Injection Molding Operation Control Using State Expression Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing injection molding apparatus systems struggle with determining optimal operation quantities that are versatile and adaptable to environmental changes, often requiring extensive relearning and large amounts of training data, and fail to diagnose optimal operation quantities effectively.
Innovation Solution
An operation quantity determination device that includes an observation unit, a state expression unit, and an operation quantity output unit, which acquires observation data, generates a state expression map, and outputs operation quantities based on this map, enabling robust adaptation to new environments through model-based reinforcement learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If model-free reinforcement learning is used to determine operation quantity, then the system can learn optimal policies, but when the environment changes, the policy becomes inapplicable and requires complete relearning from scratch
Solution Approach 1:
The patent pre-defines multiple operation quantity determination models corresponding to different environment types before actual operation. When the environment changes, the system can directly select and switch to the appropriate pre-trained model without complete relearning, thus maintaining policy applicability while adapting to new environments.
Solution Approach 2:
The patent changes the parameter of learning approach from model-free to model-based reinforcement learning. By introducing environment models and pre-defining multiple determination models with different parameters, the system achieves both reliability and adaptability - the policy remains applicable through model guidance while adapting through model selection and partial relearning.
2Adaptability or versatility
If complete relearning is performed when environment changes, then the policy can adapt to new conditions, but a huge amount of training data and training work is required
Solution Approach 1:
The patent pre-trains multiple operation quantity determination models for different environment types in advance. When environmental changes occur, the system can quickly switch between pre-trained models or perform only partial relearning on the selected model, dramatically reducing the time and training data required compared to complete relearning from scratch.
Solution Approach 2:
The patent implements a dynamic model selection mechanism that can switch between different pre-trained models based on environmental conditions. This dynamic approach allows the system to adapt to new environments quickly by selecting the most appropriate pre-trained model, avoiding the need for extensive relearning and reducing both time and computational resource requirements.
3Adaptability or versatility
If model-based reinforcement learning with pre-defined models is used, then robustness against environmental changes is improved, but the device complexity increases
Solution Approach 1:
The patent creates multiple operation quantity determination models that can handle different environment types, making the system universally applicable across various conditions. By designing models with common structures that can be selected and switched based on environment type, the patent achieves environmental robustness while managing complexity through standardized, multi-functional model architecture.
Data Source
AI summary
An operation quantity determination device for determining an operation quantity related to a molding machine, includes an observation unit configured to acquire observation data obtained by observing a physical quantity related to molding by the molding machine when the molding is executed, a state expression unit configured to generate a state expression map expressing a state of the molding machine based on the observation data acquired by the observation unit, and an operation quantity output unit configured to output the operation quantity based on the state expression map generated by the state expression unit.