Classification device, classification method, and classification program

Abstract

To provide a classification apparatus configured to automatically classify operations of an operator to reduce the time required by the operator to acquire proficiency in the operations, a classification method, and a classification program.SOLUTION: A classification apparatus, a classification method, and a classification program are configured to: receive first measurement data related to an operation of a first operator, and calculate first feature quantity that characterizes the first measurement data; and estimate a first label corresponding to the first measurement data, based on an evaluation model which outputs the first label of the first measurement data, for an input including the first feature quantity. The evaluation model is a learning model generated by machine learning based on training data using a set of second feature quantity that characterizes second measurement data related to an operation of a second operator and a second label specifying a cluster to be obtained by performing cluster analysis on the second feature quantity.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to a classification device, a classification method, and a classification program.

Background Art

[0002] Patent Document 1 discloses a technique for generating a learning model by performing machine learning. Here, the learning model uses, as input data, learning data based on time-series data related to welding work of a plurality of workers with different skill levels or the state of construction tools operated by the workers, and uses the skill level of the workers as output data.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, according to the technique disclosed in Patent Document 1, it is impossible to identify the welding work styles existing among a plurality of workers having the same skill level. Here, what is called a style indicates the characteristics of how to use the body and how to move the tools, and workers of the same style perform similar work operations. Therefore, when assigning a welding work instructor to a learner who learns welding work through OJT (On the Job Training), there is a risk that an instructor having a style different from the welding work style of the learner will be assigned. As a result, there is a problem that the efficiency of the learner learning welding work decreases and the period required for education becomes longer.

[0005] The present disclosure has been made in view of the above-described situation. That is, an object of the present disclosure is to provide a classification device, a classification method, and a classification program that can automate the classification of the work operations of a worker and shorten the time until the worker becomes proficient in the work operations.

Means for Solving the Problems

[0006] The classification device according to the present disclosure includes a reception unit, a calculation unit, and an estimation unit. Here, the reception unit receives first measurement data related to the work operation of a first worker. The calculation unit calculates a first feature amount that characterizes the first measurement data. The estimation unit estimates the first label corresponding to the first measurement data based on an evaluation model that outputs the first label of the first measurement data for an input including the first feature amount. And the evaluation model is a learning model generated by machine learning based on teacher data. The teacher data is a set of a second feature amount that characterizes second measurement data related to the work operation of a second worker and a second label that designates a cluster obtained by performing cluster analysis on the second feature amount.

[0007] The work operation may include a welding operation. And the first measurement data may include at least any one of the tip coordinates of the welding torch, the posture information of the welding torch, the current value, the voltage value, the posture information of the workpiece related to the welding operation, the time ratio of the welding operation in the work operation, and the number of work interruptions in the work operation.

[0008] The calculation unit may use the power spectral density calculated based on at least any one of the tip coordinates of the welding torch, the posture information of the welding torch, the current value, the voltage value, and the posture information of the workpiece related to the welding operation as the first feature amount.

[0009] The calculation unit may use the volume of the ellipsoidal sphere representing the trajectory drawn by the tip coordinates as the first feature amount.

[0010] The above calculation unit may calculate the first feature amount by performing principal component analysis or factor analysis on the above first measurement data.

[0011] The classification device according to the present disclosure may further include a preprocessing unit. Here, the preprocessing unit may calculate preprocessed data by performing at least any one of smoothing processing by moving average, first-order differential processing, second-order differential processing, and SNV (Standard Normal Variate) conversion processing on the above first measurement data. The above calculation unit may calculate the first feature amount based on the above preprocessed data.

[0012] The above evaluation model may be represented by a neural network or a support vector machine.

[0013] The above evaluation model may be represented by an unlearned class estimation probability neural network.

[0014] The classification device according to the present disclosure may further include a display unit that presents the above first label to the user.

[0015] The classification device according to the present disclosure may further include a storage unit and a display unit. Here, the storage unit may store the identification information of the above second operator related to the second feature amount belonging to the cluster specified by the above second label in association with the above second label. The display unit may present the identification information of the above second operator associated with the same above second label as the above first label to the user.

