WELDING SYSTEM AND WELDING RESULT JUDGMENT DEVICE

The welding system automatically determines welding quality using a machine-learning-based estimation model, extracting feature information from monitoring data to overcome manual threshold setting and empirical rule dependencies, enhancing accuracy and adaptability.

JP7595548B2Active Publication Date: 2024-12-06AMADA CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
JP2021171198
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2021-10-19
Publication Date
2024-12-06
Estimated Expiration
2041-10-19

AI Technical Summary

Technical Problem

Existing welding monitoring systems require manual threshold setting and empirical rules for quality determination, necessitating operator intervention and reconfiguration with changes in welding conditions, such as workpiece material.

Method used

A welding system that automatically extracts feature quantity information from monitoring data using a welding result estimation model, trained with machine learning, to determine welding quality without manual threshold setting or empirical rules, utilizing latent variables to reduce correlation with processing conditions.

Benefits of technology

Enables accurate and automated quality determination of welding during the process, improving estimation accuracy by reducing reliance on manual adjustments and empirical methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007595548000002
    Figure 0007595548000002
  • Figure 0007595548000003
    Figure 0007595548000003
  • Figure 0007595548000004
    Figure 0007595548000004
Patent Text Reader

Abstract

To perform quality determination using feature quantity information extracted on the basis of monitoring information about welding.SOLUTION: A welding system according to one aspect comprises: a welding device; a welding monitor device that outputs monitoring information about welding by the welding device; a learning device that inputs, as teacher data, welded-result information relevant to a welded result obtained after welding and feature quantity information extracted from monitoring information at the time of welding during which the welded result is obtained, and generates and outputs a welded-result estimation model, on the basis of the teaching data; and a welded-result determining device that extracts the feature quantity information from the monitoring information outputted at the time of welding during which the welded result should be determined, inputs the information, as data for estimation, to the welded-result estimation model, and outputs a result of quality determination of the welding, where the learning device detects values of a plurality of preset feature elements, from the monitoring information including time-series data to calculate contribution ratios to the welded result, and determines feature quantity information on the basis of the calculated contribution ratios.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a welding system and a welding result determination device. [Background technology]

[0002] Conventionally, in laser welding, a technique has been known in which an optical sensor measures changes in the light intensity of visible light, infrared light, reflected light, etc. generated from the molten part, and a welding monitor device monitors changes in the welding condition and welding quality based on the waveform shape of the time-series data output from the optical sensor (see Patent Document 1).

[0003] In this welding monitoring device, for example, a threshold value for the light intensity indicated by the waveform shape of the monitored time series data is preset, and the joining condition of the metal materials is inferred and the quality of the weld is judged as good or bad (good or bad judgment) based on whether the measured light intensity value is within the range between the upper and lower limits of the threshold value. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 3275988 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the device disclosed in the above-mentioned Patent Document 1, an operator must carry out preliminary processing or use experience to make pass / fail judgments on multiple check items such as setting a threshold value, judging the validity of the threshold value, setting a confirmation range for the waveform shape of time-series data, etc. In addition, if the welding processing conditions change when the material of the workpiece is different, it is necessary to reset the threshold value, etc.

[0006] One aspect of the present invention is a welding system and welding result judgment device that can judge the quality of welding, including during welding, by using feature information that contributes to the judgment of the quality of welding, extracted based on welding monitoring information, without an operator having to perform preliminary processing or make the judgment based on empirical rules. [Means for solving the problem]

[0007] A welding system according to one aspect of the present invention includes a welding device that welds workpieces, a welding monitoring device that outputs monitoring information obtained by monitoring a progress of welding by the welding device using a sensor, a learning device that inputs, as teacher data, welding result information related to a welding result obtained after welding by the welding device and feature amount information extracted from the monitoring information at the time of welding when the welding result was obtained, and creates and outputs a welding result estimation model based on the teacher data, and a welding result judgment device that extracts feature amount information from the monitoring information output from the welding monitoring device at the time of welding when the welding result is to be judged, inputs the feature amount information as estimation data to the welding result estimation model created by the learning device, and outputs a pass / fail judgment result of the welding, wherein the learning device detects values ​​of a plurality of preset feature elements from the monitoring information including time-series data, calculates contribution degrees of the detected values ​​of the plurality of feature elements to the welding result, and determines the feature amount information based on the calculated contribution degrees from the values ​​of the plurality of feature elements.

[0008] According to a welding system according to an aspect of the present invention, a welding result estimation model is created using, as teacher data, welding result information related to the welding result obtained after welding by a welding device and feature amount information extracted from monitoring information during welding when the welding result was obtained. Also, feature amount information is extracted from monitoring information during welding when the welding result should be judged and input to the welding result estimation model as estimation data, and a quality judgment result of welding is output. The feature amount information is determined based on the contribution degree of each feature amount information to the welding result, which is calculated by detecting values ​​of a plurality of feature elements set in advance from monitoring information including time series data. Therefore, it is possible to accurately judge the quality of welding, including during welding, by using the feature amount information that contributes to the quality judgment of welding, which is automatically extracted based on the time series data included in the welding monitoring information. Also, it is possible to learn a more accurate welding result estimation model and output a quality judgment result, which reflects monitoring information in which the correlation with the objective variable and explanatory variable can be reduced as much as possible by the feature amount information. Effect of the Invention

[0009] According to one aspect of the present invention, it is possible to determine the quality of welding, including during welding, by using feature information that contributes to determining the quality of welding and is extracted based on welding monitoring information, without the worker having to perform preliminary processing or make the determination based on empirical rules. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is an explanatory diagram showing a basic configuration of a welding system according to a first embodiment of the present invention. [Diagram 2] FIG. 2 is an explanatory diagram showing a basic hardware configuration of a learning device and / or a welding result determination device of a welding system. [Diagram 3] FIG. 3 is a graph showing an example of time series data. [Figure 4] FIG. 4 is a diagram showing an example of a graph shape of time-series fluctuation data. [Diagram 5]FIG. 5 is an explanatory diagram showing a basic configuration of a welding system according to a second embodiment of the present invention. [Figure 6] FIG. 6 is a graph showing the relationship between the welding temperature and elapsed time for explaining auxiliary variables in an example of the welding system according to the first embodiment. [Figure 7] FIG. 7 is a plot of regression errors when multiple regression is performed on the welding temperatures of the examples with respect to the processing conditions. [Figure 8] FIG. 8 is a graph showing the relationship between the estimated values ​​and the measured values ​​in the comparative example and the example, and the frequency distribution of the error of the estimated values ​​with respect to the estimated values. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] A welding system and a welding result determination device according to an embodiment of the present invention will be described in detail below with reference to the accompanying drawings. However, the following embodiments do not limit the invention according to each claim, and not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention. In addition, in the following embodiments, the arrangement, scale, dimensions, etc. of each component may be shown exaggerated or reduced, and some components may be omitted.

[0012] [First embodiment] Fig. 1 is an explanatory diagram showing a basic configuration of a welding system according to a first embodiment of the present invention, Fig. 2 is an explanatory diagram showing a basic hardware configuration of a learning device and / or a welding result determination device of the welding system.

[0013] As shown in FIG. 1, welding system 100 according to the first embodiment is a system capable of estimating welding results and outputting pass / fail judgment results, and includes a welding device (laser welder 40) that welds workpieces, a welding monitor device 50 that outputs monitoring information 5 obtained by monitoring the progress of welding by the welding device (laser welder 40) using a sensor 212, and welding result information 1 related to the welding result obtained after welding by the welding device (laser welder 40) and feature amount information (not shown) extracted from the monitoring information 5 at the time of welding when the welding result was obtained, as inputs, and creates and outputs a welding result estimation model 3 based on the inputs. and a welding result judgment device 20 that extracts feature amount information (not shown) from monitoring information 6 output from a welding monitoring device 50 during welding when the welding result is to be judged, inputs this feature amount information as estimation data into a welding result estimation model 3 created by the learning device 10, and outputs a welding quality judgment result (welding result estimated value 8). The learning device 10 detects values ​​of a plurality of preset feature elements (not shown) from monitoring information 5 including time-series data, calculates the contribution of the detected values ​​of the plurality of feature elements to the welding result by, for example, machine learning, and determines the feature amount information based on the contribution calculated from the values ​​of the plurality of feature elements.

