JP7619012B2 - Estimation system, learning device, estimation device, estimation method, learning method, and program - Google Patents

Description

The present invention relates to an estimation system, a learning device, an estimation device, an estimation method, a learning method, and a program.

Various techniques have been proposed in the past for detecting defects that occur in workpieces in which multiple sheet-like materials are laminated together. For example, the following Patent Document 1 discloses a technique in which an inspection light is irradiated onto the workpiece while the laminated workpiece is being transported, and defects are detected based on captured images of the workpiece irradiated with the inspection light. This makes it possible to detect defects that occur during the lamination process.

However, defects (failures) may occur in laminated workpieces after a certain time has passed since processing. For example, tunneling may occur in the workpiece. Tunneling is a phenomenon in which parts of a laminated workpiece peel off, and the peeled parts rise up in a tunnel-like shape.

JP 2020-034345 A

To improve the production quality of workpieces produced by lamination, it is important to detect defects that occur in the workpiece during lamination, as in the technology of Patent Document 1. In addition, to improve production quality, it is also important to prevent defects from occurring in the first place. Therefore, it is desirable to be able to laminate multiple materials without causing defects in the workpiece.

In view of the above-mentioned problems, the object of the present invention is to provide an estimation system, a learning device, an estimation device, an estimation method, a learning method, and a program that can suppress the occurrence of defects in laminated workpieces.

In order to solve the above-mentioned problems, an estimation system according to one aspect of the present invention is an estimation system that estimates an amount of deviation of a center position when a phenomenon occurs in which a center position of one end face of an original roll of a sheet-like workpiece formed by bonding together a plurality of materials is shifted to the outside of the original roll over time, and estimates a change in tension applied to the plurality of materials when the plurality of materials are laminated together and processed into the workpiece, for the workpiece wound around the original roll in which the phenomenon occurs, and a change in tension applied to the plurality of materials when the workpiece is wound up and processed into the original roll. The prediction model is generated by using actual measured values of processing conditions, which include at least a change in tension applied to an object , as explanatory variables, and the actual measured value of the deviation amount on the original roll on which the phenomenon occurred as a target variable, and by machine learning the correspondence between the processing conditions and the deviation amount into a learning model.The prediction model is equipped with a memory unit that stores the prediction model that outputs the deviation amount based on the processing conditions in the processing step of bonding the multiple materials together and processing them into the original roll, an acquisition unit that acquires the processing conditions, and an estimation unit that estimates the deviation amount using the prediction model based on the processing conditions acquired by the acquisition unit.

A learning device according to one embodiment of the present invention is provided with a learning unit that generates a predictive model by machine learning the correspondence between the processing conditions and the amount of shift, using actual measured values of processing conditions for the workpiece wound around the original roll in which a sheet-like workpiece made of multiple materials bonded together shifts toward the outside of the original roll over time, the actual measured values of processing conditions including at least the change in tension applied to the multiple materials when the multiple materials are bonded together and laminated to form the workpiece, and the change in tension applied to the workpiece when the workpiece is wound up and processed into the original roll, as explanatory variables, and the actual measured value of the amount of shift of the central position of the original roll in which the phenomenon occurred as an objective variable, and by machine learning the correspondence between the processing conditions and the amount of shift into the learning model.

According to one embodiment of the present invention, there is provided an estimation device that estimates a deviation amount of a center position of an end face of a sheet-like workpiece formed by bonding multiple materials together and wound into a roll, in a case where the center position of one of the end faces of the original workpiece shifts to the outside of the original workpiece over time. The estimation device includes: an acquisition unit that acquires processing conditions in a processing step in which the multiple materials are bonded together and processed into the original workpiece; and an estimation unit that estimates the deviation amount using a prediction model based on the processing conditions acquired by the acquisition unit. The prediction model uses actual measured values of the processing conditions, which include at least a change in tension applied to the multiple materials when the multiple materials are bonded together and processed into the workpiece by laminating them, for the workpiece wound around the original workpiece in which the phenomenon occurred, as explanatory variables, and an actual measured value of the deviation amount for the original workpiece in which the phenomenon occurred as a target variable, and is a model that outputs the deviation amount based on the processing conditions by training a learning model to learn a correspondence between the processing conditions and the deviation amount.

An estimation method according to one aspect of the present invention is a method for estimating an amount of deviation of a center position of one end face of a sheet-like workpiece formed by bonding multiple materials together and wound into a roll, in a case where a phenomenon occurs in which the center position of the center position of one end face of the original roll shifts to the outside of the original roll over time, the method comprising: a memory unit for storing, for the workpiece wound around the original roll in which the phenomenon occurs, a change in tension applied to the multiple materials when the multiple materials are bonded together and laminated to form the workpiece, and a change in tension applied to the workpiece when the workpiece is wound up and processed into the original roll, a prediction model generated by using actual measured values of processing conditions including at least the above as explanatory variables and actual measured values of the amount of deviation on the original roll on which the phenomenon occurred as a target variable, and machine learning the correspondence between the processing conditions and the amount of deviation into a learning model, the prediction model outputting the amount of deviation based on the processing conditions in a processing step in which the multiple materials are bonded together and processed into the original roll being stored; an acquisition unit acquiring the processing conditions; and an estimation unit estimating the amount of deviation using the prediction model based on the processing conditions acquired by the acquisition unit.

A learning method according to one embodiment of the present invention includes a learning unit that generates a predictive model by machine learning the correspondence between the processing conditions and the amount of deviation using actual measured values of processing conditions for the workpiece wound around the original roll in which a sheet-like workpiece made of multiple materials bonded together shifts over time to the outside of the original roll, the actual measured values of processing conditions including at least a change in tension applied to the multiple materials when the multiple materials are bonded together and laminated to form the workpiece, and a change in tension applied to the workpiece when the workpiece is wound up and processed into the original roll, as explanatory variables and an actual measured value of the amount of deviation of the center position of the original roll in which the phenomenon occurred, as an objective variable, and machine learning the correspondence between the processing conditions and the amount of deviation into a learning model.

