Resistance spot welding method

Abstract

To improve the accuracy of the quality determination using an expansion amount of a workpiece in resistance spot welding.SOLUTION: A resistance spot welding method includes: a preparation process of preparing a machine learning model obtained by learning a relation between a feature amount of test data in which a time series variation of an expansion amount of a workpiece in test welding is recorded and quality of a welding state of the test welding; and a determination process of determining quality of a welding state of regular welding by using regular welding data in which a time series variation of an expansion amount of the workpiece in the regular welding is recorded and the machine learning model. The feature amount includes: a first feature amount related to an inclination of a variation of an expansion amount of the workpiece in an expansion period in which the workpiece is expanded through electric conduction; and a second feature amount related to an inclination of a variation of an expansion amount of the workpiece in a contraction period in which the workpiece contracts after the expansion period.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present disclosure relates to a resistance spot welding method.

Background Art

[0002] Regarding resistance spot welding, Patent Document 1 discloses measuring the degree of thermal expansion of a workpiece during energization and the degree of contraction of the workpiece at the end of welding, and determining whether a proper nugget is formed in the workpiece based on these degrees.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] As in Patent Document 1, it is possible to determine the quality of welding by using the amount of expansion of the workpiece during resistance spot welding. It is desired to further improve the accuracy of such quality determination.

Means for Solving the Problems

[0005] The present disclosure can be realized in the following forms.

[0006] (1) According to one embodiment of the present disclosure, a resistance spot welding method is provided. This resistance spot welding method comprises a preparation step of preparing a machine learning model that has learned the relationship between feature quantities of test data recording the time-series change in the amount of expansion of a workpiece in a test weld and the quality of the welding state of the test weld; and a determination step of determining the quality of the welding state of the main weld using main welding data recording the time-series change in the amount of expansion of a workpiece in the main weld and the machine learning model. The feature quantities include a first feature quantity relating to the slope of the change in the amount of expansion of the workpiece during the expansion period in which the workpiece expands due to the application of current, and a second feature quantity relating to the slope of the change in the amount of expansion of the workpiece during the contraction period in which the workpiece contracts after the expansion period. In this format, the quality of the actual weld can be determined using a machine learning model that has learned the relationship between the features of the test data and the quality of the test weld. Since these features include a first feature and a second feature, which are features related to the slope of the change in the amount of expansion, the quality of the actual weld can be determined with high accuracy. (2) In the above configuration, the preparation step may include a machine learning model that has learned the relationship between the feature quantities and the type of disturbance in the test weld, and the determination step may include a first determination that uses the actual welding data and the machine learning model to determine whether the welding state of the actual weld is poor or not, a first determination that determines the type of disturbance in the actual weld that is not determined to be poor in the first determination, and a second determination that uses the determined type of disturbance and the actual welding data to determine whether the welding state of the actual weld that is not determined to be poor in the first determination is good or bad. (3) In the above configuration, in the second determination, the main welding data may be corrected according to the type of disturbance, the nugget diameter of the nugget in the main welding may be estimated using the corrected main welding data, and the quality of the welding state of the main welding may be determined based on the estimated nugget diameter. In this configuration, since the nugget diameter is estimated using the main welding data corrected according to the type of disturbance, the nugget diameter can be estimated with high accuracy even if the main welding includes disturbances. Based on the nugget diameter thus estimated, the quality of the main welding can be determined with higher accuracy. (4) In the above configuration, the machine learning model may be generated by machine learning using the following: test data associated with a first label representing the test weld in good condition and free from disturbances; test data associated with a second label representing the type of disturbance in the test weld in good condition and containing disturbances; and test data associated with a third label representing the test weld in poor condition. In this configuration, the first step can be easily performed by inputting the actual welding data into the machine learning model. (5) In the above configuration, the first feature quantity may include a feature quantity relating to the slope of the change in the amount of expansion of the workpiece during the first period and a feature quantity relating to the slope of the change in the amount of expansion of the workpiece during the second period following the first period. With such a configuration, the possibility of more accurately determining the quality of the welding state of the actual weld increases.

