Advanced Control Techniques For Efficient And Robust Operation Of Advanced Life Support Systems

This paper examines the structure and performance of three control strategies for a regenerative life support system constrained by mass balance equations. A novel agent-based control strategy derived from economic models of markets is compared to two standard control strategies, proportional feedback and optimal control. The control systems require different amounts of knowledge about the underlying system dynamics, utilize different amounts of information about the current state of the system, and differ in their ability to achieve system-wide performance goals. Simulations illustrate the dynamic behavior of the life support system after it is perturbed away from its equilibrium state or nominal operating point under the three different control strategies. The performance of these strategies is discussed in the context of system-wide performance goals such as efficiency and robustness. INTRODUCTION Part of the systems modeling research at NASA Ames Research Center has centered on the use of advanced control techniques to actively manage resources in advanced life support systems (ALSs). One important class of resources is the class of scarce, common-use resources. Power is an example of a scarce common-use resource—most ALS subsystems will use power to process and/or cycle mass, and for heating and cooling. Moreover, power in ALS systems is limited by the power generating capacity of the power plant. Management of ALS systems requires the integration of diverse yet tightly coupled system elements. Because system elements are coupled, scarce, common-use resources pose an additional challenge for ALS management systems. The added difficulty stems from the fact that system performance depends on meeting both subsystem and system-wide performance goals. For example, reference [3] focuses on the problem of eliminating system-wide power surges while meeting individual subsystem life support requirements under power constraints. The system-wide goal of eliminating power surges may reduce the required size of the power supply by reducing the need for excess capacity. It is with a view towards meeting performance goals, at both the subsystem and system levels, that we examine three control strategies aimed at intelligent resource allocation. The control strategies differ in their information structure—the amount of information necessary to calculate their controls. At one extreme is a proportional feedback control about individual system states. This is a completely decentralized information control strategy in that each system state is concerned with its own performance without regard to that of others. When the control task is such that independent subsystem operation can be tolerated, decentralized control is easy to implement and can integrate a diversity of system elements. However, one of the drawbacks of a highly decentralized control system is that it can be difficult to manage common-use resources like power and to achieve system-wide performance goals. These types of considerations require a certain amount of coordination that is not present in a decentralized information structure. At the other extreme is optimal control, which uses a global cost function to calculate feedback controls. This approach is centralized in the sense that information about all states is used in the calculation of each individual control. While in theory this approach perfectly integrates system elements to achieve a global system performance goal, in practice it is brittle (i.e. sensitive to uncertainty in or changes to system components) since it requires extensive knowledge of the control object. Furthermore, it can be computationally intensive, and it is not necessarily easy to specify a global cost function that captures the system performance criteria. Market based methods fall in between these two informational extremes. In the market based approach examined here a decentralized control strategy, in this case a proportional feedback around individual states, is supplemented with one or more signals of global scope. These signals are termed prices, and they communicate scarcity of common-use resources. The local controls are designed to respond to these price signals by adjusting their demand for the resources. In this way information about the global state of the system impacts individual control operation. The combination of centralized and decentralized elements make this approach more 'plastic' (tolerant to structural changes) while at the same time capable of addressing global performance concerns. However, this approach often requires balancing the costs of additional communication and computation against the global performance that can be achieved under a centralized information structure. The purpose of this paper is to examine example control systems with these three information structures and to describe their performance characteristics. The approach is simulation based. SYSTEM MODEL The systems modeling group at the NASA Ames Research Center has developed a suite of detailed simulation models of the BIO-Plex Advanced Life Support Test Bed [4] Here we consider a simplified mass balance model of the Air Revitalization System (ARS) (see Figure 1) as our simulation testbed for analyzing the performance of the different control systems. The key elements of the model are the crew chamber, which shares its atmosphere with the solids processing system; the biomass production chamber, which contains wheat grown with a 24 hour photoperiod and with a constant profile of crop ages; a solid polymer electrolysis (SPE) unit, which produces Oxygen from water; and two buffer tanks which hold Oxygen and Carbon Dioxide. The model tracks Oxygen and Carbon Dioxide only and does not account for system pressure. All flows are on a per hour basis. STATE EQUATIONS – The underlying dynamics of the system can be written x = f (x,u) where x represents the vector valued state and u represents the (vector of) inputs. Given an initial starting Figure 1 ARS Representation condition, the equations of state completely determine the evolution of the system and the mass in the various compartments of the model system at any point in time. There are seven state variables and seven controls. State variables: x1, x2 are the crew chamber molar fractions of O2 and CO2, respectively x3, x4 are the plant chamber molar fractions of O2 and CO2, respectively x5 is the amount (in mols) of O2 in the O2 tank x6 is the amount (in mols) of CO2 in the CO2 tank x7 is a moving average of light received by the crop in PPF Control variables: u1 is the molar flow of O2 from the O2 tank to the crew chamber u2 is the molar flow of crew air to the CO2 scrubber (we assume that the scrubber does not saturate.) u3 is the molar flow of plant air to the O2 scrubber (we assume that the scrubber does not saturate.) u4 is the molar flow of CO2 from the CO2 tank to the plant chamber u5 is the molar flow of feces to the SPS (incinerator) u6 is the molar flow of water to the SPE u7 is the light level (PPF) Parameters: h is the aggregate constant rate of human O2 uptake in mols/hr ( = 6) Vh is the volume (in mols) of the crew chamber (34366) Vp is the volume (in mols) of the plant chamber (16768) State equations: Vh x1 = u1 − 52.75u5 − h Vh x2 = 0.8894 h + 42u5 − x2u2 Vp x3 = 1.1 0.00294x4 (0.0002 + x4 2 ) 2 u7 − x3u3 Vp x4 = u4 − 0.00294x4 (0.0002 + x4 2 ) 2 u7 Eq. 1 x5 = x3u3 −u1 + 0.5u6 x6 = x2u2 − u4 x7 = −0.5x7 + 0.5u7 These equations of state (with the exception of state 7) represent a mass balance on the system of Figure 1. The relationship between plant chamber molar fraction of CO2, light level and CO2 uptake, given by: 0.00294x4 (0.0002 + x4 2 ) 2 u7 is derived from an energy cascade model of plant photosynthesis [5]. (add this reference) SYSTEM EQUILIBRIA – The equilibria of the system x e , u e occur when x = 0 . Although there are many equilibria, the requirements of human and plant physiology, as well as considerations of system buffering and operational margins, narrow the relevant number. For example, ( x , u = 0 ) is an equilibrium, albeit not a particularly interesting one. Consequently, choice of a nominal operating point is a key element in system performance. We consider the following equilibrium as the baseline or benchmark for the mode: x e = [0.23 0.0003 0.23 0.001 500 50