Description
Optimal Aircraft Engine Tuner Selection in Python
Presented by Jeffrey Armstrong
A numerical algorithm for designing on-board aircraft engine diagnostics has been implemented in Python. Employing the optimization techniques within SciPy, the code performs a search for an optimal vector of parameters for estimating engine variables, including exhaust temperatures and thrust. The algorithm exploits the numerical strengths of Python and SciPy for speed and interoperability.
Abstract
An emerging field of aircraft engine diagnostics is the inclusion of on-board engine performance tracking algorithms. These algorithms utilize data provided by a limited number of engine sensors to determine the current engine performance, which tends to degrade over time. However, estimating engine performance instantaneously is problematic due to the limited number of sensors normally available on a commercial aircraft engine.
One common practice is to estimate and track engine performance in software using a Kalman filter, a mathematical construct for tuning a numerical model to better track actual measurements (1). A new technique has been devised to optimize the design of this filter in aircraft engine applications (2). An optimization procedure to aid in the design of the filter has been implemented in Python and exercised against the significant number of minimization and optimization strategies available in SciPy. The talk focuses on the design of this optimization procedure in Python. The object-oriented nature of Python offers benefits over alternative numerical languages; speed, availability, and maintainability played central roles in the selection of Python as the implementation language. The availability of the multiprocessing module allowed for full utilization of modern multi-core CPUs, in contrast with often limited commercial numerical computing packages, further improving computational speed.
Some difficulties were encountered during this design exercise. Discussion of these obstacles and their eventual solution is presented. Specifically, iterative solvers for the discrete algebraic Riccati equation and the discrete Lyapunov equation had to be authored in Python (3,4). Additional framework for working with discrete state-space control systems was created, exploiting the object-oriented features of the language (5).
The Python implementation was able to verify the solution of the optimization problem. Comparison with an alternative, reference MATLAB implementation will be presented briefly. The results of this research is planned to be presented at the American Society for Mechanical Engineers Turbo Expo 2011 Conference in June, 2011 (6). The algorithm design in Python is meant to showcase the ability to perform controls engineering tasks in the Python language efficiently.
- “Kalman Filter,” Wikipedia: http://en.wikipedia.org/wiki/Kalman_filter
- Simon, D. L. and Garg, S., “Optimal Tuner Selection for Kalman Filter-Based Aircraft Engine Performance Estimation”. Journal of Engineering for Gas Turbines and Power. March 2010, Vol. 132.
- “Algebraic Riccati Equation,” Wikipedia: http://en.wikipedia.org/wiki/Alge braic_Riccati_equation
- “Lyapunov Equation,” Wikipedia: http://en.wikipedia.org/wiki/Lyapunov_equa tion
- “State Space,” Wikipedia: http://en.wikipedia.org/wiki/State- space
- Simon, D. L., Armstrong, J. B., "Application of an Optimal Tuner Selection Approach for On-Board Self-Tuning Engine Models," Proceedings of the ASME Turbo Expo 2011, GT2011-46408, 2011 (To Be Published).