Abstract
In order to resolve the turbulent and three-dimensional (3D) structures of combustion processes, four-dimensional (4D) diagnostics, meaning time-resolved measurements including all three spatial dimensions, are required. These 4D measurements are critical in understanding flame behaviors in a multitude of practical applications. Computed tomography (CT) is a key tool in obtaining instantaneous 3D flame measurements by using multiple two-dimensional (2D) line-of-sight projections collected from cameras to generate a 3D reconstruction of a signal emission distribution within a flame, and these measurements can be considered 4D when they are time-resolved. However, obtaining 4D combustion measurements with sufficient spatial and temporal resolution has many challenges. While improvements in camera technology and the use of additional cameras may yield higher spatial resolution, typically these factors are limited for a given experimental setup. Therefore, new techniques must be developed to maximize the spatial resolution of a given setup. Additionally, due to the large amounts of data required for 3D CT, which is compounded by the need for time-resolved measurements, improved techniques are desired that can reduce the computational costs in both time and memory while maintaining reconstruction accuracy. The primary focus of this dissertation is the development and analysis of improved algorithms to achieve the goals of reducing computational costs without a loss in accuracy, while also implementing them in such a way as to improve the spatial resolution of the reconstructions.
First, this dissertation describes an adaptive spatial discretization (ASD) technique that aims to reduce computational costs without a loss in accuracy by identifying regions within the reconstructed measurement volume that require either high or low spatial resolution and treating these regions respectively with a fine or coarse discretization. This way, only regions that require a higher resolution, mainly those regions with large spatial gradients, are solved using a fine discretization, so that the total amount of computational resources required for CT can be reduced while still maintaining accuracy in important regions. This method was validated using a phantom study using six cases that represent a range of distributions found in tomography applications.
In addition to the ASD method, a pixel masking method is described that has been applied to volumetric laser-induced fluorescence (VLIF) measurements. The masking method works on the principle that all voxels that contribute to pixels containing signal below an estimated signal floor must contain no substantial emission and can be removed from the tomographic reconstruction. This reduction of pixels and voxels reduces the computational costs of CT without adversely affecting reconstruction accuracy. This method was validated by directly comparing the results of the VLIF reconstruction to 2D planar LIF (PLIF) measurements. Both the ASD and masking techniques were used to reduce the computational costs of CT without reduction in reconstruction accuracy and also showed the ability to improve spatial resolution, satisfying two key needs of obtaining 3D and 4D combustion measurements.
In terms of practical demonstration, limited space and harsh conditions are typical and present challenges in implementation. Under many practical conditions of interest, it is difficult to obtain experimental data with enough views to perform tomography with adequate accuracy and resolution. These challenges are worsened under field conditions, outside of a laboratory environment. To enable visualization of flames under practical conditions, this work describes a proof-of-concept demonstration using fiber-based endoscopes (FBE) in conjunction with a single camera. Results showed the design could significantly reduce the equipment cost and footprint and allowed visualization of flames inside a ground vehicle testbed with limited viewing access. The time-resolved 3D measurements were used to resolve the temporal dynamics and spatial structures of the target flame under challenging experimental conditions. This experimental setup has shown to be capable of delivering 4D tomographic capabilities under challenging conditions and may serve as a promising platform for future practical measurements. With a combination of progress in both data acquisition and data processing techniques, as described earlier, this dissertation contributes to the advancement of combustion tomography and diagnostics, as well as providing techniques that may be applicable to a wide range of tomography applications.