[0007] In addition to the resistance spot welding method described above, this disclosure can also be implemented in other forms, such as a method for generating a machine learning model, a resistance spot welding apparatus, a method for controlling the resistance spot welding apparatus, a computer program for implementing the control method, and a non-temporary recording medium on which the computer program is stored. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows a schematic configuration of a resistance spot welding apparatus. [Figure 2]This is a diagram illustrating the test data. [Figure 3] This is an explanatory diagram showing examples of types of disturbances. [Figure 4] This is an explanatory diagram showing an example of the nugget diameter estimation results. [Figure 5] This is a process diagram for resistance spot welding. [Figure 6] This is a flowchart of the generation process. [Modes for carrying out the invention]

[0009] A. First Embodiment: Figure 1 shows a schematic configuration of a resistance spot welding apparatus 100. The resistance spot welding apparatus 100 melts and joins a workpiece W made up of multiple overlapping metal plates. More specifically, the workpiece W has a first metal plate W1 and a second metal plate W2 that are overlapping each other.

[0010] The resistance spot welding apparatus 100 comprises a welding gun 10, a resistance measuring device 16, and a control device 20. The welding gun 10 comprises a gun body 11, a pair of electrodes, a lower electrode 12 and an upper electrode 13, and an electrode lifting device 14. The gun body 11 has an upper arm 11T and a lower arm 11B. The lower electrode 12 is fixed to the lower arm 11B. The upper electrode 13 is fixed to the upper arm 11T via the electrode lifting device 14. The electrode lifting device 14 is configured as an electrically operated lifting device having a servo motor 15. The electrode lifting device 14 holds the upper electrode 13 and raises and lowers the upper electrode 13 by the rotational driving force of the servo motor 15. In other words, the upper electrode 13 is movable by the electrode lifting device 14 in a direction along the opposing direction between the upper electrode 13 and the lower electrode 12.

[0011] When welding a workpiece W, the resistance spot welding apparatus 100 pressurizes the workpiece W between the upper electrode 13 and the lower electrode 12, and supplies welding current to the workpiece W through both electrodes. In this way, the workpiece W melts due to Joule heating caused by the current. Subsequently, the workpiece W cools and solidifies, joining the first metal plate W1 and the second metal plate W2. The workpiece W expands as it melts due to the current and contracts as it cools after the current is applied. A nugget is formed at the joint interface of the multiple welded metal plates.

[0012] The resistance measuring device 16 measures the electrical resistance between the lower electrode 12 and the upper electrode 13. The electrical resistance measured by the resistance measuring device 16 is transmitted to the control device 20.

[0013] The resistance spot welding apparatus 100 further includes an encoder 30 and a strain gauge 40. The encoder 30 detects the amount of rotation of the servo motor 15 at predetermined intervals and transmits a signal indicating the detected amount of rotation to the control device 20. The strain gauge 40 detects the amount of displacement of the lower electrode 12 due to external force at predetermined intervals and transmits a signal indicating the amount of displacement to the control device 20. The encoder 30 and strain gauge 40 are used to detect the distance between the upper electrode 13 and the lower electrode 12.

[0014] The control device 20 is composed of a computer comprising a CPU 21 and a storage device 22. The storage device 22 stores a program 221 that controls the operation of the resistance spot welding apparatus 100, a machine learning model 222 (described later), test data 223, actual welding data 224, estimation formula data 225, and correction formula data 226. The CPU 21 executes the program 221 stored in the storage device 22 to realize various functions such as an operation control unit 71, an expansion amount calculation unit 72, a time series data generation unit 73, a correction unit 74, a nugget diameter estimation unit 75, a learning model generation unit 76, and a determination unit 77. The control device 20 may also be composed of circuits.

[0015] The operation control unit 71 controls the operation of the resistance spot welding apparatus 100.

[0016] The expansion amount calculation unit 72 calculates the expansion amount of the workpiece W by calculating the distance between the upper electrode 13 and the lower electrode 12 at predetermined time intervals from the detection value of the encoder 30 and the detection value of the strain gauge 40. In this specification, the expansion amount is the amount of change in the distance between the upper electrode 13 and the lower electrode 12 from the state where the workpiece W is sandwiched between the upper electrode 13 and the lower electrode 12 before the welding current flows through the workpiece W. The expansion amount correlates with the size of the nugget (for example, the nugget diameter which is the diameter of the nugget), which is an index of the quality of the welding state. Therefore, it is possible to determine the quality of the welding state based on the expansion amount.