An estimation method according to one aspect of the present invention is a method for estimating a deviation amount of a center position of one end face of a sheet-like workpiece formed by bonding multiple materials together and wound into a roll, in a case where a phenomenon occurs in which the center position of the center position of one end face of the original workpiece shifts to the outside of the original workpiece over time, the estimation method including: an acquisition unit acquiring processing conditions in a processing step in which the multiple materials are bonded together and processed into the original workpiece; and an estimation unit estimating the deviation amount using a prediction model based on the processing conditions acquired by the acquisition unit, wherein the prediction model uses actual measured values of the processing conditions, which include at least a change in tension applied to the multiple materials when the multiple materials are bonded together and laminated to process the workpiece into the workpiece, and a change in tension applied to the workpiece when the workpiece is wound up and processed into the original workpiece, as explanatory variables, and an actual measured value of the deviation amount in the original workpiece in which the phenomenon occurred as a target variable, and outputs the deviation amount based on the processing conditions, generated by machine learning the correspondence between the processing conditions and the deviation amount into a learning model.

A program according to one embodiment of the present invention causes a computer to function as a learning unit that generates a predictive model by machine learning the correspondence between the processing conditions and the amount of deviation using actual measured values of processing conditions, which include at least the change in tension applied to the multiple materials when the multiple materials are laminated together and processed into the workpiece, and the change in tension applied to the workpiece when the workpiece is wound up and processed into the original roll, as explanatory variables, and the actual measured value of the deviation amount of the center position of the original roll in which the phenomenon occurred, as a target variable, and by machine learning the correspondence between the processing conditions and the amount of deviation into a learning model.The program causes a computer to function as a learning unit that generates a predictive model by machine learning the correspondence between the processing conditions and the amount of deviation, using actual measured values of processing conditions, which include at least the change in tension applied to the multiple materials when the multiple materials are laminated together and processed into the workpiece, as explanatory variables, and the change in tension applied to the workpiece when the workpiece is rolled up and processed into the original roll, as explanatory variables, and the actual measured value of the deviation amount of the center position of the original roll in which the phenomenon occurred, as a target variable,

A program according to one embodiment of the present invention is a program for estimating the amount of deviation of a center position of one end face of a sheet-like workpiece formed by bonding multiple materials together and wound into a roll, in the case where the center position of the center position of one end face of the original workpiece shifts to the outside of the original workpiece over time. The program causes a computer to function as an acquisition unit that acquires processing conditions in a processing step in which the multiple materials are bonded together and processed into the original workpiece, and an estimation unit that estimates the amount of deviation using a prediction model based on the processing conditions acquired by the acquisition unit. The prediction model uses actual measured values of the processing conditions, which include at least a change in tension applied to the multiple materials when the multiple materials are bonded together and processed into the workpiece by laminating them, for the workpiece wound around the original workpiece in which the phenomenon occurred, as explanatory variables, and an actual measured value of the amount of deviation in the original workpiece in which the phenomenon occurred as a target variable, and is a model that outputs the amount of deviation based on the processing conditions, which is generated by machine learning the correspondence between the processing conditions and the amount of deviation into a learning model.

The present invention makes it possible to prevent defects in laminated workpieces.

FIG. 1 is a diagram showing an example of the configuration of a laminating device according to an embodiment of the present invention. FIG. 2 is a diagram illustrating an example of tunneling according to the embodiment. FIG. 13 is a diagram illustrating an example of lateral deviation according to the present embodiment. FIG. 1 is a diagram illustrating an example of a configuration of an estimation system according to an embodiment of the present invention. 2 is a block diagram showing an example of a functional configuration of the learning device according to the present embodiment. FIG. FIG. 11 is a diagram illustrating an example of pre-processing according to the embodiment. 2 is a block diagram showing an example of a functional configuration of the estimation device according to the present embodiment. FIG. 10 is a flowchart showing an example of a flow of processing performed by the learning device according to the present embodiment. 4 is a flowchart showing an example of a flow of processing performed by the estimation device according to the present embodiment.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
In the drawings, mutually orthogonal X-axis, Y-axis, and Z-axis are shown as necessary. On each axis, the direction in which the arrow extends is referred to as the "positive direction," and the direction opposite to the positive direction is referred to as the "negative direction."

1. Configuration of Laminating Device
A laminating device is a machine that processes multiple materials by bonding them together to form a laminate.
The configuration of the laminating device according to the present embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of the configuration of the laminating device according to the present embodiment. As shown in Fig. 1, the laminating device 10 includes a base material raw roll 11, rolls 12a to 12h, an adhesive coating section 13, a drying device 14, a bonding material raw roll 15, a laminating section 16, and a workpiece raw roll 17.

The base material roll 11 is a roll of sheet-like base material 110 (one example of a material) wound around a core. The base material 110 is, for example, paper, a resin film such as polypropylene, polyethylene, or polyester, a metal foil such as aluminum foil or copper foil, or a laminated film in which a metal foil is vapor-deposited onto such a resin film. The base material 110 unwound from the base material roll 11 is supplied to the adhesive coating section 13 via roll 12a.

The adhesive coating unit 13 applies adhesive 130 to the substrate 110. For example, when the substrate 110 passes between rolls 12b and 12c, the adhesive coating unit 13 applies adhesive 130 to the substrate 110 via roll 12b. The substrate 110 to which adhesive 130 has been applied is supplied to the drying device 14 via roll 12d.

The drying device 14 is a device that dries the substrate 110 that passes through the inside. The drying device 14 dries the substrate 110 supplied from the adhesive coating section 13 via roll 12d. The drying device 14 supplies the dried substrate 110 to the laminating section 16 via rolls 12e and 12f.

