Prediction / estimation system, learning device, and prediction / estimation device

Abstract

To improve accuracy of prediction and estimation utilizing more data.SOLUTION: A learning device according to one aspect includes: a latent variable calculation unit which, when the learning device inputs teacher data on an objective variable, teacher data on an explanatory variable which acts on the objective variable, and teacher data on an auxiliary variable which is a variable other than the objective variable and correlates with the explanatory variable and which is observed when the explanatory variable is given, calculates teacher data on a latent variable from at least the teacher data on the auxiliary variable and outputs latent variable calculation information; and a prediction and estimation model learning unit which, when the learning device inputs the teacher data on the objective variable, the teacher data on the explanatory variable, and the teacher data on the latent variable calculated by the latent variable calculation unit, creates a prediction and estimation model of the objective variable.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a prediction / estimation system, a learning device, and a prediction / estimation device. [Background technology]

[0002] In the prediction and estimation of objective variables using supervised machine learning, objective variables and explanatory variables are generally used as training data for the prediction and estimation model. In addition, auxiliary variables (auxiliary variables), which are variables other than objective variables that are available at the time of prediction and estimation, may be used as variables that have a causal relationship that results when the explanatory variables are the cause.

[0003] As an example of a method for training such a prediction / estimation model, the following is known: First, the chemical substance values ​​of multiple experimental samples are used as training data for the objective variable, and a portion of the near-infrared spectral data converted from the collected spectra of multiple experimental samples, which is the training data for the explanatory variables, is randomly selected to be used as a calibration set and the other portion as a validation set.

[0004] Then, the calibration set and validation set are subjected to principal component analysis to obtain a spectral feature space, and the samples that are most similar to each other in the calibration set and validation set in this spectral feature space are selected as a calibration subset using the Mahalanobis distance method, and collected as a syndrome set.

[0005] Then, the number of principal components is extracted from the syndrome set as training data for auxiliary variables, and a regression model is constructed as the input layer of a backpropagation (BP) neural network to train a prediction / estimation model (see Patent Document 1). [Prior art documents] [Patent documents]

[0006] [Patent Document 1] China Patent Application Publication No. 107655850 Summary of the Invention [Problem to be solved by the invention]

[0007] However, in the system disclosed in the above Patent Document 1, a regression model is constructed using the number of principal components, which is the training data for auxiliary variables, instead of the near-infrared spectrum data, which is the training data for explanatory variables, to learn a prediction / estimation model. In order to improve the accuracy of prediction / estimation, it is desirable to learn a prediction / estimation model using more training data. For this reason, it is also conceivable to learn a prediction / estimation model using both explanatory variables and auxiliary variables as training data.

[0008] However, in the above-mentioned system, the auxiliary variables have a correlation with the explanatory variables, which tends to deteriorate the accuracy of prediction and estimation. This tendency is widely known as multicollinearity, especially in multiple regression analysis.

[0009] One aspect of the present invention is a prediction / estimation system, a learning device, and a prediction / estimation device that are capable of improving the accuracy of prediction / estimation. [Means for solving the problem]

[0010] A prediction / estimation system according to one aspect of the present invention includes a learning device that receives as input teacher data for a dependent variable, teacher data for explanatory variables acting on the dependent variable, and teacher data for auxiliary variables that are variables other than the dependent variable and are correlated with the explanatory variable and are observed when the explanatory variable is given, and creates a prediction / estimation model for the dependent variable; and a prediction / estimation device that receives as input prediction / estimation data for the explanatory variables for predicting / estimating the dependent variable and prediction / estimation data for the auxiliary variables for predicting / estimating the dependent variable, and calculates a predicted / estimated value of the dependent variable based on the prediction / estimation model created by the learning device, and the learning device receives teacher data for latent variables from at least the auxiliary variable teacher data. the prediction and estimation device includes a first latent variable calculation unit that calculates data and outputs latent variable calculation information, and a prediction and estimation model learning unit that inputs teacher data of the objective variable, teacher data of the explanatory variables, and teacher data of the latent variables calculated by the first latent variable calculation unit, and creates a prediction and estimation model of the objective variable, and the prediction and estimation device includes a second latent variable calculation unit that calculates prediction and estimation data of a latent variable from prediction and estimation data of at least the auxiliary variables, based on the latent variable calculation information, and a prediction and estimation unit that calculates a prediction and estimation value of the objective variable from the prediction and estimation data of the latent variable calculated by the second latent variable calculation unit and the prediction and estimation data of the explanatory variables, based on the prediction and estimation model.

[0011] According to a prediction / estimation system according to an embodiment of the present invention, teacher data of a response variable, teacher data of an explanatory variable, and teacher data of an auxiliary variable are input, teacher data of a latent variable is calculated from at least the teacher data of the auxiliary variable, and latent variable calculation information and a prediction / estimation model of the response variable are created. Therefore, by using teacher data of a latent variable in which the correlation between the explanatory variable teacher data or between the explanatory variables is reduced more than that of the auxiliary variables, it is possible to use more teacher data and to create a more accurate prediction / estimation model by eliminating the correlation between the teacher data. Furthermore, from at least the prediction / estimation data of the auxiliary variables, prediction / estimation data of the latent variable is calculated based on the latent variable calculation information, and a prediction / estimation value of the response variable is calculated based on the calculated prediction / estimation data of the latent variable and the prediction / estimation data of the explanatory variables based on the prediction / estimation model. For this reason, by using prediction / estimation data for latent variables that have a lower correlation between the prediction / estimation data for the explanatory variables or between each other than the auxiliary variables, it is possible to use a larger amount of prediction / estimation data and to obtain more accurate predictions / estimates of the dependent variable by eliminating correlations between the prediction / estimation data. Effect of the Invention

[0012] According to one aspect of the present invention, the accuracy of prediction and estimation can be improved by utilizing more data. [Brief description of the drawings]

[0013] [Figure 1] FIG. 1 is an explanatory diagram showing the basic configuration of a prediction / estimation 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 prediction / estimation device of the prediction / estimation system. [Diagram 3] FIG. 3 is an explanatory diagram showing the basic configuration of a prediction / estimation system according to the second embodiment of the present invention. [Figure 4] FIG. 4 is an explanatory diagram showing the basic configuration of a prediction / estimation system according to the third embodiment of the present invention. [Diagram 5] FIG. 5 is an explanatory diagram showing the basic configuration of a prediction / estimation system according to a fourth embodiment of the present invention. [Figure 6] FIG. 6 is an explanatory diagram showing the basic configuration of a prediction / estimation system according to the fifth embodiment of the present invention. [Figure 7] FIG. 7 is an explanatory diagram showing a basic configuration of a system for estimating the welding strength of laser welding to which a prediction / estimation system according to a sixth embodiment of the present invention is applied. [Figure 8] FIG. 8 is an explanatory diagram showing the basic configuration of a learning device for a nondestructive measurement system to which a prediction / estimation system according to a seventh embodiment of the present invention is applied. [Figure 9] FIG. 9 is an explanatory diagram showing the basic configuration of a nondestructive testing device of a nondestructive measurement system. [Figure 10] FIG. 10 is an explanatory diagram showing a basic configuration of a quality judgment and prediction system for the cutting quality of a laser cutting device to which a prediction and estimation system according to an eighth embodiment of the present invention is applied. [Figure 11] FIG. 11 is a graph showing the relationship between the welding temperature and the elapsed time to explain auxiliary variables in an example in which the prediction / estimation system according to the fourth embodiment shown in FIG. 5 is applied to actual laser welding. [Figure 12] FIG. 12 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

[0014] Hereinafter, a prediction / estimation system, a learning device, and a prediction / estimation device according to embodiments of the present invention will be described in detail with reference to the attached drawings. However, the following embodiments do not limit the invention according to each claim, and not all combinations of features described in the embodiments are necessarily essential to the solution of the invention. In addition, in the following embodiments, the scale and dimensions of each component may be exaggerated, and some components may be omitted.

[0015] [First embodiment] Fig. 1 is an explanatory diagram showing the basic configuration of a prediction and estimation system according to a first embodiment of the present invention, and Fig. 2 is an explanatory diagram showing the basic hardware configuration of a learning device and / or a prediction and estimation device of the prediction and estimation system.

