in the art that various variations or modifications can be envisioned within the scope of the claims, and it should be understood that these variations or modifications also fall within the technical scope of the present invention. Furthermore, the various components of the aforementioned embodiments may be arbitrarily combined without departing from the spirit of the invention.

[0161] It should be noted that the present application is based on the Japanese patent application (Japanese Patent Application No. 2020-121581) filed on July 15, 2020, and the contents thereof are incorporated herein by reference.

[0162] Description of Reference Numerals

[0163] 1…Layered modeling system;

[0164] 100…Layered modeling device;

[0165] 101…Shape sensor;

[0166] 102… welding torch;

[0167] 103…base;

[0168] 104… welding robot;

[0169] 106…Robot controller;

[0170] 107…power supply;

[0171] 108…weld seam;

[0172] 200…information processing device;

[0173] 201…Modeling Control Department;

[0174] 202…power supply control unit;

[0175] 203…feed control unit;

[0176] 204…DB (database) management department;

[0177] 205: shape data acquisition unit;

[0178] 206…Learning Data Management Department;

[0179] 207…Learning Processing Department;

[0180] 208 ... welding condition derivation unit;

[0181] W…Layered shapes;

[0182] M…filling material.

Claims

1. A machine learning device for performing machine learning on welding conditions when shaping a laminated object by depositing a filler material and laminating a weld seam, characterized in that: The machine learning device includes a learning processing unit that performs a learning process for generating a learned model. The learned model takes two pieces of weld shape data or a difference between two pieces of shape data as input data and outputs a difference between welding conditions corresponding to the difference between the two pieces of shape data. The learned model is used when a difference between first shape data and second shape data is input to derive a difference for adjusting welding conditions when forming the first shape data, wherein the first shape data is shape data of a weld seam based on the design data of the stacked object, and the second shape data is shape data of a weld seam formed using the welding conditions when forming the first shape data based on the design data of the stacked object.

2. The machine learning device according to claim 1, wherein The shape data includes at least any one of the height, width, root angle, and surface irregularities of the weld seam.

3. The machine learning device according to claim 1, wherein The welding conditions include at least one of a feed speed of the filler material, a welding speed, a welding current or voltage, a target position on a base where the stacked object is formed, an input heat during forming, and a swing control condition.

4. The machine learning device according to claim 3, wherein: The welding conditions also include the number of stacking paths or the base material temperature.

5. The machine learning device according to claim 1, wherein The learning processing unit generates a learned model for each control mode of a power supply used when forming the stacked object.

6. The machine learning device according to claim 5, wherein: At least one of a voltage value, a current value, or a current pulse supplied in accordance with the welding conditions differs for each control mode of the power supply.

7. The machine learning device according to claim 1, wherein The learning processing unit performs the learning process using a supervised learning method using a neural network.

8. A stacking molding system for molding a stacked molded object by depositing a filler material and stacking weld seams, characterized in that: The stacking molding system has: a production unit that produces shape data of a weld seam as first shape data based on the design data of the stacked shaped object; a determination unit configured to determine welding conditions when forming the first shape data; an acquiring unit that acquires, as second shape data, shape data of a weld formed using the welding conditions determined by the determining unit; a deriving unit that inputs the first shape data and the second shape data, or the difference between the first shape data and the second shape data, into a learned model generated by performing a learning process that takes two pieces of shape data of a weld seam or a difference between two pieces of shape data as input data and takes a difference between welding conditions corresponding to the difference between the two pieces of shape data as output data, thereby deriving a difference between the welding conditions corresponding to the first shape data and the welding conditions corresponding to the second shape data; as well as An adjusting unit adjusts the welding condition determined by the determining unit using the difference derived by the deriving unit.

9. The stacking molding system according to claim 8, characterized in that: The deriving unit derives a difference for adjusting the welding conditions determined by the determining unit when the difference between the first shape data and the second shape data is equal to or greater than a predetermined threshold value.

10. The stacking molding system according to claim 8, characterized in that: The determination unit determines the welding conditions when forming the first shape data using a database in which the shape of the weld seam and the welding conditions are previously associated.

11. The stacking molding system according to claim 8, characterized in that: The acquisition unit acquires shape data of the weld seam by measuring the weld seam with a sensor.

12. A machine learning method for determining welding conditions when shaping a laminated object by depositing a filler material and laminating a weld seam, characterized in that: The machine learning method includes a learning processing step in which a learning process is performed to generate a learned model. The learned model takes two pieces of weld shape data or a difference between two pieces of shape data as input data and takes a difference between welding conditions corresponding to the difference between the two pieces of shape data as output data. The learned model is used when a difference between first shape data and second shape data is input to derive a difference for adjusting welding conditions when forming the first shape data, wherein the first shape data is shape data of a weld seam based on the design data of the stacked object, and the second shape data is shape data of a weld seam formed using the welding conditions when forming the first shape data based on the design data of the stacked object.

13. A method for adjusting welding conditions in a stacking molding system for molding a stacked molded object by depositing a filler material and laminating weld seams, characterized in that: The method for adjusting the welding conditions includes: a manufacturing step of manufacturing shape data of a weld seam as first shape data based on the design data of the stacked shaped object; a determining step of determining welding conditions when forming the first shape data; an acquiring step of acquiring, as second shape data, shape data of a weld formed using the welding conditions determined in the determining step; a derivation step of inputting the first shape data and the second shape data, or the difference between the first shape data and the second shape data, into a learned model generated by performing a learning process that uses two shape data of a weld seam or a difference between two shape data as input data and uses a difference between welding conditions corresponding to the difference between the two shape data as output data, thereby deriving a difference between the welding conditions corresponding to the first shape data and the welding conditions corresponding to the second shape data; as well as An adjustment step of adjusting the welding conditions determined in the determination step using the difference derived in the deriving step.

14. A storage medium, wherein: The storage medium stores a program for causing a computer to execute a learning process step, wherein a learning process is performed to generate a learned model, the learned model taking two shape data of a weld seam when shaping a stacked object or a difference between the two shape data as input data, and taking a difference between welding conditions corresponding to the difference between the two shape data as output data. The learned model is used when a difference between first shape data and second shape data is input to derive a difference for adjusting welding conditions when forming the first shape data, wherein the first shape data is shape data of a weld seam based on the design data of the stacked object, and the second shape data is shape data of a weld seam formed using the welding conditions when forming the first shape data based on the design data of the stacked object.

15. A storage medium, wherein: The storage medium stores a program for causing a computer to execute the following steps: a manufacturing step of manufacturing shape data of the weld seam as first shape data based on design data of a laminated object to be shaped by depositing a filler material and laminating the weld seam; a determining step of determining welding conditions when forming the first shape data; an acquiring step of acquiring, as second shape data, shape data of a weld formed using the welding conditions determined in the determining step; a derivation step of inputting the first shape data and the second shape data, or the difference between the first shape data and the second shape data, into a learned model generated by performing a learning process that uses two shape data of a weld seam or a difference between two shape data as input data and uses a difference between welding conditions corresponding to the difference between the two shape data as output data, thereby deriving a difference between the welding conditions corresponding to the first shape data and the welding conditions corresponding to the second shape data; as well as An adjustment step of adjusting the welding conditions determined in the determination step using the difference derived in the deriving step.

Images