Human-Expert Data Aggregation for Situation-Based Automation of Regenerative Life Support Systems
Drayer, Gregorio E.
Howard, Ayanna M.
MetadataShow full item record
Regenerative life support systems (RLSS) introduce novel challenges for the development of automation systems given the emerging behaviors that result from incremental system closure. Switching control paradigms offer the ability to manage such uncertainty by allowing flexibility into the control path, enabling for autonomy modes that depend on the situation of the system. Previous research proposed a granular approach that combines sensor information to define operation conditions and act upon them. It makes use of fuzzy associative memories (FAM) to define the pairs (Situation, Controller) that assign control actions to each situation. The FAM are composed granules that represent situations in which the autonomous system may operate. One of the challenges of this approach is the combinatorial explosion that arises for large numbers of sensors. Human-system interaction offers a solution to this problem and, for such purpose, this paper elaborates on the aggregation of human-expert data to obtain the granular structure of the FAM. The aggregation process consists of an optimization process based on particle swarms. The result is a three dimensional array with parameters that define n-dimensional non-interactive granules. Two alternatives are presented in this paper: (1) a four-dimensional optimization algorithm to obtain normal fuzzy sets, and (2) a five-dimensional alternative that results in subnormal fuzzy sets. The results were obtained with simulations of an aquatic habitat that serves as a small-scale model of a RLSS. The discussion elaborates on which of the two alternatives may be better suited for applications in situation assessment and automation.