Adaptive Feedforward Control for Actively Isolated Spacecraft Platforms

Active vibration isolation systems are being considered to improve the performance of spacecraft instruments and sensors. Because of uncertainties inherent in on-orbit operation, adaptive control strategies and algorithms have relevance to these systems. In this paper, analysis of the algorithms, numerical simulation, and laboratory test data are used to evaluate adaptive feedforward control. Of particular interest are performance characteristics and limitations of the filtered-x LMS (FXLMS) algorithm and its finite impulse response (FIR) filter implementation. Combination feedback/feedforward control and the Augmented Error algorithm are two means investigated to extend the capabilities of FXLMS by desensitizing the algorithm to the specific dynamics of the plant. Several experiments were conducted on a laboratory testbed which serves as the prototype for a planned active vibration isolation flight demonstration. Adaptive Vibration Isolation The jitter requirements for certain spacecraft sensors and instruments are stringent, resulting in submicron motion specifications. In the presence of reaction wheels, motors, and other articulating devices, the level of vibration on the satellite may exceed requirements. Vibration isolation is one means of reducing the effect of motion at critical locations. A well-designed isolation system would begin with passive control and add active and finally adaptive control as needed to counter specific disturbance environments or meet particular instrument specifications. * [email protected]; 415-494-7351; www.csaengineering.com; [email protected] ©1997 by Eric H. Anderson and Jonathan P. How Vibration isolation systems separate vibration sources from vibration-sensitive components. There are two basic types of isolation, classified by the component being isolated: 1) instrument or payload isolation and 2) base isolation. Both options have potential use on-board spacecraft. The immediate motivation for this paper is payload isolation. Most isolators in use are compliant passive systems which support vibrating machines. Passive isolators effectively reduce transmission of high frequency energy, but are ineffective below the suspension frequencies of the isolated system. These mounts are also difficult to tailor to narrowband applications involving transmission of single or multiple tones. One extension is adaptive passive isolators — inherently passive devices adjusted periodically by an external control system. Examples include stiffening a mount to counter maneuvering loads in a vehicle, and then softening the mount during cruise. Adaptation hi this case adds versatility, but the approach is still constrained by fundamental performance limitations associated with passive isolation. Fully active isolation systems introduce fundamental advantages over their passive or adaptive passive counterparts. The transmissibility of the isolator can be tailored to selectively attenuate important inputs without the passive constraint relating isolation corner frequency and static deformation. Overall performance improvements can be significant. Active isolation is already common in terrestrial applications, but this paper is guided by the potential use on spacecraft, where the flight heritage is short and the number of applications is increasing. In one case, a three-axis active isolation system was demonstrated for a vibrating cryocooler [1]. Other active systems have been developed for microgravity isolation within modules of manned vehicles, on the Shuttle Orbiter [2] and the Space Station [3]. More recent systems [4] will isolate Copyright© 1997, American Institute of Aeronautics and Astronautics, Inc. specific spacecraft instruments. The disturbances on-board operational spacecraft include reaction wheels, thrusters, solar array drives, motors, solar arrays, cryocoolers, and other instruments. Their input may be large enough to disrupt operation of the more sensitive instruments and sensors onboard satellite imaging and communications systems. An active mount can isolate either the disturbance generator or the sensitive equipment. Unfortunately, the benefits of active isolation are partially offset by the potential stability and performance difficulties associated with implementing the controllers in a space environment that is challenging to simulate accurately on the ground. Furthermore, the disturbance environment is often poorly characterized. However, many of these issues can be overcome using adaptive control to modify an active control structure, function, or form to improve performance based on a new understanding of the system dynamics or properties (See Table 1). Adaptive control makes possible the full performance benefits of an active isolation system. The application of adaptive control to vibration suppression in general is increasing, driven by the availability of low cost digital signal processors (DSPs). Vibration isolation is a more tractable problem than general vibration control because of the potentially lower order dynamics and inherent limitations on the number of physical paths for energy flow. Sommerfeldt [5] approached adaptive isolation from an active noise control perspective. His initial laboratory demonstration used an off-line FIR filter to model the secondary path dynamics and an on-line FIR filter for control. Performance was demonstrated for single tones, two tones, and limited broadband frequency regions. This was later replaced with on-line estimation of the FIR models of MIMO systems [5]. In the simplest cases, he demonstrated that use of a pure lag to model the secondary path was adequate. The primary emphasis of his work was on the interaction between the on-line estimation and FXLMS algorithms. Spanos et al. [6] demonstrated isolation on a sixstrut laboratory hexapod system and are participating in a flight demonstration system [4]. Beginning from a feedback background, and then moving toward FXLMS algorithms, their approach makes use of accurate identified models of the inverse plant dynamics. Haynes et al. [7] used an adaptive FIR feedforward controller to minimize the acceleration error while simultaneously running a second FIR filter to model the plant dynamics. Relatively high order FIR filters were used to achieve an average of 15 dB attenuation between 50-250 Hz. Other promising recent work has applied a neural framework to vibration suppression Table 1: Applications for adaptive active isolation.