The adhesive raw roll 15 is a roll of sheet-like adhesive 150 (an example of a material) wound around a winding core. The adhesive 150 is, for example, paper, a resin film such as polypropylene, polyethylene, or polyester, a metal foil such as aluminum foil or copper foil, or a laminated film in which a metal foil is vapor-deposited onto such a resin film. The adhesive 150 unwound from the adhesive raw roll 15 is supplied to the lamination section 16.

The laminating unit 16 is composed of rolls 12g and 12h, and bonds together a plurality of materials passing between the rolls 12g and 12h. For example, the laminating unit 16 bonds together a base material 110 supplied from the drying device 14 and a bonding material 150 supplied from a bonding material roll 15. The bonded base material 110 and bonding material 150 are also referred to as a processed product 170.

The workpiece 170 is wound around the winding core 171 via the roll 12i (cooling roll). The sheet-like workpiece 170 wound around the winding core 171 in a roll form is the workpiece roll 17. The center position CP indicates the center position of the end face of the end face of the workpiece roll 17.

As described above, the laminating device 10 laminates the substrate 110 and the bonding material 150 (multiple materials) to form the workpiece 170 (lamination process), and then rolls up the workpiece 170 to form the workpiece roll 17. This process is called the laminating process. Note that the number of materials laminated by the laminating device 10 is not limited to two, the substrate 110 and the bonding material 150, and may be three or more.

A certain time after processing, defects (failures) may occur in the laminated workpiece 170 and the original workpiece roll 17. For example, a defect called tunneling may occur in the workpiece 170, and a defect called lateral shift (bamboo shoot phenomenon) may occur in the original workpiece roll 17.
Tunneling is a phenomenon in which the laminated workpiece 170 partially peels off and the peeled off portion rises up in a tunnel shape. Tunneling can occur due to distortion caused by differences in the return of each material due to differences in the shrinkage rates of the laminated materials.
Lateral shifting is a phenomenon in which the center position of one end face of the workpiece original roll 17 shifts outward over time, forming a bamboo shoot shape. Specifically, the center position CP of one end face of the workpiece original roll 17 shifts outward over time. Lateral shifting can be caused by aging, which promotes hardening of the adhesive 130 by placing the workpiece original roll 17 for a predetermined time in a space such as a warehouse that is kept at a predetermined temperature. Lateral shifting can also be caused by placing the workpiece original roll 17 in a space at room temperature.

Here, an example of tunneling will be described with reference to Fig. 2. Fig. 2 is a diagram showing an example of tunneling according to the present embodiment. The left diagram of Fig. 2 shows a state of the workpiece 170 before tunneling occurs. The right diagram of Fig. 2 shows a state of the workpiece 170 after tunneling occurs.
Before the occurrence of tunneling, the base material 110 and the adhesive material 150 are adhered to each other as shown in the left diagram of Fig. 2. However, after the occurrence of tunneling, the base material 110 and the adhesive material 150 are partially peeled off from each other as shown in region A1 in the right diagram of Fig. 2, and the peeled off portion is raised in a tunnel shape.

Here, an example of lateral shift will be described with reference to Fig. 3. Fig. 3 is a diagram showing an example of lateral shift according to this embodiment. The left diagram of Fig. 3 shows the state of the workpiece original web 17 before the lateral shift occurs. The right diagram of Fig. 3 shows the state of the workpiece original web 17 after the lateral shift occurs.
Before the occurrence of lateral displacement, as shown in the left diagram of FIG. 3, the center position CP of the end face of the original workpiece web 17 is the same as the position P1 of the end face.
3, after the occurrence of the lateral shift, the winding core 171 moves to the outside of the workpiece web 17 (positive direction of the Z axis) due to the lateral shift, and the center position CP of the end face of the workpiece web 17 moves from P1 to P2. The difference between the position P2 after the center position CP moves and the position P1 before the movement is the amount of shift.

When lateral shift occurs in the workpiece original roll 17, tunneling may occur in the workpiece 170. Therefore, if the occurrence of lateral shift in the workpiece original roll 17 can be suppressed, the occurrence of tunneling in the workpiece 170 can also be suppressed. The occurrence of lateral shift in the workpiece original roll 17 is influenced by the processing conditions when the workpiece 170 is processed. Therefore, if the user knows the correspondence between the processing conditions and the lateral shift, the user can identify the cause of the lateral shift. By identifying the cause of the lateral shift, the user can adjust the processing conditions in the lamination process so that the lateral shift does not occur. As a result, if the lateral shift does not occur, the occurrence of tunneling can be suppressed.
Therefore, the estimation system of this embodiment estimates the amount of deviation of the center position CP when lateral deviation occurs in the workpiece original roll 17 based on the processing conditions during lamination. The estimation by the estimation system allows the user to understand the correspondence between the processing conditions and the amount of deviation, identify the cause of the lateral deviation, and adjust the processing conditions for lamination so that lateral deviation does not occur in the workpiece original roll 17. This suppresses the occurrence of lateral deviation in the workpiece original roll 17, and also suppresses the occurrence of tunneling in the workpiece 170.

2. Configuration of Estimation System
An example of the configuration of the estimation system according to the present embodiment will be described with reference to Fig. 4. Fig. 4 is a diagram showing an example of the configuration of the estimation system according to the present embodiment. As shown in Fig. 4, the estimation system 1 includes a lamination device 10, a learning device 20, and an estimation device 30.

The laminating device 10 is a device that processes by laminating multiple materials together. The laminating device 10 is connected to the learning device 20 so that it can communicate with the learning device 20. The actual values of the processing conditions measured by a sensor device provided in the laminating device 10 are transmitted to the learning device 20.

