Identification device, identification method, and identification program

Abstract

This invention provides an identification device, an identification method, and an identification program that enable easy identification of welding conditions in actual operational situations where welding conditions are checked. [Solution] The identification device, identification method, and identification program use a controller connected to an input unit into which response data regarding the welded area is input. The controller determines the identification data corresponding to the response data based on a model that outputs identification data indicating whether or not the welding condition of interest is occurring at the welded area in response to the input based on the response data. The model is generated based on training data consisting of training response data regarding the training welded area and training identification data indicating whether or not the welding condition of interest is occurring at the training welded area.

Description

Technical Field

[0001] The present disclosure relates to an identification device, an identification method, and an identification program.

Background Art

[0002] Patent Document 1 discloses a technique for predicting the size of a defect using a plurality of learned models that take welding parameters related to welding as input and output parameters indicating the size of a defect related to welding.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] According to the technique described in Patent Document 1, since it is necessary to pre-acquire welding parameters for the target welding, it is impossible to predict the size of a defect for a welding performed without acquiring the welding parameters. In addition, it is impossible to identify various welding states other than the size of the defect. Moreover, a large number of welding tests and confirmation of welding states are required for generating the learned models. Therefore, the technique described in Patent Document 1 has a problem that it is impossible to easily realize the identification of welding states in the actual operation scene for confirming welding states.

[0005] The present disclosure has been made in view of the above problems. The object is to provide an identification device, an identification method, and an identification program that can easily realize the identification of welding states in the actual operation scene for confirming welding states.

Means for Solving the Problems

[0006] The identification device, identification method, and identification program relating to this disclosure use a controller connected to an input unit into which response data regarding a welded area is input. The controller determines identification data corresponding to the response data based on a model that outputs identification data indicating whether or not a welding condition of interest is occurring at the welded area in response to input based on the response data. The model is generated based on training data consisting of training response data regarding a training welded area and training identification data indicating whether or not a welding condition of interest is occurring at the training welded area.

[0007] The model may be generated by decision tree analysis.

[0008] The model may be generated by machine learning.

[0009] The models may be prepared for each welding state and generated independently for each welding state.

[0010] A model may be prepared for each welding state, and the controller may determine identification data based on a combination of outputs from multiple models.

[0011] The weld condition may include at least one of the following: hot cracking, cold cracking, blowholes, pits, tungsten inclusions, slag inclusions, impurity contamination, poor penetration, poor fusion, overlap, undercuts, underfill, arc strikes, reheat cracking, stress corrosion cracking, sensitization, oxidation, and grain coarsening.

[0012] The system may further include a database for storing correspondence records linked to welding conditions, and an output unit. The controller may extract correspondence records related to welding conditions based on determined identification data and output the extracted correspondence records via the output unit. [Effects of the Invention]

[0013] According to this disclosure, an identification device, an identification method, and an identification program can be provided that enable easy identification of welding conditions in actual operational situations where welding conditions are checked. [Brief explanation of the drawing]

[0014] [Figure 1] This is a block diagram showing the configuration of an identification device according to an embodiment of this disclosure. [Figure 2] This flowchart shows the processing procedure of the identification device (during model generation). [Figure 3] This is a flowchart showing the processing procedure of the identification device (when identifying the welding condition). [Figure 4] This figure shows an example of training data. [Figure 5] This figure shows an example of a model represented by a decision tree. [Figure 6] This figure shows an example of determining classification data based on the outputs from multiple models. [Modes for carrying out the invention]

[0015] Several exemplary embodiments will be described below with reference to the drawings. Common parts in each drawing are denoted by the same reference numerals, and redundant explanations will be omitted.

[0016] [Configuration of the identification device] Figure 1 is a block diagram showing the configuration of an identification device according to an embodiment of the present disclosure. The identification device 20 comprises an input unit 21, an output unit 23, and a controller 25. The controller 25 is connected to the input unit 21 and the output unit 23 so as to be able to communicate with them. The identification device 20 may also be connected to an operation unit 10 and a database 30.