[0017] The time-series data generation unit 73 generates time-series data recording the time-series change in the expansion amount of the workpiece W based on the expansion amount calculated by the expansion amount calculation unit 72. More specifically, the time-series data generation unit 73 generates test data 223 and actual welding data 224 as the time-series data. The actual welding data 22 (should be 224) is data recording the time-series change in the expansion amount in the actual welding for manufacturing the product. The test data 223 is data recording the time-series change in the expansion amount in the test welding performed prior to the actual welding.

[0018] Figure 2 is an explanatory diagram of test data 223. More specifically, Figure 2 is a graph with the horizontal axis representing elapsed time and the vertical axis representing the expansion amount of workpiece W. In Figure 2, as examples of test data 223, test data TD1, test data TD2, and test data TD3 are shown. Test data TD1 is data regarding a test weld with a good welding state and without disturbances described later. Test data TD2 is data regarding a test weld with a good welding state and including disturbances. More specifically, test data TD2 is data regarding a test weld including a "gap" described later as a disturbance. Test data TD3 is data regarding a test weld with a poor welding state. Although not shown in the figure, the present welding data 224 is recorded as data substantially similar to the test data 223 shown in Figure 2.

[0019] Labels related to the results of the test welds for each test data 223 are associated with test data TD1 to test data TD3, respectively. A first label is associated with test data TD1. The first label is a label representing a test weld with a good welding state and without disturbances. A second label is associated with test data TD2. The second label is a label representing the type of disturbance in a test weld with a good welding state and including disturbances. For example, the second label associated with test data TD2 represents "gap" as the type of disturbance. A third label is associated with test data TD3. The third label is a label representing a test weld with a poor welding state.

[0020] In the present embodiment, when a test weld includes multiple types of disturbances, a second label representing the more dominant type of disturbance in that test weld is associated with the test data 223 of that test weld. In other embodiments, when a test weld includes multiple types of disturbances, for example, a second label representing that the test weld includes various types of disturbances may be associated with the test data 223 of that test weld.

[0021] As shown in Figure 2, in each test data 223, the time-series change in the amount of expansion is recorded over at least the expansion period PE and the contraction period PC. The expansion period PE refers to the period during which the workpiece W expands due to the application of current. More specifically, during the expansion period PE, the workpiece W expands due to the application of current through the upper electrode 13 and the lower electrode 12 while under pressure from the upper electrode 13 and the lower electrode 12. The contraction period PC refers to the period after the expansion period PE during which the workpiece W contracts. More specifically, during the contraction period PC, the workpiece W is cooled and contracts as it is pressurized by the upper electrode 13 and the lower electrode 12 while the current is stopped. Each test data shown in Figure 2 also includes the time-series change in the amount of expansion during the pre-pressure period PP. During the pre-pressure period PP, the workpiece W is pressurized by the upper electrode 13 and the lower electrode 12 while no current is applied.

[0022] The learning model generation unit 76 generates a machine learning model 222 by performing machine learning using the features of the test data 223. The machine learning model 222 is a trained model that has learned the relationship between the features of the test data 223 and the quality of the welding state of the test welds. Various machine learning models such as random forests, support vector machines (SVMs), and neural networks can be used as the machine learning model 222. Details of the method for generating the machine learning model 222 in this embodiment will be described later.

[0023] The features of the test data 223 used to generate the machine learning model 222 include a first feature and a second feature. The first feature is a feature relating to the slope of the change in expansion amount during the expansion period PE shown in Figure 2. The second feature is a feature relating to the slope of the change in expansion amount during the contraction period PC. Furthermore, the first feature in this embodiment includes a third feature and a fourth feature. The third feature is a feature relating to the slope of the change in expansion amount during the first period PE1. The fourth feature is a feature relating to the slope of the change in expansion amount during the second period PE2 following the first period PE1. More specifically, the first period PE1 in this embodiment is the period from the expansion start timing t1, when the expansion period PE begins, to the current value change timing tc. The second period PE2 is the period from the current value change timing tc to the expansion end timing t2, when the expansion period PE ends. The current value change timing tc represents the timing in the expansion period PE when the magnitude of the welding current is changed. In this embodiment, the welding current supplied to the workpiece W during the second period PE2 is greater than the welding current supplied to the workpiece W during the first period PE1. The expansion start timing t1 can also be described as the timing at which the pre-pressure period PP ends. Furthermore, the expansion end timing t2 can also be described as the shrinkage start timing t3, when the shrinkage period PC begins.