The laminating device 10 is provided with a sensor device for measuring the actual values of the processing conditions from when the multiple materials are laminated to when they are processed into a roll of raw material. The laminating device 10 may be provided with multiple types of sensor devices. Furthermore, the laminating device 10 may be provided with multiple sensor devices for each type. The actual values of each processing condition at multiple locations may be measured.

The processing conditions include, for example, the tension (winding tension) applied to each material or workpiece 170 during the processing step, the temperature at which the substrate 110 is dried in the drying device 14, the surface temperature of each material or workpiece 170, the speed at which each material or workpiece 170 moves within the laminating device 10, the amount of adhesive 130 applied to the substrate 110, etc.

The tension is measured by, for example, a tension sensor. The tension sensor is provided, for example, on each of the rolls 12a to 12i, and measures the tension applied to each material or workpiece 170 passing through each roll. The tension measured by the tension sensor is, for example, the tension between rolls that are spaced apart. Specifically, the space between rolls that are spaced apart is between roll 12a and roll 12b, between roll 12c and roll 12d, between roll 12f and roll 12g, between roll 12g (or roll 12h) and roll 12i, etc.
The temperature is measured, for example, by a temperature sensor. The temperature sensor is provided, for example, near the entrance, inside, or near the exit of the drying device 14, and measures the temperature at each location. The temperature sensor may be provided at any location of the laminating device 10 so as to be able to measure the surface temperature of each material, the workpiece 170, etc.
The speed is measured, for example, by a speed sensor. The speed sensor may be provided at any location in the laminator 10 so as to be able to measure the speed at which each material or workpiece 170 moves within the laminator 10.
The amount of adhesive 130 applied to the substrate 110 is measured by, for example, a weight sensor or an imaging sensor. The weight sensor is provided, for example, so as to be able to measure the change in weight of the adhesive 130 in the adhesive coating section 13, and measures the amount of change in weight of the adhesive 130 before and after application of the adhesive 130 as the amount of application. The imaging sensor is provided, for example, so as to be able to capture an image of the surface position of the adhesive 130 in the container, and captures an image of the surface position before and after application of the adhesive 130. The amount of application can be calculated from the amount of change in the surface position of the adhesive 130 using the image captured by the imaging sensor.

The learning device 20 generates a prediction model by having the learning model perform supervised learning. The learning device 20 is, for example, a computer device such as a cloud device, a server device, or a PC (personal computer). The prediction model here is a model that predicts the amount of deviation when lateral deviation occurs in the workpiece original roll 17 from the processing conditions. The learning device 20 performs supervised learning on the learning model, using, for example, actual measured values of the processing conditions, including at least changes in tension applied to each material and the workpiece 170 in the processing process, as explanatory variables, and actual measured values of the deviation amount in the workpiece original roll 17 where lateral deviation occurs as objective variables. The actual measured values of the processing conditions are, for example, values measured by a sensor device provided in the laminating device 10, and are input from the laminating device 10. The actual measured value of the deviation amount is, for example, a value measured by a user, and is input by the user. The learning device 20 is connected to the estimation device 30 so as to be able to communicate with it.

The estimation device 30 uses the prediction model generated by the learning device 20 to predict the amount of lateral shift that occurs in the workpiece roll 17. The estimation device 30 is, for example, a cloud device, a server device, a computer device such as a PC, etc. The estimation device 30 is connected to the learning device 20 so as to be able to communicate with it.

3. Functional configuration of the learning device
Next, an example of the functional configuration of the learning device 20 according to this embodiment will be described with reference to Fig. 5. Fig. 5 is a block diagram showing an example of the functional configuration of the learning device 20 according to this embodiment. As shown in Fig. 5, the learning device 20 includes a communication unit 21, an input unit 22, a control unit 23, and a storage unit 24.

The communication unit 21 has a function of transmitting and receiving various information. For example, the communication unit 21 receives processing conditions transmitted from the laminating device 10. The communication unit 21 outputs the received information to the control unit 23. The communication unit 21 also transmits a prediction model to the estimation device 30.

The input unit 22 has a function of accepting input. The input unit 22 outputs the accepted input to the control unit 23. The input unit 22 can be realized, for example, by an input device such as a touch panel, a keyboard, or a mouse that the learning device 20 has as hardware. The input unit 22 accepts, for example, an input of an actual measured value of the amount of deviation by a user, and outputs the input to the control unit 23.

The control unit 23 has a function of controlling the overall operation of the learning device 20. The control unit 23 is realized, for example, by causing a central processing unit (CPU) or a graphic processing unit (GPU) provided as hardware in the learning device 20 to execute a program.
As shown in FIG. 5 , the control unit 23 includes a learning information acquisition unit 230 , a preprocessing unit 231 , a learning unit 232 , and an output processing unit 233 .

The learning information acquisition unit 230 acquires learning information. The learning information is information used for learning when performing supervised learning. The learning information is, for example, the actual measured values of the processing conditions and the actual measured values of the deviation. The learning information acquisition unit 230 acquires the actual measured values of the processing conditions input from the communication unit 21 and the actual measured values of the deviation input from the input unit 22 as learning information, and outputs the acquired learning information to the pre-processing unit 231.

The pre-processing unit 231 performs pre-processing (pre-processing) for supervised learning. Specifically, the pre-processing unit 231 generates a data set for learning. The data set for learning is a combination of explanatory variables and target variables. The pre-processing unit 231 sets the actual measured values of the processing conditions as explanatory variables. The pre-processing unit 231 sets the actual measured values of the deviation amounts corresponding to the explanatory variables as target variables. The pre-processing unit 231 generates a data set for learning by combining the explanatory variables and target variables. The pre-processing unit 231 performs pre-processing for each lot of the workpiece raw roll 17.

