Machine learning approach for calculating the viscosity of ternary nitrate based molten salts

The growing production of clean electricity from solar resources has led to employ molten salts as heat transfer fluids for thermal energy storage technologies. Thus, the study of thermodynamic and physicochemical properties of multi-component molten salts are critical factors for the design of such units. From one side, binary salt mixtures’ thermodynamic properties are already accessible through critically assessed data. However, the evaluation of physicochemical properties, such as viscosity, requires experimental measurements for specific molten salt compositions. Computational models based on machine learning techniques intend to accelerate the time invested in experimental procedures to simulate and, even more, predict data with high accuracy. Thus, in this work, three machine learning techniques are proposed to predict the viscosity of nitrate-based ternary molten salts. Training data for LiNO3, NaNO3, and KNO3 (unary, binary, and ternary systems) will be collected from the literature to train Decision Trees, Support Vector Machines, and Artificial Neural Networks. Then, the coefficient of determination and the root of mean square error will be calculated to evaluate and compare the performance of models.

Faculty Supervisor:

Jean-Philippe Harvey

Student:

Partner:

Universidad Veracruzana

Discipline:

Engineering

Sector:

Education

University:

Polytechnique Montréal

Program: