Middeck Active Control Experiment
The Middeck Active Control Experiment (MACE) is a United States Space Shuttle flight experiment manifest for launch on STS-67 in February, 1995. MACE (Figure la) was designed by the Space Engineering Research Center at the Massachusetts Institute of Technology, in collaboration with Payload Systems Incorporated, the NASA Langley Research Center, and Lockheed Missiles and Space Company. The goal is to explore approaches to achieving high precision pointing and vibration control of future spacecraft and satellites. In particular, MACE extends the bandwidth of conventional rigid body instrument pointing and attitude controllers to include the flexible modes of the satellite. Since the success of such flexible control is intimately dependent upon the accuracy of the spacecraft model used for control design, MACE is essentially a spacecraft modeling validation effort where success is determined by the control performance and predictability that is achieved in earth orbit. MACE builds upon the concept of the Middeck 0-Gravity Dynamics Experiment (MODE), which flew on STS-40, STS-48 and STS-62 as a dynamics test facility to characterize fluid, Space Station structure, and crew motion dynamics in zero- gravity. MACE augments the MODE facility with real-time, digital control capabilities.
MACE Test Article to be floated in the Shuttle Middeck.
Two instruments containing rate gyros are mounted to either end of the flexible structural bus using two-axis, direct drive gimbals. A three axis reaction wheel assembly (RWA) and a two axis piezoelectric bending strut are located near the center. Each strut is instrumented with strain gauges while each gimbal axis contains an angle encoder. A power and data umbilical, extending from the RWA to the bottom right corner of the photo, connects the actuators and sensors to the Experiment Support Module (ESM) shown in Figure lb. The ESM contains all experiment sequencing, real-time control, signal conditioning, power amplification, data storage, and crew interface functions in a standard middeck locker. The real-time control supports 20 sensors, 9 actuators, and 80 state compensators at a control rate of 500 Hz. The control objective is to maintain the inertial pointing of one instrument while the other is undergoing either broad or narrowband excitation and slew maneuvers. Since, in the past, control system performance has been limited by the flexibility in the system (due to a phenomenon known as control - structures interaction), MACE represents a space flight validation of new technology which has the potential for revolutionizing the performance of space-based systems which cannot afford the massive, rigidizing support structure that would otherwise be required. This translates into earth- observing instruments with dramatically improved imaging capabilities for monitoring earth resources, pollution, ozone depletion and meteorological and oceanographic patterns. This technology also benefits astronomical instruments envisioned to detect objects at the edge of the universe and locate planets around other stars. More down to earth, this technology is already finding application in aircraft gust alleviation, computer disk drive head vibration suppression, noise control, sensitive instrument isolation, and precision machining. However, acceptance in the spacecraft community requires a comprehensive flight validation of this technology. Hence, the MACE program is designed to provide this flight validation.
Figure 1b. The MACE Electronic Support Module mounted in a standard middeck locker.
History of Control-Structures Interaction
Control-Structures Interaction (CSI) occurs when control detrimentally interacts with flexibility in the system. Such interaction is caused by mismodelling or lack of consideration for flexibility. The U.S. Space Program has a history of problems related to CSI, which have ranged from degrading spacecraft performance to causing catastrophic loss of the system. Problems with spacecraft flexibility started as early as the first U. S. satellite, Explorer I (1958). Unexpected energy dissipation in the flexibility of the four whip antennas on the spin stabilized satellite caused it to tumble. Attitude oscillation caused by the control system interacting with boom and solar panel flexibility occurred on OGO III(1966), OVI-lo (1966), DMSP (1972), and Mariner 10 (1978) which was almost lost. Thermal warping and snapping proved a major source of agitation in Alouette I (1962), Explorer XX (1964), OGO IV (1966), Voyager (1977), and Landsat (1982, 1984). Unstable interaction between the control system and liquid propellant slosh modes occurred on Leasat (1984). Flexibility in the docking unit between the Apollo Command Module and the Lunar Module was modeled in order to ensure that the gimbaled thruster would not cause instability during trans-lunar injection. Approximately half the operational time of the shuttle Remote Manipulator System (RMS) is spent waiting for flexible motion to decay. Finally, pogo, which is an unstable interaction between thrust and compressibility in the propellant system, has plagued many launch vehicles. The contemporary solution to the CSI problem has been to analyze the system and limit the performance or bandwidth of the control to not include this flexibility. Therefore, flexibility has placed a limit on the performance of systems, particularly in space where rigidizing structure is obtained at high launch cost. Therefore, any excitation of the flexibility, due to thermal snapping of solar arrays, bearing noise or imbalances in reaction wheels, or scanning and slewing of instruments and manipulators, directly degrades performance. This was the case for many recent spacecraft such as Hubble, UARS, and the Shuttle RMS. The MACE program explores Controlled Structures Technology (CST) as a means for controlling rather than avoiding flexibility in space systems, thereby penetrating this artificial performance barrier. An extensive survey of historical occurrences of CSI (The Batelle Report, March 1989) culminated in the following recommendations:
- Structural analysts need more accurate/less computationally intensive models.
- Low-gravity test capability is nonexistent. Multibody testing is difficult. Testing needs to be done in a closed-loop fashion.
- The capability to experiment in space and qualify hardware and control techniques is crucial.
MACE has responded to these recommendations.
Challenge of Controlled Structures Technology
Many CSI problems are not identified until after the system has been placed in operation. At this point, there is little that can be done to alleviate the problem. The concept of Controlled Structures Technology (CST) is to explicitly consider and control the structural flexibility in the system and to do so at an early stage in the design process when many more solutions are available. CST represents a marriage between high fidelity dynamic modeling and robust multivariable control system design. The more accurate the control design model, the more performance that can be achieved. The more robust the control, the larger the inaccuracies in the model that can be tolerated by the control in achieving improved performance. The challenge is to strike the appropriate balance between model refinement and robust control, particularly for a system which can only be tested in an environment (ground) other than that in which it will operate (space).
MACE Objective and Approach
The objective of MACE is to act as a pathfinder for a qualification procedure for flexible, precision-controlled spacecraft. For future vehicles which cannot be dynamically tested on the ground in a sufficiently realistic zero-gravity simulation, this procedure will increase confidence in the eventual orbital performance of such spacecraft. In the overall approach, illustrated in Figure 2, both finite element and measurement modeling techniques have been investigated to determine the advantages and limitations of each. A 1-g finite element model was developed which includes gravity and suspension effects. The accuracy of the FEM is improved through modal identification and model updating (Step A). Closed-loop updating (Step B) is performed because this improves the model from the perspective of control. Often, small errors in the open-loop dynamics can lead to large discrepancies between the experimental and predicted closed-loop behavior. A similar process is performed using generally more accurate 1-g measurement models which are obtained by fitting a state space system to the transfer functions measured through the control hardware. These measurement models tend to provide accurate estimates of the modal parameters of the test article, which can be used to further update the physical parameters in the 1-g FEM (Step C). Controllers are also designed based on the measurement models and, by comparing the performance obtained with that achieved using the finite element based controllers, the designer can understand the cost-benefit of further FEM refinement (Step D). One key advantage of a finite element model is that it is developed using analytic techniques, and thus can be used to predict the on-orbit system behavior (Step E). Note that the finite element updates are performed on the physical parameters of the model, which enables an explicit removal of gravity and suspension effects. This approach is in contrast to updating a particular 1-g state space model which implicitly contains the gravity and suspension effects. Of course, one would expect a variety of errors to still remain in the finite element model predictions for 0-g, and thus the need for robust control.
The activities listed in the bottom half of the approach figure occur during the mission. The 16 day Shuttle Endeavor STS-67 mission will contain six MACE operation days split into three main phases. A system identification will be performed during the first phase to obtain time response data which will be downlinked over the Ku-Band system. The data will be used to assess the accuracy of the FEM predictions and to develop 0-g measurement models. In the second phase, numerous controllers that have been designed prior to flight using the 0-g FEM will be implemented. During this period of the mission (approximately 72 hours), new 0-g models and robust controllers will be developed on the ground. These compensators will then be uplinked and implemented during the last phase of the experiment. The results of the preprogrammed controllers will enable an assessment of the accuracy of 0-g FEM predictions as they pertain to precision control. The control redesign based on the measurement models (Step F) will help identify the limitations of predicting 0-g closed-loop behavior from analysis and ground testing and the performance benefits that can be realized through on-orbit identification and control redesign. This discussion demonstrates the level of interaction between model development and control design. In the process, it also illustrates the need for an efficient control design methodology.
Contributions of the MACE Program
Space Systems The performance achieved by controlling the flexibility in the system is compared to standard industry practice where instrument pointing servo control is closed with a bandwidth roughly equal to one tenth of the frequency of the first flexible mode. The MACE program extends this bandwidth, and therefore performance, in two steps. First, the bandwidth of the instrument pointing servos are increased while maintaining 6 db of gain margin and 30 degrees of phase margin. Increasing the bandwidth to just above the frequency of the first flexible mode results in an order of magnitude improvement in inertial payload pointing. Second, dynamic CST compensation is closed around the extended bandwidth servo control to achieve an additional order of magnitude improvement in inertial payload pointing. The measured 1-g performance of this second layer of control is shown in Figure 3 by comparing the open and closed-loop performance autospectra. In all, an approximate 40 db improvement in inertial payload pointing has been achieved in 1-g tests over that obtained through standard industry practice.