[0016] As shown in Fig. 1, a prediction / estimation system 100 according to the first embodiment is a system for obtaining predicted / estimated values ​​8 of various types of dependent variables. The prediction / estimation system 100 includes a learning device 10 that receives as input teacher data 1 of dependent variables, teacher data 2 of explanatory variables acting on dependent variables, and teacher data 3 of auxiliary variables that are variables other than dependent variables and have a correlation with the explanatory variables and are observed when the explanatory variables are given, and creates a prediction / estimation model 7 of the dependent variables by machine learning, for example, and a prediction / estimation device 20 that receives as input prediction / estimation data 4 of explanatory variables for predicting / estimating the dependent variable, and prediction / estimation data 5 of auxiliary variables for predicting / estimating the dependent variable, and calculates a predicted / estimated value 8 of the dependent variable based on the prediction / estimation model 7 created by the learning device 10.

[0017] The learning device 10 includes a first latent variable calculation unit 11 that calculates teacher data (not shown) of latent variables that have a smaller correlation (reduced correlation) between the teacher data 2 of explanatory variables or between the teacher data 2 of explanatory variables or between the teacher data 2 of explanatory variables than the teacher data 3 of auxiliary variables from at least the teacher data 3 of auxiliary variables, and outputs latent variable calculation information 6, and a prediction / estimation model learning unit 12 that inputs the teacher data 1 of dependent variables, the teacher data 2 of explanatory variables, and the teacher data of latent variables calculated by the first latent variable calculation unit 11, and performs, for example, machine learning to create a prediction / estimation model 7 of the dependent variable.

[0018] The prediction / estimation device 20 includes a second latent variable calculation unit 21 that calculates prediction / estimation data (not shown) of a latent variable that has a smaller correlation (reduced correlation) between the prediction / estimation data 4 of the explanatory variables or between the prediction / estimation data 4 of the explanatory variables or between the prediction / estimation data 4 of the explanatory variables or between the prediction / estimation data 5 of the auxiliary variables based on latent variable calculation information 6 from the prediction / estimation data 5 of the auxiliary variables at least, and a prediction / estimation unit 22 that calculates a prediction / estimation value 8 of the dependent variable based on a prediction / estimation model 7 from the prediction / estimation data of the latent variable calculated by the second latent variable calculation unit 21 and the prediction / estimation data 4 of the explanatory variables.

[0019] As described above, the prediction / estimation system 100 is a system capable of obtaining prediction / estimated values ​​of various types of objective variables. When the prediction / estimation system 100 is applied to an estimation system for the welding strength of laser welding, which will be described later, for example, the objective variable is the welding strength F (N / mm 2 ) are included. In addition, for example, explanatory variables include welding conditions (laser output value, laser irradiation time, etc.). Furthermore, auxiliary variables having a correlation with explanatory variables include, for example, physical quantities observed during or after welding (welding temperature, sound, light, color, etc.). Note that the objective variables, explanatory variables, and auxiliary variables used in the prediction / estimation system 100 are not limited to these exemplified ones.

[0020] In addition, the teacher data of the latent variables and the prediction / estimation data of the latent variables calculated by the first latent variable calculation unit 11 of the learning device 10 and the second latent variable calculation unit 21 of the prediction / estimation device 20 are, for example, those in which the auxiliary variables, or the auxiliary variables and the explanatory variables, have been subjected to dimensionality reduction or dimensionality compression so as to reduce correlation between the explanatory variables or between each other more than the auxiliary variables.

[0021] That is, in this embodiment, the latent variable is a variable that characterizes the objective variable in the same way as the explanatory variable, and is a variable that is not directly observed but is estimated from the observed variables (auxiliary variables). In other words, the latent variable means a variable that characterizes the observed variables and is dimension-reduced or dimension-compressed from at least the auxiliary variables.

[0022] The explanatory variables, objective variables, latent variables, and latent variable calculation information 6 used in the prediction / estimation system 100 may each be composed of one or more variables or parameters. In this specification, "objective variable prediction / estimation" means "objective variable prediction" or "objective variable estimation." "Objective variable prediction" means assuming an objective variable that is thought to be realized in the future, and "objective variable estimation" means estimating an objective variable that is realized at present but cannot be directly observed. "Correlation" is not limited to a linear relationship between multiple variables, but means a relationship between multiple variables in which a change in one variable causes a change in the other variables, regardless of whether it is linear or nonlinear.

[0023] The first latent variable calculation unit 11 of the learning device 10 calculates teacher data of latent variables from data including at least the teacher data 3 of auxiliary variables among data including the teacher data 1 of the objective variables, the teacher data 2 of the explanatory variables, and the teacher data 3 of the auxiliary variables input to the learning device 10. The first latent variable calculation unit 11 calculates the teacher data of latent variables using, for example, a predetermined algorithm. The algorithm includes, 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). Note that the algorithm is not limited to these. The latent variable calculation information 6 means various information that is determined in the first latent variable calculation unit 11 when calculating teacher data for latent variables and is used when calculating additional latent variables, and includes, for example, various parameters, various formulas and / or algorithms.

[0024] The prediction / estimation model learning unit 12 of the learning device 10 inputs the teacher data 1 of the objective variable, the teacher data 2 of the explanatory variable, and the teacher data of the latent variable calculated by the first latent variable calculation unit 11 input to the learning device 10, and creates a prediction / estimation model 7 of the objective variable by, for example, machine learning. The learning method includes, but is not limited to, algorithms such as regression, classification, clustering, discrimination, interpolation, feature extraction, and time series modeling. The prediction / estimation model 7 learned by the prediction / estimation model learning unit 12 includes various parameters, various formulas, and / or algorithms for prediction / estimation that characterize the learning model. The learning device 10 then outputs at least the latent variable calculation information 6 and the prediction / estimation model 7 to the prediction / estimation device 20.

[0025] On the other hand, the second latent variable calculation unit 21 of the prediction / estimation device 20 calculates prediction / estimation data of the latent variables from prediction / estimation data 5 of at least the auxiliary variables, based on the latent variable calculation information 6. Therefore, when the latent variable calculation information 6 includes an algorithm, the second latent variable calculation unit 21 calculates prediction / 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 11 of the learning device 10. In this way, by calculating the latent variables using the same algorithm as that used during learning, even during prediction / estimation, prediction / estimation using the same prediction / estimation model 7 becomes possible.

[0026] A prediction / estimation section 22 of the prediction / estimation device 20 calculates and outputs a predicted / estimated value 8 of the objective variable based on the prediction / estimation data 4 of the explanatory variables input to the prediction / estimation device 20, the prediction / estimation model 7, and the prediction / estimation data of the latent variable calculated by the second latent variable calculation section 21.

[0027] The latent variable calculation information 6 passed from the learning device 10 to the prediction / estimation device 20 may be selected as appropriate for the calculation of latent variables in the learning device 10 or the prediction / estimation device 20. For example, when the calculation of teacher data for 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, latent variable parameters excluding principal component vectors in which the contribution rate of teacher data 3 of a specific auxiliary variable is higher than a predetermined value may be selected. This allows latent variables that have a smaller correlation with explanatory variables or between them than auxiliary variables to be used as teacher data and prediction / estimation data together with explanatory variables, making it possible to avoid deterioration of prediction / estimation accuracy caused by multicollinearity.

[0028] In the prediction / estimation system 100, each of the teacher data 1 of the objective variables, the teacher data 2 of the explanatory variables, the teacher data 3 of the auxiliary variables, the prediction / estimation data 4 of the explanatory variables, and the prediction / estimation data 5 of the auxiliary variables may be stored in, for example, a stationary or portable storage device or storage medium (not shown). Each of the data may be transmitted and received via an information and communication medium such as the Internet, and may be raw data acquired by a measuring device such as a sensor (not shown). The latent variable calculation information 6 and the prediction / estimation model 7 may be input / output between the learning device 10 and the prediction / estimation device 20 via the above-mentioned storage medium or information and communication medium.

[0029] 2, the learning device 10 and / or the prediction / estimation device 20 of the prediction / estimation system 100 includes, as basic hardware configuration, for example, a CPU 201, a RAM 202, a ROM 203, an HDD (hard disk drive) 204, an SSD (solid state drive) 205, and a memory card 206. The learning device 10 and / or the prediction / estimation device 20 also includes, for example, an input I / F (interface) 207, an output I / F (interface) 208, and a communication I / F (interface) 209. The components 201 to 209 are connected to each other via a bus 200.

