HOLISTIC EVALUATION OF SENSOR FIDELITY IN DIVERSE MEASUREMENT SYSTEMS AND APPLICATIONS TO LARGE SCALE EXPERIMENTAL CAMPAIGNS
Abstract
Experimental testing of operational or prototype devices and systems involves multiple diversesensor systems with anywhere from hun"dreds to thousands of individual sensors. These large-scaletests can be expensive and time-sensitive. As such, any delays due to fa"ulty instrumentation canhave serious consequences. Equally serious is the possibility of discovering a sensor failure aftertest co"mpletion, since time and effort will have been spent collecting what amounts to noise. Anylarge-scale sensor system requires method"s to ensure that all the individual sensors are working asintended. Most commercial sensor systems contain rudimentary error detect"ion for sensors withina given system, but these methods typically have no way of incorporating information about theambient condit""ions under which they were run, or, more importantly, information from the outputof other systems which are used in conjunction. Th""is is potentially a significant problem, as mostlarge sensor systems are made of smaller, unique sensor subsets. By combining the i""nformationfrom diverse sensor systems and the related metadata into a global error detection process, we canmeasure the extent to" which sensors across systems are correlated and use that correlationinformation to produce more powerful error detection capabilities. We propose a rigorousstatistical approach to sensor systems evaluation that is based on Gaussian processes (GP). Ourprior wor"k, considering at first, homogeneous systems like phased arrays, and then more diversesystems in our 2016 experiment, have led us t"o statistical approaches which are able to identifyfaults and anomalies in data sets through use of expected correlation metrics an"d response surfacetechniques. These methods, mainly the Gaussian process approach, have shown significantpromise in application al"though several issues remain unresolved. The primary technical challengeis defining the proper response distance between sensors. T"his is not a trivial task and may requirethe system to ~learn~ over the course of an experiment, over the history of a system or fa""cility.Application to sensors of varying dimensionality is also a major obstacle. Finally, this powerfultechnique has yet been int"egrated into a rapid analysis procedure. Support is requested that wouldenable the team to investigate these open matters as well as additional uses of the GP procedurefor improving the planning and execution of experimental campaigns. This includes theintersec"tion of our sensor evaluation scheme with areas of uncertainty quantification, facility andtest planning, as well as adaptive measu"rement processes.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 29, 2017
- Source ID
- N000141712944
Entities
People
- Eric Alden Smith
Organizations
- Office of Naval Research
- United States Navy
- Virginia Tech