Vibration Event Detection and Classification in an Instrumented Building

Accelerometers deployed within smart structures produce a wealth of vibration data that can be analyzed to infer information about the types of acceleration events that are occurring within the structure. In the case of monitored smart buildings, some of these acceleration events are linked to occupant behavior, such as walking, operating machinery, closing doors, etc. The identification and classification of such events has many potential applications within a smart structure or city. Understanding occupant patterns could be beneficial for operations, retail, or HVAC management, as it could be used to monitor occupancy flow with a relatively sparse sensor network. It may also have detrimental implications in terms of cybersecurity, where such information could be mined for malicious practices if unauthorized access to the data was obtained. This work presents methods for the detection and classification of vibration events in an experimental smart building, Goodwin Hall at Virginia Tech. Goodwin Hall's 200+ accelerometer network is used to gather acceleration data, from which vibration events are automatically detected and clustered. The presence of a vibration event is detected from a raw acceleration signal with an adaptive RMS threshold method. A feature vector is then created for each extracted event as areas under regions of the FFT of the event's acceleration signal. The feature vectors are then mapped into a low-dimensional space using principal component analysis, where they are clustered with various unsupervised algorithms. These processes have shown to be successful when gathering vibration events from a single-sensor setup, but pose challenges when expanded to a multi-sensor network. Because of this, expanded applications such as a semi-supervised classifier for events detected anywhere in the building are currently still under development. This semi-supervised process, combined with the known location of each sensor would allow inferences to be drawn about the frequency of different activity types in regions of the building not captured in the labeled data. Future work intends to address these multi-sensor challenges with adjustments to the algorithm process.

Name thesis:degree_name

Master of Science

Level thesis:degree_level

masters

Discipline thesis:degree_discipline

Mechanical Engineering

Department dc:contributor.department

Mechanical Engineering

Grantor dc:publisher

Virginia Tech

Year dc:date.issued

2022

Author dc:creator

• Hupfeldt, William George

Chair dc:contributor.committeechair

• Sarlo, Rodrigo

Committee members dc:contributor.committeemember

• Sandu, Corina
• Tarazaga, Pablo Alberto
• Kurdila, Andrew J.

• Unsupervised Classification
• Event Detection
• Event Classification
• Occupancy Monitoring
• Distributed Sensor Systems
• Smart Buildings

Statement dc:rights

• Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International

Licence dc:rights.uri

• http://creativecommons.org/licenses/by-nc-nd/4.0/

Language dc:language.iso

en

Dc Identifier Other

vt_gsexam:33868

OAI identifier oai:identifier

oai:vtechworks.lib.vt.edu:10919/108844