Welding condition search device and welding condition search method

Abstract

The objective is to provide a welding condition search device and welding condition search method that can more efficiently search for appropriate welding conditions by providing a machine learning model with real data and / or simulation data when searching for welding conditions to obtain desired welding results. [Solution] The welding condition search device 100 includes a receiving unit 110 that receives a desired welding result, a machine learning model 120 that searches for welding conditions that can obtain the desired welding result using the welding conditions and the welding results corresponding to those welding conditions, and an output unit 140 that outputs the welding conditions searched by the machine learning model 120. The machine learning model 120 uses at least one of the following: real data including the welding conditions and the welding results when welding is actually performed with those welding conditions, simulation data generated using a simulation model, and reference data in different environments.

Description

Technical Field

[0004] , , , , ,

[0006] , , , ,

[0005] , , , , ,

[0001] The present invention relates to a welding condition search device and a welding condition search method.

Background Art

[0002] Generally, when welding, welding results such as various bead widths and spatter amounts can be obtained according to welding conditions such as the feeding speed of the welding wire and the welding current. That is, in order to obtain a desired welding result, it is important to set appropriate welding conditions.

[0003] Also, in Patent Document 1, a technique for generating appropriate waveform information to obtain a desired welding result (spatter amount) by searching for design variables representing waveform information (pulse waveform) in which the spatter amount (welding result) is reduced using Bayesian optimization is disclosed.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] By the way, a method of searching for appropriate welding conditions using a machine learning model by performing machine learning from data on welding conditions and welding results corresponding to the welding conditions can be considered. However, in order to perform machine learning, data on welding conditions and welding results corresponding to the welding conditions are required.

[0006] In this case, if there is little data to use for machine learning, for example, it is possible to obtain appropriate welding conditions by repeatedly searching using real data obtained from actual welding, but this requires many searches and takes a lot of time to obtain the optimal welding conditions. On the other hand, there is the problem that preparing a large amount of data in advance is burdensome.

[0007] Therefore, the present invention aims to provide a welding condition search device and a welding condition search method that can more efficiently search for appropriate welding conditions by providing real data and / or simulation data to a machine learning model when searching for welding conditions to obtain desired welding results. [Means for solving the problem]

[0008] A welding condition search device according to one aspect of the present invention is a welding condition search device for searching for welding conditions that can obtain a desired welding result, comprising: a receiving unit for receiving a desired welding result; a machine learning model for searching for welding conditions that can obtain a desired welding result using the welding conditions and the welding results corresponding to those welding conditions; and an output unit for outputting the welding conditions searched by the machine learning model, wherein the machine learning model uses at least one of the following: real data including welding conditions and welding results when welding is actually performed with those welding conditions; simulation data including welding conditions generated using a simulation model and welding results for those welding conditions; and reference data including welding conditions in different environments and welding results when welding is actually performed with those welding conditions.

[0009] In this embodiment, the receiving unit receives the desired welding result, the machine learning model searches for welding conditions that can obtain the desired welding result using the welding conditions and the corresponding welding results, and the output unit outputs the welding conditions searched by the machine learning model. The machine learning model uses at least one of the following: real data including welding conditions and welding results when welding is actually performed with those welding conditions; simulation data including welding conditions generated using a simulation model and welding results for those welding conditions; and reference data including welding conditions in different environments and welding results when welding is actually performed with those welding conditions. Therefore, it can search for appropriate welding conditions more efficiently.

[0010] In the above embodiment, if simulation data is used, the degree of data variability may be added.

[0011] According to this embodiment, in order to add a degree of variability to the simulation data, data that takes into account noise and other factors in the actual data can be used.

[0012] In the above embodiment, the degree of variation may be calculated based on the standard deviation of multiple real data points.

[0013] According to this embodiment, the degree of variation is calculated based on the standard deviation of multiple real-world data, so more appropriate data can be used as simulation data.

[0014] In the above embodiment, a selection unit may be further provided to select whether to use real data, simulation data, or both real data and simulation data.

[0015] According to this embodiment, the selection unit can choose to use real data, simulation data, or both real and simulation data, allowing for the use of more appropriate data depending on the situation.

[0016] In the above embodiment, if the machine learning model uses both real data and simulated data, the influence of the simulated data may be adjusted.

[0017] According to this embodiment, the influence of the simulation data is adjusted, so that real data and simulation data can be used more appropriately depending on the nature and accuracy of the simulation data.

