S8-O/LT6-1 - Introducing Artificial Neural Networks as a Specific Enthalpy approximator for a course on introductory Thermodynamics

1. Innovative Practice Work In Progress
Aneet Dharmavaram Narendranath1
1 Michigan technological university

Through the example provided in this work-in-progress paper, mechanical engineering students can be exposed to the heuristic creation of feedforward artificial neural networks (ANN), their training, validation and quantification of their accuracy, in the context of a course in thermodynamics.

As big data and machine learning continue to permeate and affect the viscera of society, new challenges and career opportunities emerge.  Organizations such as NSF, McKinsey global institute, Gartner global newsroom, IBM, to name a few, have published projections on the global impact big data and machine learning on the job market and how these technologies are the “next frontier in innovation”.

The author is constructing computational examples that incorporate data preparation, neural networks as function approximators or classifiers into traditional mechanical engineering courses. Such examples when infused into a traditional mechanical engineering (ME) curriculum would allow graduates with an ME degree to be better prepared for the world of data and machine learning.

Using a validated ANN, the evaluation of the thermodynamic property of specific enthalpy for water and steam, given a thermodynamic state in temperature and pressure is the focus of this paper.  The data used to train the neural network is generated using the equations of state provided in the IAPWS IF-97 industrial standard for water and steam.  The effect of number and type of layers on the accuracy of the network and the effect of data pre-processing, on the accuracy of the network can be studied.