Team Name
Helpful Airplane Retune Reducing Intelligent System
Timeline
Spring 2020 – Summer 2020
Students
- Joshua Appenzeller
- Silvia Chapa
- Justin Cooper
- Elias Jester
- Alexander LeBoeuf
Sponsor
L3Harris Technologies
Abstract
H.A.R.R.I.S. developed a neural network that aims at more efficiently calculating specific parameters and deltas that are needed to accurately simulate aerodynamic models for training use. The advantage of neural networks is, ideally, only using a fraction of the time and effort needed to calculate these specific deltas, while also needing less interference from engineers. Currently, it is estimated that it takes around 3000 working hours between a team of engineers to tune the current system in place which yields a large amount of area for improvement.
Background
Currently, the Pilot in loop system is used to tune these deltas. Essentially, a set of initial parameters is set, and a specific delta or characteristic of the simulation is selected to be tuned. The pilot, engineer, and the system then go through a process of making incremental adjustments and checking if the change was closer or further from a realistic scenario. This is done until the problem has been eliminated. The process is then repeated for every remaining item to be tuned or issue that needs resolving not accounting for any situation in which solving one problem creates another or makes it worse. This system is time consuming, which is the problem our project aims to resolve.
Project Requirements
- System will export output in spreadsheet format
- Computation time shall be reduced compared to existing system
- System will be delivered in a virtual machine environment
- An integrated master schedule should be derived and followed
- Meeting Agenda, Minutes, and Presentation Materials shall be submitted according to meetings
- No engineers should be required for computation
- Source code shall be well-documented internally
- Artificial Intelligence/Machine Learning Methodology Report shall be submitted according to the integrated master schedule
System Overview
The objective of this project was to create a system to automatically tune these parameters and deltas without the need of a team of engineers and the amount of work hours that come with that. This will be accomplished by using a neural network and machine learning techniques to more effectively tune these simulations with minimal human interaction. A neural network requires some initial parameters, such as the aerodynamic model and conditions, to begin and will result in what should be a more accurate delta that the simulation would use.
Results
The neural network processes the initial parameters in order to derive a single variable. Utilizing sample data supplied by the sponsor, the network has been observed to produce accurate results 95-99% of the time.
Future Work
Due to the limited amount of time this project was allotted, not all types of aerodynamics tests were implemented. A future development would be to implement all aerodynamics tests and potentially the ability to calculate multiple variable deltas in a single run. In addition, a graphical user interface addition has been proposed in order to make the system more user friendly.
Project Files
System Requirements Specification
Architectural Design Specification
References
[1] Flight simulation training device initial and continuing qualification and use 14 cfr part 60 (2016).
[2] Saakaar Bhatnagar, Yaser Afshar, Shaowu Pan, Karthik Duraisamy, and Shailendra Kaushik. Prediction of aerodynamic flow fields using convolutional neural networks. Computational Mechanics, 64(2):525–545, Jun 2019.
[3] Peter R. Grant and Lloyd D. Reid. Protest: An expert system for tuning simulator washout filters. Journal of Aircraft, 34(2):152–159, 1997.