[0017] The operation unit 10 is an input device through which a user can perform operations. For example, the operation unit 10 is a keyboard, a mouse, a trackball, a touch panel, or the like. The operation unit 10 is not limited to the examples listed here. The content of the user's operation input via the operation unit 10 is transmitted to the controller 25.

[0018] For example, the user may input answer data by operating via the operation unit 10. Here, the "answer data" is data indicating features regarding the welding site. The "answer data" is data indicating which of the two choices, "yes" or "no", is given as an answer to a predetermined question. Further, the "answer data" may include data indicating that an option such as "don't know / blank / no answer" which is not involved in the generation of identification data is selected.

[0019] For example, the predetermined question is a question with the following content. · "Is there a welding defect in the center of the bead?" · "Does the defect appear on the surface of the welding site?" · "Is there a defect in the weld metal?" · "Does the shape of the defect correspond to a crack?" · "Is the defect linear?" · "Is the welding material a nickel-based alloy?" · "Is it a stainless steel material?" · "Does it appear black in the radiographic test?"

[0020] The predetermined question is not limited to the examples listed here. The predetermined question may be various questions used to confirm the "welding state" described later in an actual operation scenario.

[0021] The response data may be represented by the numerical values ​​"1" and "0" corresponding to the answers "yes" and "no" to a given question, respectively. Alternatively, the response data may be represented by the symbols "○" and "×" corresponding to "yes" and "no". In addition, the response data may include numerical values ​​or symbols corresponding to options that do not participate in the generation of identification data, such as "I don't know / blank / no answer". The method of representing the response data is not limited to the examples given herein.

[0022] In addition, the user may input training data through the operation unit 10. Here, "training data" refers to data consisting of training response data and training identification data related to the training welding area.

[0023] A training welding site is a welding site where the presence or absence of the welding condition of interest is known by a method other than the identification method of this disclosure. Training identification data is identification data that indicates whether or not the welding condition of interest is occurring in the training welding site.

[0024] For example, "welding conditions" can include "hot cracking," "cold cracking," "blowholes," "pits," "tungsten inclusions," "slag inclusions," "impurity contamination," "poor penetration," and "poor fusion."

[0025] "Hot cracking" refers to the occurrence of cracks in the high-temperature range during welding and cooling. Hot cracking is distinguished from cold cracking and reheat cracking. While it most often occurs in the final solidification zone of the weld metal, it can also occur in the heat-affected zone (HAZ) in materials such as stainless steel. Furthermore, in multi-layer welding, hot cracking can occur in areas heated by subsequent passes.

[0026] "Cold cracking" refers to a condition where cracks occur at temperatures below, for example, 200-300 degrees Celsius. While cold cracking can sometimes be caused solely by shrinkage strain (stress) at the weld site, it can also be caused by the action of hydrogen that dissolves and penetrates the steel when heated, diffusing and accumulating within it.

[0027] A "blowhole" is a void within the weld metal. For example, a blowhole occurs when gas generated or invading within the weld metal is not released into the atmosphere during solidification and becomes trapped within the weld metal.

[0028] A "pit" is a condition where gas holes generated within the weld metal are released onto the bead surface and solidify into holes. The presence or absence of gas sources, shielding conditions, and droplet transfer phenomena are closely related to the quality of the welding procedure.

[0029] "Tungsten inclusion" refers to a welding defect that occurs when welding with tungsten or a tungsten alloy electrode results in a portion of the electrode melting and mixing into the weld bead due to excessive current or other causes.

[0030] "Slag inclusion" refers to a welding defect caused by slag being mixed into the weld metal. For example, slag inclusion can occur due to poor fusion, such as insufficient melting of the weld interface.

[0031] "Impregnation" refers to a condition where welding defects occur due to the inclusion of impurities in the weld metal. For example, impurity elements such as phosphorus and sulfur, which cause solidification cracking in low-alloy steels, stainless steels, and high-alloys, have extremely small solid-liquid partition coefficients. Therefore, they tend to segregate at solidification grain boundaries and also lower the melting point. Relatively large amounts of impurities segregate in the final solidification zone or between columnar crystals of the weld metal, which tends to reduce ductility and toughness. Various elements can be cited as problematic in "impregnation," including copper, oxygen, nitrogen, phosphorus, sulfur, hydrogen, and carbon.

