Inductive modeling for predicting maximum turbine inlet temperatures

Engine, aircraft, and environmental data collected during the operation of eight GE TF-39 engines installed in US Air Force C-5 Galaxy cargo aircraft were used in a two part program to demonstrate, test, and evaluate the capability of polynomial neural networks (PNNs) to predict maximum turbine inlet temperatures (Max TITs). Input parameters were measured before, during, and shortly after the takeoff throttle advance. The average absolute value of the prediction error for Max TITs averaging 787 degrees F was 8.5 degrees F based on as many as 13 input variables available immediately before the takeoff throttle advance. Better accuracy was then achieved by collecting data from one input variable (TIT) for several seconds after the takeoff throttle advance before making Max TIT predictions. These results, based on models derived from the performance of eight engines on two aircraft during a total of five flights, suggest the possibility of significantly improving aviation safety and reducing engine maintenance by enabling on-board engine monitoring equipment to notify aircrews of off-normal engine performance in sufficient time to avoid operating the engines at potentially destructive TITs. However, additional data for PNN training and evaluation is needed to develop models applicable to the entire C-5 fleet