[0030] The CPU 201 controls the learning device 10 and / or the prediction and estimation device 20 by executing various programs stored in the RAM 202, the ROM 203, the HDD 204, the SSD 205, etc. In the learning device 10, the CPU 201 executes a learning program to realize the functions of the first latent variable calculation unit 11 and the prediction and estimation model learning unit 12. In the prediction and estimation device 20, the CPU 201 executes a prediction and judgment program to realize the functions of the second latent variable calculation unit 21 and the prediction and estimation unit 22. The prediction and estimation system 100 may be configured to be controlled in its entirety by the CPUs 201 of the learning device 10 and the prediction and estimation device 20 working together.

[0031] 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.

[0032] For example, a sensor 212 is connected to the input I / F 207 to acquire detection information. The sensor 212 includes various sensors such as a temperature sensor, an optical sensor, an acoustic sensor, and an image sensor. A touch panel 211 that functions as an operation unit or input unit of the learning device 10 and / or the prediction / estimation device 20 is connected to the input I / F 207, and receives information associated with an operation input from a user of the prediction / estimation 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.

[0033] A display 210 serving as, for example, a display device is connected to output I / F 208, and various information is output to be displayed on the monitor of learning device 10 and / or prediction / estimation device 20. Touch panel 211 may be provided on display 210. Furthermore, learning device 10 and / or prediction / estimation device 20 may be indirectly or directly connected, via communication I / F 209, to a server device and external devices connected to a network such as the Internet (not shown).

[0034] In the prediction and estimation system 100 of the first embodiment, the first latent variable calculation unit 11 of the learning device 10 calculates teacher data of the latent variables from data including at least teacher data 3 of the auxiliary variables, and the prediction and estimation model learning unit 12 learns and creates a prediction and estimation model 7 using the data including the teacher data of the latent variables. Then, the learning device 10 provides the latent variable calculation information 6 determined by the calculation of the teacher data of the latent variables to the prediction and estimation device 20. The second latent variable calculation unit 21 of the prediction and estimation device 20 to which the latent variable calculation information 6 has been provided calculates prediction and estimation data of the latent variables from data including at least prediction and estimation data 5 of the auxiliary variables using the same algorithm as the algorithm used by the first latent variable calculation unit 11, and the prediction and estimation unit 22 calculates the prediction and estimation value 8 of the objective variable using the data including the prediction and estimation data of the latent variables. As a result, according to the prediction and estimation system 100 of the first embodiment, the accuracy of prediction and estimation can be improved by converting the auxiliary variables into latent variables.

[0035] [Second embodiment] Fig. 3 is an explanatory diagram showing the basic configuration of a prediction / estimation system according to a second embodiment of the present invention. In the following description including Fig. 3, the same components as those in the first embodiment and its modified examples are given the same reference numerals, and therefore duplicated descriptions will be omitted below.

[0036] 3, in a prediction / estimation system 100A according to the second embodiment, a first latent variable calculation unit 11 of a learning device 10 receives as input teacher data 2 of explanatory variables and teacher data 3 of auxiliary variables, calculates teacher data of latent variables (not shown), and outputs latent variable calculation information 6. Furthermore, a second latent variable calculation unit 21 of a prediction / estimation device 20 receives as input prediction / estimation data 4 of explanatory variables and prediction / estimation data 5 of auxiliary variables, and calculates prediction / estimation data of the latent variables (not shown) based on the latent variable calculation information 6.

[0037] That is, the prediction / estimation system 100A of the second embodiment differs from the learning device 10 of the prediction / estimation system 100 of the first embodiment in that the data input to the first latent variable calculation unit 11 of the learning device 10 for calculating the teacher data of the latent variables includes teacher data 2 of the explanatory variables in addition to teacher data 3 of the auxiliary variables. Also, the prediction / estimation device 20 differs from the prediction / estimation device 20 of the prediction / estimation system 100 of the first embodiment in that the data input to the second latent variable calculation unit 21 of the prediction / estimation device 20 for calculating the prediction / estimation data of the latent variables includes prediction / estimation data 4 of the explanatory variables in addition to prediction / estimation data 5 of the auxiliary variables.

[0038] According to the second embodiment, the first latent variable calculation unit 11 of the learning device 10 and the second latent variable calculation unit 21 of the prediction / estimation device 20 calculate the teacher data and prediction / estimation data for each latent variable by using data related to the explanatory variables. This enables highly accurate prediction / estimation using latent variables with reduced correlation between the explanatory variables and auxiliary variables. Furthermore, according to the second embodiment, since more data is used to calculate the latent variables, it is possible to calculate the latent variables based on higher dimensional and more advanced algorithms. This makes it possible to further improve the accuracy of prediction / estimation using auxiliary variables while avoiding deterioration of prediction / estimation accuracy due to correlation.

[0039] [Third embodiment] Fig. 4 is an explanatory diagram showing the basic configuration of a prediction / estimation system according to a third embodiment of the present invention. As shown in Fig. 4, in a prediction / estimation system 100B according to the third embodiment, a second latent variable calculation unit 21 of a prediction / estimation device 20 calculates prediction / estimation data 23 of latent variables based on teacher data 2 of explanatory variables, teacher data 3 of auxiliary variables, and teacher data 13 of latent variables calculated by a first latent variable calculation unit 11 of a learning device 10, in addition to prediction / estimation data 4 of explanatory variables and prediction / estimation data 5 of auxiliary variables.

[0040] That is, the prediction / estimation system 100B of the third embodiment differs from the prediction / estimation system 100A of the second embodiment in that the second latent variable calculation unit 21 of the prediction / estimation device 20 includes, as input data for calculating prediction / estimation data 23 for latent variables, teacher data 2 for explanatory variables, teacher data 3 for auxiliary variables, and teacher data 13 for latent variables in addition to prediction / estimation data 5 for auxiliary variables and prediction / estimation data 4 for explanatory variables.

[0041] According to the third embodiment, the second latent variable calculation unit 21 of the prediction / estimation device 20 calculates the prediction / estimation data 23 of the latent variables using the teacher data 2, 3 of the explanatory variables and auxiliary variables as well. Therefore, it is possible to calculate the prediction / estimation data 23 of the latent variables using more data than in the second embodiment. This makes it possible to apply a more advanced algorithm and further improve the accuracy of prediction / estimation using auxiliary variables while avoiding deterioration of prediction / estimation accuracy due to correlation.

[0042] [Fourth embodiment] 5 is an explanatory diagram showing a basic configuration of a prediction / estimation system according to a fourth embodiment of the present invention. In a prediction / estimation system 100C according to the fourth embodiment, a first latent variable calculation unit 11 of a learning device 10 calculates teacher data (not shown) of latent variables using regression analysis and principal component analysis as a predetermined algorithm (latent variable calculation information 6). In addition, a second latent variable calculation unit 21 of a prediction / estimation device 20 calculates prediction / estimation data (not shown) of latent variables using the algorithm (latent variable calculation information 6) used in the first latent variable calculation unit 11.

[0043] More specifically, as shown in FIG. 5, in the prediction / estimation system 100C according to the fourth embodiment, the first latent variable calculation unit 11 of the learning device 10 includes a first regression analysis unit 14 that performs regression analysis of teacher data 3 of auxiliary variables with teacher data 2 of explanatory variables to create a regression model, and a first principal component analysis unit 16 that performs principal component analysis of regression errors 15 of the regression analysis by the first regression analysis unit 14 to calculate teacher data (not shown) of the latent variables. The latent variable calculation information 6 includes parameters and regression equations of the regression model, and principal component vectors and eigenvalues ​​of the principal component analysis. In addition, the second latent variable calculation unit 21 of the prediction / estimation device 20 includes a second regression analysis unit 24 that performs regression analysis of prediction / estimation data 5 of the auxiliary variables with prediction / estimation data 4 of the explanatory variables based on the latent variable calculation information 6 to create a regression model, and a second principal component analysis unit 26 that performs principal component analysis of regression errors 25 by the second regression analysis unit 24 based on the latent variable calculation information 6 to calculate prediction / estimation data (not shown) of the latent variables.

[0044] That is, the prediction and estimation system 100C of the fourth embodiment differs from the prediction and estimation system 100A according to the second embodiment in that the first latent variable calculation unit 11 of the learning device 10 and the second latent variable calculation unit 21 of the prediction and estimation device 20 include a first regression analysis unit 14 and a first principal component analysis unit 16, and perform regression analysis and principal component analysis to calculate teacher data for latent variables, and in that the second latent variable calculation unit 21 includes a second regression analysis unit 24 and a second principal component analysis unit 26, and performs regression analysis and principal component analysis to calculate prediction and estimation data for latent variables.