[0016] The classification method according to the present disclosure receives first measurement data related to the work operation of a first worker, and calculates a first feature amount characterizing the first measurement data. Then, based on an evaluation model that outputs a first label of the first measurement data for an input including the first feature amount, a first label corresponding to the first measurement data is estimated. Here, the evaluation model is a learning model generated by machine learning based on teacher data. The teacher data is a pair including a second feature amount characterizing second measurement data related to the work operation of a second worker, and a second label designating a cluster obtained by performing cluster analysis on the second feature amount.

[0017] The classification program according to the present disclosure causes a computer to execute a step of receiving first measurement data related to the work operation of a first worker, and a step of calculating a first feature amount characterizing the first measurement data. Then, based on an evaluation model that outputs a first label of the first measurement data for an input including the first feature amount, the computer is caused to execute a step of estimating the first label corresponding to the first measurement data. Here, the evaluation model is a learning model generated by machine learning based on teacher data. The teacher data is a pair including a second feature amount characterizing second measurement data related to the work operation of a second worker, and a second label designating a cluster obtained by performing cluster analysis on the second feature amount.

Advantages of the Invention

[0018] According to the present disclosure, it is possible to automate the classification of the work operations of workers, and by guiding workers of the same school as the worker who performs learning, it is possible to shorten the time until the worker becomes proficient in the work operations.

Brief Description of the Drawings

[0019]

Figure 1

Figure 2

Figure 3

Best Mode for Carrying Out the Invention

[0020] Hereinafter, several exemplary embodiments will be described with reference to the drawings. In each figure, the same reference numerals are assigned to the common parts, and duplicate descriptions are omitted.

[0021] [Configuration of Classification Device] FIG. 1 is a block diagram showing the configuration of a classification device according to an embodiment of the present disclosure. As shown in FIG. 1, the classification device 20 includes a receiving unit 21, a database 23 (storage unit), a controller 25, an operation unit 27, and a display unit 29. The controller 25 is connected so as to be communicable with the receiving unit 21, the database 23, the operation unit 27, and the display unit 29.

[0022] In addition, the operation unit 27 and the display unit 29 may be provided in the classification device 20 itself, or may be installed outside the classification device 20 and connected to the classification device 20.

[0023] The receiving unit 21 is connected so as to be communicable with the sensor 10 wirelessly or by wire. The receiving unit 21 receives measurement data (first measurement data, second measurement data) related to the work operations of the operator (first operator, second operator) from the sensor 10. In addition, the receiving unit 21 may receive the identification information of the operator related to the measurement data and a time stamp indicating the date and time when the measurement data was acquired, together with the measurement data.

[0024] Here, the sensor 10 and the measurement data acquired by the sensor 10 will be described. The sensor 10 acquires measurement data related to the work operations of the operator. Specifically, for example, when the work operation includes a welding operation, the sensor 10 acquires, as measurement data, the tip coordinates of the welding torch, the posture information of the welding torch, the current value, the voltage value, and the posture information of the workpiece related to the welding operation, etc., related to the welding operation.

[0025] The sensor 10 may be a camera that captures artificial markers. Examples of artificial markers include AR (Augmented Reality) tags. The sensor 10 may, for example, capture an AR tag attached to a welding torch or a workpiece related to a welding operation, and calculate the tip coordinates of the welding torch, the posture information of the welding torch, and the posture information of the workpiece related to the welding operation from the position and inclination of the AR tag.

[0026] In addition, the measurement data may be time-series data, or may be other data such as the working time for performing the work operation, the time ratio of the welding operation in the working time, and the number of work interruptions in the welding operation.

[0027] In addition to the case where the work operation includes a welding operation, the work operation may include a painting operation, a polishing operation, etc. Depending on the content of the work included in the work operation, the sensor 10 may acquire, as measurement data, the jig used for the work and the position information and posture information of the workpiece that is the object of the work.

[0028] The database 23 records the measurement data acquired by the sensor 10. The database 23 may record the identification information of the worker related to the measurement data. Also, the database 23 may record a time stamp indicating the date and time when the measurement data was acquired. Furthermore, the database 23 may record various parameters that are the premise for calculating the label corresponding to the measurement data, or an evaluation model used in the processing of the controller 25.