[0014] Welding monitoring device 50 of welding system 100 is a device capable of outputting monitoring information 5, 6 obtained by monitoring (status observation) the progress of welding by laser welder 40 with sensor 212, and refers to a device capable of monitoring the welded portion of the workpiece with sensor 212 and acquiring output information from sensor 212 as monitoring information 5, 6. The monitoring information 5, 6 output from welding monitoring device 50 includes, for example, information including time-series data of physical quantities (welding temperature, sound, light, color, etc.) observed during laser welding.

[0015] The welding system 100 of the first embodiment is a system applied to laser welding using latent variables (teaching data and estimation data) that have a smaller correlation with the processing condition information 2, 4 than the auxiliary variables (feature amount information) calculated based on the monitoring information 5, 6 from the welding monitor device 50. In this welding system 100, for example, the teaching data of the objective variable input to the learning device 10 is welding result information 1. The welding result information 1 includes, for example, cause information from which a judgment result of good welding has been derived and / or cause information from which a judgment result of poor welding has been derived. The cause information refers to information that is the cause of the judgment result of good welding or poor welding (presence or absence of spatter generation, presence or absence of focal position deviation, etc.). The welding result information 1 also includes, for example, the weld strength F (N / mm 2 The weld strength is obtained, for example, by destructive testing, and the destructive testing may be selected from the destructive tests listed on the website of the Welding Information Center of the Japan Welding Society, such as JIS Z 2241 Metallic Material Tensile Test Method or any other listed JIS standard that can be used for welding tests and inspections, and the test and evaluation methods are also those listed in the JIS standard.

[0016] The teacher data for the explanatory variables includes processing condition information 2 including various processing conditions such as processing conditions for good processing (laser power (output value), laser irradiation time, etc.) and processing conditions for poor processing in previously performed welding. The processing condition information 4 includes processing conditions for good processing and / or poor processing set in the laser welder 40. The objective variables, explanatory variables, and auxiliary variables used in the welding system 100 are not limited to these exemplified ones. An auxiliary variable means an auxiliary variable, which is a variable other than the objective variable and is available at the time of prediction / estimation as a variable having a causal relationship that becomes a result when the explanatory variable is the cause.

[0017] 1 is a laser welder 40, and the teacher data input to the learning device 10 and the estimation data input to the welding result judgment device 20 include processing condition information 2, 4, which includes at least one processing condition of the material of the workpiece, the thickness of the workpiece, the laser power, the laser irradiation time, the time from the start of laser irradiation to reach the peak power, the time from the peak power to the end of laser irradiation, the fiber diameter, the lens focal length, the focal position, and the laser diameter at the irradiation point. Also, the monitoring information 5, 6 includes time-series fluctuation data of the laser power and time-series fluctuation data of the radiated near-infrared light intensity at the irradiation point.

[0018] In addition, the welding result judgment device 20 of the welding system 100 of the first embodiment includes a notification unit (not shown) that notifies the pass / fail judgment result (contents) in a manner that allows at least one of visual and audible confirmation, and the pass / fail judgment result includes at least one of information related to the welding strength of the weld and information related to the cause.

[0019] Here, the laser welder 40 is a device that irradiates a workpiece with a laser beam under processing conditions designated by, for example, an external input, and performs welding. The laser welder 40 is configured to be able to acquire various conditions under which the laser beam is irradiated through various communications (network communications or direct connection communications). The welding result includes the welding strength, spot diameter, and welding quality. The welding quality and the like may be an index that evaluates the metal structure from the appearance or cross section, or an index that evaluates the corrosion resistance. Therefore, the welding system 100 of the first embodiment may be used as an estimation system for the welding strength included in the pass / fail judgment result of the laser welding by the laser welder 40, for example. In addition, the welding result information 1 including the pass / fail judgment result of the welding result may include the spot diameter, so the welding system 100 may also be used as an estimation system for the spot diameter included in the pass / fail judgment result of the laser welding.

[0020] The learning device 10 includes a first auxiliary variable calculation unit 12, a first latent variable calculation unit 13, and a welding result learning unit 11. The first auxiliary variable calculation unit 12 extracts feature information from the teacher data of the monitoring information 5 output from the welding monitor device 50 and calculates it as teacher data of the auxiliary variable. The first latent variable calculation unit 13 calculates teacher data (not shown) of the latent variable having a smaller correlation with the teacher data of the processing condition information 2 than the teacher data (feature information) of the auxiliary variable from the teacher data of the processing condition information 2 and the teacher data of the auxiliary variable, and outputs latent variable calculation information 7. The welding result learning unit 11 inputs the teacher data of the processing condition information 2 input to the learning device 10, the teacher data of the welding result information 1 related to the welding result of the welded part obtained by, for example, destructive inspection after welding by the laser welding machine 40 based on the processing conditions, and the teacher data of the latent variable, and performs machine learning using, for example, a predetermined machine learning algorithm to create a welding result estimation model 3. The first latent variable calculation unit 13 calculates a latent variable having a correlation with the monitoring information 5, and outputs the latent variable calculation information 7 determined during the calculation. In this embodiment, the latent variable is a variable that characterizes the objective variable like the explanatory variable, and is a variable that is not directly observed but is estimated from the observed variable (auxiliary variable). In other words, the latent variable means a variable that characterizes the observed variable and is dimension-reduced or dimension-compressed from at least the auxiliary variable. Therefore, the latent variable is one-dimensional or two-dimensional information that has a correlation with the processing condition information 2 and is obtained by reducing or compressing the dimension of the monitoring information 5. Note that the term "correlation" is not limited to a linear relationship between multiple variables, but means a relationship between multiple variables in which if one variable changes, the other variables change, regardless of whether it is linear or nonlinear.

[0021] In addition, the first auxiliary variable calculation unit 12 detects values ​​of a plurality of preset characteristic elements, which will be described later, from the monitoring information 5 including time-series data, and calculates the contributions of the detected values ​​of the plurality of characteristic elements to the welding result, for example by machine learning, and determines the feature information based on the contributions calculated from the values ​​of the plurality of characteristic elements.

[0022] The welding result learning unit 11 inputs the processing condition information 2 including the processing conditions that resulted in good and / or bad processing input to the learning device 10, the welding result information 1 including the cause information that derived the judgment result of good and / or bad processing, and the latent variables calculated by the first latent variable calculation unit 13, and creates a welding result estimation model 3 that can judge the quality of welding by machine learning using a predetermined machine learning algorithm, for example. Note that the spot diameter calculated by the combination of the lens, fiber, and focal position, for example, is used for machine learning in the welding result learning unit 11. The learned welding result estimation model 3 is composed of software that runs on a computer, for example, and in this example, a program that outputs a welding result estimation value 8 indicating the quality of welding when feature amount information extracted from the monitoring information 6 output from the welding monitor device 50 during welding to judge the welding result is input as estimation data. In addition, the welding result estimation model 3 may be, for example, a program that outputs a welding result estimation value 8, which is information related to welding strength, when processing condition information 4 is input as estimation data, and that outputs a welding result estimation value 8, which is a pass / fail judgment including cause information, when monitoring information 6 is input as estimation data.