[0045] In the prediction / estimation system 100C of the fourth embodiment, the first regression analysis unit 14 of the first latent variable calculation unit 11 of the learning device 10 performs regression analysis (RA) on teacher data 3 of the auxiliary variables with teacher data 2 of the explanatory variables to create a regression model. The regression error 15 of the regression model created by the first regression analysis unit 14 is subjected to principal component analysis (PCA) in the first principal component analysis unit 16, thereby calculating teacher data of the latent variables. At this time, for example, a principal component vector with a small correlation between the explanatory variables and the auxiliary variables is selected as the principal component parameter. As a result, the first latent variable calculation unit 11 of the learning device 10 can calculate teacher data of the latent variables with a small correlation with the teacher data 2 of the explanatory variables.

[0046] In the prediction / estimation model learning unit 12, a prediction / estimation model 7 is created by, for example, machine learning using data including teacher data of latent variables with small correlation calculated in this way. This makes it possible to avoid deterioration of prediction / estimation accuracy caused by correlation between teacher data 3 of auxiliary variables and teacher data 2 of explanatory variables. In this case, the regression model included in the latent variable calculation information 6 may include data such as regression coefficients. Furthermore, the principal component parameters included in the latent variable calculation information 6 may include data such as eigenvalues ​​of principal component vectors and sample variance matrices.

[0047] Then, in the second latent variable calculation section 21 of the prediction / estimation device 20, latent variable calculation information 6 including the parameters and regression equation of the regression model in the first latent variable calculation section 11, and the principal component parameters of the principal component analysis is used. Using this latent variable calculation information 6, in the second regression analysis section 24, a regression analysis (RA) is performed on the prediction / estimation data 5 of the auxiliary variable with the prediction / estimation data 4 of the explanatory variable, and a regression model is created. The regression error 25 of the regression model created in the second regression analysis section 24 is subjected to a principal component analysis (PCA) in the second principal component analysis section 26, whereby prediction / estimation data of the latent variable is calculated. Therefore, in the prediction / estimation device 20, it is possible to calculate prediction / estimation data of a latent variable that has a small correlation with the prediction / estimation data 4 of the explanatory variable, using the same latent variable calculation information 6 as in the learning device 10.

[0048] The prediction / estimation unit 22 calculates the predicted / estimated value 8 of the dependent variable using data including the prediction / estimation data of the latent variables with low correlation calculated in this manner. This makes it possible to reduce the deterioration of prediction / estimation accuracy caused by the correlation between the prediction / estimation data 5 of the auxiliary variables and the prediction / estimation data 4 of the explanatory variables, and to obtain the predicted / estimated value 8 of the dependent variable with higher accuracy.

[0049] [Fifth embodiment] FIG. 6 is an explanatory diagram showing a basic configuration of a prediction / estimation system according to a fifth embodiment of the present invention. As shown in FIG. 6, in a prediction / estimation system 100D according to the fifth embodiment, a first latent variable calculation unit 11 of a learning device 10 calculates teacher data 13 of latent variables by a Gaussian process latent variable model (GPLVM), and latent variable calculation information 6 includes hyperparameters of the Gaussian process latent variable model (GPLVM). Furthermore, a prediction / estimation model learning unit 12 of the learning device 10 performs, for example, machine learning by Gaussian process regression (GPR) to create a prediction / estimation model 7 of a response variable. Furthermore, a second latent variable calculation unit 21 of a prediction / estimation device 20 calculates prediction / estimation data 23 of a latent variable by a Gaussian process latent variable model (GPLVM), and a prediction / estimation unit 22 of the prediction / estimation device 20 calculates a prediction / estimation value 8 of a response variable by Gaussian process regression (GPR).

[0050] That is, the prediction and estimation system 100D of the fifth embodiment differs from the prediction and estimation system 100B of the third embodiment in that the first latent variable calculation unit 11 of the learning device 10 and the second latent variable calculation unit 21 of the prediction and estimation device 20 use a Gaussian process latent variable model (GPLVM) algorithm to calculate the teacher data 13 of the latent variables and the prediction and estimation data 23 of the latent variables, and in that the prediction and estimation model learning unit 12 of the learning device 10 and the prediction and estimation unit 22 of the prediction and estimation device 20 use a Gaussian process regression (GPR) algorithm to learn the prediction and estimation model 7 of the dependent variable and to calculate the predicted and estimated value 8 of the dependent variable.

[0051] According to the fifth embodiment, the first latent variable calculation unit 11 and the second latent variable calculation unit 21 calculate the teacher data 13 of the latent variables and the prediction / estimation data 23 of the latent variables using a Gaussian process latent variable model (GPLVM), and the prediction / estimation model learning unit 12 and the prediction / estimation unit 22 learn and calculate the prediction / estimation model 7 and the prediction / estimation value 8 using Gaussian process regression (GPR). Therefore, it is possible to achieve the same effects as the prediction / estimation system 100B of the third embodiment, and to realize a system specialized in prediction / estimation using a nonlinear model instead of a linear model such as principal component analysis (PCA) or independent component analysis (ICA).

[0052] [Sixth embodiment] Fig. 7 is an explanatory diagram showing a basic configuration of a laser welding strength estimation system to which a prediction / estimation system according to a sixth embodiment of the present invention is applied. As shown in Fig. 7, a welding strength estimation system 100E according to the sixth embodiment can be configured as an estimation system to which the configurations of the prediction / estimation systems 100 to 100D according to the first to fifth embodiments can be applied.

[0053] In the estimation system 100E of the sixth embodiment, the objective variable is the welding strength (welding strength data 1A) or welding quality (not shown) of laser welding by the laser welder 9, the explanatory variable is the welding conditions of laser welding (welding condition data 2A), and the auxiliary variable is a physical quantity (welding temperature data 3A) that can be observed during or after laser welding. In addition, for example, welding condition data 4A set in the laser welder 9 is cited as predictive / estimation data for the explanatory variable, and welding temperature data 5A, which is one of the physical quantities observed during or after laser welding by the laser welder 9, is cited as predictive / estimation data for the auxiliary variable.

[0054] The welding strength data 1A is, for example, a welding strength F (N / mm2) determined by the allowable stress and material strength of the welded portion of the workpiece in accordance with the type of workpiece to be laser-welded by the laser welding machine 9 based on the following welding condition data 2A. 2The welding quality represents a predetermined welding quality (excellent, good, fair, poor, etc.) based on the welding strength F determined according to the type of workpiece to be laser-welded, for example, as described above.

[0055] The welding condition data 2A includes various information such as the specifications of the workpiece (material, thickness, etc.), the laser output value of the laser welder 9, the laser irradiation time, the focal position of the laser light, pulse conditions such as the frequency and duty ratio of the laser light, the welding speed, the time from the start of irradiation to the peak power, the elapsed time from the peak power to the end of irradiation, the lens focal length, the lens focal position, the laser diameter at the irradiation point, the fiber diameter, and the spot diameter (combination of the lens, fiber, and focal position), and are arbitrary parameters set for laser welding. The welding temperature data 3A indicates, for example, the welding temperature detected during laser welding based on the welding condition data 2A.

[0056] The welding condition data 4A is data related to the welding conditions actually set in the laser welder 9, and is configured similarly to the welding condition data 2A. The welding temperature data 5A indicates the welding temperature detected and acquired during actual laser welding by the welding temperature sensor 17 provided in the laser welder 9. The welding temperature sensor 17 is included in the above-mentioned sensor 212. The welding strength data 1A, the welding condition data 2A, and the welding temperature data 3A are stored in a readable state in, for example, an external storage device or storage medium (not shown) connected to a network 90. ​​The network 90 may be any medium for transferring data, and may be not only a network such as the Internet or a LAN, but also a bus constituting a device, a USB, a portable storage medium such as a HDD, or the like, and there is no particular restriction on the form of the network.

[0057] The estimation system 100E includes a learning server 30 functioning as the above-mentioned learning device 10, and an estimation edge server 40 functioning as the above-mentioned prediction / estimation device 20, which are communicatively connected to each other via a network 90. ​​The estimation system 100E also includes a storage device 50 that stores latent variable calculation information 6 and (at least one of) a welding strength estimation model 7A. In this embodiment, the learning server 30 also includes a storage unit 50A having the same function as the storage unit 50. Although not shown in the figure, a storage unit equivalent to the storage unit 50A may be included in the estimation edge server 40.

