NASA SBIR 2003 Solicitation


PROPOSAL NUMBER: 03- II F2.02-9170
SUBTOPIC TITLE: Multi-agent and Human-centric Systems Technologies
PROPOSAL TITLE: Agent-Based Health Monitoring System

PRINCIPAL INVESTIGATOR/PROJECT MANAGER (Name, E-mail, Mail Address, City/State/Zip, Phone)
Leonard Haynes
15400 Calhoun Drive, Suite 400
Rockville, MD 20855-2785
U.S. Citizen or Legal Resident: Yes

We propose combination of software intelligent agents to achieve decentralized reasoning, with fault detection and diagnosis using PCA, neural nets, and maximum entropy methods. The goal of the work is to achieve integrated system health management and self-reliant systems, including integration with the maintenance and logistics scheduling systems to achieve fully automated end-to-end solutions. At low levels the agents will evaluate raw sensor signals to detect and diagnose the cause of anomalies. At the next higher level, the agents will combine the diagnostic results from multiple lower level agents to detect and diagnose anomalies in the interaction between components or subsystems. If there is a maintenance action or a spare part indicated by the prognosis, a Task Agent and/or a Spare Parts agent will be spawned to interact with the appropriate agent-based Scheduling System to insure that the requirements are met. Agents at each level are also responsible for performing graceful degradation in the event of a failure at their level. At the low level, we have demonstrated that the PCA algorithm can greatly reduce the amount of diagnostic data that must be shared between hierarchical levels. We have also demonstrated other algorithms for anomaly detection, diagnosis, and diagnostic data-fusion.

The primary commercial application of the work herein proposed is in protecting power grids from exactly the type of failure that resulted in the August 2003 multi-day loss of power in nearly the entire northeast of the US. We envision intense interest in technology to protect the power grid from massive shutdown. The problem is fundamentally one of decentralized control of many power generation plants and complex power distribution hardware. Software agents have proven themselves to be very more effective, robust, and more easily implemented than conventional centralized solutions.

The NASA target for this work is the new Space Exploration Initiative and the first BAA named the Advanced Space Technology Program. For the new Space Exploration Initiative, prognostics and diagnostics integrated with automatic resource allocation and scheduling will be essential for many reasons. 1) Maintenance and repair operations will often occur in space where resupply of parts may take months or be infeasible. It is essential to make sure all logistical requirements are met. Logistics system failures are mission and possibly life critical. Latent faults must be detected; prognosis must be automatic and accurate. 2) The total system needed to support missions to Mars will be so complex that the current largely manual scheduling is not feasible or would be so error prone that missions could fail for lack of the required components, spares, or tools because of these errors. Schedules must be generated that allow for maximum contingency planning to prevent continual replanning. 3) Space based, moon based, or Mars based space ports will be very complex. Current systems for diagnosis that generate a single good/bad indication from each sensor individually are inadequate because they do not detect subtle changes that are indicative of latent faults, and because of the time and manual effort required to analyze and determine the root cause of a problem.