[0023] The learning method includes various algorithms such as regression, classification, clustering, discrimination, interpolation, feature extraction, and time series modeling, but is not limited to these. The welding result estimation model 3 learned by the welding result learning unit 11 includes various parameters, various formulas, and / or algorithms for prediction and estimation that characterize the learned model. Then, the learning device 10 outputs the latent variable calculation information 7 and the welding result estimation model 3 to the welding result judgment device 20.

[0024] The first latent variable calculation unit 13 calculates the teacher data of the latent variables using, for example, a predetermined algorithm. This algorithm is also used for machine learning and estimation in the learning device 10 and the welding result determination device 20 (estimation device 30). As the algorithm, for example, at least one of regression analysis (RA), principal component analysis (PCA), singular value decomposition (SVD), linear discriminant analysis (LDA), independent component analysis (ICA), and Gaussian process latent variable model (GPLVM) can be used. The algorithm is not limited to these.

[0025] The latent variable calculation information 7 means various information that is determined in the first latent variable calculation unit 13 when calculating teacher data of the latent variables and is used when calculating additional latent variables, and includes, for example, various parameters, various formulas, and / or algorithms. The feature amount information extracted by the first auxiliary variable calculation unit 12 is calculated based on the monitoring information 5, and therefore can be used as teacher data for the auxiliary variables. The extraction of feature amount information by the first auxiliary variable calculation unit 12 will be described later.

[0026] On the other hand, the welding result determination device 20 is a welding result determination device that inputs monitoring information 6 obtained by monitoring a progress state of welding by a welding device (laser welder 40) that welds workpieces with a sensor 212 from a welding monitor device 50, uses welding result information 1 related to a welding result obtained after welding by the welding device (laser welder 40) and feature amount information extracted from the monitoring information 5 at the time of welding when the welding result was obtained as teacher data, extracts feature amount information from the monitoring information 6 output from the welding monitor device 50 at the time of welding for which the welding result should be determined, inputs this feature amount information to the welding result estimation model 3 as estimation data, and outputs a quality determination result of the welding. The welding result estimation model 3 detects values ​​of a plurality of preset feature elements from the monitoring information 5, 6 including time-series data, calculates the contribution degree of the detected values ​​of the plurality of feature elements to the welding result, and determines the feature amount information based on the contribution degree calculated from the values ​​of the plurality of feature elements. The welding result judgment device 20 stores therein the latent variable calculation information 7 and the welding result estimation model 3 output from the learning device 10, and includes an estimation device 30 and a notification unit (not shown). The estimation device 30 includes a second auxiliary variable calculation unit 32, a second latent variable calculation unit 33, and a welding result estimation unit 31. The second auxiliary variable calculation unit 32 extracts feature amount information from the estimation data of the monitoring information 6 output from the welding monitoring device 50 and calculates it as estimation data of the auxiliary variable. The second latent variable calculation unit 33 calculates estimation data of the latent variable (not shown) that has a smaller correlation with the estimation data of the machining condition information 4 than the estimation data of the auxiliary variable (feature amount information) from the estimation data of the machining condition information 4 and the estimation data of the auxiliary variable (feature amount information) based on the latent variable calculation information 7. For example, when the latent variable calculation information 7 includes an algorithm, the second latent variable calculation unit 33 calculates the estimation data of the latent variables using the same algorithm as that used to calculate the teacher data of the latent variables by the first latent variable calculation unit 13 of the learning device 10. In this way, by calculating the latent variables using the same algorithm as that used during learning, even during estimation, estimation using the same welding result estimation model 3 becomes possible.

[0027] The welding result estimation unit 31 estimates the welding result of the welding process performed by the laser welder 40 from the estimation data of the processing condition information 4 set in the laser welder 40 input to the welding result judgment device 20 and the estimation data of the latent variables based on the welding result estimation model 3 stored in the welding result judgment device 20, and outputs a welding result estimation value 8 indicating a pass / fail judgment. The notification unit visualizes or audibly notifies the user of the content of the pass / fail judgment result (welding strength, cause of welding defect, etc.) output by the welding result estimation unit 31 of the estimation device 30. Specifically, the welding result estimation unit 31 inputs the estimation data of the processing condition information 4 and the estimation data of the latent variables to the welding result estimation model 3, estimates the welding result, and outputs it. The welding result judgment device 20 can determine, for example, the welding strength and / or spot diameter of the welding (hereinafter referred to as "welding strength, etc.") based on the welding result estimation value 8 output from the welding result estimation unit 31 of the estimation device 30. Regarding the weld strength, the welding result estimate 8 refers to information that can numerically express, for example, the tensile strength (unit: N) and the distribution range of the strength of a certain weld point (welded portion). Regarding the spot diameter, the welding result estimate 8 refers to information that can numerically express, for example, the size of the irradiation range of the laser light on the surface of a certain weld point (welded portion) or the size of the melted portion of the welded portion.

[0028] The latent variable calculation information 7 transferred from the learning device 10 to the welding result judgment device 20 may be selected as appropriate for the calculation of latent variables by the learning device 10 and the estimation device 30. For example, when the calculation of teacher data of latent variables in the learning device 10 is performed by principal component analysis, the latent variable parameters may be principal component vectors and eigenvalues, and in addition to these, latent variable parameters excluding principal component vectors in which the contribution rate of teacher data of a specific auxiliary variable (monitoring information 5) is higher than a predetermined value may be selected. This allows a latent variable having a smaller correlation with the explanatory variable than the auxiliary variable to be used as the teacher data and estimation data together with the explanatory variable, thereby making it possible to avoid deterioration of estimation accuracy due to multicollinearity.

[0029] In addition, when performing time series data analysis, machine learning, and welding analysis and estimating pass / fail judgment results in the learning device 10 and the welding result judgment device 20, in addition to the above-mentioned algorithms, various algorithms can be used, such as logistic regression (LR), support vector machine (SVM), discriminant analysis (DA), random forest (RF), ranking support vector machine (RSVM), gradient boosting (GB), naive Bayes (NB), and K-nearest neighbor algorithm (K-NN).

[0030] Here, the algorithm of the time series data analysis in the welding system 100 refers to a process of automatically extracting and determining feature amount information that is characteristic of the time series data (time series fluctuation data of laser power and time series fluctuation data of the radiated near-infrared light intensity at the irradiation point) included in the monitoring information 5, 6 by, for example, calculation processing. The algorithm of the welding analysis refers to a calculation process for performing a quality judgment (defective welding judgment) of welding by, for example, performing a statistical process including machine learning based on the feature amount information. The first latent variable calculation unit 13 may calculate one-dimensional or two-dimensional information obtained by reducing or compressing the dimension of the teacher data of the processing condition information 2 and the teacher data (feature amount information) of the auxiliary variable as the teacher data of the latent information. The second latent variable calculation unit 33 may calculate one-dimensional or two-dimensional information obtained by reducing or compressing the dimension of the estimation data of the processing condition information 4 and the estimation data (feature amount information) of the auxiliary variable as the estimation data of the latent information. The teacher data and estimation data for the latent variables are those in which the auxiliary variables, or the auxiliary variables and the explanatory variables, have been subjected to dimensionality reduction or dimensionality compression, for example, as described above, so as to reduce the correlation between the auxiliary variables and the explanatory variables more than between the auxiliary variables.

[0031] In the first embodiment of the welding system 100, the first auxiliary variable calculation unit 12 of the learning device 10 calculates teacher data (feature information) of the auxiliary variables from data including the monitoring information 5, the first latent variable calculation unit 13 calculates teacher data of the latent variables from the teacher data of the processing condition information 2 and the feature information and outputs latent variable calculation information 7, and the welding result learning unit 11 inputs the teacher data of the processing condition information 2, the teacher data of the welding result information 1, and the teacher data of the latent variables, and creates a welding result estimation model 3 by, for example, machine learning. Then, a second auxiliary variable calculation unit 32 of the estimation device 30 of the welding result judgment device 20 calculates estimation data (feature information) for the auxiliary variable from data including the monitoring information 6, a second latent variable calculation unit 33 calculates estimation data for the latent variable based on the latent variable calculation information 7 from the estimation data for the processing condition information 4 and the feature information, and a welding result estimation unit 31 outputs a welding result estimation value 8 based on the welding result estimation model 3 from the estimation data for the processing condition information 4 and the estimation data for the latent variable.