[0029] Also, the database 23 may record teacher data for generating an evaluation model. Here, the teacher data is data that combines a feature amount calculated based on measurement data and a label corresponding to the feature amount. For example, it may be data that combines a feature amount calculated based on measurement data related to a skilled worker (second worker) who has a certain level of skill or more regarding the work operation, and a label calculated separately for the feature amount. The calculation of the feature amount and the generation of the teacher data will be described later.

[0030] The operation unit 27 is an input device through which a user of the classification device 20 can perform operations. For example, the operation unit 27 is a keyboard, a mouse, a trackball, a touch panel, or the like. The operation unit 27 is not limited to the examples listed here. The operation content of the user input through the operation unit 27 is transmitted to the controller 25.

[0031] The display unit 29 displays the information received from the controller 25. Further, the display unit 29 receives a label corresponding to the measurement data determined by the controller 25 and presents it to the user. In addition, the display unit 29 may receive the identification information of another operator having the same label as the label corresponding to the measurement data and present it to the user.

[0032] For example, the display unit 29 may be a display that displays graphics and characters by combining a plurality of display pixels, or may be a rotating lamp, a buzzer, or the like. The display unit 29 is not limited to the examples listed here.

[0033] The controller 25 (control unit) is a general-purpose microcomputer including a CPU (Central Processing Unit), a memory, and an input / output unit. A computer program (classification program) for functioning as the classification device 20 is installed in the controller 25. By executing the computer program, the controller 25 functions as a plurality of information processing circuits (251, 253, 255, 257) provided in the classification device 20.

[0034] In the present disclosure, an example of realizing a plurality of information processing circuits (251, 253, 255, 257) by software is shown. However, it is also possible to prepare dedicated hardware for executing each of the information processes shown below to configure the information processing circuits (251, 253, 255, 257). Further, the plurality of information processing circuits (251, 253, 255, 257) may be configured by individual hardware.

[0035] As shown in FIG. 1, the controller 25 includes a preprocessing unit 251, an evaluation model setting unit 253, a calculation unit 255, and an estimation unit 257 as a plurality of information processing circuits (251, 253, 255, 257).

[0036] The preprocessing unit 251 performs preprocessing on the measurement data acquired by the sensor 10. For example, when the work operation includes a welding operation, the measurement data acquired from the camera includes spike noise caused by the arc light during welding. Therefore, the preprocessing unit 251 may perform preprocessing including spike noise removal.

[0037] Specifically, the preprocessing unit 251 performs smoothing processing by moving average, first-order differential processing, second-order differential processing, SNV (Standard Normal Variate) conversion processing, etc. to clarify the characteristics of the peaks in the time-series data.

[0038] The smoothing process is used for the purpose of reducing noise, and the noise is reduced by replacing the numerical value at a certain measurement timing in the time-series data with the average of the numerical values at the previous and subsequent measurement timings. The greater the number of numerical values before and after used for smoothing, the higher the noise reduction effect.

[0039] The first-order differential processing is used to eliminate the influence of the baseline. The second-order differential processing is used to eliminate the influence of the change in the baseline that can be expressed by a first-order function of the wavelength, and further to clarify the difference between the peaks in the time-series data. When the second-order differential processing is performed on the time-series data, the relative relationship between the peak positions and the peak heights is maintained, so quantitative analysis can be performed based on the time-series data after the second-order differential processing.

[0040] The SNV (Standard Normal Variate) conversion process is used to suppress the baseline shift in the time-series data. The time-series data is treated as a set of numerical values, and spectral conversion is performed so that the average of the set is 0 and the variance is 1.

[0041] Alternatively, the preprocessing unit 251 may remove outliers using a Hampel filter to remove spike noise included in the time-series data, or may perform linear interpolation to fill in missing data points.

[0042] Further, the preprocessing unit 251 may perform preprocessing on the time-series data using multiple scattering correction processing, offset processing, trend removal, resampling, FFT (Fast Fourier Transform), Wavelet transform, or the like.