In addition, the preprocessing unit 231 processes the actual measured values of the processing conditions, and the processed information is used as an explanatory variable. In a general method of data aggregation in preprocessing, the average, maximum value, minimum value, variance value, etc. are calculated. In a general method, the characteristics of the data (value of the data) cannot be reflected in the aggregated data, and the performance of the machine learning model required for factor analysis may not be obtained. Therefore, in the preprocessing of this embodiment, the change rate (slope) of an arbitrary section of the data is calculated so that the characteristics of the data are not lost even when the data is aggregated. This makes it possible to obtain the performance of the machine learning model required for factor analysis. In addition, the user can easily understand what the change rate calculated by the preprocessing indicates. Therefore, the user can quickly make a decision when taking action based on the analysis results.

Specifically, the preprocessing unit 231 calculates the average value of the actual measurement values measured by all the sensor devices installed in the lamination device 10 for each type of sensor device and for each lot. The preprocessing unit 231 includes the calculated average value in the explanatory variable instead of the actual measurement value.

Note that the preprocessing unit 231 may also perform preprocessing other than the calculation of the average value on a particular measured value.
For example, the pre-processing unit 231 performs pre-processing to calculate a variance value for each lot for the actual measured value of tension applied to each material or workpiece 170 in the processing step included in the actual measured value of the processing condition. Then, the pre-processing unit 231 includes the calculated variance value in the explanatory variable instead of the actual measured value of tension.
Furthermore, the pre-processing unit 231 performs pre-processing to calculate the rate of change in tension in units of lots for the actual measured values of tension applied to each material or workpiece 170 in the processing step included in the actual measured values of the processing conditions in units of arbitrary sections. The arbitrary sections in this embodiment are, for example, sections separated by a predetermined winding length unit. The winding length is, for example, the length wound by the workpiece raw roll 17. The unit of the winding length is, for example, meters (m). Note that the arbitrary sections may be sections separated by a predetermined unwinding length unit, or may be sections separated by a predetermined time unit.
Specifically, the pre-processing unit 231 calculates the rate of change in tension based on the time-series change in the actual measured value of tension. Then, the pre-processing unit 231 includes the calculated rate of change in tension in the explanatory variables instead of the actual measured value of tension.
The pre-treatment unit 231 also performs pre-treatment to calculate, for each lot, the maximum and minimum values of the actual measured temperature at which the substrate 110 is dried in the drying device 14. The pre-treatment unit 231 then includes the calculated maximum and minimum values of the temperature in the explanatory variables instead of the actual measured temperature value.

The pre-processing unit 231 may also perform pre-processing to calculate a correlation coefficient indicating the correlation between the actual measured values of tension applied to each material or the workpiece 170 in the processing step included in the actual measured values of the processing conditions and the actual measured value of the radius of the workpiece original roll 17. The pre-processing unit 231 may include the calculated correlation coefficient in the explanatory variables.
The tension applied to the workpiece when it is wound is generally set to decrease with an increase in the winding radius (i.e., the radius of the workpiece roll 17). The radius of the workpiece roll 17 increases with the passage of time. Therefore, there is an inverse correlation between the actual measured values of the tension applied to each material or workpiece 170 and the actual measured value of the radius of the workpiece roll 17. The radius of the workpiece roll 17 is based on the center position CP.

Here, an example of the pre-processing performed by the pre-processing unit 231 will be described with reference to Fig. 6. Fig. 6 is a diagram showing an example of the pre-processing according to this embodiment. The graph shown in Fig. 6 shows the time series change in the actual measured value of the tension applied to each material and the workpiece 170 in the processing step. The horizontal axis of the graph shown in Fig. 6 shows the time t, and the vertical axis shows the tension x.
The pre-processing unit 231 calculates the slope in a section from time _t1 to time _t2 (a section divided by a predetermined winding length unit), for example, as the rate of change. Specifically, the pre-processing unit calculates the slope by the following formula (1).
Slope=(x ₂ −x ₁ )/(t ₂ −t ₁ ) (1)

Note that, although FIG. 6 shows an example in which the pre-processing unit 231 calculates the slope in one section, the present invention is not limited to such an example. For example, the pre-processing unit 231 may divide the graph into multiple sections in units of a predetermined winding length and calculate the slope of each divided section.

The learning unit 232 generates a prediction model by performing supervised learning on the learning model using a learning data set. For example, the learning unit 232 uses a data set in which the actual measurement values of the processing conditions when the workpiece original roll 17 with lateral displacement is processed are used as explanatory variables, and the actual measurement value of the displacement amount in the workpiece original roll 17 with lateral displacement is used as a target variable, and machine-learns the correspondence between the processing conditions and the displacement amount in the learning model. In this way, the learning unit 232 generates a prediction model. The learning model here is, for example, a deep learning model using CNN (Convolutional Neural Network) or the like. In the following, an example will be described in which the learning model is CNN, but this is not limited to this. The learning model may be a model that is applied to existing machine learning, such as DCNN (Deep CNN), decision tree, hierarchical Bayes, or SVM (Support Vector Machine).

The explanatory variables in the learning dataset are input to the learning model. The learning unit 232 adjusts the parameters of the learning model using a method such as the backpropagation method so that the output of the learning model obtained by inputting the explanatory variables (predicted value of the deviation) approaches the objective variable (actual measured value of the deviation) corresponding to the explanatory variables. The learning unit 232 performs supervised learning on the learning model by adjusting the parameters of the learning model so that the output of the learning model obtained by inputting the explanatory variables approaches the objective variable for each of the prepared learning datasets. The learning unit 232 ends the learning of the learning model when it is determined that a predetermined termination condition is satisfied. The predetermined termination condition is, for example, a condition in which the number of learning times reaches a predetermined number or the error of the predicted value is equal to or less than a predetermined threshold. The learning unit 232 sets the learning model at the time of ending the learning as a prediction model. The learning unit 232 stores the parameters set in the learning model (prediction model) at the time of ending the learning in the prediction model information 240. The parameters here are variables used to generate a predictive model, such as information indicating the number of units in each layer of the CNN's input, intermediate, and output layers, the number of hidden layers, activation functions, and the like, as well as information indicating the coupling coefficients and weights that connect the nodes in each layer.

