*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Advisor: Dr. Mark Costello
Abstract
Optimization is a powerful tool for solving practical engineering problems, including identification for aerospace systems. Due to the complexity of many optimization problems, numerical methods are often employed to obtain solutions. A diverse collection of numerical optimization methods have been developed, each suited for different types of problems. Local search methods are highly efficient on convex problems, but require good initial guesses to find the global optimum. Global methods are able to search the entire parameter space to determine the global optimum, often at the cost of increased computation time. Given the fact that no one optimization algorithm performs well on every problem, the selection of the proper algorithm for a given problem is a critical decision by an engineer.
To overcome the algorithm selection problem, this work proposes a new method for automatically selecting and deploying optimizers, dubbed meta-optimization. The goal of meta-optimization is to intelligently deploy a diverse set of optimization algorithms, leveraging the strengths of each algorithm and minimizing their weaknesses in order to reliably and accurately solve challenging optimization problems with minimal user intervention. The first component of the meta-optimizer is the bank of optimization algorithms which includes numerous local and global search methods. Algorithm selection is performed in an online manner, choosing the most effective algorithms at the current stage of the solution process. The meta-optimizer must also ensure smooth transition between different algorithms and prevent premature convergence in local optima. Finally, the meta-optimizer tunes optimization algorithms parameters which degrade performance of the algorithm.
The meta-optimizer is first tested on a set of benchmark functions and then used to solve two aerospace system identification problems. The first problem considered is the estimation of parameters for a new smart projectile system. Both simulated and experimental spark range data is used to estimate the body and control mechanism parameters. The second application is the calibration of inertial measurement units (IMUs) for onboard sensing of precision guided air drop systems using simulated and experimental calibration table data.
Committee Members
