Polymeric Materials

THE OBJECTIVES

Find the best experimental condition to produce a desired material.

THE CHALLENGES

• Too many combination patterns of raw materials and blending ratio to experiment.
• Too much time to evaluate all material conditions.

THE APPROACH

• Include process information such as mixing duration, heating time and temperature etc. to the input data for creating learning model.
• Improve model accuracy using our image analysis method, which can define ROI(Region of Interest), extract the ROI area ratio, and evaluate the quantified material properties.
  
• Suggest the best experimental conditions for desired materials properties derived from learning model.

THE OBJECTIVES

Develop novel high-performance compounds with less experiments.

THE CHALLENGES

• Too many candidates of compounds to experiment.
• Difficult to narrow down candidates.
• Too small amount of experimental data.

THE APPROACH

• Use Hitachi original “Nested AI”^*1which can handle large scale of open data prepared along with customer’s experimental data.
• Convert chemical structures to numerical data to extract factors effecting on performance.

*1 Nested AI in which AI trained by actual experimental data is embedded inside AI trained by large-scale open data.

• Nested AI empowers Materials Informatics approach even with only a small amount of experimental data.
• The outer AI converted the chemical structure into numerical data.
• The inner AI extract and tune the numerical factors effecting on material performance.
• Convert back the candidates for experiment to chemical structures.