The output processing unit 233 controls the output in the learning device 20. The output processing unit 233 transmits, for example, information about the prediction model generated by the learning unit 232 to the estimation device 30 via the communication unit 21.

The memory unit 24 is configured by a storage medium that the learning device 20 has as hardware, such as a hard disk drive (HDD), a solid state drive (SSD), a flash memory, an electrically erasable programmable read only memory (EEPROM), a random access read/write memory (RAM), a read only memory (ROM), or any combination of these storage media.
The storage unit 24 stores, for example, prediction model information 240. The prediction model information 240 is information indicating a prediction model generated by the learning device 20.

4. Functional configuration of the estimation device
7 is a block diagram showing an example of a functional configuration of an estimation device according to the present embodiment. As shown in FIG. 7, the estimation device 30 includes a communication unit 31, an input unit 32, a control unit 33, a storage unit 34, and an output unit 35.

The communication unit 31 has a function of transmitting and receiving various information. For example, the communication unit 31 receives prediction model information transmitted from the learning device 20. The communication unit 31 outputs the received information to the control unit 33.

The input unit 32 has a function of accepting input. The input unit 32 outputs the accepted input to the control unit 33. The input unit 32 can be realized, for example, by an input device such as a touch panel, a keyboard, or a mouse that the estimation device 30 has as hardware. The input unit 32 accepts, for example, input of processing conditions by a user and outputs the input to the control unit 33.

The control unit 33 has a function of controlling the overall operation of the estimation device 30. The control unit 33 is realized, for example, by causing a CPU provided as hardware in the estimation device 30 to execute a program.
As shown in FIG. 7 , the control unit 33 includes an input information acquisition unit 330 (acquisition unit), a preprocessing unit 331 , an estimation unit 332 , and an output processing unit 333 .

The input information acquisition unit 330 acquires information to be input to the prediction model. For example, the input information acquisition unit 330 acquires the processing conditions input to the input unit 32. The input information acquisition unit 330 outputs the acquired processing conditions to the preprocessing unit 331.

The preprocessing unit 331 performs preliminary processing (preprocessing) for making predictions using a prediction model. Specifically, the preprocessing unit 331 performs preprocessing on the processing conditions input from the input information acquisition unit 330 to generate input data. The input data is data to be input to the prediction model when making a prediction. The preprocessing performed by the preprocessing unit 331 is similar to the preprocessing performed by the preprocessing unit 231 of the learning device 20 described above, and therefore a detailed description thereof will be omitted.

The estimation unit 332 estimates the amount of deviation using a prediction model. The estimation unit 332 estimates the amount of deviation using a prediction model based on the processing conditions acquired by the input information acquisition unit 330. The estimation unit 332 acquires prediction model information 340 by referring to the storage unit 34, and constructs a prediction model based on the acquired information. The estimation unit 332 inputs the input data generated by the pre-processing unit 331 to the constructed prediction model. The estimation unit 332 acquires an output from the prediction model by inputting the input data to the prediction model. The information output from the prediction model here is the amount of deviation that may occur in the workpiece original roll 17 processed under the processing conditions indicated by the input data. The estimation unit 332 estimates the amount of deviation output from the prediction model as the amount of deviation that may occur in the workpiece original roll 17 processed under the processing conditions indicated by the input data.

The output processing unit 333 controls the output in the estimation device 30. For example, the output processing unit 333 outputs the amount of deviation estimated by the estimation unit 332 to the output unit 35. The output processing unit 333 may also output data highly relevant to the amount of deviation estimated by the estimation unit 332 to the output unit 35. The data output (calculated) as the highly relevant data is, for example, at least one of the data used as the explanatory variables. In this embodiment, the processing conditions are used as the explanatory variables. Therefore, the specific data output as the highly relevant data is at least one of the tension applied to each material or workpiece 170, the temperature at which the substrate 110 is dried, the surface temperature of each material or workpiece 170, the speed at which each material or workpiece 170 moves in the lamination device 10, the amount of adhesive 130 applied to the substrate 110, etc.

The storage unit 34 is configured by a storage medium provided in the estimation device 30 as hardware, such as a HDD, a flash memory, an EEPROM, a RAM, a ROM, or any combination of these storage media.
The storage unit 34 stores, for example, prediction model information 340. The prediction model information 340 is information indicating a prediction model generated by the learning device 20, and is similar to the prediction model information 240.

The output unit 35 has a function of displaying various information. For example, the output unit 35 outputs the amount of deviation input from the output processing unit 333. The output unit 35 is realized, for example, by a display that the estimation device 30 has as hardware.

<5. Processing flow>
(Flow of processing performed by the learning device)
An example of the flow of processing performed by the learning device 20 according to this embodiment will be described with reference to Fig. 8. Fig. 8 is a flowchart showing an example of the flow of processing performed by the learning device 20 according to this embodiment.

8, first, the learning device 20 acquires explanatory variables and a target variable (S100). Specifically, the learning device 20 acquires actual measured values of the processing conditions as explanatory variables. The learning device 20 also acquires actual measured values of the deviations as target variables.
The learning device 20 performs preprocessing on the acquired explanatory variables (S101).
The learning device 20 generates a data set for learning by combining the preprocessed explanatory variables and the objective variables (S102).