Figure 3. Open Versus Closed-Loop Pointing Performance in 1-g.
The extended bandwidth servos, along with the CST control, are achieved at a cost: increased sensitivity to modeling errors and time varying dynamics. However, the MACE program strives to minimize the resulting impact on performance through two operational means. First, extensive ground testing is combined with analysis to derive accurate models of 0-g behavior and develop models of the residual (deterministic) errors along with the remaining uncertainty (stochastic) bounds. Second, the MACE program realizes that there is no substitute for test data in the actual operational environment to aid in maximizing closed- loop performance. Therefore, a comprehensive identification and control redesign will be performed during the STS-67 mission. Such on-orbit redesign could be used for future spacecraft to aid in maximizing performance, working around unexpected problems, and adapting to changing environmental conditions.
Structural Modeling Analytic (FEM)
In order to predict the O-g behavior of MACE, the effects of gravity and the suspension system on the structure have been identified and modeled. These include geometric effects such as sag, pretensioning of the structural members, suspension effects, gravity effects on the sensors and actuators, and nonlinear effects. The resulting 1-g model was then used to construct an input-output model of the structure. Since this model was used to design modern state-space controllers, it had to include all the sensors and actuator dynamics, the reaction and gimbal servo controls, and the effects of using a digital controller. Initially, this model was significantly in error when compared to ground data. But, through a process called updating, the correlation of the analytical model with experimental data was improved (Figure 4). Once sufficient convergence was achieved in the update process, the gravity and suspension effects were removed from the model to arrive at a prediction of the O-g behavior of MACE. An integral part of this O-g model is a prediction of the errors likely to be present in 0-g. Since no amount of updating can eliminate all the errors in the analytical model, the residual error in the 1-g model must be propagated into 0-g. These 0-g error predictions are important because they give the control designers an idea of how much robustness is required. These errors have two components: deterministic errors and stochastic uncertainties. Multiple sets of data are used to extract these two sets of information. The average difference between measured and modeled frequency is a deterministic error. The variation, across the data sets, from this deterministic error is the stochastic uncertainty which results from test article reassembly, nonlinearity, etc. Both component errors are propagated from 1-g to 0-g using a gravity perturbation procedure. Although various people in the past have identified gravity effects on structures, no one has identified and modeled them to the same extent as MACE. Also, seldom have input- output models based on the finite element method been constructed to the kind of fidelity that the MACE team has achieved. This fidelity is demonstrated in the sheer complexity of the input-output model, which retains 40 modes below 200 Hz and has 9 actuators and 20 sensors. The accuracy of this analytical model rivals that of some measurement based techniques. Perhaps the most important and useful achievement of the MACE modeling team is the development of a concrete modeling process whose products are a high quality prediction of the 0-g behavior of a structure and a prediction of the residual error likely to be in the 0-g model, both of which are suitable for modern state-space control design. Measurement Model and System Identification In the MACE program, the following identification techniques are studied:
- Eigenstructure Realization Algorithm (ERA) and Observer/Kalman filter IDentification (OKID) technique developed at NASA Langley Research Center;
- Q-MARKOV COVariance Equivalent Realization (Q-Markov Cover) algorithm developed at Purdue University;
- State Space Frequency Domain (SSFD) identification technique developed at Jet Propulsion Laboratory at the California Institute of Technology;
- Multi-input and multi-output Output Error State sPace (MOESP) model identification technique developed at Delft University of Technology in Netherlands;
- classical Least Squares algorithm for multi-input and multi-output time series model applied to flexible structures at Wright-Patterson Air Force Base; and
- Observability Range Space Extraction (ORSE) identification technique which is an extension of the ERA and Q-Markov Cover algorithms.
The studies revealed the following common feature of these techniques: to achieve high modeling accuracy, these techniques need to overparameterize the model. To obtain low, minimal order and accurate models for control design, an integrated identification technique, which combines the Frequency domain Observability Range Space Extraction (FORSE) algorithm, LOGarithmic Least Squares (LOGLS) model updating algorithm and a balanced model reduction algorithm, was chosen to generate measurement models for MACE. The unique feature of this integrated technique is that it can use a low order model to achieve the modeling accuracy which normally requires a much higher order model with significant overparameterization. Figure 4 demonstrates the accuracy that this technique can achieve using a low order model.
