Defect generation prediction method and defect generation prediction device

Abstract

A modeling plan support method uses a mathematical model to establish a relationship between input information including each item of a material of a modeled object, a welding condition of a deposited bead, and a welding track, and output information including defect information of the modeled object in a case where the modeled object is formed by layering modeling under the condition of the input information. A database is created using the mathematical model, the defect information of the modeled object is retrieved and calculated in the database, and the defect information is prompted. The items of the input information each have a plurality of input sub-items different from each other. The output information has a plurality of individual defect information corresponding to the input sub-items, respectively. In generating the mathematical model, the input sub-items of the input information are related to the individual defect information, respectively, by the mathematical model.

Description

TECHNICAL FIELD

[0001] The present application relates to a defect occurrence prediction method and a defect occurrence prediction device when manufacturing a shaped object using a deposited weld. BACKGROUND

[0002] In recent years, the demand for part manufacturing based on additive manufacturing using a 3D printer has increased, and research and development toward practical use of shaping using a metal material is being promoted. A 3D printer that performs additive manufacturing of a metal material, for example, uses a heat source such as a laser or an electric arc to melt and solidify a metal powder body or a metal wire, and stacks a welded metal (a deposited weld) to manufacture a shaped object of a desired shape.

[0003] However, in additive manufacturing using a metal material, material properties such as a metal structure and hardness vary depending on manufacturing conditions. Therefore, the properties of the metal material that constitutes the shaped object can significantly vary from expected properties. In the conventional welding technology, therefore, the properties of the shaped object in the case where the shaped object is manufactured under specified manufacturing conditions are predicted based on insights obtained empirically, trial and error, and the like, and the manufacturing conditions are adjusted in such a way that a desired shape and properties are obtained.

[0004] Also, in Patent Literature 1, it is disclosed that, in order to materialize the use of information based on the above-described experience and trial and error on a computer, for example, a test cross-sectional image of a welded object and a test welded object are prepared, and in a process of judging the suitability of each specification of the welded object such as strength, ductility, hardness, toughness, and granular structure based on these, a case of using machine learning is disclosed. In addition, in Patent Literature 2, a technique is disclosed in which the good / bad of a shaped object is predicted based on the curve characteristics of welding current, welding voltage, filler material feeding speed, and the like.

[0005] PRIOR ART DOCUMENTS

[0006] PATENT LITERATURE

[0007] Patent Literature 1: Japanese Patent Application Laid-Open No. 2019-5809

[0008] Patent Literature 2: Japanese Patent Application Laid-Open No. 2019-162666 SUMMARY

[0009] PROBLEMS TO BE SOLVED BY THE INVENTION

[0010] However, since the additive manufacturing process is more complex than the simple welding process, it is considered that the prediction of properties based on the material of the shaped object manufactured by additive manufacturing is difficult. In addition, in the manufacturing method based on additive manufacturing, the degree of freedom of the manufacturing conditions is very high, and the combinations of the properties of the shaped object are diverse, and a large amount of calculation processing is required at the time of prediction of properties.

[0011] In particular, regarding the discovery of defects in the layered structure, since the shape of the molded article manufactured by the layered molding is complex, there are cases where it is difficult to apply contact-type internal inspection such as ultrasonic flaw detection. In addition, regarding a molded article of a large size, it is difficult to apply a flaw detection test based on X-rays. There is a problem in that the inspection of the molded article, including the discovery of defects after molding, is difficult. Therefore, a method of predicting defects simply without using a measurement method such as ultrasonic flaw detection or X-ray flaw detection is sought.

[0012] Therefore, an object of the present application is to provide a defect occurrence prediction method and a defect occurrence prediction device that can efficiently predict the occurrence of defects in a molded article with less effort and support the making of a more appropriate molding plan for the molded article.

[0013] Solution to the problem

[0014] The present application is constituted by the following structure.

[0015] (1) A defect occurrence prediction method that predicts the occurrence of defects when a molded article is manufactured by stacking a deposited bead layer formed by melting and solidifying a filler material supplied from a welding head in a desired shape, wherein

[0016] The defect occurrence prediction method includes the following steps:

[0017] generating a mathematical model that relates input information including each item of a material of the molded article, a welding condition, and a welding track to output information including defect information of the molded article in a case where layered molding is performed under the conditions of the items of the input information;

[0018] making a database that indicates the correspondence relationship of the input information and the output information using the mathematical model;

[0019] inputting each item of the material of the molded article, the welding condition, and the welding track to the database, and retrieving and calculating the defect information of the molded article in the database; and

[0020] presenting the calculated defect information of the molded article,

[0021] the items of the input information each have a plurality of input sub-items that are different from each other,

[0022] the output information has a plurality of individual defect information corresponding to the input sub-items,

[0023] in the step of generating the mathematical model, the input sub-items of the input information are related to the individual defect information by the mathematical model.

[0024] (2) A defect occurrence prediction method of predicting defect occurrence when a molded object is manufactured by causing a deposited bead layer modeled by melting and solidifying filler material supplied from a welding head to be stacked in a desired shape,

[0025] The defect occurrence prediction method includes:

[0026] a first mathematical model that establishes a relationship between input information including each item of material of the molded object, welding conditions, and a welding track, and intermediate output information including a temperature history of the molded object, a characteristic amount of a molten pool shape at the time of deposition bead formation, and information of a bead height or a bead width of the deposition bead in a case where the molded object is modeled under conditions of each item of the input information, and a second mathematical model that establishes a relationship between the intermediate output information and output information including defect information of the molded object;

[0027] a database that indicates a correspondence relationship between the input information and the output information is created using the first mathematical model and the second mathematical model;

[0028] each item of the material of the molded object, the welding conditions, and the welding track is input to the database, and defect information of the molded object is searched for and calculated in the database; and

[0029] the calculated defect information of the molded object is presented,

[0030] the items of the input information each have a plurality of input sub-items that are different from each other,

[0031] the intermediate output information has individual intermediate values corresponding to the input sub-items,

[0032] the output information has a plurality of individual defect information corresponding to the individual intermediate values,

[0033] in the process of generating the first mathematical model and the second mathematical model, the input sub-items are related to the individual intermediate values by the first mathematical model, and the individual intermediate values are related to the individual defect information by the second mathematical model.

[0034] (3) A defect occurrence prediction device of predicting defect occurrence when a molded object is manufactured by causing a deposited bead formed by melting and solidifying filler material supplied from a welding head to be stacked in a desired shape,

[0035] The defect occurrence prediction device includes:

[0036] a mathematical model generation unit that generates a mathematical model in which input information including each item of material, welding condition, and welding track of the molded article is related to output information including defect information of the molded article in a case where the molded article is manufactured by performing layering molding under conditions of the input information;

[0037] a database creation unit that creates a database indicating a correspondence relationship between the input information and the output information using the mathematical model;

[0038] a search unit that searches the database using each item of material, welding condition, and welding track of the molded article input to the database, and calculates defect information of the molded article; and

[0039] an output unit that prompts the calculated defect information of the molded article,

[0040] the items of the input information each have a plurality of input sub-items different from each other,

[0041] the output information has a plurality of individual defect information corresponding to the input sub-items,

[0042] in the process of generating the mathematical model, the input sub-items of the input information are related to the individual defect information by the mathematical model.

[0043] (4) A defect occurrence prediction device that predicts defect occurrence when a molded article is manufactured by layering a deposited bead formed by melting and solidifying a filler material supplied from a welding head to a desired shape, wherein

[0044] the defect occurrence prediction device includes:

[0045] a mathematical model generation unit that generates a first mathematical model in which input information including each item of material, welding condition, and welding track of the molded article is related to intermediate output information including a temperature history of the molded article, a characteristic amount of a molten pool shape at the time of formation of the deposited bead, or information of a bead height or a bead width of the deposited bead in a case where the molded article is manufactured by performing layering molding under conditions of the input information, and a second mathematical model in which the intermediate output information is related to output information including defect information of the molded article;

[0046] a database creation unit that creates a database indicating a correspondence relationship between the input information and the output information using the first mathematical model and the second mathematical model;

[0047] a search unit that searches the database using each item of the material, welding condition, and welding track of the molded article input to the database, and calculates defect information of the molded article; and

[0048] an output unit that presents the calculated defect information of the molded article,

[0049] the items of the input information each have a plurality of input sub-items different from each other,

[0050] the intermediate output information has individual intermediate values corresponding to the input sub-items,

[0051] the output information has a plurality of individual defect information corresponding to the individual intermediate values,

[0052] in the process of generating the first mathematical model and the second mathematical model, the input sub-items are related to the individual intermediate values by the first mathematical model, and the individual intermediate values are related to the individual defect information by the second mathematical model.

[0053] Effects of the Invention

[0054] According to the present invention, defects are simply predicted without using a complicated measurement method, and a more appropriate molding plan of a molded article can be made. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 is a whole structure diagram of a molding system that manufactures a molded article.

