Date of Award
5-2024
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Mechanical and Civil Engineering
First Advisor
Hamidreza Najafi
Second Advisor
Troy Nguyen
Third Advisor
Ryan T. White
Fourth Advisor
Ashok Pandit
Abstract
Surface heat flux measurement is essential in various industrial applications including atmospheric entry of space vehicles, machining processes, non-destructive testing and more. While direct measurement of surface heat flux is known to be a very difficult, if not infeasible, task, acquiring temperature measurement from sub- surface locations are often viable. The problem of evaluating the surface heat flux using internal temperature measurements is known as an inverse heat conduction problem (IHCP). IHCPs are considered as mathematically ill-posed as they are highly sensitive to measurement error. This research presents a solution based on the implementation of artificial neural networks (ANNs) to solve IHCPs in multidimensional domains with multiple unknown surface heat fluxes and temperature dependent material properties in a near real-time fashion. Several numerical test cases are developed in COMSOL Multiphysics, facilitating the solution to the direct problem. The results are then used to train, test and cross-validate the proposed IHCP solution techniques. Two different approaches are developed: (a) using a single neural network with inputs from n temperature sensors to estimate n unknown transient surface heat fluxes, (b) using n separate neural networks each of which with one input (from the closest temperature sensor) to estimate one unknown surface heat flux. Both approaches are evaluated through several test cases. Additionally, parametric study was performed to explore the impact of the sensor placement on the performance of the proposed approaches. The network used in this study was a feedforward ANN using backpropagation. The results were promising and show that heat fluxes can be successfully evaluated through the implementation with good accuracy particularly when the second approach was used. Higher accuracy was also achieved when the temperature sensors were placed closer to the surface as the model. The results from this study can be used for computationally-efficient surface heat flux evaluation in various applications.
Recommended Citation
Martinez Mundarain, Andrea Carolina, "Artificial Neural Networks as the Solution of Inverse Heat Conduction Problems in Multidimensional Domains" (2024). Theses and Dissertations. 1438.
https://repository.fit.edu/etd/1438