The learning device 20 inputs explanatory variables from the learning data set to the learning model (S103).
The learning device 20 adjusts the parameters of the learning model so that the difference between the output obtained from the learning model and the objective variable in the data set becomes small (S104).
The learning device 20 judges whether or not a predetermined learning termination condition is satisfied (S105). If the learning termination condition is satisfied (S105/YES), the learning device 20 confirms the learning model at that time as a prediction model and stores the parameter values and the like set in the model as prediction model information 240 (S106). On the other hand, if the learning termination condition is not satisfied (S105/NO), the learning device 20 returns to S103 and repeats the learning.

(Flow of processing performed by the estimation device)
An example of the flow of processing performed by the estimation device 30 according to this embodiment will be described with reference to Fig. 9. Fig. 9 is a flowchart showing an example of the flow of processing performed by the estimation device 30 according to this embodiment.

9, first, the estimating device 30 acquires input information (S200). Specifically, the estimating device 30 acquires processing conditions as explanatory variables.
The estimation device 30 performs pre-processing on the acquired input information (S201).

The estimation device 30 inputs the input data, which is the input information preprocessed, into the prediction model (S202).
The estimation device 30 estimates the amount of deviation output from the prediction model as the amount of deviation in the original workpiece web 17 where lateral deviation occurs (S203).

As described above, the estimation system 1 according to this embodiment is a system that estimates the amount of deviation of the center position when lateral deviation occurs in a roll of a sheet-like workpiece in which multiple materials are bonded together. The estimation system 1 stores a prediction model that outputs the amount of deviation based on the processing conditions in the processing step in which multiple materials are bonded together and processed into a roll of the workpiece. The estimation system 1 also acquires the processing conditions, and estimates the amount of deviation using the prediction model based on the acquired processing conditions.

This configuration allows the user to easily grasp the processing conditions when lateral shift occurs (when the shift amount is not equal to 0) and when no lateral shift occurs (when the shift amount is equal to 0). Based on each processing condition that the user has grasped, the user sets the processing conditions when no lateral shift occurs in the laminating device. This allows the laminating device to process a workpiece without lateral shift and without tunneling due to lateral shift.

Therefore, the estimation system 1 according to this embodiment makes it possible to suppress the occurrence of defects in laminated workpieces.

The embodiment of the present invention has been described above.
In the above embodiment, the learning device 20 having the function of generating a prediction model and the estimation device 30 having the function of performing estimation using the prediction model are independent of each other, but the present invention is not limited to such an example. For example, the functions of the learning device 20 and the estimation device 30 may be realized by a single device having both functions. Furthermore, the learning device 20 may have the functions of the estimation device 30, and the estimation device 30 may have the functions of the learning device 20.

In addition, a part or all of the learning device 20 and the estimation device 30 in the above-mentioned embodiment may be realized by a computer. In that case, a program for realizing this function may be recorded in a computer-readable recording medium, and the program recorded in the recording medium may be read into a computer system and executed to realize the function. In addition, the term "computer system" here includes hardware such as an OS and peripheral devices. In addition, the term "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into a computer system. Furthermore, the term "computer-readable recording medium" may also include a medium that dynamically holds a program for a short period of time, such as a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line, and a medium that holds a program for a certain period of time, such as a volatile memory inside a computer system that is the server or client in that case. The above program may be for implementing some of the functions described above, or may be capable of implementing the functions described above in combination with a program already recorded in the computer system, or may be implemented using a programmable logic device such as an FPGA (Field Programmable Gate Array).

The above describes in detail an embodiment of the present invention with reference to the drawings, but the specific configuration is not limited to the above, and various design changes can be made without departing from the spirit of the present invention.

LIST OF SYMBOLS 1 Estimation system 10 Laminating device 11 Substrate raw roll 12a to 12i Roll 13 Adhesive coating section 14 Drying device 15 Laminated material raw roll 16 Laminating section 17 Workpiece raw roll 20 Learning device 21 Communication section 22 Input section 23 Control section 24 Memory section 30 Estimation device 31 Communication section 32 Input section 33 Control section 34 Memory section 35 Output section 110 Substrate 130 Adhesive 150 Laminated material 170 Workpiece 171 Winding core 230 Learning information acquisition section 231 Preprocessing section 232 Learning section 240 Prediction model information 330 Input information acquisition section 331 Preprocessing section 332 Estimation section 333 Output processing section 340 Prediction model information

Claims (11)

