CONCEPTUAL AND STATISTICAL PROBLEMS AND APPROACHES IN GAS TURBINE ENGINE HEALTH MONITORING

摘要

Current approaches to gas turbine engine (GTE) health monitoring involve either (a) simple determination of whether individual GTE variables of interest (e.g., pressures temperatures, rotational speeds, vane angles) are exceeding or approaching predetermined limits, or (b) complex physics-based computer simulation modeling of GTE aero/thermo/mechanical performance. Purely statistical approaches to GTE modeling and health monitoring tend to be viewed with skepticism by GTE engineers, but nevertheless are conceptually attractive based on the epistemological philosophy that statistical models let GTE data "speak for themselves" rather than superimposing approximations of "truth" on the data (in the sense that, e.g., the ideal gas law is an approximation to truth). However, the adoption of a purely statistical approach to GTE health monitoring has its own set of conceptual and mathematical problems. This paper discusses three of these problems in the context of GTE health monitoring: defining prognosis, dynamic hypothesis testing, and determining an overall probability of abnormality for dynamic hypothesis test results for a given GTE.