[0032] "Poor penetration" refers to a condition where the actual penetration is insufficient compared to the design penetration. For example, in fillet welding, a defect where the root portion remains unmelted is an example of "poor penetration." "Poor penetration" can reduce joint strength and can also lead to notches, which are a factor in brittle fracture.

[0033] "Fusion failure" refers to a condition where the weld interface does not fuse sufficiently with the base metal. For example, it can occur when the base metal and the weld metal, or the weld metals themselves, do not fuse together completely, resulting in gaps. This is more likely to occur in downward welding or horizontal multi-layer welding, particularly in the areas where the base metal is in contact with the weld, and in the overlapping areas of weld beads during split welding.

[0034] Other "welding conditions" include "overlap," "undercut," "underfill," "arc strike," "reheat cracking," "stress corrosion cracking," "sensitization," "oxidation," and "grain coarsening."

[0035] "Overlap" refers to a condition where the weld metal does not fuse with the base metal at the weld toe, resulting in an overlapping portion. For example, overlap is a condition where the weld bead toe and the base metal do not fuse well together. In fillet welding, overlap can occur due to excess molten metal sagging due to gravity.

[0036] "Undercut" refers to a condition where a groove is present at the weld toe when welding over the base metal or a previously welded surface. This can occur when the correct amount of metal is not obtained. Therefore, undercuts can be suppressed by adjusting the welding current and welding speed, or by performing weaving.

[0037] "Underfill" refers to a condition where there is a large gap between the butt joints of steel plates at the joint, resulting in insufficient melting and a weld bead thickness that is thinner than the thickness of the steel plate. When the weld bead is recessed due to "underfill," stress concentration can occur, potentially leading to fracture or cracking.

[0038] An "arc strike" is a defect caused by a momentary arc jump on the base metal. In arc welding, a faulty arc that is not melted during subsequent welding remains in the base metal and can cause cracking.

[0039] "Reheat cracking" refers to a condition where cracks occur when a welded area is reheated. Reheat cracking can occur during post-weld heat treatment for stress relief or during post-heating for purposes such as dehydrogenation. It is also sometimes called precipitation brittle cracking, SR (Stress Relief) cracking, or short-time creep cracking.

[0040] Stress corrosion cracking is a type of cracking that occurs when three factors—material, environmental, and mechanical—are present. Stress corrosion cracking can occur when a certain level of tensile stress is continuously applied in a specific combination of material and environment.

[0041] "Sensitization" refers to the process where Cr occurs at the grain boundaries in a metal that has been exposed to a specific temperature range for a long period of time. 23 This condition, where C6 precipitates and the chromium concentration near grain boundaries decreases, results in reduced corrosion resistance in corrosive environments such as hot nitric acid solutions. For example, "sensitization" is likely to occur in austenitic stainless steels such as SUS304 and SUS316.

[0042] "Oxide" refers to the oxidation of the weld metal, or the state in which metal oxides are incorporated into the weld metal. When molten metal comes into contact with air, oxygen in the air can chemically react with the molten metal, and oxygen may dissolve into the molten metal. If the cooling and solidification process is rapid, oxides may remain in the weld metal.

[0043] "Grain coarsening" refers to a condition in the heat-affected zone where the crystal grains within the weld metal become coarser due to heating. When grain coarsening occurs, the hardness and toughness of the metal can change. The region where the crystal grains coarse is present is prone to hardening and cracking.

[0044] "Welding condition" may include "poor shielding," "improper base metal dilution," or "excessive or insufficient heat input." "Welding condition" is not limited to the examples given herein.

[0045] Figure 4 shows an example of training data. In Figure 4, training data T1 to Tm are shown, and for each training data, the answer data for questions Q1 to Qn is indicated by "1" or "0". In the following, an answer data of "1" means that the answer to the question is "yes", and an answer data of "0" means that the answer to the question is "no".