[0056] Figure 2 is a schematic structure block diagram of a robot control device.

[0057] Figure 3 is a schematic structure block diagram of a molding control device.

[0058] Figure 4 is an explanatory diagram showing a manufacturing step of a molding program in which a layered molding is performed.

[0059] Figure 5 is an explanatory diagram showing a step of constructing a database.

[0060] Figure 6 is a flowchart showing a step of constructing a database.

[0061] Figure 7 is a flowchart showing a manufacturing step of an initial database used in a first step.

[0062] Figure 8 (A) of FIG. 20 is an explanatory diagram showing a case where input information is related to output information using a mathematical model,Figure 8 (B) is an explanatory diagram showing a database in which input information and output information are associated.

[0063] Figure 9 (B) is an explanatory diagram showing a database in which input information and output information are associated.

[0064] Figure 10 (B) is an explanatory diagram showing a database in which input information and output information are associated.

[0065] Figure 11 (B) is an explanatory diagram showing a database in which input information and output information are associated.

[0066] Figure 12 (B) is an explanatory diagram showing a database in which input information and output information are associated.

[0067] Figure 13 (B) is an explanatory diagram showing a database in which input information and output information are associated.

[0068] Figure 14 (B) is an explanatory diagram showing a database in which input information and output information are associated. Figure 14 (B) is an explanatory diagram showing a database in which input information and output information are associated. Figure 14 (B) is an explanatory diagram showing a database in which input information and output information are associated.

[0069] Figure 15 (B) is an explanatory diagram showing a database in which input information and output information are associated.

[0070] Figure 16 (B) is an explanatory diagram showing a database in which input information and output information are associated.

[0071] Figure 17 (B) is an explanatory diagram showing a database in which input information and output information are associated. Figure 17 (B) is an explanatory diagram showing a database in which input information and output information are associated. Figure 17 (B) is an explanatory diagram showing a database in which input information and output information are associated.

[0072] Figure 18 (B) is an explanatory diagram showing a database in which input information and output information are associated. Figure 18 (B) is an explanatory diagram showing a database in which input information and output information are associated. Figure 18 (B) is an explanatory diagram showing a database in which input information and output information are associated.

[0073] Figure 19 is an explanatory diagram showing a case where a plurality of databases that establish a relationship between input information and output information are selectively used. DETAILED DESCRIPTION

[0074] Hereinafter, an embodiment of the present application will be described in detail with reference to the drawings.

[0075] Here, a case where a deposited bead layer built up by melting and solidifying a filler material supplied from a welding head is molded into a desired shape by a molding device is described as an example, but the molding method and the structure of the molding device are not limited thereto. For example, other molding methods such as a powder sintering build-up molding method can be employed.

[0076] <Structure of molding system>

[0077] Figure 1 is a whole structure diagram of a molding system that manufactures a molded article. The molding system 100 of this structure is provided with a molding device 11 and a molding control device 13 that controls the molding device 11.

[0078] The molding device 11 is provided with a welding robot 17 that is provided with a welding head having a welding torch 15 at a front end shaft, a robot control device 21 that drives the welding robot 17, a filler material supply part 23 that supplies a filler material (welding wire) M to the welding torch 15, and a welding power source 25 that supplies a welding current.

[0079] (Molding device)

[0080] The welding robot 17 is a multi-joint robot, and the welding torch 15 at the front end of the welding robot 17 is supported so as to be continuously supplied with the filler material M. The position and the posture of the welding torch 15 can be arbitrarily set three-dimensionally within the range of the degrees of freedom of the robot arm, in accordance with an instruction from the robot control device 21.

[0081] A shape sensor 32 and a temperature sensor 30 that move integrally with the welding torch 15 are provided at the front end shaft of the welding robot 17.

[0082] The shape sensor 32 is a non-contact sensor that measures the shape of the deposited bead 28 formed, and measures the shape around the bead formation position as needed. The measurement by the shape sensor 32 can be performed simultaneously with the formation of the deposited bead, or can be performed at different timings before and after the bead formation. As the shape sensor 32, a laser sensor that detects a three-dimensional shape by the position of reflected light of laser light irradiated, or the time from the irradiation timing to the detection of the reflected light can be used. The shape sensor 32 is not limited to the laser sensor, and can be a sensor of other detection methods.

[0083] The temperature sensor 30 is a contact type sensor such as a radiation thermometer, a thermal imager, and the like, and detects the temperature (temperature distribution) of an arbitrary position of the molded article.

[0084] The welding torch 15 is a welding torch for gas shielded arc welding having a protection nozzle not shown and supplying a protection gas from the protection nozzle. As the arc welding method, any one of a consumable electrode type such as a covered arc welding or a carbon dioxide gas arc welding, a non-consumable electrode type such as a TIG welding or a plasma arc welding can be used, and is appropriately selected depending on the layered molded article to be produced.

[0085] For example, in the case of the consumable electrode type, an electrode conducting nozzle is arranged inside the protection nozzle, and a filler material M to which a melting current is supplied is held in the electrode conducting nozzle. The welding torch 15 generates an arc from a front end of the filler material M under a protection gas atmosphere while holding the filler material M.

[0086] The filler material supply section 23 is provided with a reel 29 on which the filler material M is wound, and a welding wire feed sensor 31 that measures the feed amount of the filler material M sent from the reel 29 to the feeding mechanism and the welding torch 15. The filler material M is sent from the filler material supply section 23 to a feeding mechanism (not shown) mounted to a robot arm or the like, and is fed to the welding torch 15 by being fed forward and backward by the feeding mechanism as necessary.

[0087] As the filler material M, all commercially available welding wires can be used. For example, a welding wire defined by a MAG welding and a MIG welding solid wire for mild steel, high tensile steel, and low temperature steel (JIS Z3312), an arc welding flux-cored wire for mild steel, high tensile steel, and low temperature steel (JIS Z 3313), or the like can be used. Also, a filler material M of aluminum, aluminum alloy, nickel, nickel-based alloy, or the like can be used depending on the characteristics to be obtained.

[0088] Also, when the filler material M continuously fed as described above is melted and solidified by the arc, a deposited bead 28 that is a molten solid of the filler material M is formed on the base plate 27. The base plate 27 is a metal plate such as a steel plate, but is not limited to a plate shape, and can be a block, a rod, a cylinder, or the like, or other shapes.

[0089] (Robot control device)

[0090] The robot control device 21 drives the welding robot 17 to move the welding torch 15, and melts the continuously supplied filler material M by the welding current and the welding voltage from the welding power source 25.

[0091] Figure 2 is a schematic configuration block diagram of the robot control device 21.

[0092] The robot control device 21 is a computer device configured to include an input / output interface 33, a storage unit 35, and an operation panel 37.

[0093] The welding robot 17, welding power supply 25, and shaping control device 13 are connected to the input / output interface 33. The storage unit 35 stores various information, including the driver program described later. The storage unit 35 is composed of storage devices such as ROM, RAM, hard disks, SSDs (Solid State Drives), CDs, DVDs, and various memory cards, and is capable of inputting and outputting various types of information. The operation panel 37 can be an input console or other information input unit, or it can be a teaching input terminal (teaching programmer) for the welding robot 17.

[0094] The modeling control device 13 sends a modeling program corresponding to the model to be created to the robot control device 21. The modeling program consists of multiple command codes and is created based on appropriate algorithms according to various conditions such as the shape data (CAD data, etc.), material, and heat input of the model.

[0095] The robot control unit 21 executes the modeling program stored in the storage unit 35, driving the welding robot 17, the filler material supply unit 23, and the welding power source 25 to form weld beads 28 according to the modeling program. Specifically, the robot control unit 21 drives the welding robot 19 to move the welding torch 15 along the track (welding track) set in the modeling program, and drives the filler material supply unit 23 and the welding power source 25 according to the set welding conditions, using an electric arc to melt and solidify the filler material M at the tip of the welding torch 15. This forms weld beads 28 on the base plate 27. By repeatedly forming weld bead layers adjacent to each other and stacking another weld bead layer on top of that layer, a desired three-dimensional shape is created.

[0096] The styling control device 13 also functions as a defect prediction device that provides defect information during the generation of the styling program. It should be noted that the styling control device 13 can also be configured separately from the styling device 11 and remotely connected to the styling device 11 via a network, communication mechanism, storage medium, or the like. The styling program can be generated by other devices besides the styling control device 13 and transmitted via communication.

[0097] (Generation of the modeling process)

[0098] Next, the structure of the styling control device 13 and the specific steps until the styling control device 13 generates the styling program will be explained.

[0099] Figure 3This is a schematic structural block diagram of the styling control device 13.

[0100] The modeling control device 13 is a computer device that is the same as the robot control device 21, and is configured to include a CPU 41, a storage unit 43, an input / output interface 45, an input unit 47, and an output unit 49.

