Materials informatics and machine learning for R&D
Materials informatics uses data and machine learning to predict how a material's composition and processing affect its properties, so R&D teams test fewer candidates to reach a target. Abdul Rahman builds materials informatics workflows with Python, PyTorch, Pymatgen, MatGL and PyCalphad, including Bayesian optimization loops and CALPHAD thermodynamic models.
What's included
- Review of your existing experimental data and what it can predict
- Property-prediction models (for example random forests and graph neural networks)
- Bayesian optimization to choose the next experiments
- CALPHAD phase-diagram and thermodynamic calculations with PyCalphad
- Hand-over of code and documentation your team can keep running
Who it's for
- Alloy and coating development teams
- Labs with years of untapped experimental data
- Startups screening compositions before building a lab
- Teams exploring high-entropy alloys and new chemistries
How it works
- Discovery call: A free 30-minute video call to understand the part, the material and what you need to decide.
- Proposal: A short written scope with the analyses, the data or samples needed, the timeline and the price.
- Analysis: Data and images are shared securely online; physical testing is arranged with a lab near you when needed.
- Report and debrief: A written report with findings and recommendations, followed by a call to walk your team through it.
Questions about materials informatics & ml
What is materials informatics?
Materials informatics applies data science and machine learning to materials science. It uses data from experiments, simulations and databases to predict how composition and processing affect properties, which reduces the number of experiments needed to develop new materials.
How much data do I need to start?
Less than many teams expect. Bayesian optimization is designed for small, expensive datasets and can guide experiments starting from a few dozen data points, especially when combined with physics-based features and CALPHAD calculations.
What is CALPHAD?
CALPHAD (CALculation of PHAse Diagrams) is a method for computing phase diagrams and thermodynamic properties of multi-component materials from assessed databases. It predicts which phases form at a given composition and temperature.