Limit Cycle Oscillation Prediction Using Artificial Neural Networks

Abstract

A static artiŽ cial neural network in the form of a multilayer perceptron is investigated to determine its ability to predict linear and nonlinear  utter response characteristics. The network is developed and trained using linear  utter analysis and  ight-test results from a Ž ghter test. Eleven external store carriage conŽ gurations are used as training data, and three conŽ gurations are used as test cases. The network was successful in predicting the aeroelastic oscillation frequency and amplitude responses over a range of Mach numbers for two of the test cases. Predictions for the third test case were not as good. Several network sizes were investigated, and it was found that small networks tended to overgeneralize the training data and are not capable of accurate prediction beyond the sample space. Conversely, networks that were too large, or trained to error levels that were extreme, tended to memorize the training data, and are also unable to produce adequate predictions beyond the sample space. The results of this study indicate that relatively simple networks using small training sets can be used to predict both linear and nonlinear  utter response characteristics.