SUCCESS STORIES USING MATERIALS INFORMATICS
MI (Materials Informatics) is attracting a lot of attention not only because it can make development more efficient, but also because it is related to new and innovative developments.
This section presents examples of the challenges companies are facing and how they have solved and achieved results.
Table of Contents
FORWARD AND INVERSE ANALYSIS EXAMPLES
The analysis usually relied on the instincts of researchers and past experiences.
However, the combination of material selection and blending ratio and the process conditions are too vast to easily predict.
The problem was that the development time would become enormous because of repeating experiments, guessing based on the developer's intuition and experience.
"If possible, we want to speed up development and reduce the number of experiments." they thought.
SOLUTION WITH MI
A machine learning model is constructed using existing experimental data as teacher data.
This model can apply for forward and inverse analyses.
In the forward analysis, we were able to reduce the number of experiments by obtaining the expected physical properties from complex formulations, and by conducting reliable experiments without performing less accurate ones.
Depending on the current situation of the customer, some cases were reducing the number of experiments to about 1/4.
In addition, by performing an inverse analysis, determining the blending ratio of raw materials based on the desired physical properties, and creating a candidate experimental plan, it became possible to search for previously unforeseen and unknown materials.
TEXT MINING CASES
We had a customer who wanted to add publicly known patent and literature data to experimental data.
However, it took time and people to check all the large amounts of patents and literature, and that made it difficult to get started.
You can extract the information you want to get from patents and papers using text mining.
With a little teacher data, we create a word extraction model for text mining.
Using that model, we were able to extract data from many patents and papers and add them to experimental data.
IMAGE ANALYSIS EXAMPLES
Since only a limited number of technicians check an electron microscope and judge whether an image is good or bad, they felt it as a challenge that taking an enormous amount of time to judge many images.
Feature quantities are extracted from teacher images to build a learning model for image discrimination.
Using this model, automatic image discrimination has become possible, making it possible to quickly sort and judge a lot of electron microscope images.
MI OF HITACHI HIGH-TECH
Hitachi High-Tech has several services to help development using MI work more efficiently.
We can provide both analytical support services and analytical environment provision services.
Please see the following pages for details.
Additional MI information is available below.