Winding condition generating apparatus, winding apparatus, winding defect level prediction value generating apparatus, winding condition calculating method, winding method, and winding defect level prediction value generating method

Patent metadata

Field	Value
Publication number	US-11960992-B2
Application number	US-202017025069-A
Country	US
Kind code	B2
Filing date	Sep 18, 2020
Priority date	Mar 29, 2018
Publication date	Apr 16, 2024
Grant date	Apr 16, 2024

Abstract

A winding condition generating apparatus includes: an input unit; an output unit; and a condition calculation unit. A winding condition calculation unit includes a learning model created by machine learning using a combination of a winding parameter and a winding condition in producing a wound web that satisfies a target winding quality as training data, and calculates a winding condition of a new wound web using the learning model, from a winding parameter of a new wound web input through the input unit. The output unit outputs the winding condition. The winding parameter includes a web width, a web transport velocity, and a web winding length. The winding condition includes a tension of the web at the start of winding and a tension of the web at the end of winding.

First claim

What is claimed is:

1. A winding condition generating apparatus comprising at least one processor configured to:
calculate a winding condition of a new wound web from a winding parameter of the new wound web, using a learning model created by machine learning with a combination of a winding parameter and a winding condition in producing a wound web that satisfies a target winding quality as training data; and
output the winding condition,
wherein the winding parameter includes a web width, a web transport velocity, and a web winding length,
wherein the winding condition includes a tension of the web at the start of winding and a tension of the web at the end of winding, and
wherein in a case where with respect to a set of the winding conditions obtained in producing the web, which is the training data, a value of each winding condition item is denoted by C ni , an allowable quality range value set to the value C ni of the winding condition item is denoted by T ni and the number of items to which the allowable quality range value T ni is set is denoted by N, a range obtained by the following Equation for each item is added to the winding conditions, and 3 N −1 pieces of the training data or a part of the training data is assigned as additional training data,
C k =C ni ±0.5× T ni .

2. The winding condition generating apparatus according to claim 1 ,
wherein the winding parameter includes at least one of a diameter of a winding core around which the web is wound, a name of a line in which the wound web is produced, a thickness of the web, a difference between a maximum thickness and a minimum thickness in a web width direction, or a modulus of elasticity of the web, and
wherein the winding condition includes at least one of a diameter of the web at the end of winding, a knurling height, a pressure of an air press that presses the web, or a pressing force of a touch roller that presses the web.

3. The winding condition generating apparatus according to claim 1 ,
wherein the target winding quality is non-occurrence of a web winding misalignment defect and a web damage defect.

4. The winding condition generating apparatus according to claim 1 ,
wherein the machine learning includes a neural network and deep learning.

5. The winding condition generating apparatus according to claim 1 ,
wherein the winding condition includes any one of a tension function expressed with respect to a radial coordinate of a winding roll, an air press pressure function for pressing the web, or a function of a pressing force of a touch roller.

6. The winding condition generating apparatus according to claim 1 ,
wherein the at least one processor is configured to cause the learning model to perform the machine learning using the combination of the winding parameter and the winding condition in producing the wound web that satisfies the target winding quality, as the training data.

7. A winding apparatus that winds a web using the winding condition calculated by the winding condition generating apparatus according to claim 1 .

8. A winding defect level prediction value generating apparatus comprising at least one processor configured to:
calculate a winding defect level prediction value from a winding parameter and a winding condition of a new wound web, using a learning model created by machine learning with a combination of a winding parameter, a winding condition and a winding defect level value in producing a wound web as training data; and
output the winding defect level prediction value,
wherein the winding parameter includes at least one of a web width, a web transport velocity, a web winding length, a diameter of a winding core around which the web is wound, a name of a line in which the wound web is produced, a web thickness, a difference between a maximum thickness and a minimum thickness in a web width direction, or a modulus of elasticity of the web,
wherein the winding condition includes at least one of a tension of the web at the start of winding, a tension of the web at the end of winding, a knurling height, a pressure of an air press for pressing the web, or a pressing force of a touch roller for pressing the web,
wherein the winding defect level prediction value includes a web winding misalignment value and a web damage defect level, and
wherein in a case where with respect to a set of the winding conditions obtained in producing the web, which is the training data, a value of each winding condition item is denoted by C ni , an allowable quality range value set to the value C ni of the winding condition item is denoted by Tin and the number of items to which the allowable quality range value T ni is set is denoted by N, a range obtained by the following Equation for each item is added to the winding conditions, and 3 N −1 pieces of the training data or a part of the training data is assigned as additional training data,
C k =C ni ±0.5 ×T ni .

9. A winding condition generating apparatus comprising:
a defect level calculation model that is a learning model in the winding defect level prediction value generating apparatus according to claim 8 , receives an input of a winding condition, and outputs a web winding misalignment value and a web damage defect level; and
at least one processor configured to:
change using each sum of the web winding misalignment values and the web damage defect levels that are the output of the defect level calculation model as an objective function, and using the winding condition as a design variable, the design variable through evolutionary computation until the objective function becomes minimum; and
output a winding condition that is a design variable in a case where the objective function becomes minimum, as the winding condition.

10. A winding condition calculating method at least comprising:
a step of creating a learning model by performing machine learning using a combination of a winding parameter and a winding condition in producing a wound web that satisfies a target winding quality as training data;
a step of inputting a winding parameter of a new wound web; and
a step of calculating a winding condition of the new wound web using the learning model from the winding parameter,
wherein the winding parameter includes a web width, a web transport velocity, and a web winding length, and
wherein the winding condition includes a tension of the web at the start of winding and a tension of the web at the end of winding;
further comprising:
a step of adding, in a case where with respect to a set of the winding conditions obtained in producing the web, which is the training data, a value of each winding condition item is denoted by C ni , an allowable quality range value set to the value C ni of the winding condition item is denoted by T ni and the number of items to which the allowable quality range value T ni is set is denoted by N, a range obtained by the following Equation for each item, to the winding conditions, and assigning 3 N −1 pieces of the training data or a part of the training data, as additional training data,
C k =C ni ±0.5× T ni .

11. The winding condition calculating method according to claim 10 ,
wherein the winding parameter includes at least one of a diameter of a winding core around which the web is wound, a name of a line in which the wound web is produced, a thickness of the web, a difference between a maximum thickness and a minimum thickness in a web width direction, or a modulus of elasticity of the web, and
wherein the winding condition includes at least one of a diameter of the web at the end of winding, a knurling height, a pressure of an air press that presses the web

Assignees

• Fujifilm Corp

Inventors

• Matsumura Yoshihito

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• B65H18/26

  Mechanisms for controlling contact pressure on winding-web package, e.g. for regulating the quantity of air between web layers · CPC title

• B65H23/005

  Sensing web roll diameter (warning or safety devices responsive to a predetermined diameter B65H26/08) · CPC title