[0046] For example, in training data T1, the columns for questions Q1 and Q2 are "1", meaning that the answer to questions Q1 and Q2 is "yes" for the welding area related to training data T1. In training data T1, the columns for questions Q3 and Qn are "0", meaning that the answer to questions Q3 and Qn is "no" for the welding area related to training data T1.

[0047] Furthermore, status numbers D1 to D4 are indicated for each training data. For example, if the status number D1 column is "1", it means that the "welding state" assigned to status number D1 has occurred in the training welding area related to the training data. Conversely, if the status number D1 column is "0", it means that the "welding state" assigned to status number D1 has not occurred in the training welding area related to the training data.

[0048] Figure 4 shows state numbers D1 to D4, indicating the presence or absence of four "welding states." The types of "welding states" shown for each training data are not limited to four, but may be other than these. Also, the "welding states" associated with a single training data are not limited to one of several, but may be multiple.

[0049] For example, in training data T1, the status number D1 column shows "1," meaning there is one "welding state" associated with training data T1. Similarly, in training data Tm, the status numbers D1 and D4 columns both show "1," meaning there are two "welding states" associated with training data Tm.

[0050] The database 30 stores the response data and training data. For example, the database 30 may be various types of storage devices such as a hard disk, flash memory, ROM, RAM, optical disk, or magnetic tape. The database 30 may be included in the controller 25 itself, or it may be located outside the controller 25.

[0051] The database 30 may store models generated based on training data.

[0052] In addition, the database 30 may store corresponding records linked to welding conditions. Here, "corresponding records" may be records of past work performed to correspond to the welding condition of interest, or past literature related to the welding condition of interest. By presenting the "corresponding records" to the user, knowledge about the welding conditions linked to the "corresponding records" can be conveyed to the user.

[0053] The input unit 21 receives training data used for model generation and response data regarding welding locations. For example, the input unit 21 may be connected to the operation unit 10 and the database 30. The acquired training data and response data are sent to the controller 25.

[0054] The output unit 23 outputs various types of information generated by the controller 25. For example, the output unit 23 outputs the model and identification data generated by the controller 25.

[0055] The output unit 23 may be connected to the database 30, or to a display device (not shown). The information output from the output unit 23 may be presented to the user by a display device (not shown), or it may be stored in the database 30.

[0056] The controller 25 is a general-purpose computer equipped with a CPU (Central Processing Unit), memory, and an input / output unit. The controller 25 has a computer program (identification program) installed on it that functions as the identification device 20. By executing the computer program, the controller 25 functions as one of the multiple information processing circuits (251, 253, 255) of the identification device 20.

[0057] This disclosure provides an example of implementing multiple information processing circuits (251, 253, 255) using software. However, it is also possible to configure the information processing circuits (251, 253, 255) by preparing dedicated hardware for each of the information processing operations described below. Alternatively, the multiple information processing circuits (251, 253, 255) may be configured using separate hardware.

[0058] As shown in Figure 1, the controller 25 includes a model generation unit 251, a determination unit 253, and an extraction unit 255, which are multiple information processing circuits (251, 253, 255).

[0059] Furthermore, if the controller 25 only performs the processing during model generation (Figure 2), the controller 25 may be equipped with a model generation unit 251 but not with a determination unit 253 and an extraction unit 255. Also, if the controller 25 only performs the processing during welding state identification (Figure 3), the controller 25 may be equipped with a determination unit 253 and an extraction unit 255 but not with a model generation unit 251.

[0060] The model generation unit 251 generates a model based on training data that outputs identification data indicating whether or not a welding condition of interest is present at the welded area, in response to input based on the answer data.

[0061] For example, the model generation unit 251 generates a model using decision tree analysis. More specifically, the model generation unit 251 calculates the impurity before and after splitting multiple training data sets. Splitting multiple training data sets means focusing on a single question and dividing it into training data where the column for the question of focus is "0" and training data where the column for question Q1 is "1". The question that reduces impurity the most before and after the split is then set as the branching condition in the decision tree.

