(342c) Using Simulation-Based Optimization Approach for Space Missions' Process Synthesis and Design

Authors: 
Aydogan, S., Purdue University
Orcun, S., Purdue University
Blau, G., Purdue University
Reklaitis, G. V., Purdue University


A computational architecture, SIMulation-based OPTimization (SIMOPT) that combines combinatorial optimization within a simulation environment, is proposed to solve the process synthesis and design problem for a life support system for manned space missions. Life support systems provide a habitat that ensures health, safety and maximum efficiency of crewmembers during the mission. Two applications, Mars Surface Habitat Lander and the evolution of a Mars Base, will be presented to demonstrate the use of the proposed architecture.

The main driver for a life support system for manned space missions is the crew. Specifically, the life support system should satisfy the crewmember requirements or demands for the basic life support elements, such as O2, H2O and food. The wastes generated by the crewmembers determine the load that needs to be processed (either stabilized & sanitized or the basic life support elements are recovered & recycled for consumption) by the system. Ongoing research is being conducted to identify, develop and test the life support system technologies that would meet these requirements. As a result, there are numerous technologies that could be used for the same task. However, synthesizing an integrated life support system from these technologies with an appropriate deployment schedule for the technologies selected and predicting the behavior of the integrated process present a significant challenge.

The problem of selecting the optimum technology list for a manned space mission, given an objective such as minimum cost or maximum reliability, is quite similar to the process synthesis problem faced in chemical engineering. Given the products that one wants to produce and the raw materials to start with, process synthesis methods try to determine the best combination of processes to obtain the products in the most cost effective way. Once the technology list is determined, the second problem faced is determining how the integrated system will perform under the uncertain conditions, what the inventory levels should be and whether or not the system would be able to deliver the predetermined amounts of products under abnormal conditions.

The SIMOPT architecture has been successfully implemented and used for several applications that involve systems subject to dynamic and stochastic disturbances, such as supply chain management [1-3] and research & development pipeline management [4-5]. The SIMOPT approach [6] uses the concept of timelines to generate multiple unique realizations of the controlled evolution of the system under study. A typical SIMOPT timeline involves several applications of the Deterministic Optimization (DO) algorithm to determine optimized values for system degrees of freedom. In between each DO optimization on a timeline, the simulation is used to determine the integrated system behavior which accounts for uncertainty. The SIMOPT approach is implemented using an inner loop and an outer loop. In the inner loop, discrete event simulation advances along a timeline with decisions determined by DO algorithm. The simulation is used to introduce uncertainty in the model parameters or exceptional discrete events and hence the performance of the simulated system will differ from that predicted by the DO optimization algorithm. When the decisions made from the deterministic optimization modules are no longer applicable due to uncertainty, the simulation is paused and the system is re-optimized preserving the state of the system. The new decisions are fed back to the simulation for the remainder of the timeline and the simulation is resumed again preserving the system state. The various timelines arise from the randomness introduced through changes in parameter values, and exceptional random events that are imposed on the simulation. In the outer loop however information gathered from different timelines is used to update the parameters of deterministic optimization modules used within the inner loop. In this fashion, the SIMOPT approach can be used to solve the process synthesis and design problem in the presence of uncertainty.

The proposed framework for the process synthesis and design for a manned space mission (Figure 1) includes a superstructure optimization module, a simulation module and a data analysis module. Given the crew member requirements and loads, and available technologies and their properties, the superstructure optimization module determines the optimum technology list and its deployment schedule that would minimize the mission cost without violating the crew requirements. The simulation module is used to predict the behavior of the integrated system (technologies selected by the optimization module) under uncertainties such as technology performance variation (in recovery rates or process efficiencies) and random events e.g. technology malfunction. Trigger events are the technology malfunctions. In case of a trigger event, the simulation is stopped and the state of the system and the information gathered up to that time is fed to the optimization to revise the technology list and its deployment schedule. The simulation is resumed with the new list and schedule. This loop (inner loop in Figure 1) is continued till the end of the simulation time, resulting in one timeline. As an example, assume that the optimization algorithm determines to install technology i at time t. Throughout the course of the simulation, suppose, the technology does not behave as predicted due to a malfunction. Then, the simulation is paused and optimization is run again to determine the decisions that will be applied for the remainder of the simulation run (timeline). This cycle is repeated as required until the end of the timeline. In the outer loop, data analysis module uses the information gathered from statistically significant number of timelines to update the technology parameters, such as the reliability, recovery rates or process efficiencies, in the optimization module. At the termination of SIMOPT framework, a technology list, its deployment schedule and main life support elements' supply amounts (inventory levels) is determined that would give the minimum cost for the mission with associated reliability of the mission (99% confidence interval). Here, we refer to the reliability of the mission as the mean failure time of any technology in the system.