[0018] In the above embodiment, the influence of the simulation data may be adjusted according to the similarity between the real data and the simulation data.

[0019] According to this embodiment, the influence of the simulation data is adjusted according to the similarity between the real data and the simulation data, so that the simulation data can be used more appropriately and efficiently. As a result, the number of times welding conditions need to be searched to obtain the desired welding result is reduced, and appropriate welding conditions can be obtained more efficiently.

[0020] A welding condition search method according to one aspect of the present invention is a welding condition search method performed by a welding condition search device that searches for welding conditions that can obtain a desired welding result, and includes: a reception step of receiving a desired welding result; a data storage step of storing data of welding conditions and welding results corresponding to said welding conditions in memory; a welding condition search step of searching for welding conditions that can obtain a desired welding result using a machine learning model based on the data stored in memory; and an output step of outputting the welding conditions searched in the welding condition search step, wherein the machine learning model uses at least one of the following: actual data including welding conditions and welding results when welding is actually performed with said welding conditions; simulation data including welding conditions generated using a simulation model and welding results for said welding conditions; and reference data including welding conditions in different environments and welding results when welding is actually performed with said welding conditions.

[0021] According to this aspect, in the reception step, a desired welding result is received. In the data storage step, data on the welding conditions and the welding result corresponding to the welding conditions is stored in the memory. In the welding condition search step, based on the data stored in the memory, a welding condition under which a desired welding result can be obtained is searched using a machine learning model. Since the machine learning model uses at least any one of actual data including welding conditions and the welding result when actually welding under the welding conditions, simulation data including welding conditions generated using a simulation model and the welding result for the welding conditions, and reference data including welding conditions in different environments and the welding result when actually welding under the welding conditions, appropriate welding conditions can be searched more efficiently.

Effect of the Invention

[0022] According to the present invention, when searching for welding conditions to obtain a desired welding result, a welding condition search device and a welding condition search method capable of providing actual data and / or simulation data to a machine learning model and searching for appropriate welding conditions more efficiently can be provided.

Brief Description of the Drawings

[0023] [Figure 1] It is a functional block diagram for explaining each function in the welding condition search device 100 according to an embodiment of the present invention. [Figure 2] It is a diagram for explaining the relationship between the actual data and the simulation data used in the welding condition search device 100 according to an embodiment of the present invention. [Figure 3] It is a flowchart showing the processing flow of the welding condition search method M100 executed by the welding condition search device 100 according to an embodiment of the present invention. [Figure 4] It is a diagram for explaining the relationship between the actual data and the reference data used in the welding condition search device 100 according to an embodiment of the present invention. [Figure 5]This figure illustrates the relationship between multiple real-world data and simulation data used in a welding condition search device 100 according to one embodiment of the present invention. [Modes for carrying out the invention]

[0024] The embodiments of the present invention will be described below in detail with reference to the drawings. The embodiments described below are merely examples of how to implement the present invention and are not intended to limit the scope of the invention. Furthermore, to facilitate understanding of the explanation, the same reference numerals are used for identical components in each drawing whenever possible, and redundant explanations may be omitted.

[0025] <One Embodiment> [Configuration of the welding condition search device] Figure 1 is a functional block diagram illustrating the functions of a welding condition search device 100 according to one embodiment of the present invention. As shown in Figure 1, the welding condition search device 100 comprises a reception unit 110, a calculation unit 120, a storage unit 130, an output unit 140, a communication unit 150, and a selection unit 160, and searches for welding conditions that will yield the desired welding result. The welding machine 10 performs welding based on the welding command from the welding condition search device 100, and the welding condition search device 100 stores the results and further searches for the next welding conditions.

[0026] In this embodiment, bead width is used as an example of a welding result, and wire feeding speed is used as an example of a welding condition (parameter) for obtaining a desired bead width.

[0027] The reception unit 110 receives the desired welding result. For example, the reception unit 110 receives the desired bead width entered by the user. The user may enter the desired bead width using an input terminal, or using an input unit (interface) provided on the welding condition search device 100.

[0028] The calculation unit 120 searches for welding conditions that will yield the desired welding result, using the welding conditions and the corresponding welding results. For example, the calculation unit 120 uses an optimization algorithm (machine learning model) to search for a wire feeding speed that will yield the desired bead width, based on the wire feeding speed as a welding condition and the welding result (bead width) at that welding condition (wire feeding speed) stored in the memory unit 130.