[0043] The calculation unit 255 calculates feature amounts (first feature amount, second feature amount) that characterize the measurement data based on the measurement data. The calculation unit 255 may calculate the feature amounts based on the measurement data (preprocessed data) after the preprocessing is performed by the preprocessing unit 251.

[0044] For example, the calculation unit 255 performs principal component analysis or factor analysis on the multi-dimensional measurement data to reduce the number of dimensions constituting the measurement data and calculates the feature amounts for a predetermined dimension.

[0045] When the working operation includes a welding operation, the calculation unit 255 calculates the power spectral density based on at least any one of the tip coordinates of the welding torch, the posture information of the welding torch, the current value, the voltage value, and the posture information of the workpiece related to the welding operation related to the welding operation, and may use the power spectral density as a feature amount.

[0046] Alternatively, the calculation unit 255 may calculate the volume of an ellipsoid that represents the trajectory drawn by the tip coordinates of the welding torch, and may use the volume as a feature amount. Here, the "ellipsoid that represents the trajectory" means an ellipsoid that includes the trajectory drawn by the tip coordinates of the welding torch during the welding operation. The volume serves as an index indicating the degree of movement of the welding torch during the welding operation. Therefore, the calculation unit 255 may use the volume as a feature amount.

[0047] Based on the feature quantity (second feature quantity) calculated based on the measurement data (second measurement data) and the label (second label) corresponding to the feature quantity (second feature quantity), the evaluation model setting unit 253 performs machine learning to generate an evaluation model using the training data which is a set of the feature quantity and the label.

[0048] Here, the label included in the training data is a label that designates a cluster obtained by performing cluster analysis in advance on a plurality of measurement data based on the feature quantity. More specifically, the evaluation model setting unit 253 calculates the similarity between the feature quantities calculated for each measurement data for a plurality of measurement data. Then, it executes hierarchical cluster analysis to classify workers who have become measurement data including feature quantities determined to be similar from the similarity as belonging to the same cluster. Note that the number of clusters obtained by cluster analysis may be set in advance by the user, or may be automatically set based on the variation of the feature quantity or the like.

[0049] As a method for generating an evaluation model by machine learning, for example, a method using one or a combination of two or more of neural network, support vector machine, Random Forest, XGBoost, LightGBM, PLS regression, Ridge regression, and Lasso regression can be mentioned. The method for generating an evaluation model by machine learning is not limited to the examples listed here.

[0050] A case where the evaluation model is represented by a neural network will be described. In this case, the evaluation model setting unit 253 calculates the error or likelihood of the output obtained when the feature quantity is input to the neural network and the label corresponding to the input feature quantity. Then, the evaluation model setting unit 253 adjusts the parameters defining the neural network so that the error is minimized or the likelihood is maximized. As a result, the neural network learns the features representing the training data.

[0051] The neural network includes an input layer to which data of feature quantities are input, an output layer from which output values are output, and at least one or more hidden layers provided between the input layer and the output layer, and signals are propagated in the order of the input layer, the hidden layer, and the output layer. Each layer of the input layer, the hidden layer, and the output layer is composed of one or more units. Units between layers are connected to each other, and each unit has an activation function (for example, sigmoid function, rectified linear function, softmax function, etc.). A weighted sum is calculated based on a plurality of inputs to the unit, and the value of the activation function with the sum value as a variable becomes the output of the unit.

[0052] The evaluation model setting unit 253 adjusts the weights when calculating the weighted sum in each unit among the parameters defining the neural network, thereby minimizing the error between the output of the neural network and the label, or maximizing the likelihood of the output. In order to minimize the error regarding the output of the neural network or maximize the likelihood for a plurality of teacher data, the maximum likelihood estimation method or the like can be applied.

[0053] In order to minimize the error regarding the output of the neural network, for example, the evaluation model setting unit 253 may use the gradient descent method, the stochastic gradient descent method, or the like. The evaluation model setting unit 253 may use the error backpropagation method for gradient calculation in the gradient descent method and the stochastic gradient descent method.

