Modeling Angular Dynamics On Miniature Unmanned Helicopter

Miniature unmanned helicopter (MUH) has numerous advantages, such as small size, light weight, strong ability to remain in concealment, good flexibility, easy to achieve hovering and hedgehopping, and so on. It is widely used in both military and civilian field and being studied by institutions and organizations all over the world in recent years.In of field of research in MUH, the development of the dynamical model plays a crucial part. One main difficulty lied in modeling the helicopter includes high order, resulted in a great number of parameter to be estimated. Another difficulty is the data collected are not so reliable as sensors on board are working in a highly vibration environment. Even worse, some critical parameters can not be measured directly. The present theory on Helicopter mainly focuses on the ordinary full-size helicopter and so of great difference for MUH. Therefore, it is a great need to combine the helicopter theory and fight test data which reflects the features of MUH. In research for modeling the MUH, the angular dynamics receive most attention because it is the "core" dynamics which dominated the performance of MUH and the most complex part of its dynamics. This dissertation attempts to derive an angular dynamical model using both mechanism reasoning and the information from data. The main work of this dissertation is as follow:1. The author is aiming to construct a linear angular dynamics model of MUH in hover with its longitudinal and lateral dynamics coupled. First, by addressing the characteristics of the hover task, a continuous linear model based on mechanism is systematically derived. The model derived from mechanism reasoning is not suitable for control application for its high order. Therefore, the continuous model is discretized and using forward stepwise regression to reduce the number of items entering the final model. The model is derived from a systematic procedure by proposing the simplification assumption, deriving a linear model, discretizing, selecting items and estimating the parameters by backward stepwise regression, and testing the resulting model, proposing new assumption until we achieve a reliable model.2. We consider constructing a nonlinear model using Support Vector Machines (SVMs). We first give an introduction of Statistical Learning Theory and Support Vector Machines and its application to modeling. Then we choose the items to enter the model by considering the mechanism of the angular dynamics. At last, we give the comparison of the original and predicted output on the test data set which demonstrates its feasibility.