Use of genetic algorithms and transient models for life-support systems analysis

Direct experimentation with physical systems is slow, expensive, and must wait until the physical system is built. Simulations allow for the testing of different system configurations before hardware is built. This is useful when malfunctions are reliably expected, costs are high, or when the integrated systems are unavailable. Thus, it is important that such simulations are transient and integrate all major subsystems and activities. This paper describes a habitat model that is transient, discrete, stochastic, and nonstationary. It models most of the components of a life-support system including the crew, crops, water and air recovery systems, extravehicular activities, and power. Because the model accepts nonstationary input, it can be used to test habitat configurations and components before building an actual habitat Malfunctions can be injected at any time. A genetic algorithm is used here to find an optimal habitat configuration for a 90-day mission to the moon. Approximately 20,000 system configurations were considered in a series of experiments considering several lunar scenarios. It has been shown that a tuned genetic algorithm is capable of preliminary sizing of several life-support components. In addition, the fitness function serves as proxy for equivalent system mass, a metric related to launch costs.