Virginia Space Grant Consortium Aerospace Graduate Research Fellowship Application Academic Year 2001-2002 Development of a Satellite Adaptive Attitude Controller Submitted by: Andrew J. Turner Virginia Polytechnic Institute and State University Advisor: Dr. Christopher Hall VSGC Grad Application Development of a satellite adaptive attitude controller Abstract In this project we will develop an adaptive neural network attitude controller for use in satellite missions. A network of synaptic nodes will ‘learn’ the satellite flight characteristics and necessary commands to achieve the mission requirements. Furthermore, the controller will adapt to physical uncertainties, and system failures using on-line training algorithms. The system will first be tested in a MatLab-Freeflyer computer simulation, and then on a dynamic air bearing spacecraft simulator. Following successful ground testing, the adaptive control system will be used in the control system for the Virginia Tech nanosatellite, HokieSat. Project Objective The principal purpose of this project is to develop an adaptive system which controls a satellite’s attitude while accounting for physical uncertainties and system failures. This objective is broken down into the following list of tasks: 1. Design an adaptive neural network controller that adequately controls a satellite’s attitude Using existing methods of neural network development [1], an adequate controller can be designed that will control the three axis angles, and three rotation rates of a satellite’s attitude [2][3][4][5]. However, unlike a traditional satellite, this controller is intended for use on a nanosatellite approximately 18 inches in diameter that uses three torque-coils for control. Traditional control laws are difficult to implement since torque-coils can only produce a moment perpendicular to the local magnetic field, which may not be in the desired control direction. Therefore the controller must determine a control strategy that will allow the spacecraft to obtain any attitude orientation, regardless of the local magnetic field. Andrew Turner 1 VSGC Grad Application Development of a satellite adaptive attitude controller Figure 1 shows a specific example of a neural network controller that uses an optimizer to determine the control inputs to the system [5]. The desired attitude is given by (t), and the optimizer, using the information from the neural network, determines the control signal u(t). The system response, y(t), and the expected response, yˆ (t ) , are differenced to produce the error, (t). (t) is used in a training algorithm that determines the appropriate neural network updates to produce an adequate control response. This updating allows for online training. This loop continues as the satellite is controlled while the neural network is continuously updated. u(t) System y (t ) (t) Neural Network Optimizer + yˆ (t ) e(t) + (t) Figure 1: Neural Network Controller Block Diagram 2. Determine system response to uncertainties and system failures By modeling the satellite system in a simulated environment such as MatLab, the neural network controller can be written in MatLab script and tested. This model will be used with the FreeFlyer satellite simulation interface to fully simulate the satellite’s orbit and attitude over the mission lifetime. Changing such parameters as atmospheric drag, moments of inertia, and control torques will allow the neural network response to these effects to be characterized. Andrew Turner 2 VSGC Grad Application Development of a satellite adaptive attitude controller 3. Develop network algorithms to employ on-line training and adapt to system changes The neural network controller shown above will use on-line adaptive, or “Self-Learning” algorithms to adjust the synaptic node weights. However, the back-propagation algorithms that determine the actual values of the weight updates between successive time steps must be determined and optimized. A set of tests will be developed that can fully characterize the controller response to system failures. Using this characterization, the online training algorithms will be modified to produce a satisfactory satellite response to the simulated uncertainties and failures. 4. Test controller on a ground-based spacecraft simulator Colleagues at Virginia Tech are currently developing a Distributed Spacecraft Attitude Control System Simulator (DSACSS). The simulator uses a network of hardware systems, including computers, sensors, and actuators, mounted on three-axis air-bearing tables to model an orbiting satellite. The adaptive control system developed in this project will be tested using the DSACSS. The simulated network will be implemented in C/C++ and installed on the flight computer. A series of tests will be designed to measure the control effectiveness, and to characterize the response to system failures and uncertainties. These results will be compared to the simulated results produced in the MatLab-Freeflyer code. 5. Determine controller effectiveness on an actual nanosatellite Virginia Polytechnic Institute and State University is building a 15-kg