Artificial neural networks (ANNs) have been widely used in many engineering applications. This dissertation applies ANNs in the field of radiation heat transfer and applied optics. The topics of interest in this dissertation include both forward and inverse problems.
Forward problems involve applications in which numerical simulation is expensive in terms of time consummation and resource utilization. Artificial neural networks can be applied in these problems for speeding up the process and reducing the required resources. The Monte Carlo ray-trace (MCRT) method is the state-of-the-art approach for modeling radiation heat transfer. It has the disadvantage of being a complex and computationally expensive process. In this dissertation, after first identifying the uncertainties associated with the MCRT method, artificial neural networks are proposed as an alternative whose computational cost is greatly reduced compared to traditional MCRT method.
Inverse problems are concerned with situations in which the effects of a phenomenon are known but the cause is unknown. In such problems, available data in conjunction with ANNs provide an effective tool to derive an inverse model for recovering the cause of the phenomenon. Two problems are studied in this context. The first is concerned with an imager for which the readout power distribution is available and the viewed scene is of interest. Absorbed power distributions on a microbolometer array making up the imager is produced by discretized scenes using a high-fidelity Monte Carlo ray-trace model. The resulting readout array/scene pairs are then used to train an inverse ANN. It is demonstrated that a properly trained ANN can be utilized to convert the readout power distribution into an accurate image of the corresponding discretized scene. The recovered scene of the imager is helpful for monitoring the Earth's radiant energy budget.
In the second problem, the collection of scattered radiation by a sun-photometer, or aureolemeter, is simulated using the MCRT method. The angular distribution of this radiation is summarized using the probability density function (PDF) of the incident angles on a detector. Atmospheric water cloud droplets are known to play an important role in determining the Earth's radiant energy budget and, by extension, the evolution of its climate. An extensive dataset is produced using an improved atmospheric scattering model. This dataset is then used to train and test an inverse ANN capable of recovering water cloud droplets properties from solar aureole observations. / Doctor of Philosophy / This dissertation is intended to extend the research in the field of theoretical and experimental radiation heat transfer and applied optics. It is specifically focused on efforts for more precisely implementing the radiation heat transfer, predicting the temperature evolution of the Earth's ocean-atmosphere system and identifying the atmospheric properties of the water clouds using the tools of Machine learning and artificial neural networks (ANNs). The results of this dissertation can be applied to the conception of advanced radiation and optical modeling tools capable of significantly reducing the computer resources required to model global-scale atmospheric radiation problems. The materials of this thesis are organized for solving the following three problems using ANNs:
1: Application of artificial neural networks into radiation heat transfer:
The application of artificial neural networks), which is the basis of AI methodologies, to a variety of real-world problems is an on-going active research area. Artificial intelligence, or machine learning, is a state-of-the-art technology that is ripe for applications in the field of remote sensing and applied optics. Here a deep-learning algorithm is developed for predicting the radiation heat transfer behavior as a function of the input parameters such as surface models and temperature of the enclosures of interest. ANN-based algorithms are very fast, so developing ANN-based algorithms to replace ray trace calculations, whose execution currently dominates the run-time of MCRT algorithms, is useful for speeding up the computational process.
2. Numerical focusing of a wide-field-angle Earth radiation budget imager using an Artificial Neural Network:
Traditional Earth radiation budget (ERB) instruments consist of downward-looking telescopes in low earth orbit (LOE) which scan back and forth across the orbital path. While proven effective, such systems incur significant weight and power penalties and may be susceptible to eventual mechanical failure. This dissertation intends to support a novel approach using ANNs in which a wide-field-angle imager is placed in LOE and the resulting astigmatism is corrected algorithmically. The application of this technology is promising to improve the performance of freeform optical systems proposed by NASA for Earth radiation budget monitoring.
3: Recovering water cloud droplets properties from solar aureole photometry using an ANNs:
Atmospheric aerosols are known to play an important role in determining the Earth's radiant energy budget and, by extension, the evolution of its climate. Data obtained during aerosol field studies have already been used in the vicarious calibration of space-based sensors, and they could also prove useful in refining the angular distribution models (ADMs) used to interpret the contribution of reflected solar radiation to the planetary energy budget. Atmospheric aerosol loading contributes to the variation in radiance with zenith angle in the circumsolar region of the sky. Measurements obtained using a sun-photometer have been interpreted in terms of the aerosol single-scattering phase function, droplet size distribution, and aerosol index of refraction, all of which are of fundamental importance in understanding the planetary weather and climate. While aerosol properties may also be recovered using lidar, this dissertation proposes to explore a novel approach for recovering them via sun-photometry. The atmospheric scattering model developed here can be used to produce the extensive dataset required to compose, train, and test an artificial neural network capable of recovering water cloud droplet properties from solar aureole observations.
| Identifer | oai:union.ndltd.org:VTETD/oai:vtechworks.lib.vt.edu:10919/101900 |
| Date | 14 January 2021 |
| Creators | Yarahmadi, Mehran |
| Contributors | Mechanical Engineering, Mahan, James R., Nguyen, Vinh, Huxtable, Scott T., Vick, Brian L. |
| Publisher | Virginia Tech |
| Source Sets | Virginia Tech Theses and Dissertation |
| Detected Language | English |
| Type | Dissertation |
| Format | ETD, application/pdf, application/x-zip-compressed |
| Rights | In Copyright, http://rightsstatements.org/vocab/InC/1.0/ |
Page generated in 0.0034 seconds