An estimation system for estimating a shift amount of a center position of an end face of one of the original rolls, which is a sheet-like workpiece formed by bonding a plurality of materials together and wound into a roll, in a case where the center position of the end face of one of the original rolls shifts toward the outside of the original roll over time, comprising:
a prediction model generated by using actual measured values of processing conditions including at least a change in tension applied to the plurality of materials when the plurality of materials are laminated together and processed into the workpiece and a change in tension applied to the workpiece when the workpiece is wound up and processed into the original web as explanatory variables and an actual measured value of the deviation amount in the original web in which the phenomenon occurred as a target variable, and machine learning the correspondence between the processing conditions and the deviation amount in a learning model, the prediction model outputting the deviation amount based on the processing conditions in a processing step in which the plurality of materials are laminated together and processed into the original web;
An acquisition unit that acquires the processing conditions;
an estimation unit that estimates the deviation amount using the prediction model based on the processing conditions acquired by the acquisition unit;
An estimation system comprising: A learning unit for generating the prediction model;
Further equipped with
The estimation unit estimates the deviation amount using the prediction model trained by the learning unit through machine learning.
The estimation system according to claim 1 . a pre-processing unit that performs pre-processing for each lot of the raw web based on the actual measured values of the processing conditions;
Further equipped with
The learning unit uses actual measured values of the machining conditions subjected to the preprocessing as the explanatory variables.
The estimation system according to claim 2 . The pre-processing unit performs the pre-processing to calculate a rate of change of the tension in any interval unit for an actual measured value of the tension applied to the material in the processing step, the actual measured value being included in the processing condition,
The learning unit uses the rate of change as the explanatory variable.
The estimation system according to claim 3 . a learning unit which generates a prediction model by machine learning the correspondence between the processing conditions and the amount of deviation using actual measured values of processing conditions including at least a change in tension applied to the plurality of materials when the plurality of materials are laminated together and processed into the workpiece and a change in tension applied to the workpiece when the workpiece is wound up and processed into the original roll, for the workpiece wound around the original roll in which the phenomenon has occurred, as explanatory variables and an actual measured value of the deviation amount of the center position of the original roll in which the phenomenon has occurred, as a target variable, and
A learning device comprising: An estimation device for estimating a shift amount of a center position of an end face of one of the original rolls, which is a sheet-like workpiece formed by bonding a plurality of materials together and wound into a roll, in a case where the center position of the end face of one of the original rolls shifts toward the outside of the original roll over time,
An acquisition unit that acquires processing conditions in a processing step in which the plurality of materials are laminated together and processed into the original roll;
an estimation unit that estimates the deviation amount using a prediction model based on the processing conditions acquired by the acquisition unit;
Equipped with
The prediction model uses actual measured values of the processing conditions, which include at least a change in tension applied to the plurality of materials when the plurality of materials are laminated together and processed into the workpiece, and a change in tension applied to the workpiece when the workpiece is wound up and processed into the original web, as explanatory variables, and an actual measured value of the deviation amount in the original web in which the phenomenon occurred as a target variable, and is generated by machine learning the correspondence between the processing conditions and the deviation amount into a learning model, and outputs the deviation amount based on the processing conditions.
Estimation device. A method for estimating the amount of deviation of a center position of an end face of a roll of a sheet-like workpiece formed by bonding together a plurality of materials, the center position of one end face of the roll being shifted toward the outside of the roll over time, the method comprising the steps of:
a memory unit uses actual measured values of processing conditions for the workpiece wound around the original roll in which the phenomenon occurred, including at least a change in tension applied to the multiple materials when the multiple materials are laminated together and processed into the workpiece, and a change in tension applied to the workpiece when the workpiece is wound up and processed into the original roll, as explanatory variables, and an actual measured value of the deviation amount in the original roll in which the phenomenon occurred, as a target variable, and a prediction model generated by machine learning the correspondence between the processing conditions and the deviation amount in a learning model, the prediction model outputting the deviation amount based on the processing conditions in a processing step in which the multiple materials are laminated together and processed into the original roll;
An acquisition unit acquires the processing conditions;
an estimation unit estimating the deviation amount using the prediction model based on the machining conditions acquired by the acquisition unit;
An estimation method, comprising: a learning unit using actual measured values of processing conditions including at least a change in tension applied to the plurality of materials when the plurality of materials are laminated together and processed into the workpiece by laminating them, and a change in tension applied to the workpiece when the workpiece is wound up and processed into the original roll, as explanatory variables and an actual measured value of the amount of deviation of the center position of the original roll in which the phenomenon has occurred, as a target variable, and generating a prediction model by having a learning model learn a correspondence relationship between the processing conditions and the amount of deviation, for a phenomenon in which a center position of one end face of the original roll is shifted outward over time in a roll of the original roll in which the phenomenon has occurred, for the workpiece wound around the original roll in which the phenomenon has occurred, as explanatory variables and an actual measured value of the amount of deviation of the center position of the original roll in which the phenomenon has occurred, as a target variable;
Including, how to learn. A method for estimating the amount of deviation of a center position of an end face of a roll of a sheet-like workpiece formed by bonding together a plurality of materials, the center position of one end face of the roll being shifted toward the outside of the roll over time, the method comprising the steps of:
An acquisition unit acquires processing conditions in a processing step in which the plurality of materials are laminated together and processed into the original roll;
an estimation unit estimating the deviation amount using a prediction model based on the processing conditions acquired by the acquisition unit;
Including,
The prediction model uses actual measured values of the processing conditions, which include at least a change in tension applied to the plurality of materials when the plurality of materials are laminated together and processed into the workpiece, and a change in tension applied to the workpiece when the workpiece is wound up and processed into the original web, as explanatory variables, and an actual measured value of the deviation amount in the original web in which the phenomenon occurred as a target variable, and is generated by machine learning the correspondence between the processing conditions and the deviation amount into a learning model, and outputs the deviation amount based on the processing conditions.
Estimation method. Computer,
a learning unit which generates a prediction model by machine learning the correspondence between the processing conditions and the amount of deviation using actual measured values of processing conditions including at least a change in tension applied to the plurality of materials when the plurality of materials are laminated together and processed into the workpiece and a change in tension applied to the workpiece when the workpiece is wound up and processed into the original roll, for the workpiece wound around the original roll in which the phenomenon has occurred, as explanatory variables and an actual measured value of the deviation amount of the center position of the original roll in which the phenomenon has occurred, as a target variable, and
A program that functions as a A program for estimating the amount of deviation of a center position of an end face of one of the rolls of a sheet-like workpiece formed by bonding a plurality of materials together and wound into a roll, in which the center position of the end face of one of the rolls shifts toward the outside of the roll over time, the program comprising:
Computer,
An acquisition unit that acquires processing conditions in a processing step in which the plurality of materials are laminated together and processed into the original roll;
an estimation unit that estimates the deviation amount using a prediction model based on the processing conditions acquired by the acquisition unit;
Function as a
The prediction model uses actual measured values of the processing conditions, which include at least a change in tension applied to the plurality of materials when the plurality of materials are laminated together and processed into the workpiece, and a change in tension applied to the workpiece when the workpiece is wound up and processed into the original web, as explanatory variables, and an actual measured value of the deviation amount in the original web in which the phenomenon occurred as a target variable, and is generated by machine learning the correspondence between the processing conditions and the deviation amount into a learning model, and outputs the deviation amount based on the processing conditions.
program.

Images