[0101] The storage unit 43 is composed of ROM, which is a non-volatile storage area, and RAM, which is a volatile storage area. The aforementioned shape sensor 32, temperature sensor 30, filler material supply unit 23 with welding wire feed sensor 31, welding power supply 25, robot control device 21, input unit 47, and output unit 49 are connected to the input / output interface 45.

[0102] The input section 47 includes input devices such as a keyboard and mouse, and the output section 49 includes display devices such as a monitor or output terminals that transmit output signals.

[0103] In addition, the styling control device 13 also includes a basic information table 51 (described in detail later), a mathematical model generation unit 53, a database creation unit 55, a styling planning unit 57, and a retrieval unit 59. Each of these components operates according to instructions from the CPU 41, thus performing its respective function.

[0104] Figure 4 This is an explanatory diagram showing the steps involved in creating a layered design.

[0105] First, through the operator from Figure 3 The input unit 47 of the modeling control device 13 shows data on the material, shape, and welding conditions of the model to be manufactured. Based on the input data, the modeling control device 13 creates a modeling plan in a way that achieves the desired characteristics of the model. For example, it generates a model based on shape data, divides the generated model into layers according to a specified weld bead height, and determines various conditions such as the material of the weld beads, weld bead width, and weld bead formation sequence (welding path) by filling each layer with weld beads. Various methods exist for determining these welding paths, etc., and this method is not limited to it.

[0106] Next, referring to a pre-prepared database 61 that represents the correspondence between various manufacturing conditions and defect information of the manufactured model, defects that may occur in the model when it is manufactured using the prepared model plan are predicted. Here, defects are broadly defined as internal characteristics of the model that include defect information.

[0107] If it is predicted that defects may occur in the predicted design, a new design plan is created by adjusting the various manufacturing conditions described above. Furthermore, if the design based on the new design plan does not contain defects, or if the characteristics including defects meet the desired characteristics, a design program corresponding to that design plan is created. The resulting design program is then sent to… Figure 1 The robot control device 21 shown. The robot control device 21 executes the sent modeling program and stacks the modeling objects to form a model.

[0108] In the styling planning support method and apparatus of the present invention, the database 61 used for predicting and determining whether the styling object achieves the desired characteristics is efficiently constructed with less effort through the styling program. As a result, accurate and rapid determination of styling plans can be made, thus enabling support in a way that allows for the creation of more appropriate styling plans according to the site conditions.

[0109] <Example of First Database Structure>

[0110] Next, the construction method of the aforementioned database 61 will be explained.

[0111] Figure 5 This is an explanatory diagram illustrating the steps of constructing database 61. Here, a mathematical model is used to establish a relationship between input information, including the material of the model, welding conditions of the weld beads, and local weld tracks, and output information, including defect information of the model layered under the conditions of the input information. This relationship-establishing process is repeated through machine learning, and based on the obtained mathematical model, a database is created... Figure 4 The predictions and judgments shown refer to database 61.

[0112] Specifically, the shaping control device 13 generates a shaping plan based on the input data such as the material, shape, and welding conditions of the shaped object. Then, through the next first and second steps, it calculates the characteristic values ​​of the shaped object if it is created using this shaping plan.

[0113] In the first step, the styling control device 13 refers to an initial database 63 that pre-registers the relationship between styling plans and characteristic values, and predicts the occurrence of defects in the styling object created by layering styling plans.

[0114] In the second step, the modeling control device 13 drives the robot control device 21 as per the modeling plan, while the modeling device 11 stacks the model. Test samples are cut from the model and tested (observed) to measure mechanical strength, metal structure, etc.

[0115] The predicted characteristics of the objects based on the same design plan are compared with the experimental results to generate a mathematical model 62 in a way that minimizes the difference between the two. Database 61 is then created using this mathematical model 62. The creation of the initial database 63 and database 61 mentioned above is handled by... Figure 3 The database creation unit 55 shown can perform the work, but it can also be performed by a device other than the modeling control device 13.

[0116] Here, we will explain the process of using machine learning to generate a mathematical model 62 from the prediction results of the first step based on the characteristics of the initial database 63 and the experimental results of the second step, and then using the mathematical model 62 to create a database 61.

[0117] Figure 6 This is a flowchart illustrating the steps involved in constructing database 61.

[0118] First, shape data (based on 3D-CAD shape data) is created by determining the object to be manufactured (S11). A design plan is then created based on the shape data of the object (S12). The design plan includes multiple slice data obtained by determining the specified stacking direction axis of the object model and performing layer division, the shape of the weld beads in each slice data, and the welding conditions for forming the weld beads.

[0119] Next, based on the created design plan, defects in the design are predicted in the first step (S13). The predicted defects include, for example, the presence or absence of defects, the location of defects, the size of defects, and the presence or absence of sputtering.

[0120] Defects in the model are predicted using an initial database 63. The initial database 63 is based on basic information table 51, which represents the correspondence between various manufacturing conditions and the test results of the models produced, derived from experience and insights gained from past models. Figure 3 To make it.

[0121] Figure 7 This is a flowchart illustrating the steps involved in creating the initial database 63 used in the first step.

[0122] First, information on parameters for the database (e.g., the number of passes, the order of formation of weld beads (welding path), the cross-sectional shape of weld beads, etc.) is extracted from the pre-prepared basic information table 51, and this is prepared as learning data (S21).

[0123] Next, a relationship is established between the prepared learning data and the characteristic values ​​of the corresponding model objects through an initial mathematical model (S22). That is, by repeatedly performing machine learning on multiple learning data sets and their corresponding characteristic values ​​of the model objects, an initial mathematical model representing the relationship between the two is generated. Here, "mathematical model" refers to a quantitative representation of the characteristics of the model objects, allowing for the standardization of their behavior and the computational simulation of their properties. In other words, the mathematical model is a computational model created based on experimental data sets collected in experiments and associated using a prescribed algorithm. This computational model can be optimized in a way that closely matches the experimental data by assuming a prescribed function, or it can be created by incorporating input and output information through machine learning. Examples of specific algorithms include support vector machines, neural networks, or random forests.

[0124] Then, using the generated initial mathematical model, the characteristic values ​​of the model corresponding to multiple learning data are predicted, and these predicted values ​​are registered in the table components of the initial database 63 (S23) in correspondence with the learning data. In this way, the initial database 63 is created.

[0125] On the other hand, in the second step ( Figure 6 In sections S14 to S16), the sculpted object is created based on the established sculpting plan. That is, the sculpting plan section 57 ( Figure 3 Create a styling program (S14) corresponding to the styling plan, and drive the process through the execution of this styling program. Figure 1 The shaping device 11 shown shapes the object (S15). Then, a test sample is cut out from the obtained object, and various properties of the test sample are tested (S16).

[0126] Then, the predicted characteristics of the model obtained in the first step are compared with the experimental results obtained in the second step (S17). If the difference between the predicted and experimental results is large, [further steps are taken]. Figure 5 The mathematical model 62 shown (equivalent to the initial mathematical model used in the creation of the initial database 63) is corrected in a way that reduces the difference between the two (S18). That is, the experimental results for the input information are used as training data, and the mathematical model 62 is subjected to machine learning in a way that makes the prediction results close to the experimental results. It should be noted that if the difference between the prediction results and the experimental results is small, the mathematical model 62 is not corrected, but machine learning can still be performed in a way that helps improve the accuracy of the mathematical model 62. In this way, the mathematical model 62 becomes a learned model that has performed machine learning on the relationship between the input information and the output information.

[0127] Then, using mathematical model 62, obtained by further machine learning of the initial mathematical model, the characteristic values ​​(output information) of the model corresponding to any number of conditions (input information) are predicted. The set conditions are associated with the predicted characteristic values, which become the table components of database 61. In this way, based on the initial database 63, mathematical model 62 is used to make corrections, and database 61 is constructed in which the prediction results for specific conditions are in good agreement with the experimental results (S19).

[0128] In this way, by using mathematical model 62 to predict output information based on multiple input information, the missing parts of the experimental results can be supplemented, thereby simply increasing the amount of information in database 61 and improving the accuracy of prediction.

[0129] Next, the specific method for constructing database 61 using mathematical model 62 will be explained in more detail.

[0130] Figure 8 (A) is an explanatory diagram illustrating the use of mathematical models to establish a relationship between input and output information. Figure 8 (B) is an explanatory diagram showing a database that establishes a relationship between input and output information.

[0131] As one of the input information, we will take the filling material that becomes the material in the model as an example for explanation. For example... Figure 8 As shown in (A), multiple filler materials A, B, C, ... can be selected as filler materials. When a model is made using each filler material A, B, C, ..., defects that may occur in the resulting model are called defect information A for filler material A, defect information B for filler material B, defect information C for filler material C, ...

[0132] In this case, a mathematical model A is used to establish a relationship between the defect information A of the model with filling material A, the defect information B of the model with filling material B is used to use mathematical model B, and the defect information C of the model with filling material C is used to use mathematical model C.