[0024] Figure 2 shows various features of test data TD1 as an example of features of test data 223. More specifically, Figure 2 shows the features of test data TD1 as slope s1, slope s2, slope s3, slope s4, maximum expansion EM, minimum expansion Em, and difference value Dv.

[0025] Slope s1 represents the slope of the change in the amount of expansion from the expansion start timing t1 to the expansion end timing t2. More specifically, slope s1 is calculated as the difference between the amount of expansion at the expansion end timing t2 and the amount of expansion at the expansion start timing t1, divided by the elapsed time between the two timings. Slope s2 represents the slope of the change in the amount of expansion from the contraction start timing t3 to the contraction start timing t4 when the contraction period PC ends. More specifically, slope s2 is calculated as the difference between the amount of expansion at the contraction start timing t4 and the amount of expansion at the contraction start timing t3, divided by the elapsed time between the two timings. Slope s3 represents the slope of the change in the amount of expansion from the expansion start timing t1 to the current value change timing tc. More specifically, slope s3 is calculated as the difference between the amount of expansion at the current value change timing tc and the amount of expansion at the expansion start timing t1, divided by the elapsed time between the two timings. The slope s4 represents the slope of the change in expansion amount from the current value change timing tc to the expansion end timing t2. More specifically, the slope s4 is calculated as the difference between the expansion amount at the expansion end timing t2 and the expansion amount at the current value change timing tc, divided by the elapsed time between the two timings. The maximum expansion amount EM represents the maximum expansion amount during the expansion period PE. The minimum expansion amount Em represents the minimum expansion amount during the contraction period PC. The difference value Dv represents the difference between the maximum expansion amount EM and the minimum expansion amount Em, i.e., the change in expansion amount during the contraction period PC.

[0026] In the various features of the test data TD1 shown in Figure 2, slopes s1, s3, and s4 correspond to the first feature, respectively. Slope s2 corresponds to the second feature. Slope s3 corresponds to the third feature. Slope s4 corresponds to the fourth feature. Note that the first feature may be any feature relating to the slope of the change in the amount of expansion during the expansion period PE, and in other embodiments, it does not have to be a value that represents the slope of the change in the amount of expansion itself. For example, the first feature may be a value used to calculate the slope of the change in the amount of expansion during the expansion period PE. The same applies to the second, third, and fourth features.

[0027] Furthermore, the machine learning model 222 in this embodiment has already learned the relationship between the features of the test data 223 and the types of disturbances in the test welding.

[0028] Figure 3 is an explanatory diagram illustrating examples of disturbance types. Figure 3 shows examples of disturbance types including "edge," "gap," "press," "upward," "axial tilt," "axial misalignment," "sealer," "tip wear," "plate thickness reduction," and "vertical wall." Each disturbance shown in Figure 3 acts to reduce the amount of expansion relative to the measured nugget diameter in resistance spot welding that includes the disturbance, compared to spot welding without the disturbance. As will be discussed later, each disturbance in resistance spot welding that includes the disturbance acts to reduce the amount of expansion relative to the measured nugget diameter.

[0029] "Edge" refers to the central axis TX of the upper electrode 13 and the central axis BX of the lower electrode 12 being offset to the horizontal edge of the workpiece W by a distance greater than a reference distance from the desired position HP. "Gap" refers to the size of the gap gp between the metal plates to be welded being greater than a reference amount at the start of welding. "Downward pressure" refers to the upper end position Ed of the workpiece W being offset downward from the reference position SEd by a distance greater than a reference amount at the start of welding. "Upward pressure" refers to the position Ed at the start of welding being offset upward from the reference position SEd by a distance greater than a reference amount. "Axis inclination" refers to the central axis TX or central axis BX being inclined at a distance greater than a reference amount from the angle perpendicular to the plate surface of the workpiece W. "Axis misalignment" refers to the horizontal positional misalignment between the central axis TX and central axis BX being greater than a reference amount. "Sealer" refers to welding being performed with a conductive rust inhibitor (sealer) Sr applied to the area of ​​the workpiece W to be welded. "Tip wear" refers to welding in which the tips of the upper electrode 13 or lower electrode 12 are worn beyond a predetermined standard during welding. "Thickness reduction" refers to welding in which the thickness d1 of the workpiece W used for welding is smaller than a predetermined standard thickness ds by a standard amount. "Vertical wall" refers to welding being performed near the vertical wall Vw of a workpiece W that has a vertical wall Vw. The vertical wall Vw refers to the part of the workpiece W that is erected so as to intersect with the plate surface of the workpiece W. Note that "no disturbance" shown in Figure 3 refers to welding that does not involve any of the above disturbances.

