Adaptive machine learning system for optimizing injection molding

Overview

The proposed solution is a cutting-edge adaptive machine learning system designed to optimize and predict cycle times in injection molding processes. Traditional machine learning models struggle with distribution shifts when applied to new environments, leading to inaccurate predictions. Our system overcomes this by integrating an optimal experiment design component with an adaptive time-series prediction model. This approach ensures accurate cycle time predictions by learning from real-time data collected from specific machines in their operational settings.

Technical specifications

Key features:

• Optimal Experiment Design: Utilizes block design theory to systematically vary machine settings, enabling efficient and reliable data collection.
• Adaptive Time-Series Prediction: Employs distribution-free uncertainty quantification techniques to predict cycle times, adapting to new data environments and reducing errors associated with distribution shifts.
• Scalability: Designed to function across various machine settings, manufacturers, and materials, maintaining accuracy and efficiency.

Technology readiness level

The technology is currently at TRL 4, indicating that the core components have been developed and validated in a laboratory environment. Further implementation and testing in real-world settings are planned to advance the readiness level.