[0062] For example, the "impurity" calculated by the model generation unit 251 may be either "Gini impurity" or "cross-entropy." "Impurity" is, so to speak, an indicator of "how cleanly" the training data is divided, and the "impurity" used when creating a decision tree is not limited to the examples given here.

[0063] Furthermore, the algorithms used to create decision trees are not limited to the examples mentioned above. For example, algorithms for creating decision trees include CART (Classification and Regression Trees), ID3 (Iterative Dichotomiser 3), C4.5, and CHAID (Chi-squared Automatic Interaction Detection). Decision trees may also be created using algorithms other than those listed above.

[0064] Figure 5 shows an example of a model represented by a decision tree. For example, in the model MD shown in Figure 5, questions Q3, Q4, and Q5 are set as branching conditions in the decision tree.

[0065] In Figure 5, if question Q3 is "0", it can be inferred that a welding condition with condition number D1 has occurred (the column for condition number D1 is "1"). Also, even if question Q3 is "1", if question Q5 is "1" and question Q4 is "0", it can be inferred that a welding condition with condition number D1 has occurred. In all other cases, it can be inferred that a welding condition with condition number D1 has not occurred.

[0066] Alternatively, the model generation unit 251 may generate models using machine learning. When a model is generated using machine learning, the model is composed of a neural network. The neural network includes multiple layers, and signals propagate sequentially through these layers. Each layer consists of one or more units.

[0067] The units in each layer are connected to each other, and each unit may have an activation function (e.g., a sigmoid function, a normalized linear function, a softmax function, etc.). A weighted sum is calculated based on multiple inputs to the unit, and the value of the activation function with the sum as a variable becomes the output of the unit. For example, in machine learning, the weights used to calculate the sum in each unit of a neural network are adjusted as parameters related to the model.

[0068] The model generation unit 251 may store the connection relationships between units in the neural network, and the weights used to calculate the sum for each unit in the neural network, in a memory that is not shown. Hereafter, the "connection relationships" and "weights" between units will be referred to as "parameters".

[0069] The model generation unit 251 calculates model data to be output from the model when the teacher response data included in the training data is input to the model. The model generation unit 251 also calculates the value of a loss function based on the model data and the teacher identification data. Here, the loss function is a function that evaluates the discrepancy (error) between the model data and the teacher identification data. The loss function is configured such that the larger the discrepancy between the model data and the teacher identification data, the larger the value, and the smaller the discrepancy between the model data and the teacher identification data, the smaller the value.

[0070] The model generation unit 251 trains the model based on the calculated loss function. Specifically, the model generation unit 251 adjusts the model parameters so that the discrepancy between the model data output from the model that has been input with training response data and the training identification data is reduced. For example, the model generation unit 251 changes the model parameters in a direction that reduces the loss function. By adjusting the model parameters based on a series of training data, a model that has been trained with the training data can be obtained.

[0071] The model generation unit 251 may use methods such as gradient descent or stochastic gradient descent to adjust the parameters in a direction that minimizes the loss function. Here, backpropagation may be used for gradient calculation in gradient descent or stochastic gradient descent.

[0072] Furthermore, when training a model, techniques such as regularization, which restricts the degrees of freedom of the weights during training, may be used to mitigate overfitting to the training data. Other techniques such as dropout, which probabilistically selects units in the neural network and disables the others, may also be used. In addition, to improve generalization performance, techniques such as data regularization, data standardization, and data augmentation may be used to eliminate bias in the data.

[0073] The model generation unit 251 determines whether the termination condition for model training has been met. For example, the model generation unit 251 may determine that the termination condition has been met if the loss function is at a predetermined value for all training data.

[0074] In addition, the termination condition may be determined by the number of times the model is adjusted based on the training data. The model generation unit 251 may keep track of the number of epochs, which indicates the number of times the model has been adjusted, and determine that the termination condition has been met when the number of epochs exceeds a predetermined number.