[0030] In this embodiment, the machine learning model 222 has already learned the relationship between "edges" and "gaps" among the disturbances shown in Figure 3 and the features of the test data 223.

[0031] The determination unit 77 shown in Figure 1 uses the actual welding data 224 and the machine learning model 222 to determine whether the welding condition of the actual weld is good or bad. Details of the quality determination process for determining whether the welding condition of the actual weld is good or bad will be described later.

[0032] The nugget diameter estimation unit 75 estimates the nugget diameter formed when the resistance spot welding apparatus 100 welds the workpiece W. In this embodiment, the nugget diameter estimation unit 75 estimates the nugget diameter using the estimation formula represented by the following formula (1). This estimation formula is included in the estimation formula data 225. φ=C1×E+C2×S+C3×R+C4 (1) φ represents the estimated nugget diameter [mm]. E represents the expansion amount [mm] during the expansion period PE. More specifically, in this embodiment, the expansion amount E is the integral of the expansion amount from the expansion start timing t1 to the expansion end timing t2. S represents the contraction amount [mm] during the contraction period PC. More specifically, in this embodiment, the contraction amount S is the difference between the contraction amount at the contraction start timing t4 and the expansion amount at the contraction start timing t3. Therefore, the contraction amount S takes a negative value. R represents the electrical resistance value [Ω]. More specifically, in this embodiment, the electrical resistance value R is the electrical resistance value during the second period PE2. C1, C2, C3, and C4 each represent constants. The above estimation formula is defined using multiple regression based on, for example, experimental results of resistance spot welding without disturbances or simulation results of resistance spot welding without disturbances.

[0033] The inventors of this application have found that when test welds or actual welds include disturbances as shown in Figure 3, the relationship between the amount of expansion and measured values ​​related to the size of the nugget (e.g., measured nugget diameter) changes compared to when the test welds or actual welds do not include disturbances. More specifically, when test welds or actual welds include each of the disturbances shown in Figure 3, the amount of expansion becomes relatively smaller relative to the measured nugget diameter compared to when the test welds or actual welds do not include disturbances. Therefore, when determining the quality of a weld based on the amount of expansion in an actual weld that includes disturbances, if the quality of the weld is determined in the same way as in the case without disturbances, the accuracy of the determination of the quality of the weld may decrease.

[0034] Figure 4 is an explanatory diagram showing an example of the nugget diameter estimation results. More specifically, Figure 4 is a graph with the estimated nugget diameter on the horizontal axis and the measured nugget diameter on the vertical axis. The estimated nugget diameter in Figure 4 represents the nugget diameter calculated using equation (1) described above. In Figure 4, the results of welding without disturbances are shown by black markers, and the results of welding with disturbances are shown by white markers. As shown in Figure 4, in welding with disturbances, the estimated nugget diameter is relatively smaller than the measured nugget diameter compared to welding without disturbances. This is because, as described above, the amount of expansion is relatively smaller than the actual nugget diameter in welding with disturbances.

[0035] The correction unit 74 shown in Figure 1 corrects the welding data 224 for a given weld according to the type of disturbance present in that weld. More specifically, the correction unit 74 corrects the welding data 224 using the correction formulas included in the correction formula data 226. Each correction formula included in the correction formula data 226 is defined for each type of disturbance. A correction formula for a particular type of disturbance is defined as a correction formula to remove the effect of that disturbance from the welding data 224 for a given weld that includes that type of disturbance. For example, a correction formula for a particular type of disturbance is defined based on experimental and simulation results as a correction formula to correct the amount of expansion in the test data 223 for a test weld that includes that type of disturbance to an amount of expansion similar to the amount of expansion in the test data 223 for a test weld that does not include the disturbance. When estimating the nugget diameter of a given weld that includes a disturbance, the nugget diameter estimation unit 75 uses the integrated value of the expansion amount based on the welding data 224 corrected by the correction unit 74 as the expansion amount E.

[0036] Figure 5 is a process diagram of the resistance spot welding method in this embodiment. In S100, the machine learning model 222 is prepared. The process of preparing the machine learning model 222, as in S100, is also called the preparation process. In S100 in this embodiment, the machine learning model 222 is generated by executing a generation process described later. In the preparation process of other embodiments, for example, the machine learning model 222 may be prepared by the control device 20 acquiring the machine learning model 222 from another computer or recording medium.