[0029] The optimization algorithm (machine learning model) used here is not particularly limited; any algorithm that searches for an appropriate output for a given input will suffice.

[0030] Furthermore, the optimization algorithm (machine learning model) may use real data including welding conditions and welding results when welding is actually performed under those conditions, or it may use simulation data including welding conditions generated using a simulation model and welding results for those conditions, or it may use both.

[0031] The memory unit 130 stores actual data, including welding conditions and corresponding welding results, when welding is actually performed by the welding machine 10, and / or simulation data, including welding conditions and welding results for those welding conditions, generated using a simulation model.

[0032] For example, if the selection unit 160 described later selects to use simulation data (including cases where both simulation data and actual data are used), pre-generated simulation data may be stored in the storage unit 130. Specifically, the wire feeding speed as a welding condition and the welding result (bead width) at that welding condition (wire feeding speed) are pre-generated using a simulation model and stored in the storage unit 130 as simulation data.

[0033] Furthermore, based on the actual data and / or simulation data stored in the memory unit 130, an optimization algorithm (machine learning model) is used to search for welding conditions (wire feeding speed). Then, when welding is actually performed by the welding machine 10 under those welding conditions (wire feeding speed), the wire feeding speed and the welding result (bead width) under those welding conditions (wire feeding speed) are stored (accumulated) in the memory unit 130. In other words, each time welding is actually performed by the welding machine 10, actual data (wire feeding speed and bead width) is accumulated in the memory unit 130.

[0034] These wire feeding speeds and bead widths may be received by the receiving unit 110 each time welding is performed by the welding machine 10 and stored in the storage unit 130. Specifically, the user inputs the wire feeding speed and bead width using an input terminal (input unit), and these are stored in memory.

[0035] The output unit 140 outputs the welding conditions searched by the calculation unit 120. For example, if the desired welding result (bead width) is obtained when welding is actually performed by the welding machine 10, the output unit 140 outputs the welding conditions (wire feeding speed) at that time. Specifically, the output unit 140 may display the wire feeding speed at which the desired bead width was obtained on the input terminal, the welding condition search device 100, or the screen of the welding machine 10, or it may be made known to the user by other means.

[0036] Furthermore, since the desired welding result (bead width) is obtained when welding is actually performed by the welding machine 10, the welding conditions (wire feeding speed) are set in the welding machine 10. In other words, the function of the output unit 140 may also include outputting the welding conditions (wire feeding speed) set in the welding machine 10.

[0037] On the other hand, if the desired welding result (bead width) is not obtained when welding is actually performed by the welding machine 10, the output unit 140 outputs the welding conditions searched by the calculation unit 120 and notifies the welding machine 10 via the communication unit 150.

[0038] The communication unit 150 notifies the welding machine 10 of a welding command, including the welding conditions output by the output unit 140. For example, the communication unit 150 notifies the welding machine 10 of a welding command to set the wire feed speed as a welding condition (parameter) and then weld. As a result, the welding machine 10 sets the welding condition (wire feed speed) and actually performs welding.

[0039] The selection unit 160 selects whether to use real data, simulation data, or both real and simulation data. For example, the user may choose whether to use real data, simulation data, or both real and simulation data, or it may be automatically selected depending on the availability of simulation data. If simulation data is pre-stored in the storage unit 130, the system may choose to use the simulation data (including the case where both simulation data and real data are used).

[0040] Furthermore, when using simulation data, it is also possible to add a degree of data variability. For example, the simulation data is generated by a simulation model without actually welding using the welding machine 10, and the welding conditions (wire feeding speed) and the welding results (bead width) at those welding conditions (wire feeding speed) are theoretical values ​​calculated using mathematical formulas, functions, and AI (Artificial Intelligence).

[0041] On the other hand, actual data obtained by actually welding using the welding machine 10, including welding conditions (wire feeding speed) and welding results (bead width) at those welding conditions (wire feeding speed), contains variations (noise) depending on the welding environment (temperature and humidity, etc.), welding wire, the performance and machine variations of the welding machine 10, and other welding conditions (parameters).