[0133] Therefore, as Figure 8 As shown in (B), the database 61 is constructed by establishing relationships between each material individually: linking filler material A with defect information A, linking filler material B with defect information B, and linking filler material C with defect information C. Accordingly, the mathematical model performs machine learning separately according to the type of filler material to determine the defect information, thus allowing for accurate and detailed setting of defect information corresponding to the characteristics of the filler material. Therefore, the accuracy of defect prediction can be improved.

[0134] As for the type of filler material, it can be identified by its trade name, such as MG-51T or MG-S63B (solid welding wire manufactured by Kobe Steel), or it can be distinguished by the composition of the filler material (such as carbon content).

[0135] The example above establishes a relationship between defect information based on the type of filling material, but in actual input information, there are many other items.

[0136] Figure 9 This is an explanatory diagram illustrating how a mathematical model is used to establish a relationship between input information, which consists of multiple items, and output information.

[0137] As input information, at least the material of the object, welding conditions, and local welding paths are considered. As for the material of the weldment, in addition to the aforementioned filler material, a base plate 27 for forming the weld bead can be cited as an example. Figure 1 Components, such as structural parts not shown, that are joined with weld beads to form the constituent elements of the object.

[0138] Examples of welding conditions include at least one or a combination of the following: welding current, welding voltage, welding speed, welding track spacing, inter-pass time, target position of the weld head, welding posture of the weld head, and filler material supply rate. Here, the target position of the weld head is the position of the welding torch tip used to position the torch tip at the welding location; the welding posture of the weld head is the angle of inclination between the vertical axis and the torch axis, and the circumferential angle about the vertical axis in the direction of torch inclination. Furthermore, the welding track spacing is the distance between adjacent welding tracks, and the inter-pass time represents the time between the welding pass of one welding track and the welding pass of the next welding track.

[0139] The aforementioned interpass time affects the metal structure of the resulting weld bead.

[0140] During the formation of weld beads, when the molten mild steel filler material is rapidly cooled, it becomes a mixed structure dominated by bainite. Furthermore, when the molten mild steel filler material solidifies naturally, it becomes a structure including coarse ferrite, pearlite, and bainite. In the case of multilayer weld beads, when these structures are heated above the phase transformation point of ferrite due to the subsequent weld beads being layered, the pearlite and bainite transform into ferrite, and the coarse ferrite becomes a finer structure.

[0141] By adjusting the inter-pass time, for example, to keep the inter-pass temperature within the range of 200°C to 550°C, the next layer of weld beads is stacked while controlling the inter-pass time and heat input. Similarly, when stacking subsequent weld beads, the weld beads are heated above the phase transformation point of the ferrite. In this case, a homogenized microstructure composed of fine ferrite phases with an average grain size of 10 μm or less is obtained. Such weld beads have high hardness (e.g., 130–180 Hv on a Vickers hardness scale), good mechanical strength, and a generally uniform hardness with small deviations.

[0142] On the other hand, when depositing weld beads in layers, if the inter-pass temperature is less than 200°C, even if the weld beads are heated due to the subsequent layers, the phase transformation point of the ferrite will not be exceeded, and a homogenized microstructure consisting of fine ferrite phases cannot be obtained. For example, in the early stages of molding, due to heat dissipation from the base plate 27, the inter-pass temperature when depositing weld beads in layers is less than 200°C. In this case, the weld beads in the early stages of molding become a mixed microstructure dominated by bainite. Furthermore, when the inter-pass temperature exceeds 550°C, the weld beads are heated due to the layering of the next layer, resulting in flattening and dripping of the weld beads, preventing them from being stacked into the desired shape. Moreover, the weld beads in the later stages of molding (the uppermost layer of the molded object) are not layered by the next layer and are not reheated, so the molten filler material remains in a naturally solidified state, that is, it retains a microstructure including coarse ferrite, pearlite, and bainite.

[0143] Thus, the microstructure of the weld bead formed by the inter-pass time changes, thereby altering the defects produced in the molded object. The above describes the effect of inter-pass time on defects in the molded object; however, it is equally clear that other parameters also influence the characteristics of the molded object.

[0144] A local welding track is a welding track for the element shape obtained by cutting out a part of the shape of the object, and it is a welding track used to shape the simple shape when a complex shape is decomposed into a simple shape (element shape) (this point will be described later). The information of each welding track includes the number of weld beads formed, the order in which the weld beads are formed, and the cross-sectional shape of the weld beads.

[0145] Here, the materials, welding conditions, and local welding tracks of the aforementioned shaped parts are referred to as "projects," and the filler materials A, B, C, ... for each project, as well as the welding current, welding voltage, welding speed, ..., element shape, number of passes, number of passes, ... are referred to as "input sub-projects."

[0146] By dividing the input information into multiple sub-items, the range of input is limited. That is, as input data, it is possible to prevent settings beyond the sub-items while remaining within the recommended range for things like the welding robot 17 in the molding device 11, or the recommended conditions for using the filler material. This prevents malfunctions caused by device defects or materials, and avoids prompting for inappropriate conditions.

[0147] like Figure 9 As shown, the input information consists of multiple items, including the material of the object, welding conditions, and local welding tracks. Each item has multiple different input sub-items.

[0148] Furthermore, when the content of an item is represented by numerical values, the input data for each item can be divided into multiple sub-items, and the representative value corresponding to each sub-item can be determined as the input data. The representative value of the input sub-item can be, for example, the central value, upper limit value, or lower limit value within the input sub-item, as long as it represents the input sub-item.

[0149] Furthermore, the range of input data need not be the same as the input information that is the actual data. The database creation unit 55 inputs the input data of each input sub-item determined in this way into the mathematical model created by the mathematical model generation unit 53, and obtains the output data of each input sub-item.

[0150] Furthermore, the output data of the output information is the output value of the mathematical model corresponding to the input data. Here, for example, the input data for each input sub-item can be represented by the central value of the data range determined as the input sub-item. The database creation unit 55 creates a database by accumulating and storing the correspondence between the output data obtained by inputting the input data of each input sub-item into the mathematical model and the input data, according to the input sub-items. That is, the database creation unit 55 creates a database by accumulating the output data of each input sub-item for which the range of each item of the input data is divided into multiple intervals.

[0151] In this way, the input sub-items of each project establish a relationship with the defect information of the object through a mathematical model. It is also possible to learn multiple mathematical models through a complete combination, as described above. However, it is preferable to aggregate them into approximately one mathematical model based on specific welding conditions, welding track patterns, etc., and then adjust the parameters accordingly based on this model. Examples of such adjustments include transfer learning, which enables efficient learning by allowing the learned model to function in other areas. This reduces the amount of learning data and computational load.

[0152] Next, the shape of the element and the welding track of each element shape when determining the local welding track will be explained together with a specific example of the model.

[0153] Figure 10 This is an explanatory diagram showing the process of dividing the shape of the object to be made into multiple element shapes and determining the welding tracks for each element shape.

[0154] Here, the example of the design 65 is a design comprising a main body 65A, a first protrusion 65B connected to one side of the main body 65A, and a second protrusion 65C connected to the other side of the main body 65A. When this design 65 is divided into simple shape elements, it becomes a cylindrical first protrusion 65B, a cubic main body 65A, and a U-shaped second protrusion 65C. The division of the shape elements can be done manually or by matching them with a pre-registered pattern of simple shapes, etc.

[0155] For each segmented element shape, a welding track representing the formation sequence of the weld bead is calculated separately. That is, the welding track is determined according to the segmented element shape. The welding track for each element shape can be designed and calculated sequentially each time it is segmented into an element shape. However, since the element shape is a simple shape, multiple simple shape welding tracks (reference welding tracks) can also be pre-registered in the element database, and the welding track corresponding to the element shape can be determined by referring to the element database.

[0156] For example, if the element is cylindrical, the cylinder is divided into multiple layers, and each of these layers becomes a reference welding track for determining the formation sequence (torch path) of the weld bead. By applying this reference welding track to the first protrusion 65B, the shaping steps, i.e., welding track B, when shaping the first protrusion 65B using the weld bead can be easily determined.

[0157] Similarly, for the main body 65A and the second protrusion 65C, reference welding tracks with similar shapes are retrieved from the element database. The welding track A for the main body 65A and the welding track C for the second protrusion 65C can be easily determined from these reference welding tracks. In this way, even complex shapes can be treated as a collection of simple shapes by dividing them into element shapes, simplifying the design plan. Furthermore, the ability to predict defect occurrence based on element shapes contributes to determining the location (nearby location) of defects within the overall shape.

[0158] Figure 11 This is a flowchart illustrating the steps involved in creating a design plan by breaking down the shape of an object into its constituent shapes.

[0159] When the shape data of the object to be created is sent to Figure 3 When the modeling control device 13 shown inputs data to its modeling planning unit 57 (S31), the modeling planning unit 57 decomposes the model created based on the shape data into multiple element shapes (S32). Then, it retrieves information such as reference welding tracks and welding conditions corresponding to each decomposed element shape by searching a pre-prepared element database (not shown) (S33). The element database used here includes information on reference welding tracks and welding conditions set corresponding to the element shapes, and this information is pre-registered in the element database.