[0037] Figure 6 is a flowchart of the generation process. In S110, the learning model generation unit 76 acquires test data 223. In S120 and S130, the learning model generation unit 76 generates a machine learning model 222 by performing supervised learning using the test data 223 as training data. More specifically, in S120, the learning model generation unit 76 first extracts features from the test data 223. Next, in S130, the learning model generation unit 76 learns the relationship between the features extracted from the test data 223 and the labels associated with that test data 223. In S140, the learning model generation unit 76 determines whether machine learning is complete. In S140, for example, if learning is complete for all the test data 223 acquired in S110, the learning model generation unit 76 determines that machine learning is complete. If it determines in S140 that machine learning is not complete, the learning model generation unit 76 returns to S120.

[0038] In Figure 5, at S200, the main welding is performed. At S200, the operation control unit 71 performs the main welding, the expansion amount is calculated by the expansion amount calculation unit 72, and the main welding data 224 is generated by the time series data generation unit 73. Next, from S210 to S500, a judgment process is performed. The judgment process refers to the process of determining whether the welding condition of the main welding is good or bad using the main welding data 224 and the machine learning model 222.

[0039] In S210, the first step is executed. The first step is a process in which the welding data 224 and the machine learning model 222 are used to perform a first determination to determine whether or not the welding condition of the welding is poor, and to determine the type of disturbance in the welding that is not determined to be poor in the first determination. In this embodiment, in S210, the determination unit 77 inputs the welding data 224 into the machine learning model 222, thereby executing the first step.

[0040] From S300 to S500, the second process is executed. The second process refers to the process of performing a second determination to determine whether the welding condition of the main weld, which was determined to be non-defective in the first process, is good or bad, using the type of disturbance determined in the first process and the main welding data 224. If the welding condition of the main weld is determined to be poor in S200, steps S300 to S500 are omitted.

[0041] In S300, the welding data 224 is corrected according to the type of disturbance determined in S200. In this embodiment, in S300, the correction unit 74 refers to the correction formula data 226 based on the type of disturbance determined in S200 and corrects the welding data 224 using the correction formula corresponding to the type of disturbance.

[0042] In S400, the nugget diameter is estimated using the main welding data 224 corrected in S300. In this embodiment, in S400, the nugget diameter estimation unit 75 estimates the nugget diameter using the main welding data 224 corrected in S300 and the estimation formula included in the estimation formula data 225.

[0043] In S500, the quality of the weld is determined based on the nugget diameter estimated in S400. In this embodiment, in S500, the determination unit 77 determines whether the nugget diameter estimated in S400 is greater than or equal to a predetermined threshold, thereby determining the quality of the weld. More specifically, if the nugget diameter is greater than or equal to the threshold, the determination unit 77 determines that the weld is good. Conversely, if the nugget diameter is less than the threshold, the determination unit 77 determines that the weld is poor.

[0044] In addition, multiple main welds may be performed in S200, in which case main weld data 224 corresponding to each main weld will be generated. Furthermore, in this case, for example, S210 to S500 may be executed for each main weld after a series of main welds have been completed, or S210 to S500 may be repeatedly executed for each main weld.

[0045] The resistance spot welding method in this embodiment, as described above, includes a preparation step of preparing a machine learning model 222 that has learned the relationship between the features of test data 223 and the quality of the test weld, and a determination step of determining the quality of the actual weld using the actual welding data 224 and the machine learning model 222. The features of the test data 223 include a first feature relating to the slope of the change in the amount of expansion during the expansion period PE, and a second feature relating to the slope of the change in the amount of expansion during the contraction period PC. In this configuration, the quality of the welding state of the actual weld can be determined using the machine learning model 222. Furthermore, since the features of the test data 223 include a first feature and a second feature representing the slope of the change in the amount of expansion, the quality of the welding state of the actual weld can be determined with high accuracy. More specifically, for example, even if the lengths of the expansion period PE and contraction period PC in the test data 223 are different from the lengths of the expansion period and contraction period in the actual welding data 224, the quality of the welding state of the actual weld can be determined with high accuracy using the machine learning model 222.

