Low-dimensional spaces for the analysis of sensor network data: Identifying behavioural changes in a propulsion system

Abstract

The amount of sensors installed for equipment health monitoring on board engines, aircraft, and vehicles has increased steadily in the digital age. Developing strategies and capabilities to extract useful information from the tremendous amounts of data collected is a separate challenge that can be used to establish the indicators of system health, a necessary precursor to the implementation of condition based maintenance, that could be explored through the use of data analytics tools and methods. In this work, initial analysis of the sensor data related to a diesel engine system and specifically its turbocharger subsystem was carried out. An incident involving seizure of the turbocharger was captured by the sensor data, and hence analysis of this event provides an opportunity to identify changes in equipment indicators with a known outcome. Several data analysis tools were used, including the transformation of the original highdimensional sensor data to a low-dimensional space. The data analysis is focused on characterizing the healthy and failed states of the turbocharger system and identifying the change in behaviour of the system during that transition.