Closed-loop system for control of combustor pattern factor

A neural network (NN) controller was developed to control the exit temperature distribution in a small-scale liquid fueled combustor. Simple feedforward NN structures were developed and trained off-line to simulate the closed-loop system and optimize the training algorithm. The adaptive NN controller was trained online and its ability to control under different operating scenarios was assessed. Various training sets were used to determine accuracy when controlling to set points representing interpolated and extrapolated values. These tests revealed accuracies similar to those when controlling to set points included in the training set and indicated that the controller could develop accurate models of the combustor response even when trained with limited information. Combustor performance drift was simulated by introducing a baseline profile offset. The control system demonstrated the ability to quickly adapt and compensate for the offset, The successful implementation of this control system in gas turbine engines would result in an increased service life of hot section components and allow higher average combustor exit temperatures.