[0046] Furthermore, in this embodiment, the determination process comprises a first step and a second step. In the first step, a first determination is performed to determine whether the welding state of the actual weld is poor or not, using a machine learning model 222 that has learned the relationship between the features of the test data 223 and the type of disturbance in the test weld, and the actual welding data 224. At the same time, the type of disturbance in the actual weld that is not determined to be poor in the first determination is determined. In the second step, the quality of the welding state of the actual weld that is not determined to be poor in the first determination is determined using the determined type of disturbance and the actual welding data 224. Therefore, it is possible to suppress a decrease in the accuracy of determining the quality of the welding state of the actual weld due to disturbances. Also, for example, as in the test data TD2 in Figure 2, the slope of the expansion amount during the expansion period PE of the test data 223 that includes "gap" as a disturbance is smaller than the slope of the expansion amount during the expansion period PE of the test data 223 that does not include disturbances, such as the test data TD1. Furthermore, although not shown in the diagram, for example, the absolute value of the slope of the expansion amount during the contraction period PC for test data 223 that includes "edges" as a disturbance is larger than the absolute value of the slope of the expansion amount during the contraction period PC for test data 223 that does not include disturbances. Thus, the slope of the expansion amount change during the expansion period PE and the contraction period PC reflects the characteristics of each disturbance. Therefore, as mentioned above, by including the first and second features in the feature vector used to generate the machine learning model 222, the type of disturbance can be determined with greater accuracy in the first step.

[0047] Furthermore, in this embodiment, in the second step, the main welding data 224 is corrected according to the type of disturbance, the nugget diameter in the main weld is estimated using the corrected main welding data 224, and the quality of the welding state of the main weld is determined based on the estimated nugget diameter. In this way, since the nugget diameter is estimated using the main welding data 224 corrected according to the type of disturbance, the nugget diameter can be estimated with high accuracy even when the main weld contains disturbances. Based on the nugget diameter estimated in this way, the quality of the main weld can be determined with even higher accuracy.

[0048] Furthermore, in this embodiment, the machine learning model 222 is generated by machine learning using test data 223 associated with a first label representing a test weld with good welding condition and no disturbances, test data 223 associated with a second label representing the type of disturbance in a test weld with good welding condition and disturbances, and test data 223 associated with a third label representing a test weld with poor welding condition. Therefore, the first step can be easily executed by inputting the welding data 224 into the machine learning model 222.

[0049] Furthermore, in this embodiment, the first feature quantity includes a third feature quantity relating to the slope of the change in the amount of expansion during the first period PE1, and a fourth feature quantity relating to the slope of the change in the amount of expansion during the second period PE2. Therefore, the likelihood of more accurately determining the quality of the weld in the judgment process is increased. In particular, in this embodiment, the first period PE1 is the timing before the current value change timing tc, and the second period PE2 is the timing after the current value change timing tc. Therefore, even if the magnitude of the welding current is changed in the middle of the expansion period PE during the weld, the quality of the weld can be determined with accuracy.

[0050] B. Other embodiments: (B1) In the above embodiment, the features of the test data 223 used to generate the machine learning model 222 include the first and second features, as well as the maximum value of the expansion during the expansion period PE, the minimum value of the expansion during the contraction period PC, and the change in the expansion during the contraction period PC. In contrast, the features of the test data 223 only need to include the first and second features, and do not need to include the maximum value of the expansion during the expansion period PE, the minimum value of the expansion during the contraction period PC, or the change in the expansion during the contraction period PC. Furthermore, the features of the test data 223 may include features other than those mentioned above.

[0051] (B2) In the above embodiment, a first judgment and a second judgment are performed in the judgment step. However, the first judgment and the second judgment do not have to be performed in the judgment step. For example, in the judgment step, the quality of the welding state of the actual weld may simply be determined by inputting the actual welding data 224 into the machine learning model 222. In this case, the machine learning model 222 does not have to have learned the relationship between the features of the test data 223 and the type of disturbance in the test welding.

[0052] (B3) In the above embodiment, in the second determination, the main welding data 224 is corrected according to the type of disturbance, and the nugget diameter in the main welding is estimated using the corrected main welding data 224. In contrast, the nugget diameter does not necessarily have to be estimated using the corrected main welding data 224; for example, the nugget diameter may be estimated according to the type of disturbance by using different estimation formulas depending on the type of disturbance.