[0032] As a result, by simply inputting processing condition information 4 and monitoring information 6 into the learned welding result estimation model 3, welding system 100 can obtain a welding result estimation value 8 indicating a pass / fail judgment associated with the welding result, obtain a welding result estimation value 8 indicating the weld strength and the cause of the welding defect included in the pass / fail judgment result, and notify the pass / fail judgment result via the notification unit, thereby making it possible to judge the quality of the welding even during welding.

[0033] In addition, each data of the welding result information (teaching data of objective variables) 1 and the processing condition information (teaching data of explanatory variables) 2 in the welding system 100 may be composed of one or more variables or parameters, and may be stored in, for example, a stationary or portable storage device or storage medium (not shown). In addition, each data may be transmitted and received via an information communication medium such as the Internet, and may be raw data acquired by a measuring device such as a sensor (not shown). In addition, the welding result estimation model 3 may be input and output between the learning device 10 and the welding result judgment device 20 via the above-mentioned storage medium or information communication medium. The explanatory variables, objective variables, latent variables, and latent variable calculation information 7 used in the welding system 100 may each be composed of one or more variables or parameters. In this specification, the term "prediction and estimation of objective variables" is used to mean "prediction of objective variables" or "estimation of objective variables". "Prediction of objective variables" means assuming an objective variable that is thought to be realized in the future, and "estimation of objective variables" means estimating an objective variable that is currently realized but cannot be directly observed.

[0034] As shown in FIG. 2, the learning device 10 and / or welding result judgment device 20 of the welding system 100 includes, as basic hardware configuration, for example, a CPU (central processing unit) 201, a RAM (random access memory) 202, a ROM (read only memory) 203, a HDD (hard disk drive) 204 and / or an SSD (solid state drive) 205, and a memory card 206.

[0035] Furthermore, learning device 10 and / or welding result determination device 20 includes, for example, an input I / F (interface) 207, an output I / F (interface) 208, and a communication I / F (interface) 209. Each of the components 201 to 209 is connected to each other by a bus 200.

[0036] CPU 201 executes various programs stored in RAM 202, ROM 203, HDD 204, SSD 205, etc., to control learning device 10 and / or welding result judgment device 20. In learning device 10, CPU 201 executes a learning program to realize the functions of first auxiliary variable calculation unit 12, first latent variable calculation unit 13, and welding result learning unit 11.

[0037] Furthermore, in welding result judgment device 20, CPU 201 executes a judgment program to realize the functions of second auxiliary variable calculation unit 32, second latent variable calculation unit 33, welding result estimation unit 31, and notification unit of estimation device 30. Note that the respective CPUs 201 of learning device 10 and welding result judgment device 20 can be configured to control the entire welding system 100 by cooperating with each other.

[0038] The RAM 202 can be used as a working area for the arithmetic processing of the CPU 201. The ROM 203 stores at least the above-mentioned various programs in a readable manner. The HDD 204 and the SSD 205 store the above-mentioned various data in a readable and writable manner. The memory card 206 stores the above-mentioned various data in a readable and writable manner, and constitutes a storage medium that is detachable from each of the devices 10 and 20. The HDD 204, the SSD 205, and the memory card 206 realize the functions of the above-mentioned storage devices or storage media.

[0039] For example, a sensor 212 is connected to the input I / F 207 to acquire detection information. The input I / F 207 may also acquire processing condition information 2, 4 from the laser welding machine 40 and monitoring information 5, 6 from the welding monitor device 50. The sensor 212 includes various sensors such as a temperature sensor, an optical sensor, an acoustic sensor, and an image sensor. Note that a touch panel 211 functioning as an operation unit or input unit of the learning device 10 and / or the welding result determination device 20 is connected to the input I / F 207, and receives information associated with an operation input from a user of the welding system 100. Various input devices such as a keyboard and a mouse (including a trackball mouse) (not shown) may also be connected to the input I / F 207.

[0040] A display 210 serving as a display device, for example, is connected to output I / F 208, and various information is output to be displayed on a monitor of learning device 10 and / or welding result judgment device 20. Touch panel 211 may be provided on display 210. Furthermore, learning device 10 and / or welding result judgment device 20 may be indirectly or directly connected via communication I / F 209 to a server device, external device, etc. connected to a network such as the Internet (not shown).

[0041] Here, extraction and determination of feature amount information by the first auxiliary variable calculation unit 12 of the learning device 10 and the second auxiliary variable calculation unit 32 of the estimation device 30 of the welding result judgment device 20 will be described. As described above, the first and second auxiliary variable calculation units 12, 32 detect values ​​of multiple characteristic elements (several types of information) from the monitoring information 5, 6 including time-series data. The values ​​of the multiple characteristic elements are automatically extracted and detected in advance by computer processing or the like based on selection conditions confirmed in advance by experiments or the like. Then, the contribution of the detected values ​​of the multiple characteristic elements to the welding result is calculated by machine learning, for example. Based on the contribution calculated from the values ​​of the multiple characteristic elements thus detected, feature amount information to be the auxiliary variable is determined and calculated as teacher data for the auxiliary variable. The plurality of characteristic elements are, for example, at least one of the following: the time from the start of laser irradiation to the first inflection point at which a predetermined laser output is reached, the time from the start of laser irradiation to the start of melting, the temperature at the start of melting, the time from the end of laser irradiation to the second inflection point at which the temperature starts to drop, the time from the end of laser irradiation to the third inflection point at which solidification starts, the temperature at the start of solidification, the time from the first inflection point to the second inflection point, the temperature gradient from the first inflection point to the second inflection point, the time from the second inflection point to the third inflection point, the temperature gradient from the second inflection point to the third inflection point, and the total temperature during the laser output. Specifically, the following are assumed as the plurality of characteristic elements that can be used to extract and determine the characteristic amount information.

[0042] FIG. 3 is a graph showing an example of time series data. Examples of the time series data included in the monitoring information 5 and 6 include time series fluctuation data D1 of laser power and time series fluctuation data D2 of the intensity of radiated near-infrared light at the irradiation point, as shown in Fig. 3. In Fig. 3, the horizontal axis represents time (seconds) and the vertical axis represents temperature (×100°C), and a number of feature elements from which feature amount information in the data can be extracted and determined are shown.

[0043] As one of the values ​​of the multiple feature elements in Figure 3, "rise_period" is a value that means the time from laser on to the inflection point (prediction (70%)). This "rise_period" value is calculated from the number of measurement points (time x 5 μs) from the time the laser is on to the time it reaches 70% of the measured value at the time the laser rise is complete.

[0044] Furthermore, "inflection1_period" is a value that indicates the time from when the laser is turned on to when melting begins. The value of this "inflection1_period" is calculated from the number of measurement points (time x 5 μs) from the time the laser is turned on to inflection point 1 at the completion of the laser rise. "rise_value" is a value that indicates the inflection point (prediction (70%)). The value of this "rise_value" is calculated as 70% of the measured value at the time the laser rise is completed. "inflection1_value" is a value that indicates the temperature at the start of melting (melting point). The value of this "inflection1_value" is calculated as the measured value at inflection point 1.