[0042] It is preferable to use the simulation data generated by the simulation model with a degree of variation that may occur when actually welding using the welding machine 10. Specifically, the degree of variation may be calculated based on the standard deviation of multiple real data, or it may be calculated by setting the standard deviation by the user. Furthermore, the degree of variation may be calculated by setting what percentage (for example, 5%) of the data contains what amount of noise (for example, a deviation of ±0.1 mm in bead width, a deviation of ±0.1 cm / min in wire feeding speed, etc.) as random noise.

[0043] [Data for optimization algorithms (machine learning models)] Figure 2 is a diagram illustrating the relationship between real data and simulation data used in a welding condition search device 100 according to one embodiment of the present invention. As shown in Figure 2, it has a real data field and a simulation data field.

[0044] In the actual data field, for example, when welding is actually performed by the welding machine 10, the wire feeding speed as a welding condition and the welding result (bead width) at that welding condition (wire feeding speed) are obtained, and actual data is generated. Based on this actual data, an optimization algorithm (machine learning model) can be used to search for welding conditions (wire feeding speed) that will yield the desired welding result (bead width).

[0045] In the simulation data field, simulation data is generated by a simulation model without actually using the welding machine 10 to perform welding, showing the wire feeding speed as a welding condition and the welding result (bead width) at that welding condition (wire feeding speed). Furthermore, the above-mentioned degree of variation may be added to this simulation data. Based on this simulation data, an optimization algorithm (machine learning model) can be used to search for welding conditions (wire feeding speed) that will yield the desired welding result (bead width).

[0046] Furthermore, based on both real-world and simulated data, an optimization algorithm (machine learning model) can be used to search for welding conditions (wire feed speed) that will yield the desired welding result (bead width).

[0047] In this case, the influence of the simulation data may be adjusted. Here, influence refers to the magnitude of the impact that the simulation data in the simulation data field has on the real data field, and can be expressed as the distance between the simulation data field and the real data field.

[0048] Depending on the distance, it is determined what percentage (confidence level) of the simulation data should be treated as relative to the real data. For example, if the distance is small (e.g., distance "1"), the simulation data may be treated as 90% of the real data, while if the distance is large (e.g., distance "5"), the simulation data may be treated as 50% of the real data.

[0049] The distance may be set by the user, and it is preferable to set it appropriately according to the similarity between the actual data and the simulation data. If the actual data and the simulation data are not similar, reducing the distance will increase the number of times welding conditions are searched to obtain the desired welding result. On the other hand, if the actual data and the simulation data are similar, increasing the distance will reduce the effectiveness of using the simulation data. In other words, by setting the distance appropriately according to the similarity between the actual data and the simulation data, the number of times welding conditions are searched to obtain the desired welding result can be reduced, and appropriate welding conditions can be obtained more efficiently.

[0050] [Method for Exploring Welding Conditions] Figure 3 is a flowchart showing the processing flow of a welding condition search method M100 performed by a welding condition search device 100 according to one embodiment of the present invention. As shown in Figure 3, the welding condition search method M100 includes steps S101 to S112, each of which is performed by a processor included in the welding condition search device 100.

[0051] In step S101, the welding condition search device 100 sets the desired welding result. Specifically, the reception unit 110 receives the desired bead width entered by the user using the input terminal (input unit). Then, the desired bead width is set as the desired welding result.

[0052] In step S102, the welding condition search device 100 selects whether or not to use simulation data. Specifically, the selection unit 160 may allow the user to select whether or not to use simulation data using an input terminal (input unit). Here, simulation data refers to, for example, welding conditions and welding results under said welding conditions generated by a simulation model, and in this embodiment, it refers to the wire feeding speed and the bead width corresponding to said wire feeding speed.

[0053] If "Use simulation data" is selected in step S102 ("Yes" in step S102), the process proceeds to step S103. If "Do not use simulation data" is selected ("No" in step S102), the process proceeds to step S107.

[0054] In step S103, the welding condition search device 100 adds a variability index to the simulation data. Specifically, it adds a variability index that may occur when actually welding using the welding machine 10 to the simulation data that has been generated in advance by the simulation model. Note that the simulation data may be generated or the variability index may be added to the simulation data by a device other than the welding condition search device 100.

[0055] In step S104, the welding condition search device 100 stores (accumulates) the simulation data to which the degree of variation was assigned in step S103 in a storage unit 130 such as a memory.

