Advanced machine learning to accelerate materials research
Case studies in inorganic solids, polymers and more.
- Tremendous demands for development of new materials with improved performance and greener chemistry
- Pressure to bring products to market quickly and cost-efficiently
- Rapid need for informatics-based predictive models to extend the scale of materials optimization and discovery
- Limited availability of large datasets of materials properties to effectively train models
- Amplify physics-based modeling with advanced machine learning technology and deep learning capabilities
- Leverage machine learning to extend predictions over extensive chemical space unavailable to experimental methods
- Develop customized descriptors for highly diverse material systems to improve the accuracy of predictions
- Collaborate, manage disparate data, and share predictive models across project teams with a unified cloud-based enterprise informatics platform
With the rapid development of technology in the fields of energy, aerospace, electronics, foods and more, there are increasing demands for the design and discovery of new materials. To meet these demands, materials scientists seek innovative methods to optimize chemical properties and reduce time-to-market of better performing, more sustainable products. However, relying solely on traditional trialand- error experimental methods has become too costly and too slow, and is limited by experimental conditions. Thus the materials innovation R&D cycle has long benefited from the use of physics-based simulation engines such as quantum mechanics and molecular dynamics to help lower the cost of discovering novel chemistries, structures, morphologies, and compositions of materials for a wide array of applications and industries. In the past few years, the growth of computational power and the interest in building large datasets of materials properties has led to the growing adoption of materials informatics and machine learning-powered approaches in materials science. However, such methods are highly data intensive and suffer from an inability to extrapolate beyond the chemical space of the training model. There is a pressing need for rapid, informatics-based predictive models to extend the scale of materials optimization and discovery, extract property limits and design rules, and drive the natural synergy between physics-based modeling and machine learning methods. Additionally, with the rapid advance of machine learning methods, there is an increasing need to deploy and share accurate predictive models across research organizations for use by experts and non-experts alike. Successful R&D digitization efforts require a robust, collaborative platform for assessing predictive models for use in research.