[0054] In machine learning by a neural network, generalization performance (discrimination ability for unknown data) and overfitting (a phenomenon in which the model fits well to the teacher data while the generalization performance does not improve) can be problems.

[0055] Therefore, in generating the evaluation model in the evaluation model setting unit 253, in order to mitigate overfitting, techniques such as regularization that restricts the degree of freedom of weights during learning may be used. Additionally, techniques such as dropout that probabilistically selects units in the neural network and invalidates the other units may also be used. Furthermore, in order to improve generalization performance, in the preprocessing unit 251, techniques such as data regularization, data normalization, and data augmentation that eliminate biases in the teacher data may also be used.

[0056] Alternatively, the evaluation model setting unit 253 may generate an evaluation model using an unlearned class estimation probability neural network that can consider unlearned clusters not assumed during learning based on teacher data. The unlearned class estimation probability neural network is described in detail in the literature "Takayuki Mukai, Keisuke Shima, 'Unlearned Class Estimation Probability Neural Network Based on Mixed Complementary Event Distribution', Transactions of the Institute of Measurement and Control, vol. 56, no. 12, pp. 532 - 540, 2020."

[0057] The evaluation model setting unit 253 may generate an evaluation model using a support vector machine instead of a neural network. In this case, the evaluation model is represented by the support vector machine. Machine learning by support vector machines has no problem of local solution convergence and tends to improve generalization performance.

[0058] The evaluation model generated by the evaluation model setting unit 253 is transmitted to the estimation unit 257 or the database 23 and used for estimation in the estimation unit 257.

[0059] Based on the measurement data (first measurement data), the estimation unit 257 estimates the label (first label) corresponding to the feature amount (first feature amount) calculated by the calculation unit 255 based on the evaluation model. More specifically, the estimation unit 257 acquires the evaluation model from the evaluation model setting unit 253. The estimation unit 257 calculates the output from the evaluation model with the feature amount (first feature amount) input. Then, the output is used as the estimated value of the label (first label) corresponding to the feature amount (first feature amount).

[0060] Alternatively, instead of acquiring the evaluation model from the evaluation model setting unit 253, the estimation unit 257 may acquire the generated evaluation model from the database 23.

[0061] The label (first label) corresponding to the feature amount (first feature amount) estimated by the estimation unit 257 is output to the display unit 29.

[0062] [Processing Procedure of Classification Device] Next, the processing procedure of the classification device according to the present disclosure will be described with reference to the flowcharts of FIGS. 2 and 3.

[0063] FIG. 2 is a flowchart showing the processing procedure during learning of the classification device. The processing of the flowchart shown in FIG. 2 starts when a plurality of measurement data (second measurement data) are stored in the database 23. Note that the processing of the flowchart shown in FIG. 2 may be started by a user's instruction or may be repeatedly started at a fixed cycle.

[0064] In step S101, the reception unit 21 acquires the measurement data (second measurement data).

[0065] In step S103, the preprocessing unit 251 performs preprocessing on the measurement data acquired by the sensor 10.

[0066] In step S105, the calculation unit 255 calculates the feature amount (second feature amount) based on the measurement data or the preprocessed data.

[0067] In step S107, the evaluation model setting unit 253 performs cluster analysis on a plurality of measurement data based on feature amounts.

[0068] In step S109, the evaluation model setting unit 253 generates teacher data that is a set of a feature amount (second feature amount) calculated based on measurement data (second measurement data) and a label (second label) corresponding to the feature amount (second feature amount).

[0069] In step S111, the evaluation model setting unit 253 performs machine learning based on the teacher data to generate an evaluation model.

[0070] Next, FIG. 3 is a flowchart showing a processing procedure at the time of label estimation of the classification device. The processing of the flowchart shown in FIG. 3 starts after an evaluation model is generated by the flowchart of FIG. 2. Note that the processing of the flowchart shown in FIG. 2 may be started according to a user's instruction.

[0071] In step S201, the reception unit 21 acquires measurement data (first measurement data).

[0072] In step S203, the preprocessing unit 251 performs preprocessing on the measurement data acquired by the sensor 10.

