Date of Award

2016

Document Type

Open Access Dissertation

Department

Civil and Environmental Engineering

Sub-Department

College of Engineering and Computing

First Advisor

Juan Caicedo

Abstract

In the US alone, 20% of citizens will be over the age of 65 by the year 2030, and the largest challenge facing this growing demographic is not a new disease but a simple motion - the fall. These events are the premier cause of fatal and nonfatal injuries. In fact, a fall is so common that every 17 seconds an older adult is treated for fall-related injuries, and every 30 minutes, an older adult dies from fall complications. Research has shown a positive outcome of a fall event is largely dependent on the immediate response (<30 minutes) and rescue of the person.

This work explores the concept of utilizing structural vibrations to detect human falls for rapid rescue response. A human-induced vibration monitoring system is developed on the principles of the ideal fall detection system. The system was installed throughout the William Jennings Bryan Dorn Veteran’s Administration Medical Center (VAMC) and a private four person family’s residence to collect real world humaninduced vibrations. Installation resulted in the recording of 16 human falls and 45,000 acceleration events, expanding the database to 220,597 events as of 2016 February 1. This and other information are recorded according to the data management plan presented within to enable future study of human activity from vibrations.

For a successful fall detection system implementation, the accelerometers need the ability to discern good signals to reduce the amount of data analyzed. A method of signal categorization using Support Vector Machines is explored to this end, with 96.8% accuracy over 100 trials. Following signal selection, the ability to detect a fall regardless of the distance from the event to the accelerometer becomes paramount, and is overcome with the introduction of the Force Estimation and Event Localization (FEEL) Algorithm. The algorithm allows structures to ‘FEEL’ an impact, such as a fall, boasting 96.4% accuracy in locating the impact in over 3575 impacts of eight different types, and a 99% confidence interval for being within -2.0% ± 1.3% of the actual force magnitude. The strength of the algorithm is that it intrinsically embeds the properties of any structure and does not require time synchronization of sensors. Since FEEL operates in the frequency domain, an Environments For Fostering Effective Critical Thinking (EFFECT) active learning module is included to aid in educating future learners.

Rights

© 2016, Benjamin Thomas Davis

Download

Share

COinS