EXPERT ANALYSIS OF TELEMETRY DATA

Abstract

We describe FAES, a knowledge-based system for postprocess interpretion of telemetry data obtained from in-water tests of the Torpedo MK48 ADCAP and recorded on tape in a telemetry format. The system is designed to automate a diagnostic application in fleet operations. A generic software solution provides the infrastructure for customization through application-specific knowledge representation. Pattern recognition provides a feature-extraction layer between the raw data and an expert system, and gives domain experts a natural and comfortable representation. Use of features abstracted from the raw data greatly reduces the complexity of encoding the rules that describe the behavior of the system under investigation. This allows the experts – not the system programmers – to control the resulting software. The approach has led to development of a system which accurately determines the cause of shutdown in torpedo tests and which will be extended to the full range of diagnostics now done manually. A slightly modified system is being used to support torpedo proofing by automating comparisons of recorded data with the weapon specification and alerting engineers to violations.