MACE has been used to evaluate the performance and robustness of a variety of modern control formulations. These include Linear Quadratic Gaussian (LQG) control, Sensitivity Weighted LQG (SWLQG), Parameter Robust LQG (PRLQG), Neo-classical control, Maximum Entropy, Multiple Model, Hinfinity, mu-synthesis, Popov, etc. This evaluation has led to various results which include: identification of the conservatism in robust control; the development of a design process using several of these formulations; a downselect of the various formulations based upon their applicability to and effectiveness in large order, experimental hardware. Two basic control evaluation procedures have been developed under the MACE program. The first is measurement based while the second is analytical. The measurement- based procedure uses open-loop, measured data, in conjunction with the designed control algorithm, to calculate the predicted closed-loop response for performance, the multivariable Nichols plot for stability, and the singular values of the output error transfer function for sensitivity to complex parameter uncertainty. The second approach involves mixed mu analysis whose framework permits the investigation of stability and performance of a system under a range of complex and real parameter uncertainties. This test is particularly useful for analysis of 0-g control designs for which measurements are not available prior to flight. The application of mixed mu analysis on MACE is unique in two respects: the scale of the problem (30 uncertainties in a 100 state system) and the use of a scaled sensitivity transfer function as the performance measure. In addition to feedback control techniques, MACE has been used to evaluate the performance of feedforward command shaping techniques. When a disturbance is generated onboard in a deterministic manner, such as instrument scanning and manipulator slewing, the effects can be reduced at the source by shaping the commanded input to minimize the residual vibration that is excited. Unlike feedback control, these command shaping techniques rely solely on system models rather than sensor measurements to reduce vibration in flexible structures. Consequently, these techniques can provide a simple and effective control alternative for systems with inadequate sensor feedback. Additionally, for systems with adequate sensor information, command shaping methods can be used along with feedback control to further improve vibration suppression and reduce actuator energy requirements. Several different feedforward command shaping methods have been evaluated for use on MACE. These include the Input Shaping, Inverse Dynamics, and LQG Trajectory Computation. As a result of this evaluation, optimal feedforward command shaping strategies have been identified, and a new approach for deriving optimal shaped actuator commands has been developed.
Oz operating the ESM
MACE FM in second configuration
MIT SERC has been very careful, through the MACE program, to conduct a comprehensive and unbiased evaluation of the control community's leading model development and robust control formulations as applied to the control of precision spacecraft. In doing so, SERC has developed model development and control design procedures which incorporate the best attributes of several of the leading techniques and has made the computational algorithms more mature and efficient due to the needs of the MACE hardware. In addition, MACE has developed not only an experiment but also a 0-g flight test facility for Shuttle and Space Station capable of conducting dynamics and control experiments on a diverse variety of test articles in the micro-gravity environment of earth orbit.
Liu, K. and Miller, D. W., "Identification of Structure Systems Using the ORSE Identification Technique," accepted to the ASME Journal of Dynamic Systems, Measurement, and Control, April, 1994.
How, J., Glaese, R., Grocott, S., and Miller, D., "Finite Element Model Based Robust Controllers for the Middeck Active Control Experiment (MACE)," presented at the 1994 American Control Conference and ASME Journal of Dynamic Systems, Measurement and Control, Baltimore, MD.
How, J. P. and Miller, D. W., "Assessment of Modelling and Robust Control Techniques for Future Spacecraft: Middeck Active Control Experiment," submitted to the Journal of the Astronautical Sciences, June 5, 1994.
Grocott, S., How, J., Miller, D., MacMartin, D. and Liu, K., "Robust Control Implementation on the Middeck Active Control Experiment (MACE)," accepted to the MAA Journal of Guidance, Control, and Dynamics, Dec., 1993.
Grocott, S. C. 0., How, J. P., and Miller, D. W., "A Comparison of Robust Control Techniques for Uncertain Structural Systems," presented at the AIAA Guidance, Navigation and Control Conference, Scottsdale, AZ August 1-3, 1994.
For further information please contact:
Dr. David W. Miller (co-Principal Investigator)
Massachusetts Institute of Technology
77 Massachusetts Ave
Cambridge, MA 02139
TEL. (617) 253-3288
FAX. (617) 258-5940
Mr. Gregory Stover (Technical Monitor)
NASA Langley Research Center
Hampton, VA 23681
TEL. (804) 864-7097
FAX. (804) 864-7009
Dr. Javier de Luis (Subcontractor)
Payload Systems Incorporated
270 Third St.
Cambridge, MA 02142
TEL. (617) 868-8086
FAX. (617) 868-6682