A stability criterion method based on neural network and its application on flutter boundary prediction

摘要

According to the idea that the Artificial Neural Network (ANN) can describe any linear and nonlinear system/object theoretically, the stability characters of the system should be contained in the parameters of ANN. This paper presents the use of a feed forward ANN for modeling the flutter test signal, in order to predict the flutter boundary through the jury criterion method. Based on Back Propagation Networks, ANN model and time-varying autoregressive moving-average (ARMA) model are investigated. Combined the Jury criterion with ANN and ARMA, a new method of Jury stability criterion is proposed from the aspect of mathematics, which means the weight values in ANN can be used to construct the Jury criterion directly. Compared with traditional time series analysis approaches, the rationality and validity of the new Jury criterion using the neural network are studied through numerical simulation. Furthermore, the feasibility and engineering applicability of paper method are verified and explored by using experimental flight flutter data.