Online Program Home
My Program

Abstract Details

Activity Number: 511 - Statistical Applications in the Physical Sciences
Type: Contributed
Date/Time: Wednesday, July 31, 2019 : 10:30 AM to 12:20 PM
Sponsor: Section on Physical and Engineering Sciences
Abstract #306869
Title: Robust Anomaly Detection in Large-Scale Multi-Type Sensor Systems
Author(s): Sierra Merkes*
Companies: Virginia Tech Statistics Department
Keywords: anomaly detection; Bayesian; PCA; robust; sensor systems; wind tunnel
Abstract:

Modern wind tunnel systems are constructed of various instruments containing hundreds and thousands of individual sensors. Each sensor requires continual quality monitoring, as corrupted data will increase experimental costs, wasted time, and lead to spurious engineering conclusions. Monitoring methodologies such as Gaussian Processes and Principal Component Analysis have been examined for anomaly detection in wind tunnel systems. However, these methodologies are limited due to continual instrumental movement, wind tunnel re-configurations, or low-levels of adverse observations in training datasets. In this presentation, we compare two robust system monitoring methodologies: (1) a method based on a robust Principal Component Analysis technique, and (2) a proposed method based on a Bayesian, heavy-tailed posterior distribution with Principal Component Analysis. Additionally, we extend our methodology to mixtures of signals in order to account for cases where sensors are intermittently collecting anomalous signals. Through the simulation and real wind tunnel experiments, we exemplify the need for anomaly detection methodologies using a robust, correlated, multi-type sensor approach.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program