Project 2: System Identification of Rotorcraft
Faculty Mentor:
Dr. Kelly Cohen
Associate Professor, School of Aerospace Systems
University of Cincinnati
Office: 735 Rhodes Hall
E-Mail:
Phone: (513)-556-3523
Graduate Student Mentor:
Mr. Wei Wei
Graduate Student in Aerospace Engineering
Office: 518 Old Chem
Email:
Phone: 513-556-9969
Project Summary
Both wild-land and structural fires can be disastrous resulting in loss of life and property. One of the approaches to enhance our ability to fight fires is to extract real-time information using unmanned aerial vehicles (UAVs) to obtain and augmented situational awareness which in turn will lead to improved decision making. In this project, we are looking into exploiting UAV/quadrotor technology for structural fire-fighting as shown in Figure 1. It’s capabilities include looking through windows of high rise buildings and seeing through smoke. This solution incorporates the following attributes: ease of operation; safety; night operations; and self-preservation.
Figure 1: Quadrotors Operating in a Structural Fire
In order for the quadrotor UAVs to be effective in this rough environment, effective autonomous flight control systems are required. Given the complex dynamics of a quadrotor, it is highly desirable to develop a model based flight controller. The 6 DOF (six degree of freedom) dynamic model is not sold with the off-the-shelf systems and it needs to be developed. An effective method, which will be utilized in this project, to obtain these dynamic models is called System Identification.
The AR Parrot drone has served as a benchmark in 2012 we will continue using it as a preliminary stage in 2013 for study purposes as we apply our skills to obtain the dynamic model of a far more effective home-built quadrotor. The main goal of this project is to apply state-of-the-art System Identification techniques to develop the dynamic model of the radio-controlled home-built Quadrotor Drone system. The research objectives are as follows:
· Objective 1 – Students will study the flight characteristics of the AR drone which will include dynamics of rotorcraft and learning to fly this radio controlled air vehicle.
· Objective 2 – Students will utilize an experimental process called "System Identification" and the CIFER software to develop the dynamic model of the AR Parrot Quadrotor Drone as a benchmark.
· Objective 3 – Students will apply the lessons learned on the AR Parrot Quadrotor Drone and apply them to developing the dynamic model of the effective (in terms of payload & endurance) home-built Quadrotor UAV.
· Objective 4 – Students will prepare a detailed engineering flight test and dynamic modeling report which will be presented to our collaborators in Industry (Modus Recte Inc).
Brief description of the equipment and computer resources to be used by the students
The AR Drone (see Figures 2-3) will be utilized for the flight tests.
Figure 2: UC students piloting the AR Parrot Quadrotor Drone (Summer REU, 2012)
Figure 3: AR Parrot Quadrotor Drone - Description of main parts
http://ardrone.parrot.com/parrot-ar-drone/usa/
The current AR Parrot Quadrotor Drone system, described in Figures 2 and 3, is radio-controlled using an iPAD or an iPhone. Within the framework of research at Most Aero Labs, under the direction of Dr. Kelly Cohen, autonomous flight controllers for morphing rotorcraft are being researched. The AR Parrot drone is a benchmark and serves as a simple and effective means to develop dynamic models and the appropriate controllers. In this REU we will focus our efforts on developing a dynamic model for the AR quadrotor drone as well as a new home built effective UAV quadrotor using an experimental process called "System Identification".
System identification is a procedure by which a mathematical description of vehicle or component dynamic behavior is extracted from test data. System identification can be thought of as an inverse of simulation. Simulation requires the adoption of (a-priori) engineering assumptions to allow the formulation of model equations. These simulation models are then used to predict aircraft or subsystem motion. In contrast, system identification begins with measured aircraft motion and "inverts" the responses to rapidly extract a model which accurately reflects the measured aircraft motion, without making a-priori assumptions or requiring a time-consuming modeling effort.
In most cases, these system identification models were used for flight control design even when physics based models were available. The methodology is well suited to rotorcraft identification due to its insensitivity to uncorrelated output noise (which produces a bias in time-domain methods), and its ability to identify unstable dynamics (which is also difficult in the time domain due to divergence). The reasons for this include:
· A model based on flight data will better match the dynamics of the actual vehicle.
· System identification is more time efficient than attempting to correct the physics based model to better match flight data.
· The identified model provides additional physical insight to the control designer (which in many cases is later used to correct the physics based model).
· Uncertainty data are readily available for the identified model.
The system identification methodology has four main steps:
1. Frequency response identification
2. State-space model fitting to the MIMO frequency response database
3. Model structure determination
4. Time domain verification.
In this REU program, we will be using a software package, CIFER®, for the System Identification process (see schematic descriptions in Figure 4). The U.S. Army and University of California, Santa Cruz (UARC) have jointly developed an integrated facility for system identification called CIFER®. CIFER® is based on a comprehensive frequency-response approach that is uniquely suited to the difficult problems associated with flight test data analysis. The foundation of the CIFER® approach is the high-quality extraction of a complete multi-input/multi-output (MIMO) set of non-parametric input-to-output frequency responses. These responses fully characterize the coupled characteristics of the system without a-priori assumptions. Advanced Chirp-Z transform and composite optimal window techniques developed and exercised with over 10 years of flight project applications provide significant improvement in frequency-response quality relative to standard Fast Fourier Transforms (FFTs). Sophisticated nonlinear search algorithms are used to extract a state-space model which matches the complete input/output frequency-response data set. Application modules within CIFER® allow the:
· Rapid identification of transfer-function models
· Spectral signal analysis
· Handling-qualities and classical servo loop analysis
· Time and frequency-domain comparisons of identification versus simulation model predictions.
The system identification methodology developed in the MOST Aero labs and used for the AR drone is presented in Figure 5.
Figure 4: CIFER® - Schematic Description
http://uarc.ucsc.edu/flight-control/cifer/index.shtml
Figure 5: System Identification Methodology