Abstract
Whatever specific methods come to be preferred in the field of structural health/integrity monitoring, the associated raw data will eventually have to provide inputs for appropriate damage accumulation models and decision making protocols. The status of hardware under investigation eventually will be inferred from the evolution in time of the characteristics of this kind of functional figure of merit. Irrespective of the specific character of raw and processed data, it is desirable to develop simple, practical procedures to support damage accumulation modeling, status discrimination, and operational decision making in real time. This paper addresses these concerns and presents an auto-adaptive procedure developed to process data output from an array of many dozens of correlated sensors. These represent a full complement of information channels associated with typical structural health monitoring applications. What the algorithm does is learn in statistical terms the normal behavior patterns of the system, and against that backdrop, is configured to recognize and flag departures from expected behavior. This is accomplished using standard statistical methods, with certain proprietary enhancements employed to address issues of ill conditioning that may arise. Examples have been selected to illustrate how the procedure performs in practice. These are drawn from the fields of nondestructive testing, infrastructure management, and underwater acoustics. The demonstrations presented include the evaluation of historical electric power utilization data for a major facility, and a quantitative assessment of the performance benefits of net-centric, auto-adaptive computational procedures as a function of scale.