[0058] The learning server 30 is configured to be able to input teacher data of the objective variable, teacher data of explanatory variables acting on the objective variable, and teacher data of auxiliary variables that are variables other than the objective variable, have a correlation with the explanatory variable, and are observed when the explanatory variable is given. The learning server 30 also includes a latent variable calculation unit (first latent variable calculation unit 31) that calculates teacher data (not shown) of latent variables that have a smaller correlation with the teacher data of the explanatory variables or between them than the teacher data of the auxiliary variables from at least the teacher data of the auxiliary variables, and outputs latent variable calculation information 6, and a prediction / estimation model learning unit (weld strength estimation model learning unit 32) that inputs the teacher data of the objective variable, the teacher data of the explanatory variables, and the teacher data of the latent variables calculated by the latent variable calculation unit (first latent variable calculation unit 31), and performs, for example, machine learning to create a prediction / estimation model of the objective variable (weld strength estimation model 7A).

[0059] On the other hand, the estimation edge server 40 is a prediction / estimation device 20 that calculates a predicted / estimated value of the dependent variable (estimated value 8A of welding strength) based on a prediction / estimation model of the dependent variable (welding strength estimation model 7A) previously learned using teacher data of the dependent variable, teacher data of explanatory variables acting on the dependent variable, and teacher data of latent variables calculated based on latent variable calculation information 6 from teacher data of auxiliary variables that are variables other than the dependent variable and are correlated with the explanatory variable and are observed when the explanatory variable is given. The estimation edge server 40 also includes a latent variable calculation unit (second latent variable calculation unit 41) which calculates prediction / estimation data for a latent variable (not shown) having a smaller correlation between the prediction / estimation data for the explanatory variables or between the prediction / estimation data for the explanatory variables than the prediction / estimation data for the auxiliary variables, based on the latent variable calculation information 6, from the prediction / estimation data for at least the auxiliary variables, and a prediction / estimation unit (welding strength estimation unit 42) which calculates a prediction / estimation value of a response variable (welding strength estimation value 8A) from the prediction / estimation data for the latent variables calculated by the latent variable calculation unit (second latent variable calculation unit 41) and the prediction / estimation data for the explanatory variables, based on a prediction / estimation model (welding strength estimation model 7A).

[0060] The learning server 30 and the estimation edge server 40 can function as the above-mentioned learning device 10 and the prediction / estimation device 20, respectively, and can therefore realize all aspects already described in the first to fifth embodiments. Therefore, in the following, descriptions that overlap with the components of the learning device 10 and the prediction / estimation device 20 will be omitted.

[0061] [Operation of the estimation system] The welding strength data 1A, welding condition data 2A, and welding temperature data 3A stored in an external storage device or storage medium are input to the learning server 30 as teacher data for the objective variable, explanatory variable, and auxiliary variable, respectively, and stored in the storage unit 50A. The learning server 30 calculates teacher data for the latent variables in the first latent variable calculation unit 31 based on at least the welding temperature data 3A among these input data. The learning server 30 also inputs the welding strength data 1A, the welding condition data 2A, and the teacher data for the latent variables calculated by the first latent variable calculation unit 31 in the welding strength estimation model learning unit 32, and performs machine learning, for example, to create a welding strength estimation model 7A. The learning server 30 stores the latent variable calculation information 6 and the welding strength estimation model 7A determined when calculating the teacher data for the latent variables in the storage unit 50A, and outputs them to the storage device 50 via the network 90. ​​The storage device 50 stores the input latent variable calculation information 6 and the welding strength estimation model 7A. The welding strength data 1A, the welding condition data 2A, and the welding temperature data 3A may be stored in, for example, the storage device 50, instead of in an external storage device or storage medium.

[0062] Meanwhile, in the laser welding machine 9, laser welding is performed based on the set welding condition data 4A, and welding temperature data 5A during laser welding of an estimation target (for example, a welded portion of a workpiece) is acquired by a welding temperature sensor 17 attached to the laser welding machine 9. The welding condition data 4A of the laser welding machine 9 and the welding temperature data 5A from the welding temperature sensor 17 are input to the estimation edge server 40 as prediction / estimation data for explanatory variables and prediction / estimation data for auxiliary variables, respectively. In addition, the latent variable calculation information 6 and the welding strength estimation model 7A stored in the storage device 50 are also input to the estimation edge server 40.

[0063] The estimation edge server 40 stores these input data in the memory unit, and calculates prediction / estimation data of latent variables from among these data, including at least the welding temperature data 5A, in a second latent variable calculation unit 41 based on latent variable calculation information 6. The estimation edge server 40 also calculates an estimated value 8A of welding strength based on a welding strength estimation model 7A from the prediction / estimation data of latent variables calculated by the second latent variable calculation unit 41 in a welding strength estimation unit 42 and the welding condition data 4A. The calculated prediction / estimation data of latent variables and the estimated value 8A of welding strength are stored in the memory unit, and the estimated value 8A of welding strength can be output from the estimation edge server 40 in an appropriately usable form such as display or printing.

[0064] In the estimation system 100E of the sixth embodiment, teacher data of the latent variables is calculated based on the welding temperature data 3A, which is teacher data of the auxiliary variables of the welding strength data 1A, and then a welding strength estimation model 7A is created using the teacher data of the latent variables. Furthermore, based on latent variable calculation information 6 determined when calculating the teacher data of the latent variables, prediction / estimation data of the latent variables is calculated from the welding temperature data 5A, which is prediction / estimation data of the auxiliary variables, and then an estimated value 8A of the welding strength is calculated based on the welding strength estimation model 7A from the welding condition data 4A, which is prediction / estimation data of the explanatory variables, and the prediction / estimation data of the latent variables.

[0065] Therefore, according to the estimation system 100E of the sixth embodiment, it is possible to estimate the weld strength of the estimation target (weld portion of the workpiece) in a non-destructive manner from data on the welding conditions of laser welding and data on the temperature during welding. In general, in laser welding, even when laser welding is performed under the same welding 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 without destroying the product. However, the estimation system 100E of the sixth embodiment employs a configuration that can improve prediction and estimation accuracy by using more data and avoid deterioration of prediction and estimation accuracy due to correlation, so that this problem can be solved and the same effects as those of the first to fifth embodiments can be achieved.

[0066] The estimation system 100E of the sixth embodiment is intended to estimate the welding strength of laser welding, but is not limited thereto. For example, it is also possible to predict and estimate various physical phenomena related to welding quality, such as spatter and porosity. In this case, the objective variable may be the welding quality of laser welding. In addition, although the welding temperature data 3A and 5A have been described as an example of the auxiliary variables, the auxiliary variables are not limited thereto. They may be physical quantities observed during welding, such as sound, light, and color during welding, or physical quantities observed after welding, such as the welding width, color, and number of spatters of the welded portion after welding. In the estimation system 100E, the learning server 30, the estimation edge server 40, and the storage device 50 can be installed, for example, in a centralized manner at the same site as the laser welding machine 9 or separately at a remote location, so that a wide variety of operations can be performed in terms of estimating the welding strength of laser welding, even assuming global deployment.

[0067] [Seventh embodiment] Fig. 8 is an explanatory diagram showing the basic configuration of a learning device of a nondestructive measurement system to which a prediction and estimation system according to a seventh embodiment of the present invention is applied. Fig. 9 is an explanatory diagram showing the basic configuration of a nondestructive testing device of a nondestructive measurement system. As shown in Figs. 8 and 9, a nondestructive measurement system 100F of the seventh embodiment can be configured as a nondestructive measurement system to which the configurations of the prediction and estimation systems 100 to 100D according to the first to fifth embodiments can be applied.

[0068] The nondestructive measurement system 100F includes a learning device 33 as shown in Fig. 8, and a nondestructive testing device 49 including a prediction and estimation device 43 as shown in Fig. 9. The learning device 33 includes a first latent variable calculation unit 34 and a prediction and estimation model learning unit 35. The prediction and estimation device 43 of the nondestructive testing device 49 includes a second latent variable calculation unit 44 and a prediction and estimation unit 45.

