CALCE-ARL Kickoffs off New Project: Autonomous Prognostic Monitoring Device
To predict remaining useful life of critical devices, components, and sub-systems on a multitude of U.S. Army platforms, CALCE proposed the development of an autonomous prognostic monitoring device with the following attributes: a small form factor, minimal power consumption, wireless data transmission enabled, and the ability to integrate with existing diagnostics and sensors. Recently CALCE was awarded two contracts by the U.S. Army Research Laboratory (ARL) to develop an autonomous monitoring device and also develop algorithms that can process the collected data to identify primary patterns or relationships within the data to reveal how the system is changing. CALCE has teamed up with Texas-based ePrognostics LLC. to develop the monitoring device for ARL.
On May 1, 2009, CALCE, ARL, and ePrognostics LLC had a kickoff meeting at the University of Maryland. The CALCE-ePrognostic sensor system is a novel monitoring device which can monitor multiple parameters used for prognostics including, but not limited to, temperature, humidity, vibration, shock, and external sensors. This device will be wireless and can be mounted in the host system easily and non-intrusively. It will be compatible with an open common architecture such as CLOE or PS-MRS so as to enable ARL to use their software algorithms and also incorporate their existing shock sensors and future sensors developed for environmental and operations profiling. The new system will have the processing efficiency necessary to expedite operational and logistical decision making without severely impacting the physical and logistical footprint.
This project ties in with another CALCE-ARL project to develop algorithms to reveal fundamental relationships within the data. Relationships within the training data provide the foundational knowledge about the system, and relationships between the observed fielded data and training data reveal how the system is changing. Thus there are critical relationships among the observation (sensory data), the knowledge (training data), and how decisions, conclusions, and estimations are made. The primary patterns or relationships that the CALCE PHM algorithms search for include correlation, covariance, residual, and inference patterns. The algorithms developed in that effort will be incorporated onto the new sensor system. The sensor systems provide data that algorithms use to compile a composite characterization of the system. The multivariate data is reduced to univariate and then analyzed using statistical pattern recognition techniques. CALCE algorithms develop a healthy baseline model of the equipment under normal usage conditions. Any anomaly identified is further trended for statistical relevance. For more details please contact Prof. Michael Pecht (pecht@calce.umd.edu)