[0056] In step S105, the welding condition search device 100 reads the data (welding conditions and welding results) stored in the storage unit 130. The data stored in the storage unit 130 may be simulation data stored in step S104, and actual data including welding conditions and welding results corresponding to those welding conditions when welding is actually performed by the welding machine 10 described later. Specifically, the calculation unit 120 reads the welding conditions (wire feeding speed) and the welding results (bead width) corresponding to those welding conditions (wire feeding speed) as simulation data and / or actual data stored in the storage unit 130.

[0057] In step S106, the welding condition search device 100 searches for the next welding condition (wire feeding speed) using an optimization algorithm (machine learning model). Specifically, the calculation unit 120 searches for a wire feeding speed that will yield the desired bead width using an optimization algorithm (machine learning model) based on the data (welding conditions and welding results) read out in step S105.

[0058] In step S107, the welding condition search device 100 sets the wire feeding speed as a welding condition. For example, if it was selected not to use simulation data in step S102 ("No" in step S102), the device may set the wire feeding speed to the median value among the wire feeding speeds set by the user or the search range (settable range). On the other hand, if the wire feeding speed is searched by the optimization algorithm (machine learning model) in steps S105 and S106, the welding condition search device 100 sets the searched wire feeding speed.

[0059] In step S108, the welding condition search device 100 actually causes the welding machine 10 to perform welding. Specifically, the communication unit 150 notifies the welding machine 10 of a welding command that includes the wire feeding speed set in step S107 as a welding condition. The welding machine 10 then actually performs welding at that wire feeding speed.

[0060] In step S109, the welding condition search device 100 acquires the welding result (bead width). Specifically, in the welding actually performed in step S108, the user confirms the welding result (bead width). The reception unit 110 then receives the bead width entered by the user using the input terminal (input unit).

[0061] Furthermore, the welding result (bead width) in the welding actually performed in step S108 may be automatically acquired, for example, by equipping the welding condition search device 100 or welding machine 10 with a camera and measuring using the camera, without requiring confirmation by the user.

[0062] In step S110, the welding condition search device 100 stores (accumulates) the welding conditions (wire feed speed) and welding results (bead width) from when welding was actually performed by the welding machine 10 in step S108. Specifically, after welding is actually performed by the welding machine 10 in step S108, the reception unit 110 receives the wire feed speed and bead width entered by the user using the input terminal (input unit). Then, in the storage unit 130, each time welding is actually performed by the welding machine 10 in step S108, the wire feed speed and the bead width at that wire feed speed are stored, and actual data is accumulated.

[0063] In this example, actual welding data from welding machine 10 is stored in the storage unit 130, but the data stored in the storage unit 130 is not limited to this. The data stored in the storage unit 130 may include simulation data stored in advance in step S104, and may include, for example, data from welding performed separately (wire feeding speed and bead width), data from welding performed with other welding machines (wire feeding speed and bead width), and other generated data (wire feeding speed and bead width).

[0064] In step S111, it is determined whether the welding result (bead width) is as desired. Specifically, it is determined whether the bead width obtained in step S109 is as desired in step S101. If it is as desired ("Yes" in step S111), the process proceeds to step S112. If it is not as desired ("No" in step S111), the process proceeds to step S105.

[0065] Furthermore, if the bead width obtained in step S109 is within the range of 100% to 120% of the desired bead width set in step S101 (tolerance range + 20%), it may be determined to be the desired bead width. In addition, this tolerance range may be set by the user in step S101, for example.

[0066] In step S112, the welding condition search device 100 outputs the welding conditions (wire feeding speed) at which the desired welding result (bead width) was obtained. Specifically, the output unit 140 displays the welding conditions (wire feeding speed) at which the desired welding result (bead width) was obtained on the screen of the input terminal, the welding condition search device 100, or the welding machine 10, or makes them known to the user by other means.

[0067] In this manner, steps S105 to S111 are repeated until the desired welding result (bead width) is obtained.

[0068] As described above, according to the welding condition search device 100 and welding condition search method M100 of one embodiment of the present invention, the receiving unit 110 receives the desired welding result (bead width), and the calculation unit 120 searches for welding conditions (wire feeding speed) that can obtain the desired welding result (bead width) using a machine learning model based on data including welding conditions (wire feeding speed) and welding results (bead width) corresponding to the welding conditions (wire feeding speed) stored in the storage unit 130. If the selection unit 160 selects to use simulation data, the machine learning model searches for welding conditions that can obtain the desired welding result based on data including simulation data of welding conditions (wire feeding speed) and welding results (bead width) generated using the simulation model. As a result, the number of times the welding conditions (wire feeding speed) are searched to obtain the desired welding result (bead width) is reduced, and appropriate welding conditions (wire feeding speed) can be obtained more efficiently.

