Reinforcement Learning for Resin Kneading Condition Selection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Determining appropriate processing conditions for resin kneading devices is challenging due to reliance on years of experience, making it difficult to achieve desired pellet states without extensive expertise.

Innovation Solution

A machine learning method that acquires state variables related to resin and processing conditions, calculates rewards, updates functions, and determines optimal conditions using reinforcement learning to automate the process, considering parameters like resin properties, feeder settings, and equipment operations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If conventional experience-based methods are used to determine kneading conditions, then the processing conditions can be determined using existing knowledge, but it becomes difficult to easily determine appropriate processing conditions for required pellet states

Engineering Contradiction:

Improveease of determining processing conditionsVSAvoidprecision of pellet state

Core Design Contradiction:

Ease of manufactureVSManufacturing precision

Solution Approach 1:

The patent replaces the mechanical system of experience-based determination with an information processing system. A machine learning model processes relationships between processing conditions and pellet states, substituting human expert knowledge with automated computational analysis. This allows the system to determine appropriate processing conditions through data-driven predictions rather than relying on accumulated human experience.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent introduces a machine learning model as an intermediary between processing conditions and pellet states. This intermediary learns the complex relationships between various parameters (screw rotation speed, barrel temperature, feed rate) and pellet quality metrics, enabling accurate prediction and optimization without direct human intervention in the determination process.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Ease of operation

If machine learning methods are introduced to determine processing conditions, then ease of determination is improved, but system complexity increases

Engineering Contradiction:

Improveease of determining processing conditionsVSAvoidcomplexity of determination system

Core Design Contradiction:

Ease of operationVSDevice complexity

Solution Approach 1:

The patent creates a universal machine learning model that can handle multiple processing conditions and pellet state predictions simultaneously. The single model encompasses relationships between various parameters (temperature, speed, feed rate) and multiple quality metrics, reducing the need for separate specialized systems for each determination task.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent performs preliminary training of the machine learning model using historical data before actual operation. By pre-learning the relationships between processing conditions and pellet states from accumulated data, the system prepares the model in advance, so that during actual operation, determining optimal conditions becomes a simple prediction task rather than requiring complex real-time analysis.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentEP3988270A1Machine learning method, machine learning device, and machine learning program

Publication Date: 2022.04.27 KOBE STEEL LTD

Figure 1
Figure 2
Figure 3

AI summary

A machine learning method includes acquiring a state variable including a physical quantity related to a pellet state of a resin and a processing condition; calculating a reward for a determination result of the processing condition based on the state variable; updating a function for determining a processing condition from a state variable based on the reward; and repeating updating of the function, thereby determining a processing condition under which the reward can be most obtained.