[0069] In the nondestructive measurement system 100F, the objective variable is the concrete strength (strength property data 1B), the explanatory variables are the influencing factors and conditions of the measurable concrete strength (influencing factor and condition data 2B), and the auxiliary variables are the elastic wave propagation characteristics at the time of measuring the strength (elastic wave propagation characteristic data 3B). For example, predictive / estimating data for the explanatory variables includes influencing factor and condition data 4B for the concrete whose strength is predicted / estimated, and predictive / estimating data for the auxiliary variables includes elastic wave propagation characteristic data 5B based on data acquired by the measuring device (acoustic sensor 18). The strength property data 1B, the influencing factor and condition data 2B for the strength, and the elastic wave propagation characteristic data 3B are input to the learning device 33, and the influencing factor and condition data 4B for the strength and the elastic wave propagation characteristic data 5B are input to the predictive / estimating device 43 of the nondestructive testing device 49.

[0070] The strength property data 1B may include, for example, compressive strength, tensile strength, bending strength, shear strength, bearing strength, bond strength, fatigue strength, etc., as values ​​representing the strength of concrete. In this embodiment, the concrete compressive strength F (N / mm 2The strength influence factor / condition data 2B includes, for example, various information such as the type of cement, which is the material of the concrete, the cement-water ratio, cement additives, age of the concrete, and temperature and humidity from the time of concrete pouring to the time of strength measurement, but is not limited to these.

[0071] The elastic wave propagation characteristic data 3B represents elastic wave propagation velocity, elastic wave reflection characteristics, etc., but is not limited to these, and various types of characteristic data obtained by nondestructive measurement in the nondestructive measurement system 100F can be applied. The strength influencing factor / condition data 4B is data related to the influencing factors / conditions of the concrete strength whose strength is predicted / estimated, and is configured similarly to the strength influencing factor / condition data 2B input to the learning device 33. The elastic wave propagation characteristic data 5B represents elastic wave propagation characteristics derived from acoustic data detected by the acoustic sensor 18, and is configured similarly to the elastic wave propagation characteristic data 3B input to the learning device 33. The acoustic sensor 18 is included in the sensor 212 described above.

[0072] [Non-destructive measurement system operation] The strength property data 1B, the strength influence factor / condition data 2B, and the elastic wave propagation characteristic data 3B are input to the learning device 33 as teacher data for the objective variable, explanatory variable, and auxiliary variable, respectively. The learning device 33 calculates teacher data for latent variables in a first latent variable calculation unit 34 based on data including at least the strength influence factor / condition data 2B and the elastic wave propagation characteristic data 3B among these input data. The learning device 33 also inputs the strength property data 1B, the strength influence factor / condition data 2B, and the teacher data for the latent variables calculated by the first latent variable calculation unit 34 in a prediction / estimation model learning unit 35, and performs, for example, machine learning to create a prediction / estimation model 7 for concrete strength. The learning device 33 outputs to the outside the latent variable calculation information 6 and the concrete strength prediction / estimation model 7 determined when calculating the teacher data for the latent variables.

[0073] On the other hand, the nondestructive testing device 49 is provided with the latent variable calculation information 6 and the prediction / estimation model 7 output from the learning device 33. The latent variable calculation information 6 and the prediction / estimation model 7 are provided in the nondestructive testing device 49 by being stored in a storage medium possessed by the nondestructive testing device 49 or a portable storage medium detachable from the nondestructive testing device 49, or by being acquired from an information transmission medium such as the Internet.

[0074] Moreover, the non-destructive testing equipment 49 is provided with elastic wave propagation characteristic data 5B derived based on the acoustic data, as inputted with acoustic data detected by the acoustic sensor 18 during concrete striking. The prediction / estimation device 43 of the non-destructive testing equipment 49 is then inputted with the strength influence factor / condition data 4B, together with the latent variable calculation information 6, the prediction / estimation model 7, and the elastic wave propagation characteristic data 5B.

[0075] The prediction / estimation device 43 calculates prediction / estimation data for latent variables in a second latent variable calculation unit 44 from data including at least the strength influence factor / condition data 4B and the elastic wave propagation characteristic data 5B among these input data, based on latent variable calculation information 6. Furthermore, the prediction / estimation device 43 calculates a concrete strength prediction / estimation value 8 based on a concrete strength prediction / estimation model 7 from the prediction / estimation data for latent variables calculated by the second latent variable calculation unit 44, the strength influence factor / condition data 4B, and the elastic wave propagation characteristic data 5B in a prediction / estimation unit 45. The calculated prediction / estimation value 8 can be output from the non-destructive testing device 49 in an appropriately usable form such as display or printing.

[0076] In the nondestructive measurement system 100F of the seventh embodiment, the teacher data of the latent variables is calculated based on the elastic wave propagation characteristic data 3B, which is the teacher data of the auxiliary variables of the strength property data 1B, and the strength influence factor and condition data 2B, which is the teacher data of the explanatory variables, and then a concrete strength prediction and estimation model 7 is created using the teacher data of the latent variables. In addition, based on the latent variable calculation information 6 determined when calculating the teacher data of the latent variables, the prediction and estimation data of the latent variables is calculated from the elastic wave propagation characteristic data 5B, which is the prediction and estimation data of the auxiliary variables, and the strength influence factor and condition data 4B, which is the prediction and estimation data of the explanatory variables, and then a prediction and estimation value 8 of the concrete strength is calculated based on the concrete strength prediction and estimation model 7 from the strength influence factor and condition data 4B, the elastic wave propagation characteristic data 5B, and the prediction and estimation data of the latent variables.

[0077] Therefore, according to the nondestructive measurement system 100F of the seventh embodiment, it is possible to predict and estimate the strength of the object to be estimated (concrete) nondestructively from data on the influencing factors and conditions of the concrete strength, which represent factors at the time of pouring, and data on the elastic wave propagation characteristics in a nondestructive test using an elastic wave method. Therefore, it is possible to improve the prediction and estimation accuracy by using more data and to avoid deterioration of the prediction and estimation accuracy due to correlation, and it is possible to achieve the same effects as those of the first to fifth embodiments. Note that, when data other than the elastic wave propagation characteristic data 3B and 5B is used as an auxiliary variable, various sensors corresponding to the auxiliary variable may be used instead of the acoustic sensor 18 to obtain the auxiliary variable.

[0078] [Eighth embodiment] Fig. 10 is an explanatory diagram showing the basic configuration of a quality judgment prediction system for the cutting quality of a laser cutting device to which a prediction / estimation system according to an eighth embodiment of the present invention is applied. As shown in Fig. 10, a quality judgment prediction system 100G according to the eighth embodiment can be configured as a quality judgment prediction system to which the configurations of the prediction / estimation systems 100 to 100D according to the first to fifth embodiments can be applied.

[0079] The quality judgment prediction system 100G includes a learning device 36 and a laser cutting device 60. The learning device 36 includes a first latent variable calculation unit 37 and a quality judgment prediction model learning unit 38. The laser cutting device 60 includes a quality judgment prediction device 46 and a data storage and transfer device 61. The quality judgment prediction device 46 includes a second latent variable calculation unit 47 and a quality judgment prediction unit 48.

[0080] In the quality judgment prediction system 100G, the objective variable is the quality judgment result of the laser cutting by the laser cutting device 60 (quality judgment result data 1C), the explanatory variable is the cutting condition of the laser cutting (cutting condition data 2C), and the auxiliary variable is a measurable physical quantity related to the laser cutting (emission intensity data 3C). For example, the predictive / estimating data of the explanatory variable is the cutting condition data 4C set in the laser cutting device 60, and the predictive / estimating data of the auxiliary variable is the physical quantity observed in the laser cutting device 60 (emission intensity data 5C).

[0081] Here, the quality judgment 59 of the cutting quality of the laser cutting device 60 refers to, for example, automatically or manually judging whether the physical state of the cut surface of the workpiece cut by the laser cutting device 60, i.e., for example, the surface roughness, the adhesion state of dross, the occurrence state of gouging (cutting defects), etc., is suitable for practical use of the workpiece. Note that dross means a deposit of molten metal or oxides that adhere to the lower surface of the cut material, and is synonymous with slag. Also, gouging in laser cutting means, for example, a state in which the laser light does not penetrate during the laser cutting and molten metal is sprayed onto the surface of the material, resulting in a dirty appearance.