[0045] Furthermore, "inflection3_value" is a value that indicates the temperature at the start of solidification (melting point). This "inflection3_value" value is calculated as the measurement value at inflection point 3, where solidification begins after the laser is stopped. "laseroff_inf3_period" is a value that indicates the time from laser off to the start of solidification. This "laseroff_inf3_period" value is calculated using the number of measurement points from laser off to inflection point 3 (time in 5 μs). "inflection3_5ms_value" is a value that indicates the first 5 ms of solidification (speed of cooling (hardness of solidified metal) x (related to molten volume)). This "inflection3_5ms_value" value is calculated as the measurement value 5 ms after inflection point 3.

[0046] Furthermore, "inflection2_value" is a value that indicates the temperature at which the temperature begins to drop. This "inflection2_value" value is calculated as the measurement value at inflection point 2, where the measurement value begins to drop after the laser is turned off. "inf2_inf3_period" is a value that indicates the time from when the temperature begins to drop to when solidification begins. This "inf2_inf3_period" value is calculated using the number of measurement points from inflection point 2 to inflection point 3 (time in 5 μs). "laseroff_inf2_period" is a value that indicates the time from when the laser is turned off to when the temperature begins to drop. This "laseroff_inf2_period" value is calculated using the number of measurement points from when the laser is turned off to when the temperature begins to drop.

[0047] In addition, "trend_start_value" means the first value (minimum) of the regression line interval. This "trend_start_value" value is calculated as the value at the time of inflection point 1 of the linear regression equation of the measured values ​​in the period from the time of inflection point 1 to the time of inflection point 2 (hereinafter referred to as the "specific period"). "trend_finish_value" means the last value (maximum) of the regression line interval. This "trend_finish_value" value is calculated as the value at the time of inflection point 2 of the linear regression equation of the measured values ​​in the specific period. "trend_period" is a value that means the time from the start of melting to the start of the temperature drop (time already calculated). This "trend_period" value is calculated at the time (μs) of the specific period. "trend_slope" is a value that means the rate of temperature increase during melting. This "trend_slope" value is calculated as the slope of the linear regression equation of the measured values ​​in the specific period.

[0048] Other feature elements, not shown in FIG. 3, include "rise_value_sum", "laserontime_value_sum", "count", "mean", "std", "min", "25%", "50%", "75%", "max", "dence1-5", etc. "rise_value_sum" is a value that means the total temperature rise. The value of this "rise_value_sum" is calculated by integrating the measured values ​​from the time the laser is turned on to "rise_period". "laserontime_value_sum" is a value that means the total temperature during laser output. The value of this "laserontime_value_sum" is calculated by integrating the measured values ​​from the time the laser is turned on to the time the laser is turned off.

[0049] Additionally, "count" is a value that indicates the time from when melting begins until the temperature starts to drop. This "count" value is calculated from the number of measurement points in a specific period (time x 5 μs). "Mean" is a value that indicates the degree of deviation from a linear trend (waviness). This "mean" value is calculated as the average value of the measured values ​​excluding the linear trend (linear regression) in a specific period. Additionally, "std" is a value that indicates the degree of deviation from a linear trend (waviness), just like "mean". This "std" value is calculated as the standard deviation of the measured values ​​in a specific period.

[0050] Additionally, "min" is a value that indicates the minimum temperature during melting. This "min" value is calculated as the minimum value of the measured values ​​in a specific period. "25%" means the 25th / 100th measured value in ascending order. This "25%" value is calculated as the 25th percentile of the measured values ​​in a specific period. "50%" means the 50th / 100th measured value in ascending order. This "50%" value is calculated as the 50th percentile of the measured values ​​in a specific period. "75%" means the 75th / 100th measured value in ascending order. This "75%" value is calculated as the 75th percentile of the measured values ​​in a specific period. "max" is a value that indicates the maximum temperature during melting. This "max" value is calculated as the maximum value of the measured values ​​in a specific period.

[0051] Furthermore, "dence1" is a value that indicates the (rough) degree of temperature fluctuation (jaggedness) based on the time-series fluctuation data D2. This "dence1" value is calculated using a spectral density of 195-585 Hz (the sum of the spectral densities of three specified frequencies). "dence2" is a value that indicates the (slightly rough) degree of temperature fluctuation (jaggedness) based on the time-series fluctuation data D2. This "dence2" value is calculated using a spectral density of 781-1171 Hz (the sum of the spectral densities of three specified frequencies).

[0052] "dence3" is a value that indicates the (medium) degree of temperature fluctuation (jaggedness) based on the time-series fluctuation data D2. The value of "dence3" is calculated using a spectral density of 1367-1757 Hz (the sum of the spectral densities of three specified frequencies). "dence4" is a value that indicates the (slightly finer) degree of temperature fluctuation (jaggedness) based on the time-series fluctuation data D2. The value of "dence4" is calculated using a spectral density of 1953-2343 Hz (the sum of the spectral densities of three specified frequencies). "dence5" is a value that indicates the (finer) degree of temperature fluctuation (jaggedness) based on the time-series fluctuation data D2. The value of "dence5" is calculated using a spectral density of 2539-2929 Hz (the sum of the spectral densities of three specified frequencies).

[0053] The first auxiliary variable calculation unit 12 and the second auxiliary variable calculation unit 32 detect, for example, values ​​of the plurality of automatically extracted feature elements from the plurality of feature elements as described above, calculate the contribution to the welding result, and determine and calculate the feature amount information based on the contribution calculated from the values ​​of the plurality of feature elements, but are not limited thereto. For example, the first auxiliary variable calculation unit 12 may detect the values ​​of the plurality of automatically extracted feature elements based on the magnitude of the contribution to the estimation of the objective variable. The contribution is weighted as the contribution to the estimation of various physical phenomena related to the welding quality, such as spatter and porosity, for the plurality of feature elements included in the time-series data of the teacher data of the monitoring information 5 by the machine learning algorithm used when the welding result learning unit 11 creates the welding result estimation model 3. The first auxiliary variable calculation unit 12 may detect the values ​​of the plurality of automatically extracted feature elements based on the contribution, and calculate the value as the feature amount information. Therefore, in the welding system 100, as described above, the graph shape formed by the processing conditions for good welding (processing condition information 2, 4) and the auxiliary variables of the processing results (monitoring information 5, 6), or the processing conditions for poor welding (processing condition information 2, 4) and the auxiliary variables of the processing results (monitoring information 5, 6) can be the teacher data. Since the monitoring information 5, 6 includes information on the laser beam emitted from the welding point (time-series fluctuation data D1 of the actually irradiated laser power, and time-series fluctuation data D2 of the radiated near-infrared light intensity at the irradiation point), when this is used as the objective variable, the processing condition information 2, 4 of each processing condition, the profile information of the laser beam, the machine specification information of the laser welder 40, etc. can also be used as the latent variable.

[0054] Then, by using the time series data analysis algorithm of welding system 100, when feature information extracted from monitoring information 6 at the time of laser welding at which the welding result from laser welder 40 should be determined and estimation data representing other processing states are input to welding result estimation model 3, which has been trained using processing condition information 2 (teacher data for explanatory variables), welding result information 1 (teacher data for objective variables), monitoring information 5 (teacher data for auxiliary variables), and teacher data for latent variables, welding result estimation model 3 can output a latent variable that shows a difference from the graph shape of the information related to the laser beam, which is the objective variable, based on the input data.

[0055] FIG. 4 is a diagram showing an example of a graph shape of time-series fluctuation data. The time-series fluctuation data D2 of the radiated near-infrared light intensity at the irradiation point included in the information on the laser beam outputted as described above is, for example, as shown in Fig. 4. In Fig. 4, the horizontal axis represents time (msec) and the vertical axis represents light intensity (counter value).

