Session: 11-62-01: Machine Learning for Thermal Transport

Paper Number: 117109

117109 - Exploring Efficacy of Machine Learning (Artificial Neural Networks) for Enhancing Reliability and Resilience of Thermal Energy Storage Platforms Utilizing Phase Change Materials for Sustainability and Mitigating Food-Energy-Water (FEW) Nexus 

Thermal Energy Storage (TES) platforms can be used to balance the difference between consumption and supply, i.e., they store thermal energy during periods of excess supply and redistribute thermal energy during periods of deficit. Phase change materials (PCMs) can help with improving resilience and reliability of TES platforms. High latent heat values accruing from PCMs enhance storage densities. This enables compact form factors for TES applications. Inorganic PCMs confer higher latent heat values than organic PCMs. However, these advantages are offset by compromised reliability of inorganic PCMs. Inorganic PCMs often require high degree of supercooling (also known as “subcooling”) to initiate nucleation (which degrades their reliability, net energy storage capacity, and power rating of the TES platform). “Cold Finger Technique (CFT)” is an effective strategy for mitigating these issues. For implementing CFT, a small portion of the PCM in the TES platform is left un-melted during the melt-cycle (this facilitates the spontaneous nucleation during the freezing cycle). Thus CFT enhances reliability of the TES platform but at a marginal cost to the net storage capacity. In CFT, power rating of the TES remains almost unaffected. In this study, machine learning (ML) techniques were leveraged to exploit the capability of CFT more effectively. Temperature transients were recorded from PCM melting experiments. This data was used to train an Artificial Neural Network (ANN) model. This deep learning technique (i.e., using multi-layer perceptron model or “MLP”) is implemented to predict the required time to reach the designated melt-fraction of the PCM. The results show that the ANN model is capable of predicting the time required to reach pre-designated value of melt-fraction with outstanding accuracy (e.g., 85% melt-fraction, that is specified by the user). The mean error of the predictions is calculated and is reported in this study. However, this approach is vulnerable to the fidelity of the data-set utilized for training the ANN (MLP) algorithm.

Highlights

• A simple experimental apparatus is configured for digital recording of spatial variation of temperature transients within a Thermal Energy Storage (TES) device using Phase Change Material (PCM) and on the surface of the containment apparatus.
• Multi-Layer Perceptron (MLP) model is utilized for developing the artificial neural network (ANN) which is then used for enhancing the performance, reliability and accuracy of predictions for exploiting Cold Finger Techniques (CFT).
• Using transient measurements from only three thermocouples, that were immersed in PCM at different levels, the ANN model can predict the progression of the melt front.
• Effect of different power input on the prediction accuracy was explored and the reliability of the ANN model was validated using experimental data (that was used for training the MLP model).

Presenting Author: Debjyoti Banerjee Texas A&M University

Presenting Author Biography: Debjyoti Banerjee (dbanerjee@tamu.edu) - in addition to his appointment as a Professor of J. Mike Walker ’66 Department of Mechanical Engineering since 2005; as James J. Cain ’51 Faculty Fellow I since 2017 and as Professor (Joint Courtesy Appointment) in the TAMU Department of Petroleum Engineering since 2011; he was appointed as an Adjunct Faculty in the Department of Medical Education at TAMU School of Medicine [SOM]. Prof. Banerjee was also the Dean’s Fellow of EnMED (the Inter-Collegiate School of Engineering Medicine). EnMED is a joint endeavor of the Houston Methodist Hospital with the Texas A&M University [TAMU] College of Engineering [COE] and the TAMU School of Medicine [SOM]. He was elected as a Fellow of the American Society of Mechanical Engineers [ASME] in 2016.

Authors:

Pinjala Sai Sudhir Texas A&M University
Debjyoti Banerjee Texas A&M University