Browsing by Author "Meade, Andrew J."
Now showing 1 - 4 of 4
Results Per Page
Sort Options
Item A Comprehensive Probabilistic Framework to Learn Air Data from Surface Pressure Measurements(Hindawi, 2015) Srivastava, Ankur; Meade, Andrew J.Use of probabilistic techniques has been demonstrated to learn air data parameters from surface pressure measurements. Integration of numerical models with wind tunnel data and sequential experiment design of wind tunnel runs has been demonstrated in the calibration of a flush air data sensing anemometer system. Development and implementation of a metamodeling method, Sequential Function Approximation (SFA), are presented which lies at the core of the discussed probabilistic framework. SFA is presented as a tool capable of nonlinear statistical inference, uncertainty reduction by fusion of data with physical models of variable fidelity, and sequential experiment design. This work presents the development and application of these tools in the calibration of FADS for a Runway Assisted Landing Site (RALS) control tower. However, the multidisciplinary nature of this work is general in nature and is potentially applicable to a variety of mechanical and aerospace engineering problems.Item Development of a Meshfree and Matrix-Free Method for Compressible Computational Fluid Dynamics(2021-04-30) Villarreal, Javier Alejandro; Meade, Andrew J.Meshfree methods aim to approximate accurate and efficient numerical solutions to partial differential equation problems without the necessity for a mesh, relieving some of the difficulties associated with mesh generation. In this thesis, a hybrid method is presented which alternates between an implicit sequential function approximation scheme and an explicit Runge-Kutta time marching scheme. In the sequential function approximation scheme, a solution is improved incrementally as a summation of Gaussian radial basis functions (RBFs) which have been optimized using a genetic algorithm. Spatial differentiation is performed under both schemes using a meshfree, local differential quadrature method. This thesis focuses on the governing equations for compressible flow problems relevant to aerospace applications, namely aerodynamics. Consequently, numerical methods from finite difference and finite volume methods are adapted to the meshless scheme to account for characteristics particular to these governing equations. The method presented is used to approximate solutions for some common validation cases and the results are compared with those found in the literature.Item Radial Basis Function Neural Network System Identification and Model Prediction Control for Aircraft(2021-04-30) Smith, Graham; Meade, Andrew J.This thesis details efforts made to perform system identification and control of aircraft using a self-assembling radial basis neural network method. First, Sequential Function Approximation (SFAX) is presented as an efficient, accurate and reliable tool for system identification and used to create and train a neural network model of an aircraft’s longitudinal and lateral-directional dynamics, using only input and output data. This is demonstrated on data from a small unconventional UAV design called ICE as well as from the T-38 jet flight test aircraft. Then, this neural network model is applied to a model predictive control scheme using MATLAB optimization routines. Finally, SFAX is used to build a direct inverse neural network controller.Item Use of Active Learning to Design Wind Tunnel Runs for Unsteady Cavity Pressure Measurements(Hindawi, 2014) Srivastava, Ankur; Meade, Andrew J.Wind tunnel tests to measure unsteady cavity flow pressure measurements can be expensive, lengthy, and tedious. In this work, the feasibility of an active machine learning technique to design wind tunnel runs using proxy data is tested. The proposed active learning scheme used scattered data approximation in conjunction with uncertainty sampling (US). We applied the proposed intelligent sampling strategy in characterizing cavity flow classes at subsonic and transonic speeds and demonstrated that the scheme has better classification accuracies, using fewer training points, than a passive Latin Hypercube Sampling (LHS) strategy.