Forecasting gas turbine Exhaust Gas Temperatures using Support Vector Machine Experts and Genetic Algorithm

One aspect of modern commercial aircraft engine maintenance involves monitoring recorded engine parameters. When these parameters exceed their respective threshold tolerances, appropriate maintenance actions are taken. Reducing these unscheduled maintenance actions would allow maintainers to more effectively plan their maintenance schedules which help in the reduction of costs. One way of accomplishing this is to learn the behavior of the statistics of parameters and the ability to reliably forecast their future values. The parameter studied in this paper is the Exhaust Gas Temperature. A hybrid algorithm called Support Vector Machine Experts and Genetic Algorithm is applied to the task of forecasting a statistic of EGT. The algorithm involves first partitioning the input space according to similar training examples through a procedure that uses a Self Organizing Map. A SVM is regressed onto each partition and the optimal parameters for SVM are found using Genetic Algorithm. This algorithm is shown to have an improved mean absolute error when forecasting EGT values over a single support vector regression machine. The study was conducted for both engines of two Boeing 737 commercial aircrafts.