dc.creator | Swischuk, Renee C | |
dc.date.accessioned | 2019-07-24T16:16:29Z | |
dc.date.available | 2019-07-24T16:16:29Z | |
dc.date.created | 2017-05 | |
dc.date.issued | 2016-08-17 | |
dc.date.submitted | May 2017 | |
dc.identifier.uri | https://hdl.handle.net/1969.1/177548 | |
dc.description.abstract | Aircraft guidance is dependent on various sensors which provide information on speed, altitude and location with respect to both the ground and the surrounding air. The pitot static system, global positioning system (GPS) and inertial navigation system (INS) are the main sources of information. The pitot static system measures total and static pressures to provide airspeed information. This system includes two ports located outside of the aircraft making them vulnerable to interference and failures. Autonomous aircraft software has not yet been developed to handle failures in this system. If an aircraft has access to redundant data streams, then it can be trained to autonomously recognize errors in the pitot static system and learn to correct them. In this work, we develop a novel machine learning approach to detecting pitot static system failures, identifying types of failures and predicting airspeed with the use of redundant flight data. Two running estimates of airspeed are kept during flight and major discrepancies between the two triggers an error identification system. This identification system computes the autocorrelation of the incoming pressure data to classify the state of the pitot static system. Exploratory dimensionality reduction and feature selection techniques are performed on the redundant data to create a library of selected sensor output from previous flights. This library is used to train a k-nearest neighbors regression model to make online airspeed predictions in the event of a pitot static system failure. We demonstrate our methodology on sample flight data from a four engine commercial jet. Having this fault resistant guidance system for an aircraft makes it possible to remain in flight and continue critical missions. | en |
dc.format.mimetype | application/pdf | |
dc.subject | data-driven methods | en |
dc.subject | error detection | en |
dc.subject | sensor correction | en |
dc.subject | offline/online methods | en |
dc.subject | feature selection | en |
dc.subject | machine learning | en |
dc.subject | UAV | en |
dc.subject | dimension reduction | en |
dc.subject | airspeed prediction | en |
dc.subject | pitot static system | en |
dc.subject | autocorrelation | en |
dc.subject | k nearest neighbors | en |
dc.subject | Sammon's Mapping | en |
dc.subject | Simulated Annealing | en |
dc.title | A Machine Learning Approach to Pitot Static Error Detection and Airspeed Prediction | en |
dc.type | Thesis | en |
thesis.degree.department | Mathematics | en |
thesis.degree.discipline | Applied Mathematical Sciences | en |
thesis.degree.grantor | Undergraduate Research Scholars Program | en |
thesis.degree.name | BS | en |
thesis.degree.level | Undergraduate | en |
dc.contributor.committeeMember | Allaire, Douglas L | |
dc.type.material | text | en |
dc.date.updated | 2019-07-24T16:16:29Z | |