Date of Award

12-2021

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Mechanical and Civil Engineering

First Advisor

Hamidreza Najafi

Second Advisor

Darshan Pahinkar

Third Advisor

Ryan White

Fourth Advisor

Ashok Pandit

Abstract

A critical component necessary for the safety and survival of space vehicles are the thermal protection systems (TPS). Especially when subject to atmospheric re-entry, the heating loads subject to the space vehicle requires the implementation of extensive protection. The TPS commonly used currently are passive ablative materials which burn away under extreme heat flux. Due to the nature of the material and the incident heat flux, it is not feasible to attain accurate, real-time measurements for the boundary conditions with sensors placed on the surface of the space vehicle. The ability to do such would allow opportunities to implement active thermal control systems which may not only increase safety margins but also increase the serviceability and lifespan of the vehicle. It is for this purpose that alternate methods to attain surface heat flux must be explored, one of which consists of placing temperature sensors internally and solving the inverse heat conduction problem (IHCP) to attain surface boundary conditions. There have been previous attempts to reliably solve the complex problem presented, promising methods of note are the use of artificial neural networks (ANN) and intelligent algorithms (IA) to predict incident heat flux in near-real-time fashion. Machine learning is an ever growing and changing field of study with multiple different algorithms and methods developing each year. Efforts to categorize and compare the performance of such methods must be undertaken to provide visibility and direction to future research especially in the application of IHCP. As a result, a comprehensive, comparative study of notable methods for the IHCP are presented herein. The methods investigated include digital filter applications in: the feed-forward Artificial Neural Network (ANN), the recurrent Nonlinear Autoregressive Exogenous Neural Network (NARX NN) and a novel method utilizing the intelligent optimization algorithms Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). Several test cases are considered, including the moving and non-moving boundary one-dimensional IHCP subject to 6 incident heat flux profiles with constant and variable material properties. developed in COMSOL Multiphysics. The performance of these networks with respect to resiliency to noise, accuracy, time-lag necessary for satisfactory results and prediction robustness dependent on the training data are presented and compared. Results of note in the study presented reveal clear benefits in the utilization of Neural Networks in solving IHCPs due to its ease of use and accuracy with minimal time-lag; however, the success in implementation of the digital filter method with genetic algorithm reveals a robust and accurate methodology if time can be spared for adequate optimization. The results demonstrated provide meaningful insights towards possible regions of focus and interest to be applied in the development of sensors in active TPS systems.

Download

Share

COinS