An official website of the United States government
For the development and optimization of molten salt reactors, nuclear fuel cycles, and energy storage materials, the temperature-dependent molten salt properties must be known over a wide range of possible compositions. Yet, for several relevant chloride and fluoride salt systems, significant gaps in data still exist, which inhibit the development of key advanced energy technologies. [1] Filling all of these gaps with high-temperature experiments is inherently challenging, especially due to the corrosive, volatile, and hazardous nature of these salts. Meanwhile, carefully validated atomistic simulations (ab initio, classical or machine learning-based) are capable of predicting thermophysical properties but are highly computationally expensive [2-5], limiting our ability to screen over large temperature-compositional spaces. In this work, we propose to circumvent these limitations by using supervised machine learning (ML) models to learn from existing bulk density data and predict densities of new and unseen mixtures.
more »
« less
This content will become publicly available on November 2, 2026
Siamese Graph Neural Networks for Melting Temperature Prediction of Molten Salt Eutectics
High-throughput screening enabled by structure-property prediction models is a powerful approach for accelerating materials discovery. However, while machine learning of structure-property models have become widespread, its application to mixtures remains limited due to increased complexity and the scarcity of available data. Machine learning methods for high-throughput screening of eutectic mixtures have been proposed in recent years, but there remain challenges due to the lack of diverse, open-access datasets and the need for feature engineering based on chemical knowledge. To overcome these limitations, we propose a method using Siamese graph neural networks trained solely on structural information, without requiring any prior chemical descriptors, to predict eutectic melting temperatures. We demonstrate on a dataset of molten salt eutectics that this approach can reach similar performance to chemistry-based models that require significantly more prior knowledge. We show that lower-order mixtures may be used to augment data on higher-order mixtures. Interestingly, our model trained on inorganic molten salts seems to learn information about the ideal mixture model. We also evaluate the efficacy of using our inorganic molten salt model for transfer learning with a variety of organic eutectic mixtures.
more »
« less
- Award ID(s):
- 2102406
- PAR ID:
- 10654594
- Publisher / Repository:
- ChemRxiv
- Date Published:
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
The demand for clean energy production and storage has increased interest in molten salt technologies, including Molten Salt Reactors (MSR). Understanding of how molten salts properties change with respect to temperature and structure is vital to establishing efficient, cost effective MSR systems. Research into these materials however has been limited due to the difficulty in accurately measuring properties of these reactive materials at elevated temperatures and controlled environment in a time efficient way. Much research has turned to molecular dynamic (MD) modeling to alleviate these issues. This research presents a custom fabricated falling ball viscometer system for measuring molten salt viscosity quickly. A model for correlating velocity to viscosity for Re < 300 was also developed for use with this system. The viscometer is demonstrated on eutectic FLiNaK and NaF-ZrF4 (53–47 mol%) up to 150 K above the respective melting points. The results are compared to MD simulations to verify their effectiveness for predicting viscosity and previously reported measurements.more » « less
-
High-throughput screening and material informatics have shown a great power in the discovery of novel materials, including batteries, high entropy alloys, and photocatalysts. However, the lattice thermal conductivity ( κ ) oriented high-throughput screening of advanced thermal materials is still limited to the intensive use of first principles calculations, which is inapplicable to fast, robust, and large-scale material screening due to the unbearable computational cost demanding. In this study, 15 machine learning algorithms are utilized for fast and accurate κ prediction from basic physical and chemical properties of materials. The well-trained models successfully capture the inherent correlation between these fundamental material properties and κ for different types of materials. Moreover, deep learning combined with a semi-supervised technique shows the capability of accurately predicting diverse κ values spanning 4 orders of magnitude, especially the power of extrapolative prediction on 3716 new materials. The developed models provide a powerful tool for large-scale advanced thermal functional materials screening with targeted thermal transport properties.more » « less
-
Virtual screening is a cost- and time-effective alternative to traditional high-throughput screening in the drug discovery process. Both virtual screening approaches, structure-based molecular docking and ligand-based cheminformatics, suffer from computational cost, low accuracy, and/or reliance on prior knowledge of a ligand that binds to a given target. Here, we propose a neural network framework, NeuralDock, which accelerates the process of high-quality computational docking by a factor of 10 6 , and does not require prior knowledge of a ligand that binds to a given target. By approximating both protein-small molecule conformational sampling and energy-based scoring, NeuralDock accurately predicts the binding energy, and affinity of a protein-small molecule pair, based on protein pocket 3D structure and small molecule topology. We use NeuralDock and 25 GPUs to dock 937 million molecules from the ZINC database against superoxide dismutase-1 in 21 h, which we validate with physical docking using MedusaDock. Due to its speed and accuracy, NeuralDock may be useful in brute-force virtual screening of massive chemical libraries and training of generative drug models.more » « less
-
Abstract The purpose of this work is to lower the computational cost of predicting charge mobilities in organic semiconductors, which will benefit the screening of candidates for inexpensive solar power generation. We characterize efforts to minimize the number of expensive quantum chemical calculations we perform by training machines to predict electronic couplings between monomers of poly‐(3‐hexylthiophene). We test five machine learning techniques and identify random forests as the most accurate, information‐dense, and easy‐to‐implement approach for this problem, achieving mean‐absolute‐error of 0.02 [× 1.6 × 10−19J],R2= 0.986, predicting electronic couplings 390 times faster than quantum chemical calculations, and informing zero‐field hole mobilities within 5% of prior work. We discuss strategies for identifying small effective training sets. In sum, we demonstrate an example problem where machine learning techniques provide an effective reduction in computational costs while helping to understand underlying structure–property relationships in a materials system with broad applicability.more » « less