nanosatellite, HokieSat, which will launch from the space shuttle in 2002. Its mission is to demonstrate formation flying with two other nanosatellites being built by the University of Washington and Utah State University, as well as to collect ionospheric measurements using GPS and 2 types of Andrew Turner 3 VSGC Grad Application Development of a satellite adaptive attitude controller electron probes [6][7]. Following successful completion of the ground testing of the controller, the system will be installed on HokieSat and tested during the mission. Justifications The majority of current satellite control technologies employ classical methods that are determined a priori. However, due to changes in the satellite system, or unknown flight characteristics, these systems may prove to be unreliable. Adaptive control systems provide a robust means of controlling a satellite in the face of these uncertainties and failures. In this research project we will develop a control system that will adapt to the satellite’s flight characteristics. By using ‘soft-computing’ techniques such as neural networks and genetic algorithms, the control system can learn and adapt to changes in the controllability of the satellite. The use of an attitude control system can greatly reduce the costs of satellite development. An adaptive control system can learn from a model of the satellite and from the actual flight mission instead of having to specify the classical control law. Furthermore, a welldesigned neural network can be reused for many missions, without having to change the base structure of the system. This reuse can greatly reduce the cost of having to develop a control system from the ground up. Also, the adaptive control system can recover from system failures that may otherwise cripple a spacecraft such as Mariner 2 and PanAmSat Corporation's Galaxy 4 [Space News, 25-31 May 1998, p. 3]. The robust satellite can recover from these failures without having to wait for human control and updating of its software. References Andrew Turner 4 VSGC Grad Application Development of a satellite adaptive attitude controller 1. D.H. Nguyen, B. Widrow. “Neural Networks for Self-Learning Control Systems.” International Journal of Control. Vol. 54, No. 6, pp. 1439, 1991. 2. P.J. Antsaklis, K.M. Passino, et al. “Towards Intelligent Autonomous Control Systems: Architecture and Fundamental Issues.” Journal of Intelligent Robotic Systems. Vol. 1, pp. 315-342, 1989. 3. K. KrishnaKumar. “Adaptive Neuro-Control for Spacecraft Attitude Control.” Neurocomputing, Vol. 9, No. 2, pp. 131-148. Oct 1995. 4. J.J. Sheen, R.H. Bishop. “Adaptive Nonlinear Control of Spacecraft.” Proceedings of the American Control Conference. Baltimore, Maryland, pp. 2867-2871. June 1994. 5. J.R. Noriega, H. Wang. “A Direct Adaptive Neural-Network Control for Unknown Nonlinear Systems and Its Application,” IEEE Transactions on Neural Networks, Vol. 9, No. 1, pp. 2734. Jan 1998. 6. M. Campbell, R. R. Fullmer, and C. D. Hall, "The ION-F Formation Flying Experiments," AAS/AIAA Space Flight Mechanics Meeting, Clearwater, FL, Jan 23-26, 2000. 7. N. Davis, J. DeLaRee, C. D. Hall, W. L. Stutzman, and W. A. Scales, "Virginia Tech Ionospheric Scintillation Measurement Mission," AIAA/Utah State University Conference on Small Satellites, Logan, UT, Aug 31- Sep 3, 1999, Paper SSC99-III-3. Andrew Turner 5 VSGC Grad Application Development of a satellite adaptive attitude controller Applicant: Andrew James Turner Education M.S. in Aerospace Engineering, Virginia Tech, expected May 2002 B.S. in Aerospace Engineering (minor-Computer Science), University of Virginia, May 2000 Research and Experience University Nanosatellite Program design team, Virginia Tech, Blacksburg, Virginia. HokieSat Attitude Dynamics & Control Team. Jun 2000 – Present Attitude Dynamics and Control Research Assistant, Virginia Tech, Blacksburg, Virginia. Fall 1998 – Present Software Development, Analytical Graphics Inc., Malvern, PA. May 1998 – Aug 1998 Research Assistant, Dynamics and Control System Laboratory, University of Virginia, Charlottesville, Virginia. Spacecraft & Robotics Simulation & Testing. Dec 1997 – Sep 1998 University of Virginia Solar Airship Project, Charlottesville, Virginia. Control System Lead, Program Lead, Vice President of Engineering. Oct 1996 – May 2000 Conference Publication “Development of a Semi-Autonomous Control System for the UVA Solar Airship Aztec,” 3rd Annual Airship Convention, Friedrichshafen, Germany, Jul 2000. Awards and Honors Research & Design Symposium Finalist. May 2000 Engineer in Training. Designated Apr 2000 Raven Society Scholar. Apr 1999 University of Virginia Academic Dean’s List. Jan 1998 – May 2000 University of Virginia Solar Airship - Most Outstanding Engineer. May 1999, 2000 Rodman Engineering Honors Student. Sept 1996 – June 2000 Plan of Study Spring 2001 Applied Numerical Methods Advanced Orbital Mechanics Virtual Environments Fall 2001 Advanced Aerodynamics/Hydrodynamics Vehicle Structures Optimization Techniques Spring 2002 Optimal Control Advanced Spaceflight Dynamics Introduction to Artificial Intelligence Andrew Turner 6