[0073] In step S205, the calculation unit 255 calculates a feature amount (first feature amount) based on the measurement data or the preprocessed data.

[0074] In step S207, the estimation unit 257 estimates a label (first label) corresponding to the feature amount based on the feature amount and the evaluation model.

[0075] In step S209, the estimation unit 257 outputs the estimated label (first label) to the display unit 29.

[0076] [Effects of the Embodiment] As described in detail above, the classification device, classification method, and classification program according to the present disclosure receive first measurement data related to the work operation of a first worker, and calculate a first feature amount that characterizes the first measurement data. Then, based on an evaluation model that outputs the first label of the first measurement data for an input including the first feature amount, the first label corresponding to the first measurement data is estimated. Here, the evaluation model is a learning model generated by machine learning based on teacher data that includes a second feature amount that characterizes second measurement data related to the work operation of a second worker and a second label that designates a cluster obtained by performing cluster analysis on the second feature amount.

[0077] Thereby, the classification of the work operation of the worker can be automated, and by guiding a worker of the same school as the worker who performs the learning, the time until the worker becomes proficient in the work operation can be shortened. In particular, it is possible to identify the schools of work operations existing among workers. As a result, it is possible to improve the work operation when a learner learns the work operation from an instructor through OJT. Furthermore, the period required for education can be shortened.

[0078] In the classification device, classification method, and classification program according to the present disclosure, the work operation may include a welding operation. And the first measurement data may include at least any one of the tip coordinates of the welding torch, the posture information of the welding torch, the current value, the voltage value, and the posture information of the workpiece related to the welding operation, the time ratio of the welding operation in the work operation, and the number of work interruptions in the work operation.

[0079] Thereby, it is possible to identify the schools of welding operations existing among workers. Therefore, when assigning an instructor for welding operations to a learner who learns welding operations, it is possible to reduce the possibility that an instructor having a school different from the school of welding operations possessed by the learner is assigned. As a result, the efficiency when the learner learns welding operations can be improved. Furthermore, the period required for education can be shortened.

[0080] The classification device, classification method, and classification program according to the present disclosure may use, as the first feature amount, the power spectral density calculated based on at least any one of the tip coordinates of the welding torch, the attitude information of the welding torch, the current value, the voltage value, and the attitude information of the workpiece related to the welding work among those related to the welding work. Thereby, it is possible to identify the welding work styles existing among the workers. In addition, the number of parameters included in the measurement data when performing machine learning can be reduced, and the efficiency of machine learning can be improved.

[0081] The classification device, classification method, and classification program according to the present disclosure may use, as the first feature amount, the volume of the ellipsoid representing the locus drawn by the tip coordinates. Thereby, it is possible to identify the welding work styles existing among the workers. In addition, the number of parameters included in the measurement data when performing machine learning can be reduced, and the efficiency of machine learning can be improved.

[0082] The classification device, classification method, and classification program according to the present disclosure may calculate the first feature amount by performing principal component analysis or factor analysis on the first measurement data. Thereby, the number of parameters included in the measurement data when performing machine learning can be reduced, and the efficiency of machine learning can be improved.

[0083] The classification device, classification method, and classification program according to the present disclosure may further calculate preprocessed data by performing at least any one of smoothing processing by moving average, first-order differential processing, second-order differential processing, and SNV (Standard Normal Variate) conversion processing on the first measurement data. Further, the first feature amount may be calculated based on the preprocessed data. Thereby, noise included in the measurement data can be removed, and machine learning can be efficiently performed. Furthermore, the accuracy in identifying the welding work styles existing among the workers can be improved.

[0084] In the classification device, classification method, and classification program according to the present disclosure, the evaluation model may be represented by a neural network or a support vector machine. Thereby, it is possible to learn the features inherent in the teacher data and identify the welding work styles existing among the workers.

[0085] In the classification device, classification method, and classification program according to the present disclosure, the evaluation model may be represented by an unlearned class estimation probability neural network. Thereby, it can be determined that it is difficult to classify measurement data having feature amounts belonging to unlearned clusters that are not assumed during learning based on the teacher data into any of the clusters included in the teacher data.