[0075] Alternatively, the model generation unit 251 may prepare a model for each welding state and generate a model independently for each welding state. For example, the model generation unit 251 may generate training data for each welding state and generate a model for each welding state based on the training data for each welding state.

[0076] Here, "training data for each welding state" refers to data created by ignoring the entries in the columns for the welding state of interest and other welding states.

[0077] For example, according to the example of training data shown in Figure 4, there are four types of welding states associated with state numbers D1 to D4. Therefore, four types of datasets will be generated as "training data for each welding state". The model generation unit 251 will then generate four models according to the number of welding states. The number of training data points included in the dataset used to generate each model will be m, corresponding to the training data T1 to Tm.

[0078] The determination unit 253 uses a model generated based on training data to determine identification data corresponding to the response data, based on the response data regarding the welding area where it is unclear whether or not the welding condition of interest is occurring. In other words, the determination unit 253 determines identification data corresponding to the response data based on a model that outputs identification data indicating whether or not the welding condition of interest is occurring at the welding area in response to an input based on response data.

[0079] For example, as shown in Figure 5, when the model is represented by a decision tree, the identification data is determined based on the questions set as branching conditions in the decision tree. For example, if question Q3 included in the answer data is "0", the identification data is determined to indicate that a welding state with state number D1 has occurred. Also, if question Q3 is "1", and question Q5 is "1" and question Q4 is "0", the identification data is determined to indicate that a welding state with state number D1 has occurred. In all other cases, the identification data is determined to indicate that a welding state with state number D1 has not occurred.

[0080] Furthermore, if the model is composed of a neural network, the model generation unit 251 determines the identification data based on the data output from the model when response data is input to the model.

[0081] If models are provided for each welding state, the determination unit 253 may determine the identification data based on a combination of outputs from multiple models.

[0082] Figure 6 shows an example of determining classification data based on outputs from multiple models. In Figure 6, for each of the response data related to the training data T1 to Tm, examples of outputs from models MD1 to MD4 are shown in range RS. The classification data determined based on the outputs shown in range RS is shown in range RT.

[0083] For example, the determination unit 253 may obtain identification data by combining the outputs included in a combination of outputs from multiple models.

[0084] In Figure 6, "1,0,0,0" is shown as the output from models MD1 to MD4 based on the training data T1. Therefore, based on the output from model MD1 being "1", the determination unit 253 may generate "D1" as the identification data. When the identification data is "D1", it means that a welding state with state number D1 has occurred.

[0085] Based on the training data Tm, the output from models MD1 to MD4 is shown as "1,0,1,0". Therefore, based on the fact that the output from model MD1 is "1" and the output from model MD3 is "1", the determination unit 253 may generate "D1 or D3" as identification data. When the identification data is "D1 or D3", it means that either a welding state with state number D1 or a welding state with state number D3 has occurred.

[0086] The extraction unit 255 extracts corresponding records related to the welding state based on the determined identification data. The extraction unit 255 also outputs the extracted corresponding records via the output unit 23.

[0087] For example, the extraction unit 255 may identify a welding condition that has been determined to be occurring at the welded site based on the identification data. The extraction unit 255 may then refer to the database 30 to extract corresponding records associated with the identified welding condition.

[0088] The extraction unit 255 may output the extracted correspondence records to a display device (not shown) via the output unit 23. The extracted correspondence records may also be presented to the user.

[0089] [Processing procedure of the identification device (during model generation)] Figure 2 is a flowchart showing the processing procedure of the classifier (during model generation). In particular, Figure 2 shows the processing of the classifier 20 when training the model based on training data.

[0090] In step S101, the model generation unit 251 acquires training data to be used for training the model.

[0091] In step S103, the model generation unit 251 generates training data for each welding state.

[0092] In step S105, the model generation unit 251 selects a welding state. It then starts generating a model related to the selected welding state.

[0093] In step S107, the model generation unit 251 calculates impurity based on the training data related to the selected welding state.

[0094] In step S109, the model generation unit 251 determines the branch in the decision tree based on the calculated impurity.

