1 NASA Glenn Research Center, Cleveland, United States; 2 University of Pittsburgh, Pittsburgh, United States; 3 National Energy Technology Laboratory, Pittsburgh, United States
Environmental barrier coatings (EBCs) are an enabling technology for the transition from superalloys to silicon carbide (SiC) ceramic matrix composites (CMCs) in gas turbine aeroengines for increased fuel efficiency. SiC-based CMCs are prone to oxidation-based degradation in the engine hot section, and rare-earth (RE) silicates are promising candidates for EBCs due to their close thermal expansion match to the composite substrate and low water vapor volatility. However, the design of EBCs is hindered by the large chemical space of candidate materials and the difficulty in obtaining material properties for engineering optimization, especially as research continues into mixed-cation or “high-entropy” RE silicates. First-principles computational methods such as density functional theory (DFT) are highly effective at calculating material properties to guide coating design but are limited by their computational cost. Machine learning (ML) is a promising technique to accelerate material property predictions compared to DFT, via either direct prediction or development of interatomic potentials (IAPs) for atomistic simulations. Machine learning models such as neural networks (NNs) can be trained to predict material properties directly from an input feature set, but they are generally limited to predicting properties they were trained on. Alternatively, IAPs can be utilized in atomistic simulations using molecular dynamics or Monte Carlo methods to predict a wider variety of properties at lower computational cost than DFT.
In this work, we demonstrate two ML approaches to accelerate the calculation of RE silicate properties relevant to EBC design. First, we present a NN model to directly predict constant pressure heat capacity, Cp, of RE silicates and oxides directly from easily obtainable unit cell parameters. The NN model performs predictions orders of magnitude faster than DFT calculations which can enable its use as a surrogate model for multiscale simulations. Second, we developed a ML-derived interatomic potential (IAP) for atomistic simulations of yttrium disilicate (Y2Si2O7) from DFT training data. Atomistic molecular dynamics (MD) simulations utilizing the IAP yield lattice properties and bond lengths in good agreement with both DFT and experimental results from x-ray diffraction. The IAP was also used to calculate finite-temperature thermodynamic properties via the finite-displacement phonon method and quasi-harmonic approximation orders of magnitude faster than DFT with good agreement to DFT and experimental results. Finally, the IAP was used in MD simulations of large supercells (~10,000 atoms) to calculate the coefficients of thermal expansion (CTE) for the β-, γ-, and δ-phases of Y2Si2O7. The IAP correctly predicted the anisotropic nature of the CTE in all three phases. The two methods presented in this work demonstrate the utility of ML for accelerating the prediction of RE silicate properties, which can in turn accelerate EBC design and optimization.
