An Aircraft's Parameter Identification Method Based On Expectation Maximization Algorithm And Gaussian Smoother

Parameter estimation of flight dynamics is of great significance on various tasks during aircraft design process and gradually become a necessary part of designing a new aircraft. With higher requirements on accuracy and efficiency of estimation results, it is a necessity to build some new estimation algorithms.There are plenty of parameter estimation methods that can be used on identifying aerodynamic parameters. However, the results that obtained by most of these algorithms in practical problem are not good enough due to the following reasons: 1) The state estimator used in these algorithms are filters, which cannot completely get the useful information from the flight data, thus it may lead to inaccurate estimation results with nonlinear systems. 2) Flight dynamic models are more complicated than before with higher dimension states and larger number of unknown parameters, and the relation between sates and unknown parameters is complex which may bring difficulties to obtain accurate results with these methods.3) Most of these methods do not consider the process noise and measurement noise during parameter identification, this may lead to large errors of the estimated results.In order to solve these problems, a new algorithm is built to identify the flight dynamic model based on the expectation maximization(EM) algorithm and cubature Kalman smoother(CKS). EM algorithm is used to estimate the unknown statistics including initial values of means and covariance of the sates as well as the process noise and measurement noise for its numerical stability. CKS is a smoother which is more accurate and effective in estimating high dimension problems, and it is used to estimate the state and unknown parameters of the system. The validation of this method with real flight data shows that the algorithm built here has some advantages such as high accuracy and fast convergence.The research work and results are as follows:1) Build the algorithm frame based on the EM algorithm as well as the Gaussian smoother. EM is used to estimate the unknown statistics and Gaussian smoother is used to estimate the state and unknown patameters.2) Design of the parameter estimation method based on EM algorithm and CKS. Firstly, some research has been done on EM, several kinds of Gaussian filters and smoothing theories; Secondly, after comparing those filters mentioned before, choose CKS as the state estimator which combines cubature Kalman filter and RTS smoothing theory; Thirdly, the proposed method is built with the combination of EM and CKS, and the calculating steps of the algorithm is given.3) Validation of the proposed method on parameter identification with real flight data. A lateral linear model and a longitudinal nonlinear model have been used respectively to validate the proposed method and the estimate results show that the proposed method is effective on the flight dynamic parameter estimation problems.4) Comparison with other methods. The same models and flight data have been used with CKS as well as the algorithm based on EM and unscented RTS smoother and the corresponding results have been obtained. The comparison between these two methods and proposed method shows that for both problems the proposed method is of high accuracy and fast convergence. Although the calculation quantity of the proposed method is larger than some of the simple algorithms, it still holds great advantage in offline estimation area.