[0095] In step S111, the model generation unit 251 determines whether the predetermined model creation termination conditions are met. If the model creation termination conditions are not met (if NO is found in step S111), the process returns to step S107.

[0096] If the model creation completion conditions are met (YES in step S111), the model with the selected welding state is set in step S113.

[0097] In step S115, the model generation unit 251 determines whether or not there are any unselected welding states. In other words, the model generation unit 251 determines whether or not there are any welding states for which a corresponding model has not been generated.

[0098] If there are any unselected welding states (if the answer is YES in step S115), the process returns to step S105, and the model generation unit 251 selects one welding state from among the unselected welding states.

[0099] If there are no unselected welding states (NO in step S115), the process shown in Figure 2 is terminated because model generation is complete for all welding states.

[0100] [Processing procedure for identification device (when identifying welding condition)] Figure 3 is a flowchart showing the processing procedure of the identification device (when identifying the welding state). In particular, Figure 3 shows the processing of the identification device 20 when determining the identification data corresponding to the response data based on the response data regarding the welding area for which the welding state is unknown.

[0101] In step S201, the decision unit 253 obtains the response data.

[0102] In step S203, the determination unit 253 selects one model.

[0103] In step S205, the decision unit 253 determines the output of the selected model when the response data is input to the model.

[0104] In step S207, the decision unit 253 determines whether or not there are any unselected models.

[0105] If there are unselected models (if the answer is YES in step S207), the process returns to step S203, and the decision unit 253 selects one model from among the unselected models.

[0106] If there are no unselected models (the answer is NO in step S207), in step S209, the determination unit 253 determines the identification data based on the output from the models.

[0107] In step S211, the determination unit 253 outputs identification data. After that, the process shown in Figure 3 is terminated.

[0108] [Effects of the Embodiment] As described in detail above, the identification device, identification method, and identification program relating to this disclosure use a controller connected to an input unit into which response data regarding a welded area is input. The controller determines identification data corresponding to the response data based on a model that outputs identification data indicating whether or not a welding condition of interest is occurring at the welded area in response to input based on the response data. The model is generated based on training data, which consists of training response data regarding a training welded area and training identification data indicating whether or not a welding condition of interest is occurring at the training welded area.

[0109] This makes it possible to easily identify the welding condition in actual operational situations where welding conditions are checked. In particular, the welding condition can be automatically estimated from data such as answers to questionnaires regarding the welding area. As a result, even inexperienced users can quickly identify the welding condition. Furthermore, it is possible to model the knowledge of experts with expertise in welding, which can be used to pass on know-how and technology.

[0110] Furthermore, in actual operational situations where welding conditions are checked, the ability to easily identify the welding condition reduces the time and cost associated with communication between workers checking the welded area and experts. The identification device, identification method, and identification program described herein make it easier to obtain initial answers on-site and reduce the need for requests for secondary answers from experts (escalation).

[0111] The model may be generated by decision tree analysis. This allows for the automatic estimation of the welding state with relatively little computation. In particular, because the model is represented by a decision tree, it is possible to identify key questions in identifying the welding state. Converting expert knowledge into a decision tree model also improves the efficiency of technology transfer from experts to less experienced users.

[0112] The model may be generated using machine learning. This allows for the automatic estimation of the welding state. Unlike models constructed using decision trees, it can also mitigate overfitting to the training data. Obtaining a model with high generalization performance can also improve user convenience.

[0113] The models may be prepared for each welding state and generated independently for each welding state. Compared to generating a model from multiple candidate welding states, generating a model for each welding state makes it easier for the model to converge and achieve sufficient accuracy. Therefore, it is possible to generate a model with sufficient prediction accuracy while reducing the computational load.

[0114] A model may be prepared for each welding state, and the controller may determine identification data based on a combination of outputs from multiple models. This allows for the generation of a model based on less training data, even for welding locations where it is not possible to narrow down the candidates for the current welding state.

[0115] In situations where even experts are unsure of the outcome, the available training data is insufficient, and conventional methods have been unable to generate models with sufficient accuracy. On the other hand, according to this disclosure, by generating a model that simply determines whether or not a welding state is occurring, it is possible to address situations where it is not possible to narrow down the candidates for the occurring welding state.

