Physics-based and Machine Learning-based Approaches to Identifying new Quaternary Oxide Alloys for Piezoelectric SONAR
Abstract
This project proposes investigations into the physical mechanisms of piezoelectric SONAR materials, in order to provide guidance toward the discovery of new materials with superior SONAR detection and broadcast properties. Piezoelectric materials couple polarization to strain, so that an oscillating electric field which induces changes in material polarization drives a mechanical response. Adding additional end-members (as in PIN-PMN-PT, which is an alloy of Pb(In½Nb½)O3-Pb(Mg1/3Nb2/3)O3-PbTiO3) or doping with other elements can provide advantages such as enhanced thermal stability or further reduction of barriers to polarization rotation. However, these additional components can have disadvantages such as increasing the dielectric loss. Optimizing material composition is therefore a difficult balancing act, which can be guided by theoretical insight and simulation. Six major research directions are envisioned: 1. Use DFT and bond-valence molecular dynamics (BVMD) to directly predict piezoelectric performance and dielectric loss at finite temperature and under finite electric field and mechanical stress conditions. This will enable unprecedented connections between dopants and functional properties under actual use conditions. 2. Directly correlate compositions and ion orders and domains with piezoelectric performance. This will provide guidance for processing conditions and the impact of microstructure. 3. Local analysis can reveal changes from global behavior: active role of domains, influence of cation inhomogeneity. 4. Machine learning algorithms can be a vital companion to physics-based investigations to accelerate progress toward the understanding of domain dynamics and guide the selection of new compositions with superior properties. 5. Use kinetic Monte Carlo (KMC) to investigate the composition variation in the growth of PIN-PMNPT single crystal and new compositions. 6. Identification of the soft vibrational modes can unveil the origin of piezoelectric performance. The mapping of local soft modes and the externalcondition- dependent changes in local soft modes will analyzed with machine learning to guide development of new SONAR piezoelectrics. A complete and rational atomistic picture is still missing regarding the main controlling factors for piezoelectric response and electromechanical coupling. The research directions in this concept paper are designed to develop new fundamental insights via a combination of first principles, multi-scale, and machine learning methods, focused on new compositions and processing conditions that will empower a better SONAR device. We propose compositional searches to add one or two material compositions to PIN-PMN-PT. The complexity of these phase diagrams and the complexity of the multi-material response suggest that material compositions with superior properties can be found but have not been found. Taken together, these approaches will link complex multi-component (ternary and quaternary) perovskite oxide compositions with functional properties directly. Complex compositions will be studied with DFT. These calculations will calibrate BMVD, enabling finite-temperature evaluation of piezoelectric SONAR responses. Physical analysis and learning algorithms will accelerate the selection of new compositions with improved properties.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 20, 2020
- Source ID
- N000142012701
Entities
People
- Andrew M Rappe
Organizations
- Office of Naval Research
- United States Navy
- University of Pennsylvania