[0160] For each extracted reference welding track, a welding track is calculated that is applied to the corresponding element shape (S34), and a modeling plan for the entire model is created by combining it with the welding conditions (S35).

[0161] The created design plan became Figure 7 The modeling plan for S12 is shown. Therefore, the modeling plan, which is decomposed into element shapes and determines the welding path and welding conditions according to the element shapes, also generates a mathematical model in the same way as described above, and constructs a database 61 to support... Figure 4 The design plan shown.

[0162] It should be noted that, regarding welding conditions, information related to welding conditions can be easily collected based on the modeling device 11, the wire feed sensor 31, the shape sensor 32, and the drive signals of the welding robot 17. These values ​​can also be used when the shape of the model is controlled by feedback as needed.

[0163] <Example of Second Database Structure>

[0164] Next, we will explain the case where intermediate output information is set in addition to the input and output information of the aforementioned database 61.

[0165] Figure 12 This is an explanatory diagram illustrating how a mathematical model is used to establish relationships between input information, intermediate output information, and output information.

[0166] The input information, including the material of the model, welding conditions, and local welding tracks, each has multiple input sub-items. The intermediate output information establishes relationships between these sub-items and their combinations through a separate first mathematical model. Furthermore, the input sub-items of the intermediate output information establish relationships with the input sub-items of the output information through a second mathematical model.

[0167] Here, the intermediate output information includes the temperature history of the object, characteristic parameters of the weld pool shape during weld bead formation, and information on the weld bead height or width. It may also include characteristic parameters of the arc shape, the presence or absence of sputtering, etc. For information on the weld pool, please refer to [reference needed]. Figure 15 This will be discussed later. Additionally, regarding the characteristics of the weld bead, please refer to... Figure 16 to Figure 18 To be described later.

[0168] First, the temperature history of the sculpture is explained. Temperature history has a significant impact on the internal state (e.g., material) of the sculpture's layered structure, including defects. Therefore, analyzing the temperature history to infer the internal state of the sculpture plays a role in predicting the likelihood and extent (e.g., defect size) of defects.

[0169] When the filler material and other materials used in the molding are heated according to welding conditions and melt and solidify along a specified welding path, the temperature history of the resulting molding (weld deposit) varies depending on the conditions described above. Therefore, the mechanical strength, metal structure, and other properties of the resulting molding also vary depending on the conditions, thus affecting the generation of defects.

[0170] When estimating defect information of a molded object, even when it is difficult to directly deduce the defect information from the various items (conditions) of the input information, it is sometimes easier to deduce the defect information if the corresponding temperature history, characteristic quantities of the weld pool shape, weld height, or weld width can be obtained. Therefore, when establishing a relationship between the input information and the defect information of the molded object as output information, a two-stage relationship is established: first, the relationship between each item of the input information and the intermediate output information; then, the relationship between the intermediate output information and the defect information of the molded object as output information (broadly speaking, including the internal state of the molded object, including defects). Based on this, a higher accuracy in establishing relationships and estimating defects can be achieved compared to directly establishing a relationship between the input and output information.

[0171] By setting temperature history, molten pool shape characteristics, weld height, or weld width as intermediate outputs, representative features of the shaping conditions can be aggregated, making it easy to establish a correspondence between each feature and defect information. Furthermore, the intermediate output information mentioned here can also be easily collected during the shaping process by setting shape and temperature sensors.

[0172] Figure 13 It is a graph showing the temperature history at a specific location of the weld bead formed during the molding process. For example... Figure 13As shown, after the repeated layers of weld beads melt and solidify to form weld beads, they are subjected to heat input due to the weld beads stacked on top of them, undergoing repeated heating (and sometimes melting if they are adjacent layers) and cooling. The temperature history peaks decrease as the weld beads are further away from a specific location from the layer above them.

[0173] When the melting point Tw of the weld bead is set to 1534°C, which is the melting point of iron (carbon steel), and the phase transformation point Tt (the A1 phase transformation point of carbon steel) is set to 723°C, the material of the solidified weld bead is roughly determined by the temperature history within the range above the phase transformation point Tt and below the melting point Tw. That is, in the laminated molding process, repeated heating and cooling occur, but the factors affecting the microstructure of the molded object are the temperature history within the aforementioned range Aw. Therefore, by extracting the characteristic quantity of the temperature history within the range (checking the temperature range) Aw above the phase transformation point Tt and below the melting point Tw, the characteristics of the molded object can be predicted.

[0174] For example, ignoring Figure 13 The peaks shown are those exceeding the melting point Tw and those below the phase transition point Tt. Furthermore, the temperatures of the lowest-temperature side maxima Pk2 closest to the phase transition point Tt and the second highest-temperature side maxima Pk1 closest to the phase transition point Tt are extracted from the peaks in the inspection temperature range Aw above the phase transition point Tt and below the melting point Tw. These temperatures of the high-temperature side maxima Pk1 and the low-temperature side maxima Pk2 are set as characteristic quantities of the temperature history, i.e., intermediate output information.

[0175] Figure 14 This is a graph showing the differences in cooling characteristics when weld beads are formed with different amounts of heat input. Figure 14 (A) is a graph showing the temperature change characteristics under conditions of relatively high heat input. Figure 14 (B) is a graph showing the temperature change characteristics under conditions of relatively low heat input.

[0176] like Figure 14 As shown in (A), even if the heat input is increased in the order of Qb, Qc, Qd from Qa, the time to cool to 350°C remains almost unchanged, becoming approximately 15 seconds in this case (see Pend). On the other hand, as Figure 14As shown in (B), when the heat input is relatively low, the cooling time is set to approximately 15 seconds to cool to approximately 300°C (refer to Pend). That is, the higher the heat input, the slower the cooling rate, and the lower the heat input, the faster the cooling rate. Therefore, the cooling rate depends on the heat input, and if the temperature of the low-temperature side maximum point Pk2 is known, the microstructure of the weld bead can be predicted. Furthermore, by combining the temperature of the low-temperature side maximum point Pk2 with the temperature of the high-temperature side maximum point Pk1 to predict microstructure and other characteristics, the prediction accuracy is improved compared to predicting only one temperature.

[0177] Thus, if the temperature history, which determines the material of the weld bead, can be determined based on the aforementioned characteristic quantities, then the material of the formed weld bead can be predicted with relatively high accuracy using this temperature history, and it can be applied to determine the likelihood of defects. Therefore, by setting the intermediate processing information as a determining factor of the material of the molded part, a relationship is established between the input information and the intermediate output information using a first mathematical model, and a relationship is established between the intermediate output information and the output information using a second mathematical model. As a result, compared to directly establishing a relationship between the input information and the output information, it is expected that the relationship between the two can be established more accurately.

[0178] Regarding this temperature history, a temperature sensor 30 can also be used in the design. Figure 1 The temperature sensor 30 can be used to monitor the temperature of the object and obtain temperature data at a specified location. The temperature sensor 30 can also work in conjunction with the shape sensor 32 to detect temperature. That is, the shape sensor 32 detects the shape of the object, and the temperature sensor 30 detects the temperature at a specific location on the object.

[0179] Alternatively, temperature simulations can be performed based on the type of filler material and welding conditions. The following shows an example of a basic formula used in temperature simulation.

[0180] [Mathematical Expression 1]

[0181] t+Δt {H}= t {H}-Δt〔C〕〔K〕 t {T}-Δt〔C〕 t {F}+Δt t {Q}…(1)

[0182] The basic formula (1) is an analytical formula for heat transfer based on the so-called Finite Element Method (FEM). The parameters of the basic formula (1) are as follows.

[0183] H: enthalpy

[0184] C: Reciprocal of the node volume

[0185] K: Thermal conductivity matrix

[0186] F: Heat flux

[0187] Q: Volumetric heating

[0188] Therefore, by setting enthalpy as an unknown variable, nonlinear phenomena such as latent heat release can be calculated with good accuracy. It should be noted that the heat input during welding is a parameter for volumetric heating or heat flux.

[0189] In the basic equation (1) of the aforementioned three-dimensional heat conduction equation, the amount of heat input during shaping (welding) can also be assigned to the welding area in accordance with the welding speed. In addition, when the weld bead is short, the heat input of the entire weld bead can also be assigned.

[0190] Next, we will explain the case where the intermediate output information is a feature quantity related to the molten pool.

[0191] Figure 15 This is an explanatory diagram showing the molten pool formed at the front end of the filler material. Figure 15 In the image, the arc center 71, the front end of the filling material 73, the front end of the molten pool 75, the left end of the molten pool 77, and the right end of the molten pool 79 are shown as features (image feature information). However, it is not limited to this; for example, the width and shape of the arc can also be set as features and inferred and extracted.

