Machine-learning guided fast optimization of materials synthesis with high quality

A fast-convergence machine learning approach for materials synthesis optimization, with minimal training data

Background: Materials synthesis is a complicated and lengthy process requiring a high degree of precision. At the same time, there is a constant need to develop new materials, adapt existing ones, and improve the quality of materials synthesis. Traditional approaches to materials synthesis optimization rely on trial-and-error methods that can be time-consuming and resource-intensive. The complex relationships between synthesis parameters and final material properties make it difficult to predict optimal synthesis conditions. Current optimization methods often require extensive experimental datasets and may not converge quickly to optimal solutions.

Technical Overview: Northeastern researchers have developed a machine-learning guided fast optimization technology that represents a revolutionary approach for materials synthesis. A new framework built from machine learning algorithms can rapidly identify optimal synthesis conditions with minimal training data. The system employs advanced optimization techniques that can quickly converge to high-quality solutions while reducing the number of experimental iterations required. The approach combines predictive modeling with active learning strategies to efficiently explore the synthesis parameter space and identify optimal conditions.

Benefits:

  • Accelerates materials synthesis process significantly, cutting up to 80% of time
  • Reduces the need for extensive experimental datasets
  • Provides fast convergence to optimal synthesis conditions
  • Maintains high material quality while reducing development time
  • Enables efficient exploration of synthesis parameter space

Application:

  • Advanced manufacturing industries, where high-quality materials are paramount
  • Development of new materials for emerging technologies
  • Optimization of existing materials synthesis processes
  • Research and development in materials science

Opportunity:

  • License
  • Research collaboration
Patent Information:
Title App Type Country Serial No. Patent No. File Date Issued Date Expire Date
FABRICATION PROCESS DESIGN USING BAYESIAN OPTIMIZATION WITH ACTIVE CONSTRAINT LEARNING Utility *United States of America 18/776,170   7/17/2024