Particle Sensing in Gas Turbine Inlets Using Optical Measurements and Machine Learning
| dc.contributor.author | Moon, Chi Young | en |
| dc.contributor.committeechair | Lowe, K. Todd | en |
| dc.contributor.committeemember | Paterson, Eric G. | en |
| dc.contributor.committeemember | Ma, Lin | en |
| dc.contributor.committeemember | Alexander, William Nathan | en |
| dc.contributor.department | Aerospace and Ocean Engineering | en |
| dc.date.accessioned | 2021-01-20T09:00:55Z | en |
| dc.date.available | 2021-01-20T09:00:55Z | en |
| dc.date.issued | 2021-01-19 | en |
| dc.description.abstract | Propulsion systems are exposed to a variety of foreign objects that can significantly damage or impact their performance. These threats can range from severe dangers such as sandstorms and volcanic eruptions, which can induce engine failure in minutes, to condensation and moisture during ground tests that can negatively impact the engine's fuel efficiency. While numerous computational and experimental studies have investigated the effects of particle ingestion on the component level, an accurate in-situ measurement technique is needed for a systematic understanding of the effects and real-time engine health monitoring. Optical measurement techniques are attractive for this application due to their non-intrusive nature. However, conventional optical particle measurement methods assume the particle to be spherical, which introduces large errors for measuring particles with complex and irregular shapes, such as sand, volcanic ash, and ice crystals. The light-particle interaction contains information on the desired parameters, such as particle shape and size. The research presented in this dissertation uses this idea for a novel particle sensor concept. Scattering and extinction of light by particles are chosen as crucial features that can identify the particle as its unique signature. Numerical tools are used to simulate the scattering and extinction for particles the sensor is expected to encounter. Machine learning models are trained using the data to use scattering and extinction as inputs and estimate the particle parameters. Different types and applications of supervised machine learning models were investigated, including a layered approach with numerous models and a generalized approach with a single neural network. The particle sensor is first demonstrated using data found in the literature. This study confirmed the importance of non-spherical particles in the library to guide the machine learning models. Further demonstrations are made at a full engine and wind tunnel scale to measure injected condensation and sand sprays, respectively. The mass flow rates of the ingested material were calculated using the model outputs and validated. | en |
| dc.description.abstractgeneral | Foreign objects ingested into gas turbines can cause serious damage and degrade their performance. Threats can range from sand, dust, and volcanic ash to condensation on ground and high altitude ice crystals. On the component level, experiments and simulations have been performed to establish the performance decrease and risks to continued operations. However, there is a need for a real-time and non-intrusive measurement technique for the ingested mass. While there are established optical methods applicable for this use, these existing methods assume the particle shape to be spherical. The light-particle interaction contains information on the desired parameters, such as particle shape and size. Optical measurements of these interactions, such as scattering and extinction, can serve as "fingerprints" that can be used to estimate particle parameters. A novel particle measurement technique utilizing supervised machine learning models is presented. The models are trained using a library containing numerically calculated scattering and extinction data. Laser scattering and extinction measurements are used as inputs for the models. This new technique is first demonstrated by sizing particles found in a particle scattering database in the literature. The method's versatility and ruggedness are then demonstrated by accurately estimating the volume flow rate of a spray nozzle spraying water into a research engine. Additionally, the mass flow of sand particles is measured using this technique in a high-speed wind tunnel, in a similar environment to an engine inlet. | en |
| dc.description.degree | Doctor of Philosophy | en |
| dc.format.medium | ETD | en |
| dc.identifier.other | vt_gsexam:28991 | en |
| dc.identifier.uri | http://hdl.handle.net/10919/101969 | en |
| dc.publisher | Virginia Tech | en |
| dc.rights | In Copyright | en |
| dc.rights.uri | http://rightsstatements.org/vocab/InC/1.0/ | en |
| dc.subject | foreign object damage | en |
| dc.subject | engine health monitoring | en |
| dc.subject | laser diagnostics | en |
| dc.title | Particle Sensing in Gas Turbine Inlets Using Optical Measurements and Machine Learning | en |
| dc.type | Dissertation | en |
| thesis.degree.discipline | Aerospace Engineering | en |
| thesis.degree.grantor | Virginia Polytechnic Institute and State University | en |
| thesis.degree.level | doctoral | en |
| thesis.degree.name | Doctor of Philosophy | en |