[0053] (B4) In the above embodiment, the quality of the welding state of the main weld that was not determined to be poor in the first judgment is determined based on the nugget diameter estimated by the nugget diameter estimation unit 75. In contrast, the quality of the welding state of the main weld does not have to be determined based on the nugget diameter. For example, the quality of the welding state of the main weld that was not determined to be poor in the first judgment may be determined by inputting the main welding data 224 corrected according to the type of disturbance determined into the machine learning model 222.

[0054] (B5) In the above embodiment, the machine learning model 222 is generated by machine learning using test data 223 associated with a first label, test data 223 associated with a second label, and test data 223 associated with a third label. However, the machine learning model 222 does not have to be generated in this way; for example, it may be generated by machine learning using test data 223 associated with a second label and test data 223 associated with a third label. Also, if the first or second step is not performed in the judgment step, the machine learning model 222 may be generated by machine learning using test data 223 associated with a first label and test data 223 associated with a third label.

[0055] (B6) In the above embodiment, the first feature includes the third feature and the fourth feature. In contrast, the first feature does not necessarily have to include the third or fourth feature.

[0056] (B7) In the above embodiment, the machine learning model 222 may have already learned the relationship between the features of the test data 223 and disturbances of types other than "gaps" and "edges".

[0057] This disclosure is not limited to the embodiments described above, and can be implemented in various configurations without departing from its spirit. For example, the technical features in the embodiments corresponding to the technical features in each form described in the summary of the invention can be replaced or combined as appropriate in order to solve some or all of the above-described problems, or to achieve some or all of the above-described effects. Furthermore, if a technical feature is not described as essential in this specification, it can be deleted as appropriate. [Explanation of Symbols]

[0058] 10...Welding gun, 11...Gun body, 11B...Lower arm, 11T...Upper arm, 12...Lower electrode, 13...Upper electrode, 14...Electrode lifting device, 15...Servo motor, 16...Resistance measuring device, 20...Control device, 21...CPU, 22...Memory device, 30...Encoder, 40...Strain gauge, 71...Motion control unit, 72...Expansion amount calculation unit, 73...Time series data generation unit, 74...Correction unit, 75...Nugget diameter estimation unit, 76...Learning model generation unit, 77...Determination unit, 100...Resistance spot welding device, 221...Program, 222...Machine learning model, 223...Test data, 224...Actual welding data, 225...Estimation formula data, 226...Correction formula data

Claims

1. A resistance spot welding method, A preparatory step involves preparing a machine learning model that learns the relationship between the feature quantities of test data recording the time-series change in the amount of expansion of the workpiece during test welding and the quality of the welding state of the test welding. The system includes a determination step that uses welding data recording the time-series change in the amount of expansion of the workpiece during the welding process, and the machine learning model, to determine whether the welding condition of the welding is good or bad. The aforementioned feature quantity includes a first feature quantity relating to the slope of the change in the amount of expansion of the workpiece during the expansion period in which the workpiece expands due to the application of electricity, and a second feature quantity relating to the slope of the change in the amount of expansion of the workpiece during the contraction period in which the workpiece contracts after the expansion period. In the preparation step, the machine learning model is prepared, which has learned the relationship between the feature quantities and the type of disturbance in the test welding. The aforementioned determination step is, A first step is to perform a first determination using the welding data and the machine learning model to determine whether or not the welding condition of the welding is poor, and to determine the type of disturbance in the welding in which the welding condition is not determined to be poor in the first determination, The system includes a second step of performing a second determination to determine whether the welding condition of the main weld, which was not determined to be poor in the first determination, is good or bad, using the determined type of disturbance and the main welding data. Resistance spot welding method.

2. A resistance spot welding method according to claim 1, A resistance spot welding method comprising: in the second determination, correcting the main welding data according to the type of disturbance; estimating the nugget diameter of the nugget in the main welding using the corrected main welding data; and determining the quality of the welding state of the main welding based on the estimated nugget diameter.

3. A resistance spot welding method according to claim 1, A resistance spot welding method is generated by machine learning using the machine learning model, which includes test data associated with a first label representing a test weld with good welding condition and no disturbances, test data associated with a second label representing the type of disturbance in the test weld with good welding condition and disturbances, and test data associated with a third label representing a test weld with poor welding condition.

4. A resistance spot welding method according to any one of claims 1 to 3, A resistance spot welding method wherein the first feature quantity includes a feature quantity relating to the slope of the change in the amount of expansion of the workpiece during a first period, and a feature quantity relating to the slope of the change in the amount of expansion of the workpiece during a second period following the first period.

Images