[0086] The classification device, classification method, and classification program according to the present disclosure may further present the first label to the user. Thereby, the user can be made to recognize the welding work styles existing among the workers.

[0087] The classification device, classification method, and classification program according to the present disclosure may further store the identification information of the second worker related to the second feature amount belonging to the cluster specified by the second label in association with the second label. Further, the identification information of the second worker associated with the same second label as the first label may be presented to the user. Thereby, a second worker suitable for guiding the first worker related to the first label can be selected.

[0088] Each function shown in the above-described embodiment can be implemented by one or more processing circuits. The processing circuit includes a programmed processor, an electric circuit, etc., and further includes a device such as an application-specific integrated circuit (ASIC), or a circuit component arranged to execute the described function.

[0089] According to the present disclosure, it is possible to automate the classification of the work operations of an operator and shorten the time until the operator becomes proficient in the work operations. Therefore, for example, it is possible to contribute to Goal 4 of the Sustainable Development Goals (SDGs) led by the United Nations, "Ensure inclusive and equitable quality education and promote lifelong learning opportunities for all."

[0090] Although several embodiments have been described, it is possible to modify or deform the embodiments based on the above disclosure. All the components of the above embodiments and all the features described in the claims may be individually extracted and combined as long as they do not conflict with each other.

Explanation of Reference Numerals

[0091] 10 Sensor 20 Classification Device 21 Receiver 23 Database (Storage Unit) 25 Controller 251 Pretreatment Unit 253 Evaluation Model Setting Unit 255 Calculation Unit 257 Estimation Unit 27 Operation Unit 29 Display Unit

Claims

1. A receiving unit that receives first measurement data related to the work operation of a first worker, A calculation unit that calculates a first feature amount characterizing the first measurement data, Based on an evaluation model that outputs a first label that designates any one of clusters obtained by performing cluster analysis on a second feature amount that characterizes second measurement data related to the work operation of a second worker for an input including the first feature amount, an estimation unit that estimates the first label corresponding to the first measurement data, A classification device comprising: The first measurement data includes at least one of a time ratio of welding work in the work operation and the number of work interruptions of the welding work, a classification device.

2. A receiving unit that receives first measurement data related to the work operation of a first worker, A calculation unit that calculates a first feature amount characterizing the first measurement data, Based on an evaluation model that outputs a first label that designates any one of clusters obtained by performing cluster analysis on a second feature amount that characterizes second measurement data related to the work operation of a second worker for an input including the first feature amount, an estimation unit that estimates the first label corresponding to the first measurement data, A classification device comprising: The work operation includes welding work, The calculation unit calculates the first feature amount based on the power spectral density calculated based on at least one of the tip coordinates of the welding torch, the attitude information of the welding torch, the current value, the voltage value, and the attitude information of the workpiece related to the welding work related to the welding work. Classification device.

3. A receiving unit that receives first measurement data related to the work operation of a first worker, A calculation unit that calculates a first feature amount characterizing the first measurement data, Based on an evaluation model that outputs a first label that designates any one of clusters obtained by performing cluster analysis on a second feature amount that characterizes second measurement data related to the work operation of a second worker for an input including the first feature amount, an estimation unit that estimates the first label corresponding to the first measurement data, A classification device comprising: The work operation includes welding work, The calculation unit calculates the first feature amount based on the volume of the ellipsoid representing the locus drawn by the tip coordinates of the welding torch related to the welding work. Classification device.

4. The classification device according to any one of claims 1 to 3, wherein the calculation unit calculates the first feature amount by performing principal component analysis or factor analysis on the first measurement data.

5. The classification device further includes a preprocessing unit that performs at least any one of smoothing processing by moving average, first-order differentiation processing, second-order differentiation processing, and SNV (Standard Normal Variate) conversion processing on the first measurement data to calculate preprocessed data. The calculation unit calculates the first feature amount based on the preprocessed data, and the classification device according to any one of claims 1 to 4.

6. The evaluation model is represented by a neural network or a support vector machine, and the classification device according to any one of claims 1 to 5.