[0069] Furthermore, in this embodiment, as explained with reference to Figure 2, the influence of simulation data was adjusted according to its similarity to real data, and an optimization algorithm (machine learning model) was used to more efficiently search for appropriate welding conditions (wire feeding speed). However, the data used in the optimization algorithm (machine learning model) is not limited to simulation data generated by a simulation model or the like. For example, instead of simulation data generated by a simulation model or the like, other real data (reference data including welding conditions in different environments and welding results when welding was actually performed under those conditions) may be used as the data used in the optimization algorithm (machine learning model).

[0070] Figure 4 is a diagram illustrating the relationship between actual data and reference data used in a welding condition search device 100 according to one embodiment of the present invention. As shown in Figure 4, it has an actual data field and a reference data field.

[0071] In the actual data field, for example, welding is actually performed by the welding machine 10 at a welding speed of 80 cm / min, thereby obtaining the wire feeding speed as a welding condition and the welding result (bead width) at that welding condition (wire feeding speed), and generating actual data. Based on this actual data, an optimization algorithm (machine learning model) can be used to search for welding conditions (wire feeding speed) that will yield the desired welding result (bead width).

[0072] For example, suppose actual data exists from the past and / or from a different welding machine in different environments. Specifically, suppose actual data exists regarding the wire feed speed as a welding condition obtained by actually welding at a welding speed of 60 cm / min and / or a welding speed of 100 cm / min, and the welding result (bead width) at that welding condition (wire feed speed).

[0073] In the reference data field, actual data is shown as reference data for different environments with wire feed speeds at welding speeds of 60 cm / min and / or 100 cm / min, and bead widths at those wire feed speeds. In other words, although this reference data is actual data, it is not data obtained by actually welding at a welding speed of 80 cm / min, so it may be treated as reference data for welding at a welding speed of 80 cm / min. Note that, unlike data generated using a simulation model, this reference data is actual data and therefore contains variability (noise), so it is not necessary to assign a degree of variability.

[0074] These reference data are stored in the storage unit 130 as accumulated data, and the calculation unit 120 reads the data stored in the storage unit 130, including these reference data (step S105), and can search for welding conditions (wire feeding speed) using an optimization algorithm (machine learning model) (step S106).

[0075] Furthermore, actual data from different environments, such as wire feeding speeds at welding speeds of 60 cm / min and / or 100 cm / min, and bead widths at those wire feeding speeds, may be used in the actual data field instead of the reference data described above.

[0076] Figure 5 is a diagram illustrating the relationship between multiple real data and simulation data used in a welding condition search device 100 according to one embodiment of the present invention. As shown in Figure 5, it has a real data field and a simulation data field. The real data field and the simulation data field show two welding conditions (parameters), namely wire feeding speed and welding speed, and the welding result (bead width) for the combination of these two welding conditions. The welding result (bead width) can be shown using, for example, color, shade, density, and a three-dimensional graph.

[0077] In the actual data field, for example, actual data is used that represents welding conditions obtained by actually welding at welding speeds of 60 cm / min and / or 100 cm / min in the past using and / or other welding machines, and the welding results (bead width) at those welding conditions (wire feeding speed).

[0078] Then, in the actual data field, welding is actually performed by the welding machine 10 at a welding speed of 80 cm / min, and the wire feeding speed as a welding condition and the welding result (bead width) at that welding condition (wire feeding speed) are obtained, and actual data is generated.

[0079] In the simulation data field, simulation data is generated by a simulation model for the wire feed speed as a welding condition and the welding result (bead width) at that welding condition (wire feed speed), without actually welding using the welding machine 10 (welding speed of 80 cm / min). Furthermore, the simulation data may be modified to include a degree of variation. In addition, simulation data for the wire feed speed as a welding condition and the welding result (bead width) at welding speeds of 60 cm / min and / or 100 cm / min, which were used in the actual data field described above, may also be used in the simulation data field.