[0082] The quality judgment result data 1C may be given as a result of the above-mentioned quality judgment 59 as discrete category information such as "optimum", "excellent", "good", "fair" or "not good", or may be given as numerical information ranging from 0 to 100, and indicates the quality of the judged laser cutting. The cutting condition data 2C may be configured to include various information such as the material and / or product name of the workpiece to be laser cut, the laser output value, the laser irradiation time, and the cutting speed, and indicates the cutting conditions to be set in the laser cutting device 60 during the laser cutting. The cutting condition data 2C may also include various information on environmental conditions such as temperature and humidity during the laser cutting, but the various information in the cutting condition data 2C is not limited to these.

[0083] The light emission intensity data 3C represents, for example, the light emission intensity during piercing performed prior to laser cutting. However, the teaching data of the auxiliary variable is not limited to the light emission intensity, and may be any physical quantity data that can be measured in association with laser cutting, such as the sound or emission spectrum during piercing. The cutting condition data 4C is data related to the cutting conditions set in association with the operation input of the cutting conditions performed by the user via the touch panel 211, for example, and is configured in the same manner as the cutting condition data 2C. The light emission intensity data 5C represents the light emission intensity detected by the optical sensor 19 provided in the laser cutting device 60 during actual piercing. The optical sensor 19 is included in the above-mentioned sensor 212.

[0084] [Operation of quality judgment prediction system] Among the quality judgment result data 1C, the cutting condition data 2C, and the emission intensity data 3C, at least the cutting condition data 2C and the emission intensity data 3C are input to the learning device 36 as teacher data of the explanatory variable and the auxiliary variable, respectively. The quality judgment result data 1C can be input to the learning device 36 as teacher data of the objective variable. The learning device 36 calculates teacher data of the latent variable in the first latent variable calculation unit 37 based on data including at least the cutting condition data 2C and the emission intensity data 3C among these input data. In addition, the learning device 36 inputs the quality judgment result data 1C, the cutting condition data 2C, and the teacher data of the latent variable calculated by the first latent variable calculation unit 37 in the quality judgment prediction model learning unit 38, and performs, for example, machine learning to create a quality judgment prediction model 7B of the cutting quality of the laser cutting. The learning device 36 outputs the latent variable calculation information 6 and the quality judgment prediction model 7B determined when calculating the teacher data of the latent variable to the laser cutting device 60.

[0085] The laser cutting device 60 is provided with the latent variable calculation information 6 and the quality judgment prediction model 7B output from the learning device 36. The latent variable calculation information 6 and the quality judgment prediction model 7B are provided in the laser cutting device 60 by being stored in a storage medium possessed by the laser cutting device 60 or a portable storage medium that is detachable from the laser cutting device 60, or by being acquired from an information transmission medium such as the Internet.

[0086] Furthermore, emission intensity data 5C during piercing detected by the optical sensor 19 and cutting condition data 4C input by operation via the touch panel 211 are input to and temporarily stored in the data storage and transfer device 61 of the laser cutting device 60. Then, the quality judgment prediction device 46 of the laser cutting device 60 receives the cutting condition data 4C and emission intensity data 5C stored in the data storage and transfer device 61, as well as the latent variable calculation information 6 and the quality judgment prediction model 7B.

[0087] The quality judgment prediction device 46 calculates prediction and estimation data of latent variables based on latent variable calculation information 6 in the second latent variable calculation unit 47 from at least the cutting condition data 4C and the emission intensity data 5C among these input data. In addition, the quality judgment prediction device 46 performs quality judgment prediction of the cutting quality based on the quality judgment prediction model 7B from the prediction and estimation data of the latent variables calculated by the second latent variable calculation unit 47, the cutting condition data 4C, and the emission intensity data 5C in the quality judgment prediction unit 48, and calculates a quality judgment prediction value 8B. The calculated quality judgment prediction value 8B can be output from the laser cutting device 60 in a suitable usable form such as display or printing. Note that the laser cutting device 60 can output various information related to the actual cutting result 58, so the above-mentioned quality judgment 59 is performed based on various information related to this cutting result 58, and the quality judgment result data 1C is created.

[0088] Furthermore, when calculating the teacher data of the latent variables and creating the quality judgment prediction model 7B, the learning device 36 may use the cutting condition data 4C and the emission intensity data 5C stored in the data storage and transfer device 61 as teacher data of the explanatory variables and auxiliary variables in addition to the cutting condition data 2C and the emission intensity data 3C. In this way, more data can be used, which contributes to improving the accuracy of prediction and estimation. Furthermore, the learning device 36 can additionally use the quality judgment result data 1C as teacher data of the objective variable, so in this case, it is possible to further improve the accuracy of prediction and estimation.

[0089] In the quality judgment prediction system 100G of the eighth embodiment, the teacher data of the latent variables is calculated based on the light emission intensity data 3C (5C) which is the teacher data of the auxiliary variables of the quality judgment result data 1C and the cutting condition data 2C (4C) which is the teacher data of the explanatory variables, and then a quality judgment prediction model 7B of the cutting quality is created using the teacher data of the latent variables. Also, based on the latent variable calculation information 6 determined when calculating the teacher data of the latent variables, the prediction and estimation data of the latent variables is calculated from the light emission intensity data 5C which is the prediction and estimation data of the auxiliary variables and the cutting condition data 4C which is the prediction and estimation data of the explanatory variables, and then a quality judgment prediction value 8B of the cutting quality is calculated based on the quality judgment prediction model 7B from the cutting condition data 4C, the light emission intensity data 5C, and the prediction and estimation data of the latent variables.

[0090] Therefore, according to the quality judgment prediction system 100G of the eighth embodiment, it is possible to predict the cutting quality of the quality prediction target (the part to be cut of the workpiece) before laser cutting based on data on the cutting conditions of the laser cutting and data on the emission intensity during cutting. In general, in laser cutting, even if the laser cutting is performed under the same cutting conditions, the cutting quality varies depending on factors such as the material of the workpiece and the variation in the cutting environment, so there is a problem that the cutting quality cannot be determined unless the laser cutting is actually performed. However, the quality judgment prediction system 100G of the eighth embodiment adopts a configuration that can improve the quality judgment prediction accuracy by using more data and avoid deterioration of the quality judgment prediction accuracy due to correlation, so that it is possible to predict the cutting quality before laser cutting to solve this problem and to achieve the same effects as the first to fifth embodiments.

[0091] In addition, in the quality judgment prediction system 100G of the eighth embodiment, the learning device 36 may be incorporated into the laser cutting device 60, and the quality judgment prediction device 46 may be an independent device separate from the laser cutting device 60 (e.g., an edge server, etc.). In the machine learning and quality judgment prediction in the learning device 36 and the quality judgment prediction device 46, various algorithms can be used, such as the above-mentioned 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), as well 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).

[0092] [Other embodiments] In the first embodiment described above, the learning method of the prediction / estimation model 7 in the learning device 10 and the prediction / estimation device 20 and the prediction / estimation method of the predicted / estimated value 8 can be realized as a program for executing these methods and / or a computer-readable storage medium storing this program. The learning method and the prediction / estimation method may be realized using various hardware resources such as an electronic circuit for executing these methods and a logic circuit using other physical media, or may be realized as a device equipped with hardware resources such as these programs, electronic circuits, and / or logic circuits. The learning method and the prediction / estimation method may be realized using various hardware resources such as an electronic computer and / or electronic circuit equipped with a storage medium storing a program executed as one system, and other logic circuits, or may be realized using various hardware resources such as one or more electronic computers and / or electronic circuits equipped with separate storage media, and other logic circuits. In addition, when executed as one system, for example, the learning method may be realized on the cloud side and the prediction / estimation method may be realized on the edge side, or vice versa.

[0093] [Example] Fig. 11 is a graph showing the relationship between the welding temperature and the elapsed time to explain auxiliary variables in an example in which the prediction / estimation system according to the fourth embodiment shown in Fig. 5 is applied to actual laser welding. Fig. 12 is a graph showing the relationship between the estimated value and the measured value in a comparative example and an example, and the frequency distribution of the error of the estimated value against the estimated value.

[0094] In the laser welding of the embodiment, the 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 prediction / estimation model 7 was created using multiple regression analysis. In the comparative example, the above-mentioned latent variables were not used.

[0095] As shown in FIG. 11, 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.