[0056] The results of the latent variables output from the welding result estimation model 3 are surrounded by dashed lines in the graph shown in FIG. 4 and represent the feature elements included in the specified feature waveform sections h1, h2, and h3, which become the values ​​of the automatically extracted and determined feature elements. The first auxiliary variable calculation unit 12 and the second auxiliary variable calculation unit 32 calculate feature amount information using the contributions calculated from the values ​​of the multiple feature elements. The feature waveform sections h1 to h3 include feature elements that represent the features of the waveform shape in the time-series fluctuation data D2 (such as jaggedness and inflection points). In the welding analysis algorithm of the welding system 100, various information included in the specified feature waveform sections h1 to h3 and information about the laser beam obtained from the monitoring information 6 are, for example, subjected to statistical processing to automatically detect welding abnormalities (such as insufficient welding strength and causes of welding defects), and are output as a welding result estimation value 8 representing a pass / fail judgment result, and are notified to, for example, a user.

[0057] The welding system 100 of the first embodiment calculates latent variables from data including monitoring information 5 (time-series fluctuation data D1, D2, feature amount information) to learn a welding result estimation model 3 for determining whether welding is good or bad, and can determine whether welding is good or bad based on the welding result estimation model 3 from the latent variables, processing condition information 4, and monitoring information 6. This makes it possible to predict and estimate the weld strength and the cause of welding defects accurately and easily with a simple configuration, while preventing deterioration of estimation accuracy due to correlation, even during welding.

[0058] In particular, when determining the cause of poor welding, welding result information 1 including cause information indicating the occurrence of spatters and focal position shift as teacher data for the objective variable is made to learn in welding result estimation model 3 together with teacher data for the explanatory variables (processing condition information 2) and teacher data for the auxiliary variables (monitoring information 5) as classification category information. Then, by inputting judgment data for the explanatory variables (processing condition information 4) and judgment data for the auxiliary variables (monitoring information 6) to welding result estimation model 3, the welding result estimation model 3 can be configured to output the cause of poor welding (welding result information 1) when it determines that the cause falls into a certain classification category of poor welding causes.

[0059] [Welding system operation] In the welding system 100, the welding result information 1 and the processing condition information 2 stored, for example, in an external storage device or storage medium, and the monitoring information 5 output from the welding monitor device 50 are input to the learning device 10 and stored in a storage unit (not shown), for example, as teacher data for the objective variable, teacher data for the explanatory variable, and teacher data for the auxiliary variable, respectively.

[0060] The learning device 10 calculates feature amount information (teaching data of auxiliary variables) in a first auxiliary variable calculation unit 12 from the input data including the teaching data of the monitoring information 5 from the welding monitor device 50. In addition, a first latent variable calculation unit 13 calculates teaching data of a latent variable having a smaller correlation with the teaching data of the machining condition information 2 than with the feature amount information from the feature amount information calculated in the first auxiliary variable calculation unit 12 and the teaching data of the machining condition information 2, and outputs latent variable calculation information 7.

[0061] Furthermore, learning device 10 inputs, in welding result learning unit 11, teacher data of welding result information 1, teacher data of processing condition information 2, and teacher data of the latent variables calculated by first latent variable calculation unit 13, and creates a welding result estimation model 3 for determining whether welding is good or bad, for example by performing machine learning. Learning device 10 stores latent variable calculation information 7 determined when calculating the teacher data of the latent variables and the created welding result estimation model 3 in a storage unit within the device, and outputs them to estimation device 30 of welding result judgment device 20 so that they can be used.

[0062] The learning device 10 may output the latent variable calculation information 7 and the welding result estimation model 3 to an external storage device (not shown), for example, via a connected network. In this case, the external storage device temporarily or permanently stores the input latent variable calculation information 7 and the welding result estimation model 3. The various data may be stored in, for example, a storage device (not shown) constituting a part of the welding system 100, instead of in the above-mentioned external storage device or storage medium.

[0063] Meanwhile, in laser welding machine 40, laser welding is performed based on set processing condition information 4. Monitoring information 6, which is state observation information during laser welding of an estimation target (e.g., a welded portion of a workpiece), is acquired by welding monitor device 50 using sensor 212 attached to laser welding machine 40. Processing condition information 4 set in laser welding machine 40 and monitoring information 6 of welding monitor device 50 are input to determination device 20 as estimation data for explanatory variables and estimation data for auxiliary variables, respectively.

[0064] The welding result determination device 20 stores various data input together with the latent variable calculation information 7 and the welding result estimation model 3 in a storage unit (not shown), and in the estimation device 30, the second auxiliary variable calculation unit 32 calculates feature amount information (estimation data of auxiliary variables) from data including estimation data of the monitoring information 6 from the welding monitor device 50. In addition, the second latent variable calculation unit 33 calculates estimation data of a latent variable that has a smaller correlation with the estimation data of the processing condition information 4 than with the feature amount information, based on the latent variable calculation information 7, from the feature amount information calculated by the second auxiliary variable calculation unit 32 and the estimation data of the processing condition information 4 set in the laser welding machine 40.

[0065] Furthermore, estimation device 30 of welding result determination device 20 outputs a welding result estimation value 8 for determining the quality of welding based on welding result estimation model 3 from the estimation data of the latent variables calculated by second latent variable calculation unit 33 in welding result estimation unit 31 and the estimation data of processing condition information 4. The welding result estimation value 8 is stored in the storage unit, and is also output from estimation device 30 in an appropriately usable form such as display or printing, and can be used by welding result determination device 20.

[0066] Note that, when calculating the feature amount information, calculating the latent variables, and creating the welding result estimation model 3, the learning device 10 may add the estimation data of the processing condition information 4 set in the laser welding machine 40 and the estimation data of the monitoring information 6 from the welding monitor device 50 as appropriate as teacher data for explanatory variables and auxiliary variables to the teacher data of the processing condition information 2 and the teacher data of the monitoring information 5, or may refer to and use the same as teacher data for explanatory variables and auxiliary variables. In this way, more data can be used to contribute to improving the estimation accuracy.

[0067] In the welding system 100 of the first embodiment, feature amount information and latent variables are calculated based on the monitoring information 5 and 6 as teacher data for the auxiliary variables of the welding result information 1, and then a welding result estimation model 3 is created using this information. In addition, feature amount information and latent variables are calculated again based on the monitoring information 6 as estimation data for the auxiliary variables using latent variable calculation information 7, and then a welding result estimation value 8 for determining whether the welding is good or bad is output based on the welding result estimation model 3 from this information and the processing condition information 4 as estimation data for the explanatory variables.

[0068] Therefore, according to the welding system 100 of the first embodiment, not only data related to the laser welding processing conditions but also feature information that contributes to determining whether the welding of the estimation target (the welded portion of the workpiece) is good or bad in a non-destructive manner can be automatically extracted from data monitored and observed during welding and used for the determination. Therefore, it is possible to determine whether the welding is good or bad even during welding, and to estimate the weld strength and the cause of the welding defect.

[0069] Generally, in laser welding, even when laser welding is performed under the same processing conditions, the weld strength varies depending on factors such as the material of the workpiece and the variation in the welding environment, so there is a problem that the weld strength cannot be measured unless the product is actually destroyed. In addition, there is also a problem that it is impossible to know the various processing conditions required to obtain the desired welding results or the cause of poor welding unless an experiment is actually performed in which a laser beam is irradiated.

[0070] However, the welding system 100 of the first embodiment uses feature amount information from the monitoring information as auxiliary variables and automatically extracts features that contribute to the pass / fail judgment, and adopts a configuration that can improve the judgment accuracy while making the pass / fail judgment including the fluctuation of the welding state as an event that occurred during welding, thereby solving such problems. In addition, since a latent variable is used, more precise estimation is possible, improving the estimation accuracy and avoiding deterioration of accuracy due to correlation.

