Machine learning method, machine learning device, machine learning program, communication method, and kneading device

Patent metadata

Field	Value
Publication number	US-12420452-B2
Application number	US-202117996397-A
Country	US
Kind code	B2
Filing date	May 14, 2021
Priority date	May 29, 2020
Publication date	Sep 23, 2025
Grant date	Sep 23, 2025

Abstract

A machine learning method includes: acquiring a state variable including at least one first evaluation parameter related to performance evaluation of a kneaded product and at least one kneading condition; calculating a reward for a decision result of the at least one kneading condition based on the state variable; updating a function for deciding the at least one kneading condition from the state variable based on the reward; and by repeating the update of the function, deciding a kneading condition under which the reward obtained becomes maximum, in which the at least one first evaluation parameter includes at least one of physical properties and shape characteristics related to the kneaded product.

First claim

The invention claimed is:

1. A machine learning method for a machine learning device to decide a kneading condition of a kneading device for kneading a polymer material to obtain a kneaded product, the kneading device including:
a chamber to which a material to be kneaded is input;
two or more rotors which knead the material input to the chamber; and
a controller in charge of control of the two or more rotors, control of a kneading time of the material, and control of an operation step of the kneading device, the machine learning method comprising:
acquiring a state variable including at least one first evaluation parameter related to performance evaluation of the kneaded product and at least one kneading condition;
calculating a reward for a decision result of the at least one kneading condition based on the state variable;
updating a function for deciding the at least one kneading condition from the state variable based on the reward; and
by repeating the update of the function, deciding a kneading condition under which the reward obtained becomes maximum,
wherein the at least one first evaluation parameter includes at least one of physical properties and shape characteristics related to the kneaded product.

2. The machine learning method according to claim 1 , wherein the at least one kneading condition is at least one of a first parameter related to the material and a second parameter related to control of the two or more rotors.

3. The machine learning method according to claim 1 , wherein
the at least one kneading condition is at least one of:
a first parameter related to the material;
a second parameter related to control of the two or more rotors; and
a third parameter related to an operation step,
the third parameter being selected from at least one of:
a kneading time in at least one of a plurality of steps;
a material input time in at least one of the plurality of steps;
at least one condition out of conditions for proceeding to a next step;
a total kneading time; and
accumulated electric power.

4. The machine learning method according to claim 1 , wherein
the kneading device further includes a weight, and
the at least one kneading condition includes a fourth parameter related to operation of the weight.

5. The machine learning method according to claim 1 , wherein
the kneading device further includes a temperature adjustment mechanism, and
the at least one kneading condition includes a fifth parameter related to temperature adjustment.

6. The machine learning method according to claim 2 , wherein the first parameter includes at least one of a mixing amount of the material, an order for inputting a component of the material, and a filling factor of the chamber with the material.

7. The machine learning method according to claim 2 , wherein the second parameter includes at least one of the number of rotations of the two or more rotors, phases of the two or more rotors, and a speed ratio of each of the rotors.

8. The machine learning method according to claim 3 , wherein in the third parameter, the condition for proceeding to a next step includes at least one of a kneading time for proceeding to a next step, a temperature of the material, a temperature of the material to be maintained in each step, instant electric power of a motor which drives the two or more rotors, accumulated electric power of the motor, an instant electric current of the motor, a torque of the motor, and a temperature of the material at discharging.

9. The machine learning method according to claim 4 , wherein the fourth parameter includes at least one of a pressing pressure of the weight at the time of pushing the material into the chamber, a position of the weight, and a speed of the weight.

10. The machine learning method according to claim 5 , wherein the fifth parameter includes at least one of a temperature of a circulating medium coming into the chamber, a temperature of a circulating medium going out from the chamber, a temperature of a circulating medium coming into the two or more rotors, a temperature of a circulating medium going out from the two or more rotors, a temperature of a circulating medium coming into a door from which the material is to be discharged, and a temperature of a circulating medium going out from the door.

11. The machine learning method according to claim 1 , wherein the state variable further includes a second evaluation parameter related to operation stability of the kneading device.

12. The machine learning method according to claim 1 , wherein the physical properties include at least one of Mooney viscosity, vulcanization properties, Payne effect, dispersion of an additive, dynamic viscoelasticity, hardness, a weight of the kneaded product, and a temperature of the kneaded product.

13. The machine learning method according to claim 1 , wherein the function is updated in real time using deep reinforcement learning.

14. The machine learning method according to claim 1 , wherein
in calculation of the reward, the reward is increased in a case where the at least one first evaluation parameter approaches a predetermined reference value corresponding to each first evaluation parameter, and the reward is decreased in a case where the at least one first evaluation parameter does not approach the reference value corresponding to each first evaluation parameter.

15. A machine learning device which decides a kneading condition of a kneading device for kneading a polymer material to obtain a kneaded product, the kneading device including:
a chamber to which a material for obtaining the kneaded product is input;
two or more rotors which knead the material input to the chamber; and
a controller in charge of control of the two or more rotors, control of a kneading time of the material, and control of an operation step of the kneading device,
the machine learning device comprising:
a state acquisition unit which acquires a state variable including at least one first evaluation parameter related to performance evaluation of the kneaded product and at least one kneading condition;
a reward calculation unit which calculates a reward for a decision result of the at least one kneading condition based on the state variable;
an update unit which updates a function for deciding the at least one kneading condition from the state variable based on the reward; and
a decision unit which, by repeating the update of the function, decides a kneading condition under which the reward obtained becomes maximum,
wherein the at least one first evaluation parameter includes at least one of physical properties and shape characteristics related to the kneaded product.

16. A non-transitory computer-readable recording medium which records a machine learning program of a machine learning device which decides a kneading condition of a kneading device for kneading a polymer material to obtain a kneaded product, the kneading device including:
a chamber to which a material for obtaining the kneaded product is input;
two or more rotors which knead the material input to the chamber; and
a controller in charge of control of the two or more rotors, control of a kneading time of the material, and control of an operation step of the kneading device, the machine learning program causing a computer to function as,
the machine learning device comprising:
a state acquisition unit which acquires a state variable including at least one first evaluation parameter related to performance evaluation of the kneaded product and at least one kneading condition;
a reward calculation unit which ca

Assignees

• Kobe Steel Ltd

Inventors

• Hamada Hikaru
• Yamane Yasuaki
• Yamada Norifumi
• Miyasaka Kazuo

Classifications

• B29B7/603

  in measured doses, e.g. proportioning of several materials · CPC title

• B29B7/242

  in measured doses · CPC title

• B29B7/286

  measuring properties of the mixture, e.g. temperature, density (B29B7/283 takes precedence) · CPC title

• B29B7/283

  measuring data of the driving system, e.g. torque, speed, power · CPC title

• B29B7/28

  for measuring, controlling or regulating, e.g. viscosity control {(B29B7/242 takes precedence)} · CPC title