[0116] The weld condition may include at least one of the following: hot cracking, cold cracking, blowholes, pits, tungsten inclusions, slag inclusions, impurity contamination, poor penetration, poor fusion, overlap, undercuts, underfill, arc strikes, reheat cracking, stress corrosion cracking, sensitization, oxidation / oxides, and grain coarsening. This allows for easy identification of the weld condition across a wide range of weld conditions.

[0117] The system may further include a database for storing correspondence records linked to welding conditions, and an output unit. The controller may extract correspondence records related to welding conditions based on determined identification data and output the extracted correspondence records via the output unit. This allows knowledge about welding conditions linked to correspondence records to be conveyed to the user. As a result, user convenience is improved.

[0118] Each of the functions described in the embodiments above may be implemented by one or more processing circuits. These processing circuits may include programmed processors, electrical circuits, and other devices such as application-specific integrated circuits (ASICs), or circuit components arranged to perform the described functions.

[0119] According to this disclosure, in actual operational situations where welding conditions are checked, the determination of the type of welding condition can be easily achieved, resulting in improved user productivity. Therefore, for example, it can contribute to United Nations Sustainable Development Goal (SDG) 8, "Promote inclusive and sustainable economic growth and full and productive employment and decent work for all."

[0120] Although several embodiments have been described, it is possible to modify or transform the embodiments based on the above disclosure. All components of the above embodiments, and all features described in the claims, may be taken individually and combined, provided that they do not conflict with each other. [Explanation of Symbols]

[0121] 10 Control section 20 Identification device 21 Input section 23 Output section 25 Controllers 30 databases 251 Model Generation Unit 253 Decision Section 255 Extraction part

Claims

1. An input section into which response data regarding the welding area is entered, Controller and An identification device comprising, The controller determines the identification data corresponding to the response data based on a model that outputs identification data indicating whether or not the welding condition of interest is present at the welding site in response to the input based on the response data. The aforementioned model is an identification device generated based on training data consisting of training response data relating to a training welding site and training identification data indicating whether or not the welding state focusing on the training welding site is occurring.

2. The identification device according to claim 1, wherein the model is generated by decision tree analysis.

3. The aforementioned model is generated by machine learning, and is an identification device according to claim 1.

4. The identification device according to claim 1, wherein the model is prepared for each welding state and generated independently for each welding state.

5. The aforementioned models are provided for each welding condition, The identification device according to claim 1, wherein the controller determines the identification data based on a combination of outputs from a plurality of models.

6. The identification device according to claim 1, wherein the welding condition includes at least one of the following: hot cracking, cold cracking, blowholes, pits, tungsten inclusions, slag inclusions, impurity contamination, poor penetration, poor fusion, overlap, undercuts, underfill, arc strikes, reheat cracking, stress corrosion cracking, sensitization, oxidation / oxides, and grain coarsening.

7. The system further includes a database for storing corresponding records linked to the welding state, and an output unit. The aforementioned controller, Based on the determined identification data, the corresponding record relating to the welding state is extracted. The extracted corresponding records are output via the output unit. An identification device according to any one of claims 1 to 6.

8. An identification method for controlling a controller connected to an input unit into which response data regarding welding locations is input, The controller determines the identification data corresponding to the response data based on a model that outputs identification data indicating whether or not the welding condition of interest is present at the welding site in response to the input based on the response data. The model is an identification method generated based on training data comprising training response data relating to a training welding site and training identification data indicating whether or not the welding state focusing on the training welding site is occurring.

9. An identification program executed by a controller connected to an input unit into which response data regarding the welding area is input, The controller includes a step of determining the identification data corresponding to the response data based on a model that outputs identification data indicating whether or not a welding state of interest is present in the welding area in response to an input based on the response data. The model is an identification program generated based on training data consisting of training response data for training welding sites and training identification data indicating whether or not the welding state focusing on the training welding site is occurring.

Images