[0071] The welding system 100 of the first embodiment is intended to determine whether or not the welding is good or bad in laser welding, but is not limited thereto. For example, it is possible to obtain information (such as which characteristic element acts and causes the occurrence of a physical phenomenon related to welding defects such as spattering and porosity) that indicates the cause (cause information) of the occurrence of the physical phenomenon, and it is also possible to estimate the welding quality associated with the welding result based on various physical phenomena related to the welding quality such as spattering and porosity. The data related to the welding quality includes, for example, at least one of data indicating a good welding and data indicating a poor welding, and the processing condition information 2 and monitoring information 5 at that time are used as teacher data for explanatory variables and auxiliary variables, and the learning device 10 performs, for example, machine learning to create a welding result estimation model 3. If the estimation device 30 uses this welding result estimation model 3 to perform estimation based on the estimation data of the processing condition information 4 and the monitoring information 6, it is possible to estimate the welding result including the quality of the welding, even during laser welding.

[0072] Furthermore, in the welding system 100, the learning device 10 and the welding result judgment device 20 can be installed, for example, centrally at the same site as the laser welder 40 or separately at a remote location. This enables a wide variety of operations, even for global deployment, regarding the judgment of the quality of laser welding.

[0073] [Second embodiment] Fig. 5 is an explanatory diagram showing a basic configuration of a welding system according to a second embodiment of the present invention. As shown in Fig. 5, a welding system 100A according to the second embodiment is similar to the welding system 100 of the first embodiment in that it uses monitoring information 5, 6 from a welding monitor device (resistance welding checker) 51 as auxiliary variables (teaching data and estimation target data) correlated with processing condition information 2, 4, but differs from the welding system 100 of the first embodiment in that it is applied to resistance welding.

[0074] That is, in the welding system 100A of the second embodiment, the welding device is a resistance welding machine (resistance welding power source and head 41). The teacher data input to the learning device 10 and the estimation data input to the welding result judgment device 20 include processing condition information 2, 4, which includes at least one processing condition of welding current, welding pressure, current application time, welding method, material of the workpiece, plate thickness of the workpiece, surface treatment of the workpiece, combination condition of the workpiece, sampling period, displacement detection resolution, and nugget diameter. The monitoring information 5, 6 includes time-series fluctuation data of the welding current and voltage and time-series fluctuation data of the welding pressure and displacement at the welding point. In addition, the multiple characteristic elements are at least one of the time from the start of current flow to the first inflection point where a predetermined welding current and voltage are reached, the time from the start of current flow to the start of melting, the temperature at the start of melting, the time from the end of current flow to the second inflection point where the temperature starts to decrease, the time from the end of current flow to the third inflection point where solidification starts, the temperature at the start of solidification, the time from the first inflection point to the second inflection point, the slope of the temperature from the first inflection point to the second inflection point, the time from the second inflection point to the third inflection point, the slope of the temperature from the second inflection point to the third inflection point, and the total temperature during current flow.

[0075] In the welding system 100A of the second embodiment, the first auxiliary variable calculation unit 12 of the learning device 10 calculates the feature amount information (teaching data of the auxiliary variables) based on the monitoring information 5, the first latent variable calculation unit 13 calculates the teaching data of the latent variables based on the feature amount information, the welding result learning unit 11 learns the welding result estimation model 3, and provides the welding result estimation model 3 and the latent variable calculation information 7 to the estimation device 30 of the welding result judgment device 20. The second auxiliary variable calculation unit 32 of the estimation device 30 of the welding result judgment device 20 calculates the feature amount information (estimation data of the auxiliary variables) based on the monitoring information 6, the second latent variable calculation unit 33 calculates the estimation data of the latent variables based on the latent variable calculation information 7 from the feature amount information and the estimation data of the processing condition information 4, and the welding result estimation unit 31 inputs the data including the estimation data of the latent variables, and judges the quality of the welding based on the welding result estimation model 3. Then, the welding result of the welding strength and the cause of the welding defect included in the welding result estimation value 8 of the obtained quality judgment result can be judged even during welding. In this way, welding system 100A of the second embodiment can also achieve the same effects as welding system 100 of the first embodiment.

[0076] [Example] Fig. 6 is a graph showing the relationship between the welding temperature and elapsed time to explain auxiliary variables in an example of the welding system according to the first embodiment. Fig. 7 is a graph plotting regression errors when performing multiple regression on the welding temperature in the example with the processing conditions. Fig. 8 is a graph showing the relationship between the estimated value and the measured value in the comparative example and the example, and the frequency distribution of errors of the estimated value with respect to the estimated value.

[0077] In the laser welding of the embodiment, the processing conditions as explanatory variables were the welding conditions (the laser output intensity and irradiation time used in the laser welding), and the objective variable was the welding strength. In the learning device 10, the welding result estimation model 3 was created using multiple regression analysis. In the comparative example, the above-mentioned latent variables were not used.

[0078] As shown in FIG. 6, according to a graph in which the welding temperature of a welding point in laser welding is plotted on the vertical axis and the elapsed time of laser welding is plotted on the horizontal axis, the temperature of the welding point starts to rise suddenly at the same time as the start of laser irradiation at the elapsed time T1. Then, at a certain point after the elapsed time T1 (elapsed time T2), the temperature rise slows down, and at the same time as the end of the laser irradiation after the elapsed time T2 (elapsed time T3), the temperature drops suddenly, and at a certain point after the elapsed time T3 (elapsed time T4), it returns to the temperature before the start of the laser irradiation. The welding temperature in laser welding changes with such characteristics. Therefore, the temperature TP1 at the point when the temperature rise slows down (elapsed time T2) and the temperature TP2 at the end of the laser irradiation (elapsed time T3) were used as auxiliary variables used to estimate the welding strength.

[0079] As shown in Figure 7, the regression error when multiple regression is performed on the welding conditions (laser output intensity and irradiation time used for laser welding) for temperature TP1, which is the start-up temperature, is plotted on the horizontal axis, and the regression error when multiple regression is performed on temperature TP2, which is the temperature at the end of laser irradiation, under the same welding conditions as above is plotted on the vertical axis, and the obtained data is subjected to principal component analysis. This results in a one-dimensional latent variable that has been dimension-reduced along the projection axis PA of the first principal component.

[0080] The estimated values ​​of weld strength and the actual measured values ​​of weld strength in the comparative example are shown in FIG. 8(a). Meanwhile, the estimated values ​​of weld strength and the actual measured values ​​of weld strength in the example in which the obtained latent variables were added as explanatory variables and multiple regression analysis was performed are shown in FIG. 8(b). In FIGS. 8(a) and (b), the horizontal axis represents the measured values ​​of weld strength when the welding condition values ​​are changed to various conditions, and the vertical axis represents the estimated values ​​of weld strength corresponding to the welding conditions. In addition, in FIGS. 8(a) and (b), both the measured values ​​and the estimated values ​​are plotted after being normalized so that the maximum weld strength is 1.

[0081] The error according to the comparative example and the error according to the embodiment were evaluated in terms of the mean absolute error (MAE) and the root mean squared error (RMSE), and the results are shown in Table 1 below.

[0082] [Table 1]

[0083] The error in the embodiment was smaller than the error in the comparative example in both MAE and RMSE. As can be seen from FIG. 8 and Table 1, it can be understood that the estimated value is closer to the measured value due to the effect of using the latent variables. Therefore, it was proven that the welding system of the embodiment can improve the accuracy of judgment even if more data including auxiliary variables and latent variables are used.

