Piezoelectric modulus prediction using machine learning and graph neural networks

Abstract

Piezoelectric materials are widely used in all kinds of industries such as electric cigarette lighters, diesel engines and x-ray shutters. However, discovering high-performance and environmentally friendly (e.g. lead-free) piezoelectric materials is a difficult problem due to the sophisticated relationships from materials' composition/structures to the piezoelectric effect. Compared to other material properties such as formation energy, band gap, and bulk modulus, it is much more challenging to predict piezoelectric coefficients. Here, we propose a comprehensive study on designing and evaluating advanced machine learning models for predicting the piezoelectric modulus from materials' composition and/or structures. We train the prediction models based on extensive feature engineering combined with machine learning models (Random Forest and Support Vector Machines) and automated feature learning based on deep graph neural networks. Our SVM model with crystal structure feature outperform other methods. We also use this model to predict the piezoelectric coefficients for 12,680 materials from the Materials Project database and report the top 20 potential high performance piezoelectric materials.

13 pages

References in corpus (6)

• A General-Purpose Machine Learning Framework for Predicting Properties of Inorganic Materials
• SchNet: A continuous-filter convolutional neural network for modeling quantum interactions
• A multi-material level set-based topology optimization of flexoelectric composites
• Global Attention based Graph Convolutional Neural Networks for Improved Materials Property Prediction
• Towards Deeper Graph Neural Networks with Differentiable Group Normalization
• Predicting Elastic Properties of Materials from Electronic Charge Density Using 3D Deep Convolutional Neural Networks