| dc.description.abstract | Sensor data augmentation for accurate system monitoring is relevant to many engineering applications, as there is often a gap between available instrumentation and measurement needs. Installing sensors can be limited due to factors such as harsh environmental conditions, the need to avoid operational distortions, and limited space. While continued efforts to develop novel sensor technologies to improve measurement density and quality are important, it is equally crucial to maximize the use of data from existing sensors and measurements. In this work, we employed physics-based methods to solve inverse heat transfer (IHT) problems. Because accurate and well-understood physics models provide strong prior knowledge, physics-based IHT can provide clear solution with use of small amount of temperature measurements. However, existing work in IHT relies on 'perfect' physics models and has been used to solve relatively simple problems such as conduction heat transfer problems. This thesis extends the IHT problem scope to thermal fluid systems, including the efficient use of sensor data and uncertainty quantification (UQ).
We leveraged high-resolution thermal-fluid experiments to demonstrate the solution of two types of IHT problems. The first problem estimates the operating conditions of the experiment based on the minimal use of sensors from high-resolution temperature data. The estimated solution is used to reconstruct the entire temperature distribution on a heating surface, while the rest of the data is used to validate the inverse problem methodology. The estimation result is supported by UQ considering measurement errors and modeling errors that adds value to the estimation. The second IHT problem consists of identifying sharp-featured 2D heat source distributions with an array of temperature sensors from a subset of experiment data. Solving IHT involved regularization prior with strong sparsity-promoting capability. The designed iterative solution optimization process finds the unknown heat source distribution as well as regularization hyperparameter. In addition, Bayesian inference enhanced the solution quality by providing UQ of the heat source magnitude.
Expanding the scope of IHT problems, we also addressed online state estimation in dynamic systems. This work focuses on a hypothetical inverse conduction problem of a transient heat source in a composite materials system. The physics modeling of system is assumed to include uncertainty arising from gap thermal resistance at material interfaces, which complicates the estimation of an internal heat source from external sensor data. To address this challenge, the IHT approach leverages future time-step measurements to correct estimates at the current time step, enabling more efficient use of limited sensor information. The approach is sampling-based and its statistics provides UQ on the quantity of interest.
While this work addresses inverse problems within specific thermal-fluid systems, the methodology is designed for broad applicability beyond these cases. It lays the groundwork for advanced sparse sensing and inverse problem-solving in thermal systems, offering a more efficient, tractable, and reliable tool for engineers and researchers addressing system monitoring with modeling uncertainty. Looking forward, these methodologies could be valuable for digital twin applications, where live sensor measurements are integrated to provide robust, real-time estimation of the state of physical systems. | |