[0084] Although several embodiments of the present invention have been described above, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are included in the scope of the invention and its equivalents described in the claims. [Explanation of symbols]

[0085] 1. Welding result information (training data for the objective variable) 2 Processing condition information (teaching data for explanatory variables) 3. Welding result estimation model 4 Processing condition information (data for estimating explanatory variables) 5. Monitoring information (teaching data for auxiliary variables) 6 Monitoring information (data for estimating auxiliary variables) 7 Latent variable calculation information 8. Estimated welding results 10 Learning Device 11 Welding Results Learning Section 12 First auxiliary variable calculation section 13 First latent variable calculation section 20 Welding result judgment device 30 Estimation device 31 Welding result estimation section 32 Second auxiliary variable calculation unit 33 Second latent variable calculation section 40 Laser Welding Machine 41 Resistance welding power supply head 50,50A Welding Monitor Device 51 Resistance welding checker 100 Welding System

Claims

1. A laser welder that welds the workpieces; a welding monitor device that monitors the progress of welding by the laser welder using a sensor and outputs monitoring information obtained by the monitoring device; a learning device that inputs, as training data, welding result information related to a welding result obtained after welding by the laser welder and feature amount information extracted from the monitoring information at the time of welding when the welding result was obtained, and creates and outputs a welding result estimation model based on the training data; a welding result judgment device that extracts feature amount information from monitoring information output from the welding monitor device during welding for which a welding result is to be judged, inputs the feature amount information as estimation data into the welding result estimation model created by the learning device, and outputs a quality judgment result of the welding, the learning device detects values ​​of a plurality of preset feature elements from the monitoring information including time-series data, calculates contributions of the detected values ​​of the plurality of feature elements to the welding result, and determines the feature amount information based on the calculated contributions from the values ​​of the plurality of feature elements; The plurality of characteristic elements are at least one of a time from the start of laser irradiation to a first inflection point at which a predetermined laser output is reached, a time from the start of laser irradiation to the start of melting, a temperature at the start of melting, a time from the end of laser irradiation to a second inflection point at which the temperature starts to decrease, a time from the end of laser irradiation to a third inflection point at which solidification starts, a temperature at the start of solidification, a time from the first inflection point to the second inflection point, a temperature gradient from the first inflection point to the second inflection point, a time from the second inflection point to the third inflection point, a temperature gradient from the second inflection point to the third inflection point, and a total temperature during the laser output. Welding system.

2. A resistance welding machine for welding the workpiece; a welding monitor device that monitors the progress of welding by the resistance welding machine using a sensor and outputs the obtained monitoring information; a learning device that inputs, as teacher data, welding result information related to a welding result obtained after welding by the resistance welding machine and feature amount information extracted from the monitoring information at the time of welding when the welding result was obtained, and creates and outputs a welding result estimation model based on the teacher data; a welding result judgment device that extracts feature amount information from monitoring information output from the welding monitor device during welding for which a welding result is to be judged, inputs the feature amount information as estimation data into the welding result estimation model created by the learning device, and outputs a quality judgment result of the welding, the learning device detects values ​​of a plurality of preset feature elements from the monitoring information including time-series data, calculates contributions of the detected values ​​of the plurality of feature elements to the welding result, and determines the feature amount information based on the calculated contributions from the values ​​of the plurality of feature elements; The plurality of characteristic elements are at least one of the following: a time from the start of current flow to a first inflection point where a predetermined welding current / voltage is reached, a time from the start of current flow to the start of melting, a temperature at the start of melting, a time from the end of current flow to a second inflection point where the temperature starts to decrease, a time from the end of current flow to a third inflection point where solidification starts, a temperature at the start of solidification, a time from the first inflection point to the second inflection point, a temperature gradient from the first inflection point to the second inflection point, a time from the second inflection point to the third inflection point, a temperature gradient from the second inflection point to the third inflection point, and a total temperature during current flow. Welding system.

3. The welding result judgment device includes a notification unit that notifies the pass / fail judgment result in a manner that allows at least one of visual confirmation and audibility confirmation, The pass / fail judgment result includes at least one of information regarding the weld strength of the weld and information regarding the cause of the weld. The welding system according to claim 1 or 2.

4. the teacher data input to the learning device and the estimation data input to the welding result determination device include processing condition information, The processing condition information includes at least one processing condition of a material of a workpiece, a plate thickness of the workpiece, a laser power, a laser irradiation time, a time from the start of laser irradiation to the time when the peak power is reached, a time from the peak power to the time when the laser irradiation is ended, a fiber diameter, a lens focal length, a focal position, and a laser diameter at an irradiation point; The monitoring information includes time-series fluctuation data of laser power and time-series fluctuation data of radiated near-infrared light intensity at an irradiation point. The welding system of claim 1 .

5. the teacher data input to the learning device and the estimation data input to the welding result determination device include processing condition information, The processing condition information includes at least one processing condition of a welding current, a welding pressure, a current application time, a welding method, a material of the workpiece, a plate thickness of the workpiece, a surface treatment of the workpiece, a combination condition of the workpiece, a sampling period, a displacement amount detection resolution, and a nugget diameter, The monitoring information includes time-series fluctuation data of welding current and voltage and time-series fluctuation data of welding pressure and displacement amount of a welding point. The welding system of claim 2 .

6. A progress status of welding by a laser welding machine that welds the workpiece is monitored by a sensor, and monitoring information obtained by the monitoring is input from a welding monitor device; a welding result estimation model is created using, as training data, welding result information related to a welding result obtained after welding by the laser welder and feature amount information extracted from the monitoring information at the time of welding when the welding result is obtained; a welding result judgment device extracting feature amount information from monitoring information output from the welding monitor device during welding for which a welding result is to be judged, inputting the feature amount information as estimation data into the welding result estimation model, and outputting a quality judgment result of the welding, the welding result estimation model detects values ​​of a plurality of preset feature elements from the monitoring information including time-series data, calculates contributions of the detected values ​​of the plurality of feature elements to the welding result, and determines the feature amount information based on the calculated contributions from the values ​​of the plurality of feature elements; The plurality of characteristic elements are at least one of a time from the start of laser irradiation to a first inflection point at which a predetermined laser output is reached, a time from the start of laser irradiation to the start of melting, a temperature at the start of melting, a time from the end of laser irradiation to a second inflection point at which the temperature starts to decrease, a time from the end of laser irradiation to a third inflection point at which solidification starts, a temperature at the start of solidification, a time from the first inflection point to the second inflection point, a temperature gradient from the first inflection point to the second inflection point, a time from the second inflection point to the third inflection point, a temperature gradient from the second inflection point to the third inflection point, and a total temperature during the laser output. Welding result judgment device.

7. A progress status of welding by a resistance welding machine that welds the workpiece is monitored by a sensor, and monitoring information obtained by the monitoring is input from a welding monitor device; a welding result estimation model is created using, as training data, welding result information related to a welding result obtained after welding by the resistance welding machine and feature amount information extracted from the monitoring information at the time of welding when the welding result is obtained; a welding result judgment device extracting feature amount information from monitoring information output from the welding monitor device during welding for which a welding result is to be judged, inputting the feature amount information as estimation data into the welding result estimation model, and outputting a quality judgment result of the welding, the welding result estimation model detects values ​​of a plurality of preset feature elements from the monitoring information including time-series data, calculates contributions of the detected values ​​of the plurality of feature elements to the welding result, and determines the feature amount information based on the calculated contributions from the values ​​of the plurality of feature elements; The plurality of characteristic elements are at least one of the following: a time from the start of current flow to a first inflection point where a predetermined welding current / voltage is reached, a time from the start of current flow to the start of melting, a temperature at the start of melting, a time from the end of current flow to a second inflection point where the temperature starts to decrease, a time from the end of current flow to a third inflection point where solidification starts, a temperature at the start of solidification, a time from the first inflection point to the second inflection point, a temperature gradient from the first inflection point to the second inflection point, a time from the second inflection point to the third inflection point, a temperature gradient from the second inflection point to the third inflection point, and a total temperature during current flow. Welding result judgment device.

Citation Information

Patent Citations

  • Mechanical learning device, laser processing system and mechanical learning method

    JP2017164801A

  • Butt Weld Monitoring Method and Butt Weld Monitoring Apparatus

    JP3275988B2

  • JPP6818970B

  • JPP6854984B

  • System and Method to Facilitate Welding Software as a Service

    US20170032281A1