Formulation Development with Artificial Intelligence

Compound development processes require significant time and numerous iterations to find the optimal formulation. 

Uncountable enables R&D labs to cut development time and reduce testing iterations

Artificial intelligence algorithms suggest the best recipes to run, avoiding the tedious, manual tweaking of individual ingredients. Complicated development processes that used to involve tens or hundreds of experiments can now be conducted in half the time utilizing advanced machine learning models.

 
 

HOW IT WORKS

Screen Shot 2018-07-23 at 10.54.56 PM.png
 

STEP 1: INGEST PREVIOUS TEST DATA

Uncountable's software incorporates multiple data sources to understand every piece of the development process, including property objectives, testing requirements, list of available ingredients, past formulations and test history. 

Past experimental data can be ingested via several methods, including via database dumps, disparate spreadsheets, or other record systems. 

 
Screen Shot 2018-07-23 at 10.54.27 PM.png
 

STEP 2: MODEL MATERIAL SPACE

Uncountable's machine learning algorithms build out a model of the material property performance "space." The use of artificial intelligence models allows the system to understand the effect that altering each subset of ingredients will have on the performance of the formulation.

Models from traditional DOE software that typically rely on linear regressions make oversimplifications and fail to capture the complexity of the development process.

 
Screen Shot 2018-07-23 at 10.55.25 PM.png
 

 

STEP 3: SUGGEST NEW RECIPES

Artificial intelligence software delivers multiple new compound formulations to the user for further testing. These suggestions are the result of a high dimensional optimization analysis, yielding optimized testing sequences and reduced trial cycles en route to the ideal material recipe.  

This obviates the need for factorial testing where every combination of values is tried. Uncountable's approach is adaptive, enabling significantly more learning per round of experiments while also moving closer to the project property goals.

 

The machine learning engine is continuously improved by feeding further trial information into its base of test history, meaning a research team gets better and better at finding breakthrough materials. 

 
 

 

Uncountable vs Traditional DoE's

 
Screen Shot 2018-07-27 at 6.27.19 PM.png
 

 

Want to learn more?

Name *
Name