| Summary: | The continuous evolvement of electromechanical systems in the last decade has increased the need for methods to efficiently determine the health of a system. In this paper, we present the development of a coherent anomaly detection system that effectively measures and provides reliable information on system health through noncontact optical distance sensing. The related hardware was designed with the same feasibility criterion that the developed system needs to be deployed in narrow enclosures such as a motor box. The system operates through continuous data analysis from the sensor-provided data streams to detect anomalies in comparison to a trained feature model obtained through averaging a large set of data obtained from a healthy system. In this work, supervised learning techniques such as linear regression, resampling, dynamic time warping, and barycenter averaging was deployed to tune the model to achieve a satisfactory trade-off between data accuracy and processing speed. The developed system is targeted for use in escalator systems, but it was hypothesized that it may be adapted to other applications with minimal effort.
|
|---|