| This dissertation takes a certain model of fighters as the research target which provided by an institute. We do some research about fault prediction of nonlinear systems mainly using neural network and design the fault detection and diagnosis scheme at the different work point. Finally, make the realization in C++ language.First, in the process of flying, the states of the fighters change continuously, so the network off-line doesn't satisfy the requirement of real time. According to the reason above, we design a fault prediction method base on RBF network on line learning algorithm, which adjust the number and the location of the hidden units and the weight dynamically and it can studies and predicts at the same time of system's running. This method overcomes the problem of poor real time of the off-line network, so for the prediction about actual systems, it has better application value.Second, for a complete prediction system, it doesn't only need to detect whether it has the fault, but also need to confirm the location further. In order to identify the fault pattern, a novel training algorithm called AFOLS is presented, which combines accelerated fuzzy C-means(AFCM) and orthogonal least squares(OLS). This method reduces the number of training datas participated in OLS and solve the problem which can't get the flowing parameters from OLS. Simulation result shows the AFOLS algorithm achieves faster speed and reduces the size of the network.Finally, the fault prediction client is established, which makes the full-envelope real-time simulation program of the flight as the server. In this platform, we realize the fault prediction method based on RBF network on line learning with C++ language and complete each state real-time prediction of the flight using roll optimum theory. According to the prediction state and the structure of the network chosen, fault diagnosis is completed and the result of fault state prediction are displayed visually.
|
PDF Full Text Request