[0096] The estimated values ​​of weld strength and the actual measured values ​​of weld strength in the comparative example are shown in Fig. 12(a). Meanwhile, the estimated values ​​of weld strength and the actual measured values ​​of weld strength in the example are shown in Fig. 12(b). In Figs. 12(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 those welding conditions. Note that in Figs. 12(a) and (b), both the measured values ​​and the estimated values ​​are plotted after being normalized so that the maximum weld strength is 1.

[0097] The frequency distribution of the error of the estimated value with respect to the measured value in the comparative example is shown in Fig. 12(c). On the other hand, the frequency distribution of the error of the estimated value with respect to the measured value in the embodiment is shown in Fig. 12(d). In Figs. 12(c) and (d), the horizontal axis shows the range of error, and the vertical axis shows the frequency corresponding to the error.

[0098] 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.

[0099] [Table 1]

[0100] The errors in the embodiment were smaller in both MAE and RMSE than those in the comparative example. Also, as is clear from Figures 12(c) and (d), the estimated values ​​in the embodiment are closer to the measured values ​​than in the comparative example, and the tail (range) of the error distribution is narrower. Therefore, it was proven that the prediction and estimation system of the embodiment improves the accuracy of prediction and estimation even when more data, including auxiliary variables, is used.

[0101] 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]

[0102] 1. Training data for the objective variable 2. Training data for explanatory variables 3. Training data for auxiliary variables 4. Prediction and estimation data for explanatory variables 5. Data for predicting and estimating auxiliary variables 6 Latent variable calculation information 7 Prediction and estimation models 8 Predictions and estimates 10 Learning Device 11 First latent variable calculation section 12 Prediction and estimation model learning section 13 Teaching data for latent variables 20 Prediction and Estimation Devices 21 Second latent variable calculation section 22 Prediction and Estimation Section 23 Data for predicting and estimating latent variables 100 Prediction and Estimation Systems

Claims

1. a learning device which receives as input teacher data of a dependent variable, teacher data of explanatory variables acting on the dependent variable, and teacher data of auxiliary variables which are variables other than the dependent variable and have a correlation with the explanatory variable and are observed when the explanatory variable is given, and creates a prediction / estimation model of the dependent variable; a prediction / estimation device that receives as input prediction / estimation data of the explanatory variables for predicting / estimating the dependent variable and prediction / estimation data of the auxiliary variables for predicting / estimating the dependent variable, and calculates a prediction / estimation value of the dependent variable based on the prediction / estimation model created by the learning device; Equipped with The learning device includes: a first latent variable calculation unit that calculates teacher data of a latent variable by subtracting a correlation between the teacher data of the auxiliary variable and the teacher data of the explanatory variable or between the teacher data of the auxiliary variables from the teacher data of at least the auxiliary variables, and outputs latent variable calculation information; a prediction / estimation model learning unit that receives as input the teacher data of the objective variable, the teacher data of the explanatory variable, and the teacher data of the latent variable calculated by the first latent variable calculation unit, and creates a prediction / estimation model of the objective variable; Including, The prediction / estimation device comprises: a second latent variable calculation unit that calculates prediction / estimation data of a latent variable by subtracting a correlation between the prediction / estimation data of the auxiliary variable and the prediction / estimation data of the explanatory variable or between the prediction / estimation data of the auxiliary variables from at least the prediction / estimation data of the auxiliary variable based on the latent variable calculation information; a prediction / estimation unit that calculates a prediction / estimation value of the objective variable based on the prediction / estimation model from the prediction / estimation data of the latent variable calculated by the second latent variable calculation unit and the prediction / estimation data of the explanatory variable; A prediction and estimation system including:

2. The first latent variable calculation unit of the learning device calculates teacher data of the latent variable based on teacher data of the explanatory variable and teacher data of the auxiliary variable, The second latent variable calculation unit of the prediction / estimation device calculates prediction / estimation data of the latent variable based on the prediction / estimation data of the explanatory variable and the prediction / estimation data of the auxiliary variable. The prediction / estimation system according to claim 1.

3. The second latent variable calculation unit of the prediction / estimation device further calculates prediction / estimation data for the latent variable based on teacher data of the explanatory variable, teacher data of the auxiliary variable, and teacher data of the latent variable. The prediction / estimation system according to claim 2.

4. The latent variable calculation information includes an algorithm, a formula, and / or a parameter; The algorithm includes a process of dimensionality reduction or dimensionality compression. The prediction / estimation system according to any one of claims 1 to 3.

5. The algorithm includes 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). The prediction / estimation system according to claim 4.

6. the first latent variable calculation unit of the learning device includes a first regression analysis unit that performs regression analysis of teacher data of the auxiliary variable with teacher data of the explanatory variable to create a regression model, and a first principal component analysis unit that performs principal component analysis of regression errors of the regression analysis by the first regression analysis unit to calculate teacher data of the latent variable, the latent variable calculation information includes parameters and a regression equation of the regression model, and principal component vectors and eigenvalues ​​of the principal component analysis, The second latent variable calculation unit of the prediction / estimation device includes a second regression analysis unit that performs regression analysis of prediction / estimation data of the auxiliary variable with prediction / estimation data of the explanatory variable based on the latent variable calculation information to create a regression model, and a second principal component analysis unit that performs principal component analysis of regression errors by the second regression analysis unit based on the latent variable calculation information to calculate prediction / estimation data of the latent variable. The prediction / estimation system according to claim 2.

7. the first latent variable calculation unit of the learning device calculates teacher data of the latent variables using a Gaussian process latent variable model (GPLVM); The latent variable calculation information includes hyperparameters of the Gaussian process latent variable model (GPLVM), the prediction / estimation model learning unit of the learning device creates a prediction / estimation model of the objective variable by Gaussian process regression (GPR); the second latent variable calculation unit of the prediction / estimation device calculates prediction / estimation data of the latent variables using the Gaussian process latent variable model (GPLVM); The prediction / estimation unit of the prediction / estimation device calculates a predicted / estimated value of the dependent variable by the Gaussian process regression (GPR). The prediction / estimation system according to claim 2.

8. The objective variable is a weld strength or a weld quality of laser welding by a laser welder, the explanatory variables are welding conditions of the laser welding, The auxiliary variable is a physical quantity that can be observed during or after the laser welding. The prediction / estimation system according to any one of claims 1 to 7.

9. The objective variable is the strength of concrete, The explanatory variables are measurable factors and conditions affecting the strength of the concrete, The auxiliary variable is an elastic wave propagation characteristic at the time of measuring the strength. The prediction / estimation system according to any one of claims 1 to 7.

10. the objective variable is a quality assessment result of laser cutting by a laser cutting device, The explanatory variables are cutting conditions of the laser cutting, The auxiliary variable is a measurable physical quantity related to the laser cutting. The prediction / estimation system according to any one of claims 1 to 7.

11. inputting teacher data of a response variable, teacher data of explanatory variables acting on the response variable, and teacher data of auxiliary variables which are variables other than the response variable, have a correlation with the explanatory variable, and are observed when the explanatory variable is given; a latent variable calculation unit that calculates teacher data of a latent variable by subtracting a correlation between the teacher data of the auxiliary variable and the teacher data of the explanatory variable or between the teacher data of the auxiliary variables from the teacher data of at least the auxiliary variables, and outputs latent variable calculation information; a prediction / estimation model learning unit that receives as input the teacher data of the objective variable, the teacher data of the explanatory variables, and the teacher data of the latent variables calculated by the latent variable calculation unit, and creates a prediction / estimation model of the objective variable; A learning device equipped with

12. A prediction / estimation device that calculates a predicted / estimated value of a dependent variable based on a prediction / estimation model of the dependent variable that is trained in advance using teacher data of a dependent variable, teacher data of explanatory variables acting on the dependent variable, and teacher data of a latent variable calculated based on latent variable calculation information from teacher data of an auxiliary variable that is a variable other than the dependent variable and is correlated with the explanatory variable and is observed when the explanatory variable is given, comprising: a latent variable calculation unit that calculates prediction / estimation data of the latent variable by subtracting a correlation between the prediction / estimation data of the auxiliary variable and the prediction / estimation data of the explanatory variable or between the prediction / estimation data of the auxiliary variables based on the latent variable calculation information from the prediction / estimation data of at least the auxiliary variable; a prediction / estimation unit that calculates a prediction / estimation value of the objective variable based on the prediction / estimation model from the prediction / estimation data of the latent variable calculated by the latent variable calculation unit and the prediction / estimation data of the explanatory variables; A prediction and estimation device equipped with the above.

Images