Webinar: Machinery Anomaly Detection Under Indeterminate Operating Conditions
Tuesday, June 19, 2018
Anomaly detection is a critical task in system health monitoring. The current practice of anomaly detection for mechanical and electromechanical systems, such as wind turbines, cooling fans, and pumps, is still unsatisfactory. Generally, features derived from the data are tracked to identify and detect anomalies. However, some features are insensitive to the change of health, and some features are redundant with each other. These insensitive and redundant features can reduce the effectiveness of the detection process. Another problem with the current status of machinery anomaly detection arises when a change in operating conditions can be mistakenly detected as an anomalous state of the system. Operating conditions, such as rotation rate of a motor or velocity of an actuator, are normally not static, and they may not be readily identifiable from the data. Their changes, if not accounted for, can contribute to false positive detection. The goal of this webinar is to describe techniques for the reduction of anomaly detection errors by developing methods to select predictive features and use them for anomaly detection under indeterminate operating conditions.
About The Presenter: Jing Tian has recently completed his doctoral dissertation under the direction Prof. Michael Pecht and Dr. Michael Azarian. Dr. Tian has worked at Lenovo research center, Center for Advanced Life Cycle Engineering (CALCE) at University of Maryland, and DEI Group in the research areas of condition-based maintenance and data-driven prognostics and health management.