[0192] As an example, when a part of the welded object has a complex-shaped protrusion, or when there is a wall of the workpiece nearby, the presence of these features restricts the movement of the welding torch, or the generated arc is pulled towards the protrusion or wall, causing a change in the arc direction. This can result in variations in characteristic quantities such as the area or position of the molten pool deviating from the normal range, or the distance between the tip of the filler material and the center of the arc increasing. By using machine learning to analyze the tendency of such variations, it is possible to predict (estimate) characteristic quantities related to the molten pool and the arc based on input information (at least one of the material of the welded object, welding conditions, trajectory plan, etc.).

[0193] Furthermore, sometimes characteristic parameters of the molten pool or arc are considered to deviate from the normal range. For example, there is a possibility that only a portion of the molten metal is heated intensely and melts and solidifies earlier, while the temperature of other portions rises more slowly. Due to the temperature rise, the previously solidified portions remelt unevenly, resulting in insufficient bond strength. In such cases, if machine learning can be applied to these tendencies, it is possible to predict (estimate) the likelihood, extent, and location of defects based on characteristic parameters such as the molten pool.

[0194] Next, we will explain the cases where the intermediate output information is related to the characteristics of the weld bead, such as weld bead height and weld bead width.

[0195] Figure 16 This is a cross-sectional view showing an example of a defect that occurs in a stacked structure of a 1-layer, 2-column weld bead.

[0196] In the layered structure shown here, a first weld bead 81 is formed on the base plate 27, and a second weld bead 83 is formed overlapping a portion of the first weld bead 81.

[0197] A cavity (nest) 85 is formed between the first weld bead 81 and the second weld bead 83. Although not shown in the figure, it is believed that slag, spatter, etc., also adhere around each weld bead. Slag and spatter form on the outer side of the laminated structure, but in the case of weld bead stacking, they can sometimes be trapped between weld beads and enter the interior of the laminated structure. In this case, it affects properties such as joint strength, durability, and mechanical strength. Thus, the characteristics of adjacent weld beads are closely related to defects within the weld bead, and may provide useful information for predicting defect formation.

[0198] Such internal defects in welds cannot be detected simply by photographing the weld's external shape; ultrasonic or X-ray flaw detection is necessary. However, according to this prediction method, by applying information from the captured images to a model shape function, the weld can be decomposed into constituent weld beads. By understanding the characteristics of each decomposed weld bead, defect occurrence can be estimated with high accuracy. The model shape function will be explained in detail here.

[0199] Figure 17 (A) is a cross-sectional view based on the simulation results of the model shape function of a 1-layer, 2-column stacked weld bead. Figure 17 (B) is shown Figure 17 A cross-sectional view of the shape of each weld bead in (A).

[0200] exist Figure 17 In (A), BD1 is the overall cross-sectional shape of a 1-layer, 2-column weld bead. This cross-sectional shape is determined by... Figure 17 The shape of the first weld bead BD1-1 shown in (B) is combined with the shape of the second weld bead BD1-2. That is, the overall cross-sectional shape of the weld bead can be decomposed into the first weld bead and the second weld bead, and the defect information of each weld bead can be estimated separately.

[0201] Figure 18 (A) is a cross-sectional view based on the simulation results of the model shape function of a 2-layer, 2-column stacked weld bead. Figure 18 (B) is shown Figure 18 A cross-sectional view of the shape of each weld bead in (A).

[0202] exist Figure 18 In (A), BD2 is the overall cross-sectional shape of the 2-layer, 2-column weld bead. This cross-sectional shape is achieved by... Figure 18 The shape of the first weld bead BD2-1 and the shape of the second weld bead BD2-2 shown in (B) are combined with the shape of the third weld bead BD2-3 and the shape of the fourth weld bead BD2-4 of the second layer. It should be noted that the third and fourth weld beads of the second layer are respectively shown as the weld bead shapes stacked up after the formation of the first and second weld beads of the first layer. That is, the overall cross-sectional shape of the deposited weld bead can be decomposed into the first to fourth weld beads, and the defect information of each weld bead can be estimated separately.

[0203] In this way, by applying the model shape function to the overall appearance shape information of the weld bead obtained from the shape detection results, the shape information such as weld bead height and weld bead width in each layer can be determined. The determined shape information is pre-registered in the database as parameters of the learning model, and a correspondence is established with the defect information obtained from the experimental results, thereby enabling more accurate defect prediction that also takes into account the characteristics of each weld bead.

[0204] <Examples of other database structures>

[0205] Figure 19 This is an illustrative diagram showing a scenario where multiple databases are selectively used to establish relationships between input and output information.

[0206] The aforementioned first database structure example uses mathematical model I to establish a relationship between input information and output information, and uses this mathematical model I to construct database DB1 (the aforementioned database 61).

[0207] In addition, the second database structure example uses mathematical model IIa to establish a relationship between input information and intermediate output information, and uses mathematical model IIb to establish a relationship between intermediate output information and output information, and uses these mathematical models IIa and IIb to construct database DB2 (the aforementioned database 61).

[0208] Furthermore, the constructed database DB1 and database DB2 were compared, and the database with more accurate output information for the input information was selected as the database. Figure 4 The database 61 shown is used. For example, comparing databases DB1 and DB2 using sets of input and output information (training data) with known correspondences, the accuracy of the output for the input is determined.

[0209] Accordingly, by constructing multiple databases and selectively using higher-precision databases, the prediction accuracy of defects (characteristics that broadly include defects) of the model can be improved, enabling the creation of more appropriate modeling plans.

[0210] Thus, the present invention is not limited to the above-described embodiments. Combining the various structures of the embodiments with each other, as well as making changes and applications based on the description and well-known techniques by those skilled in the art, are also intended by the present invention and are included within the scope of the claimed protection.

[0211] As stated above, the following matters are disclosed in this specification.

[0212] (1) A defect generation prediction method, which predicts the generation of defects when manufacturing a molded object by stacking weld beads formed by the melting and solidification of filler material supplied from the welding joint into a desired shape, wherein...

[0213] The defect generation prediction method includes the following steps:

[0214] A mathematical model is generated that establishes a relationship between input information, including the material, welding conditions, and welding track of the object, and output information, including the defect information of the object when it is layered under the conditions of the input information.

[0215] A database representing the correspondence between the input information and the output information is created using the mathematical model.

[0216] The material, welding conditions, and welding track information of the object are input into the database; the defect information of the object is retrieved and calculated from the database; and...

[0217] The obtained defect information of the object is displayed.

[0218] The input information items each have multiple distinct input sub-items.

[0219] The output information contains multiple individual defect information corresponding to the input sub-item.

[0220] In the process of generating the mathematical model, the mathematical model is used to establish relationships between the input sub-items of the input information and the individual defect information.

[0221] Based on this defect prediction method, by constructing a database related to defect generation, it is possible to predict the occurrence of defects in the model before lamination. Furthermore, even for models manufactured through lamination and with complex shapes that are difficult to inspect using contact methods such as ultrasonic testing, or for models that are too large to be inspected using X-ray-based testing, defects can be easily predicted.

[0222] (2) A defect prediction method, which predicts the defects that occur when manufacturing a molded object by causing the weld bead layer formed by the melting and solidification of the filler material supplied from the weld joint to be of a desired shape, wherein...

[0223] The defect generation prediction method includes the following steps:

[0224] A first mathematical model is generated that establishes a relationship between input information, including the material, welding conditions, and welding track of the object, and intermediate output information, including the temperature history of the object, the characteristic quantity of the molten pool shape when the weld bead is formed, and the information of the weld bead height or weld bead width when the object is layered under the conditions of the input information; and a second mathematical model is generated that establishes a relationship between the intermediate output information and output information, including the defect information of the object.

[0225] A database representing the correspondence between the input information and the output information is created using the first mathematical model and the second mathematical model;

[0226] The material, welding conditions, and welding track information of the object are input into the database; the defect information of the object is retrieved and calculated from the database; and...

[0227] The obtained defect information of the object is displayed.

[0228] The input information items each have multiple distinct input sub-items.

[0229] The intermediate output information has a unique intermediate value corresponding to the input sub-item.

[0230] The output information contains multiple individual defect information corresponding to the individual intermediate value.

[0231] In the process of generating the first mathematical model and the second mathematical model, the first mathematical model is used to establish a relationship between the input sub-items and the individual intermediate values, and the second mathematical model is used to establish a relationship between the individual intermediate values ​​and the individual defect information.

[0232] According to this defect prediction method, by setting intermediate outputs such as temperature history, molten pool characteristics, weld height, or weld width, representative features of the shaping conditions can be easily correlated with defect information. Furthermore, the intermediate output information described here can be easily collected during the shaping process using shape sensors and temperature sensors, allowing for simple implementation. For example, the predicted temperature history can be used to determine whether the internal state of the shape is prone to defects or has a higher probability of defect generation than usual. Moreover, based on characteristics directly related to welding, such as the molten pool, the likelihood of welding deviating from the normal range can be estimated, determining the presence and extent of defects. Additionally, by using model shape functions to estimate the shape information of the weld beads in each layer, the location and size of defects can be predicted, taking into account the shape, size, flattening, irregularity, and structural instability of the weld beads. Furthermore, the presence or absence of sputtering can also be considered when predicting these defects. In this way, we can make high-precision predictions of defects without significantly increasing the learning load by making simple use of the database.