7. The evaluation model is represented by an unlearned class estimation probability neural network, and the classification device according to any one of claims 1 to 6.

8. The classification device according to any one of claims 1 to 7 further includes a display unit that presents the first label to the user.

9. A storage unit that stores the identification information of the second worker related to the second feature amount belonging to the cluster in association with the cluster, A display unit that presents the identification information of the second worker associated with the cluster designated by the first label to the user, The classification device according to any one of claims 1 to 8 further includes.

10. Receiving first measurement data related to the work operation of the first worker, Calculating a first feature amount characterizing the first measurement data, Based on an evaluation model that outputs a first label that designates any one of the clusters obtained by cluster analysis of the second feature amount characterizing the second measurement data related to the work operation of the second worker for the input including the first feature amount, estimating the first label corresponding to the first measurement data A classification method, The first measurement data includes at least any one of the time ratio of the welding operation in the work operation and the number of work interruptions of the welding operation, and the classification method.

11. Receiving first measurement data related to the work operation of the first worker, Calculating a first feature amount characterizing the first measurement data, Based on an evaluation model that outputs a first label that designates any one of the clusters obtained by cluster analysis of the second feature amount characterizing the second measurement data related to the work operation of the second worker for the input including the first feature amount, estimating the first label corresponding to the first measurement data A classification method, The work operation includes a welding operation. In calculating the first feature amount, based on the power spectral density calculated based on at least any one of the tip coordinates of the welding torch, the attitude information of the welding torch, the current value, the voltage value, and the attitude information of the workpiece related to the welding operation in the welding operation, the first feature amount is calculated. Classification method.

12. Receive first measurement data related to the working operation of a first operator. Calculate a first feature amount that characterizes the first measurement data. Based on an evaluation model that outputs a first label that designates any one of the clusters obtained by cluster analysis of the second feature amount that characterizes the second measurement data related to the working operation of the second operator for the input including the first feature amount, estimate the first label corresponding to the first measurement data. A classification method, The working operation includes a welding operation. In calculating the first feature amount, based on the volume of the ellipsoid that represents the locus drawn by the tip coordinates of the welding torch in the welding operation, the first feature amount is calculated. Classification method.

13. Cause a computer to Receive a step of receiving first measurement data related to the working operation of a first operator. Calculate a step of calculating a first feature amount that characterizes the first measurement data. Based on an evaluation model that outputs a first label that designates any one of the clusters obtained by cluster analysis of the second feature amount that characterizes the second measurement data related to the working operation of the second operator for the input including the first feature amount, estimate the first label corresponding to the first measurement data. A classification program for causing execution, The first measurement data includes at least any one of the time ratio of the welding operation in the working operation and the number of work interruptions of the welding operation. Classification program.

14. Cause a computer to Receive a step of receiving first measurement data related to the working operation of a first operator. Calculate a step of calculating a first feature amount that characterizes the first measurement data. Based on an evaluation model that outputs a first label that designates any one of the clusters obtained by cluster analysis of the second feature amount that characterizes the second measurement data related to the working operation of the second operator for the input including the first feature amount, estimate the first label corresponding to the first measurement data. A classification program for causing execution, The working operation includes a welding operation. In the step of calculating the first feature amount, the first feature amount is calculated based on the power spectral density calculated based on at least any one of the tip coordinates of the welding torch, the attitude information of the welding torch, the current value, the voltage value, and the attitude information of the workpiece related to the welding operation, which are related to the welding operation. Classification program.

15. A computer, Receiving first measurement data related to the working operation of a first operator; Calculating a first feature amount characterizing the first measurement data; Estimating the first label corresponding to the first measurement data based on an evaluation model that outputs a first label designating any one of clusters obtained by performing cluster analysis on a second feature amount characterizing second measurement data related to the working operation of a second operator with respect to an input including the first feature amount; A classification program for causing the above to be executed, The working operation includes a welding operation, In the step of calculating the first feature amount, the first feature amount is calculated based on the volume of an ellipsoid representing the locus drawn by the tip coordinates of the welding torch related to the welding operation. Classification program.

Images