Two case studies, Mars Surface Habitat Lander [7] and evolution of the Mars Base [7], are studied using the described SIMOPT framework. The Surface Habitat Lander will be operational when the crew arrives to the Mars providing a crew habitat for the 600-day surface mission. The technologies that are considered for this study are limited to Physical/Chemical (P/C) or partial P/C technologies, whose behavior can be predicted with higher certainty, compared to the biochemical technologies. The best possible combination of the technologies and the deployment schedule, that would minimize the Surface Habitat Lander cost, are determined out of 24 different technology options, 14 of them dealing with carbon dioxide removal and reduction, nitrogen and oxygen generation and supplying them from earth, five of them for waste water recovery and supplying fresh water from earth, and five of them for solid waste processing and recovery. We also assume that all the food is shipped from Earth as pre-packaged food for this case. The second scenario looks at the long term Mars exploration lifecycle where the life of the Mars Base is assumed 14.6 years (consisting of 7 shifts noting that a transfer opportunity between Earth and Mars occurs once every 26 months). For the evolution of the Mars Base scenario, the technology array is not limited to only P/C or partial P/C technologies, but also includes the Biochemical technologies. For example, the food can be supplied as pre-packaged food from Earth as well as can be grown and prepared on site. The results of these two cases will be presented to demonstrate the differences in design decisions for a stationary versus an evolving system, and the similarities with a chemical process synthesis and design will be highlighted.

[1] Subramanian, V., Pekny, J.F. and Reklaitis, G.V., ?A Computational Framework for Studying Decentralized Supply Chain Dynamics?, presented in: ESCAPE-14: European Symposium on Computer Aided Process Engineering, Lisbon, Portugal, May 16-19, 2004.

[2] Jung, J.Y., Blau, G., Pekny, J.F., Reklaitis, G.V. and Eversdyk, D., ?A Simulation Based Optimization Approach to Supply Chain Management under Demand Uncertainty?, Computers and Chemical Engineering, Vol. 28, pp. 2087-2106, 2004.

[3] Wan, X., Pekny, J.F. and Reklaitis, G.V., ?Simulation-Based Optimization with Surrogate Models ? Application to Supply Chain Management?, Computers and Chemical Engineering, Vol. 29, pp. 1317-1328, 2005.

[4] Subramanian, D., Pekny, J.F. and Reklaitis, G.V., ?A Simulation-Optimization Framework for Research and Development Pipeline Management?, AIChE Journal, Vol. 47, pp. 2226-2242, 2001.

[5] Subramanian, D., Pekny, J.F., Reklaitis, G.V. and Blau, G.E., ?A Simulation-Optimization Framework for Stochastic Optimization of R&D Pipeline Management?, AIChE Journal, Vol. 49, pp. 96-112, 2003.

[6] Pekny, J. F., ?Algorithm architectures to support large-scale process systems engineering applications involving combinatorics, uncertainty, and risk management?, Computers and Chemical Engineering, Vol. 26, pp. 239-267, 2002.

[7] Hanford, A.J., and Ewert, M.K., Reference Mission Document, CTSD-ADV-383, NASA Lydon B. Johnson Space Center, Houston, Texas, 2001.

Figure 1. SIMOPT framework for life support system synthesis and design