[0233] Furthermore, combining multiple intermediate output information to predict defect occurrence can improve prediction accuracy. Additionally, increasing the variety of intermediate output information can further enhance prediction accuracy.

[0234] (3) A defect generation prediction method according to (1) or (2), wherein,

[0235] The information about the material in the input information includes information about the type of filling material.

[0236] According to this defect prediction method, there is a tendency for the viscosity of the weld bead during melting to differ depending on the type of filler material, and consequently, the cross-sectional shape of the weld bead also differs. Therefore, by creating separate mathematical models according to the type of filler material, defect prediction suitable for each filler material can be implemented. This also facilitates the efficient planning and formulation of welding conditions and track schedules suitable for each filler material.

[0237] (4) The defect generation prediction method according to any one of (1) to (3), wherein,

[0238] The welding conditions information in the input information includes at least one of the following, or a combination thereof: welding current, welding voltage, welding speed, spacing width between adjacent welding tracks, time between passes from one welding track to another among the plurality of welding tracks, target position of the weld head, welding posture of the weld head, and feed rate of the filler material.

[0239] According to this defect prediction method, by focusing on various aspects of welding conditions, such as limiting it to a few key aspects or cutting in various patterns, intermediate output information and defect generation information corresponding to the selected aspects can be output. That is, a wide variety of variations can be obtained when predicting defect generation. This also facilitates the expansion of the data set. Furthermore, since all indicators can be monitored during the modeling process, data collection is easy.

[0240] (5) The defect generation prediction method according to any one of (1) to (4), wherein,

[0241] The information about the welding track in the input information includes at least one of the following: the number of passes forming the weld bead, the number of passes, the formation order of the weld bead, and the cross-sectional shape of the weld bead.

[0242] Based on this defect, a prediction method can be generated, which can achieve the same effect as (4) above.

[0243] (6) The defect generation prediction method according to any one of (1) to (5), wherein,

[0244] The welding track includes a local welding track corresponding to the element shape obtained by cutting out a part of the overall shape of the object.

[0245] Based on this defect prediction method, by dividing the entire object into multiple element shapes and setting a trajectory plan for each element shape, complex objects can be represented as combinations of simple element shapes. This allows for the prediction of intermediate output information corresponding to the input information and defect information as the final output information with reduced computational load. In other words, by establishing a database linked to local trajectory plans, even complex objects can be easily predicted by appropriately decomposing the shapes, thus improving versatility.

[0246] Furthermore, by aggregating defect prediction information based on local track plans, it is possible to predict defects in the overall form. The decomposition of each element's shape can be done manually or by cutting it out through pattern matching with pre-registered simple shapes.

[0247] The cut elements can be layered by determining the specified stacking direction axis, and each layer can be divided into specified weld bead units to create the tracks for each weld bead. This allows for the prediction of defects based on the element shapes, thus easily determining the location of defects in the object. Furthermore, by pre-creating local weld tracks corresponding to the shapes of each element in various variations, even complex shapes can be efficiently predicted, including defects, without complicated processing, enabling the creation of appropriate design plans.

[0248] (7) The defect generation prediction method according to any one of (1) to (6), wherein,

[0249] The output information includes at least one of the following: defect size, defect shape, sputtering amount, and presence or absence of defect.

[0250] Based on this defect prediction method, important indicators related to the quality of the design can be set as output information, and database-based retrieval, i.e., defect estimation, can be performed. Therefore, before actual design, defect prediction using a database can be conducted to obtain defect information that may become important indicators related to the quality of the design. In other words, by setting these important indicators as output information, the quality of the design can be predicted efficiently and effectively.

[0251] (8) The defect generation prediction method according to any one of (1) to (7), wherein,

[0252] The mathematical model is a learned model obtained by machine learning the relationship between the input information and the output information.

[0253] Based on this deficiency, a prediction method is developed that allows for the simple construction of a mathematical model as long as data exists, with the model's accuracy improved by expanding the data with each layer of modeling. By utilizing machine learning to construct the mathematical model, the lack of experimental data can be supplemented, thereby improving prediction accuracy. Furthermore, the data corresponding to the input and output can be collected from basic shapes such as wall forms and block forms, making it easy to prepare machine learning data and resulting in a structure with excellent feasibility.

[0254] (9) The defect generation prediction method according to any one of (1) to (8), wherein,

[0255] The input range of the input information is limited to a range defined based on pre-set conditions.

[0256] According to this defect prediction method, by limiting the input range in a way that does not deviate from the recommended range of the driving molding device and the recommended conditions for the use of filler materials, it is possible to avoid input conditions that are prone to causing failures due to device malfunctions or materials. Furthermore, for example, by setting cases that deviate from the recommended range of the welding machine and the recommended conditions for the use of filler materials as the search targets, the computational load on the mathematical model can also be reduced.

[0257] (10) A defect generation prediction device that predicts defects that will occur when a molten weld bead formed by the melting and solidification of filler material supplied from a welding head is stacked into a desired shape to manufacture a molded object, wherein...

[0258] The defect generation prediction device includes:

[0259] The mathematical model generation unit generates a mathematical model that establishes a relationship between input information, including the material of the model, welding conditions, and welding track, and output information, including the defect information of the model when it is layered under the conditions of the input information.

[0260] The database creation department uses the mathematical model to create a database that represents the correspondence between the input information and the output information;

[0261] The retrieval unit searches the database using items such as the material, welding conditions, and welding path of the model, which are input into the database, and retrieves defect information of the model; and

[0262] The output unit displays the defect information of the obtained model.

[0263] The input information items each have multiple distinct input sub-items.

[0264] The output information contains multiple individual defect information corresponding to the input sub-item.

[0265] In the process of generating the mathematical model, the mathematical model is used to establish relationships between the input sub-items of the input information and the individual defect information.

[0266] Based on this defect prediction device, by constructing a database related to defect generation, it is possible to predict the generation of defects in the model before lamination. Furthermore, even for models manufactured through lamination and with complex shapes that are difficult to inspect using contact methods such as ultrasonic testing, or for models that are too large to be inspected using X-ray-based testing, defects can be easily predicted.

[0267] (11) A defect generation prediction device that predicts defects that will occur when a molten weld bead formed by the melting and solidification of filler material supplied from a welding head is stacked into a desired shape to manufacture a molded object, wherein...

[0268] The defect generation prediction device includes:

[0269] The mathematical model generation unit generates a first mathematical model that establishes a relationship between input information, including the material of the model, welding conditions, and welding track, and intermediate output information, including the temperature history of the model under the conditions of the layered modeling under the conditions of the input information, the characteristic quantity of the molten pool shape when the weld bead is formed, and the information of the weld bead height or weld bead width, and a second mathematical model that establishes a relationship between the intermediate output information and output information, including the defect information of the model.

[0270] The database creation department uses the first mathematical model and the second mathematical model to create a database representing the correspondence between the input information and the output information;

[0271] The retrieval unit searches the database using items such as the material, welding conditions, and welding path of the model, which are input into the database, and retrieves defect information of the model; and

[0272] The output unit displays the defect information of the obtained model.

[0273] The input information items each have multiple distinct input sub-items.

[0274] The intermediate output information has a unique intermediate value corresponding to the input sub-item.

[0275] The output information contains multiple individual defect information corresponding to the individual intermediate value.

[0276] In the process of generating the first mathematical model and the second mathematical model, the first mathematical model is used to establish a relationship between the input sub-items and the individual intermediate values, and the second mathematical model is used to establish a relationship between the individual intermediate values ​​and the individual defect information.

[0277] According to this defect prediction device, by setting intermediate outputs such as temperature history, molten pool characteristics, weld height, or weld bead, representative features of the shaping conditions can be easily aggregated and correlated with defect information. Furthermore, the intermediate output information described here can be easily collected during the shaping process using shape sensors and temperature sensors, thus enabling simple implementation. For example, the predicted temperature history can be used to determine whether the internal state of the shape is prone to defects or has a higher probability of defect generation than usual. Moreover, based on characteristics directly related to welding, such as the molten pool, the likelihood of welding deviating from the normal range can be estimated, thus determining the presence and extent of defects. Additionally, by using model shape functions to estimate the shape information of the weld beads in each layer, the location and size of defects can be predicted, taking into account the shape, size, flattening, irregularity of the weld bead shape, and degree of structural instability. Furthermore, the presence or absence of sputtering can also be considered when predicting these defects. In this way, we can make high-precision predictions of defects without significantly increasing the learning load by making simple use of the database.

[0278] Furthermore, combining multiple intermediate output information to predict defect occurrence can improve prediction accuracy. Additionally, increasing the variety of intermediate output information can further enhance prediction accuracy.