[0080] In this way, using real data and / or simulation data from different environments, the calculation unit 120 reads the data stored in the storage unit 130, including this real data and / or simulation data (step S105), and uses an optimization algorithm (machine learning model) to search for welding conditions (wire feeding speed) at a welding speed of 80 cm / min (step S106). The accuracy of searching for welding conditions can be improved by using various real data and / or simulation data.

[0081] Then, based on real data and simulation data, an optimization algorithm (machine learning model) can be used to search for welding conditions (wire feed rate and welding rate) to obtain the desired welding result (bead width). Here, since welding is performed at a welding rate of 80 cm / min, the welding rate as a welding condition may be fixed, and the wire feed rate may be searched.

[0082] In this embodiment, bead width was used as an example of a welding result, and wire feed speed (and welding speed) was used as an example of welding conditions (parameters) to obtain a desired bead width. However, the invention is not limited to this. For example, welding results may include bead height, weld bead, penetration, spatter amount, cycle time, undercut, bead stability, arc stability, and instantaneous arc start rate. Welding conditions may include parameters such as wire feed speed, welding current, welding voltage, welding speed, temperature, and time.

[0083] Here, bead stability refers to the average of the standard deviations of one end of the formed bead and the standard deviations of the other end. In other words, the smaller the bead stability, the less unevenness is found on both sides of the formed bead.

[0084] Arc stability is expressed as the sum of the average of the standard deviations of the positions of one end of the arc and the other end of the arc, with the wire position as the center (origin), and the absolute value of the sum of the coordinates of one end of the arc and the other end of the arc at each time interval, divided by the welding time. In other words, the smaller the arc stability, the less the arc is considered to be scattered on both sides with the wire position as the center.

[0085] Furthermore, although the welding condition search device 100 is described in this embodiment as a separate device connected (wired or wirelessly) to the welding machine 10, the functions of the welding condition search device 100 may be incorporated into the welding machine 10 to form an integrated device (welding machine).

[0086] The embodiments described above are provided to facilitate understanding of the present invention and are not intended to limit its interpretation. The elements, arrangement, materials, conditions, shapes, and sizes of the embodiments are not limited to those exemplified and can be modified as appropriate. Furthermore, it is possible to partially substitute or combine the configurations shown in different embodiments. [Explanation of symbols]

[0087] 10...Welding machine, 100...Welding condition search device, 110...Reception unit, 120...Calculation unit (machine learning model), 130...Storage unit, 140...Output unit, 150...Communication unit, 160...Selection unit, M100...Welding condition search method, S101~S112...Each step of welding condition search method M100

Claims

1. A welding condition search device for searching for welding conditions that can obtain desired welding results, A receiving unit that receives the desired welding result, A machine learning model that searches for welding conditions that yield the desired welding result using welding conditions and corresponding welding results, The system includes an output unit that outputs welding conditions explored by the machine learning model, The machine learning model uses at least one of the following: real data including welding conditions and welding results when welding is actually performed under those conditions; simulation data including welding conditions generated using a simulation model and welding results for those conditions; and reference data including welding conditions in different environments and welding results when welding is actually performed under those conditions. Welding condition search device.

2. When using the aforementioned simulation data, the degree of data variability is added. The welding condition search device according to claim 1.

3. The degree of variation is calculated based on the standard deviation of multiple actual data points. The welding condition search device according to claim 2.

4. The system further includes a selection unit that allows the user to choose whether to use the actual data, the simulation data, or both the actual data and the simulation data. The welding condition search device according to claim 1.

5. When the machine learning model uses both the real data and the simulation data, the influence of the simulation data is adjusted. The welding condition search device according to claim 4.

6. The influence of the simulation data is adjusted according to the similarity between the actual data and the simulation data. The welding condition search device according to claim 5.

7. A welding condition search method performed by a welding condition search device that searches for welding conditions that can obtain a desired welding result, A reception step for receiving the desired welding result, A data storage step involves storing data of welding conditions and corresponding welding results in memory. A welding condition search step in which a machine learning model is used to search for welding conditions that can obtain the desired welding result based on the data stored in the memory, The output step includes outputting the welding conditions found in the welding condition search step, The machine learning model uses at least one of the following: real data including welding conditions and welding results when welding is actually performed under those conditions; simulation data including welding conditions generated using a simulation model and welding results for those conditions; and reference data including welding conditions in different environments and welding results when welding is actually performed under those conditions. A method for searching for welding conditions.

Images