[0279] This application is based on Japanese Patent Application No. 2020-123859, filed on July 20, 2020, the contents of which are incorporated herein by reference.

[0280] Explanation of reference numerals in the attached figures

[0281] 11. Art installations

[0282] 13. Modeling control device

[0283] 15 Welding torches

[0284] 17 Welding Robots

[0285] 21 Robot Control Device

[0286] 23 Filler Material Supply Department

[0287] 25 Welding Power Supply

[0288] 27 Base Plate

[0289] 29 scrolls

[0290] 30 Temperature Sensor

[0291] 31 Welding wire feed sensor

[0292] 32 Shape Sensors

[0293] 33 Input / Output Interfaces

[0294] 35 Storage Department

[0295] 37. Control Panel

[0296] 39 weld bead layers

[0297] 41 CPU

[0298] 43 Storage Department

[0299] 45 Input / Output Interfaces

[0300] 47 Input Section

[0301] 49 Output Section

[0302] 51 Basic Information Table

[0303] 53 Mathematical Model Generation Department

[0304] 55 Database Production Department

[0305] 57 Styling Planning Department

[0306] 59. Retrieval Department

[0307] 61 Database

[0308] 63 Initial Database

[0309] 65. Shaped objects

[0310] 65A Main Body (Element Shape)

[0311] 65B First protrusion (element shape)

[0312] 65C Second protrusion (element shape).

Claims

1. A defect generation prediction method, which predicts the generation of defects when manufacturing a molded object by stacking weld beads formed by the melting and solidification of filler material supplied from the weld joint into a desired shape, wherein, The defect generation prediction method includes the following steps: A mathematical model is generated that establishes a relationship between input information, including the material, welding conditions, and welding track of the object, and output information, including the defect information of the object when it is layered under the conditions of the input information. A database representing the correspondence between the input information and the output information is created using the mathematical model. The material, welding conditions, and welding track of the object are input into the database, and the defect information of the object is retrieved and obtained from the database. as well as The obtained defect information of the object is displayed. The input information items each have multiple distinct input sub-items. The output information contains multiple individual defect information corresponding to the input sub-item. In the process of generating the mathematical model, the mathematical model is used to establish relationships between the input sub-items of the input information and the individual defect information.

2. The defect generation prediction method according to claim 1, wherein, The information about the material in the input information includes information about the type of filling material.

3. The defect generation prediction method according to claim 1, wherein, The welding conditions information in the input information includes at least one of the following, or a combination thereof: welding current, welding voltage, welding speed, spacing width between adjacent welding tracks, time between passes from one welding track to another among the plurality of welding tracks, target position of the weld head, welding posture of the weld head, and feed rate of the filler material.

4. The defect generation prediction method according to claim 2, wherein, The welding conditions information in the input information includes at least one of the following, or a combination thereof: welding current, welding voltage, welding speed, spacing width between adjacent welding tracks, time between passes from one welding track to another among the plurality of welding tracks, target position of the weld head, welding posture of the weld head, and feed rate of the filler material.

5. The defect generation prediction method according to any one of claims 1 to 4, wherein, The mathematical model is a learned model obtained by machine learning the relationship between the input information and the output information.

6. A defect generation prediction method, which predicts the generation of defects when manufacturing a shaped object by molding a weld bead layer formed by the melting and solidification of filler material supplied from the weld joint into a desired shape, wherein, The defect generation prediction method includes the following steps: A first mathematical model is generated that establishes a relationship between input information, including the material, welding conditions, and welding track of the object, and intermediate output information, including the temperature history of the object, the characteristic quantity of the molten pool shape when the weld bead is formed, and the information of the weld bead height or weld bead width when the object is layered under the conditions of the input information; and a second mathematical model is generated that establishes a relationship between the intermediate output information and output information, including the defect information of the object. A database representing the correspondence between the input information and the output information is created using the first mathematical model and the second mathematical model; The material, welding conditions, and welding track of the object are input into the database, and the defect information of the object is retrieved and obtained from the database. as well as The obtained defect information of the object is displayed. The input information items each have multiple distinct input sub-items. The intermediate output information has a unique intermediate value corresponding to the input sub-item. The output information contains multiple individual defect information corresponding to the individual intermediate value. In the process of generating the first mathematical model and the second mathematical model, the first mathematical model is used to establish a relationship between the input sub-items and the individual intermediate values, and the second mathematical model is used to establish a relationship between the individual intermediate values ​​and the individual defect information.

7. The defect generation prediction method according to claim 6, wherein, The information about the material in the input information includes information about the type of filling material.

8. The defect generation prediction method according to claim 6, wherein, The welding conditions information in the input information includes at least one of the following, or a combination thereof: welding current, welding voltage, welding speed, spacing width between adjacent welding tracks, time between passes from one welding track to another among the plurality of welding tracks, target position of the weld head, welding posture of the weld head, and feed rate of the filler material.

9. The defect generation prediction method according to claim 7, wherein, The welding conditions information in the input information includes at least one of the following, or a combination thereof: welding current, welding voltage, welding speed, spacing width between adjacent welding tracks, time between passes from one welding track to another among the plurality of welding tracks, target position of the weld head, welding posture of the weld head, and feed rate of the filler material.

10. The defect generation prediction method according to any one of claims 6 to 9, wherein, The first mathematical model and the second mathematical model are learned models obtained by machine learning of the relationship between the input information and the output information.

11. The defect generation prediction method according to any one of claims 1 to 4, 6 to 9, wherein, The information about the welding track in the input information includes at least one of the following: the number of passes forming the weld bead, the number of passes, the formation order of the weld bead, and the cross-sectional shape of the weld bead.

12. The defect generation prediction method according to any one of claims 1 to 4, 6 to 9, wherein, The welding track includes a local welding track corresponding to the element shape obtained by cutting out a part of the overall shape of the object.

13. The defect generation prediction method according to claim 11, wherein, The welding track includes a local welding track corresponding to the element shape obtained by cutting out a part of the overall shape of the object.

14. The defect generation prediction method according to any one of claims 1 to 4, 6 to 9, wherein, The output information includes at least one of the following: defect size, defect shape, sputtering amount, and presence or absence of defect.

15. The defect generation prediction method according to claim 11, wherein, The output information includes at least one of the following: defect size, defect shape, sputtering amount, and presence or absence of defect.

16. The defect generation prediction method according to claim 12, wherein, The output information includes at least one of the following: defect size, defect shape, sputtering amount, and presence or absence of defect.

17. The defect generation prediction method according to claim 13, wherein, The output information includes at least one of the following: defect size, defect shape, sputtering amount, and presence or absence of defect.

18. The defect generation prediction method according to any one of claims 1 to 4, 6 to 9, wherein, The input range of the input information is limited to a range defined based on pre-set conditions.

19. A defect generation prediction device that predicts defects that will occur when manufacturing a molded object by stacking weld beads formed by the melting and solidification of filler material supplied from a welding joint into a desired shape, wherein... The defect generation prediction device includes: The mathematical model generation unit generates a mathematical model that establishes a relationship between input information, including the material of the model, welding conditions, and welding track, and output information, including the defect information of the model when it is layered under the conditions of the input information. The database creation department uses the mathematical model to create a database that represents the correspondence between the input information and the output information; The retrieval unit searches the database using items such as the material, welding conditions, and welding path of the model, which are input into the database, and retrieves defect information of the model; and The output unit displays the defect information of the obtained model. The input information items each have multiple distinct input sub-items. The output information contains multiple individual defect information corresponding to the input sub-item. In the process of generating the mathematical model, the mathematical model is used to establish relationships between the input sub-items of the input information and the individual defect information.

20. A defect generation prediction device that predicts defects that will occur when manufacturing a molded object by stacking weld beads formed by the melting and solidification of filler material supplied from a welding joint into a desired shape, wherein... The defect generation prediction device includes: The mathematical model generation unit generates a first mathematical model that establishes a relationship between input information, including the material of the model, welding conditions, and welding track, and intermediate output information, including the temperature history of the model under the conditions of the layered modeling under the conditions of the input information, the characteristic quantity of the molten pool shape when the weld bead is formed, and the information of the weld bead height or weld bead width, and a second mathematical model that establishes a relationship between the intermediate output information and output information, including the defect information of the model. The database creation department uses the first mathematical model and the second mathematical model to create a database representing the correspondence between the input information and the output information; The retrieval unit searches the database using items such as the material, welding conditions, and welding path of the model, which are input into the database, and retrieves defect information of the model; and The output unit displays the defect information of the obtained model. The input information items each have multiple distinct input sub-items. The intermediate output information has a unique intermediate value corresponding to the input sub-item. The output information contains multiple individual defect information corresponding to the individual intermediate value. In the process of generating the first mathematical model and the second mathematical model, the first mathematical model is used to establish a relationship between the input sub-items and the individual intermediate values, and the second mathematical model is used to establish a relationship between the individual intermediate values ​​and the individual defect information.

Images