Hiroyuki Miyajima

Application of Intelligent Control to Material Circulation in Advanced Life Support Systems

An Advanced Life Support System (ALSS) is a life support system (LSS) for accommodating prolonged missions far from Earth. It is distinguished from an LSS for International Space Station (ISS) by Food production, Biological processing in addition to physicochemical processing, Resource recovery by recycling human waste and inedible crop portions. Typical missions assumed include manned Mars exploration, which is 150 days for the outward trip, 619 days for the Mars stay, and 110 days for the return trip for a total mission of 879 days. (Hoffman and Kaplan, NASA’s Reference Mission for Human Exploration of Mars).

Design of Intelligent Control Software for Mini-Earth

In this paper, we describe the design of intelligent control software required for operating the Closed Ecology Experiment Facilities (CEEF) that we call Mini-Earth. We will develop intelligent control software by reconfiguring functions of a Control Computer System (CCS) for controlling the CEEF into three layers that consist of planning & scheduling, task, and control levels. In the last half of this paper, we will focus on the planning & scheduling level, and describe the operation scheduling problem of a CEEF gas circulation system using a Planning and Scheduling Language (PSL) to develop an Operation Schedule Interactive Generation system (OSIG).

Applying Autonomous Control method for Material Circulation to Advanced Life Support System

An Advanced Life Support System (ALS) recycles and circulates substances within a living environment, and makes it possible to sustain life in space. Among those systems, this study addressed a system for recycling the elements C, H, and O in the system, which includes waste processing and food production. We proposed a procedure for such a system that combines automatic generation of scheduling and multi-agent reinforcement learning, based on a hierarchical control method. In this report, a procedure was applied to a simple model and a computer simulation was performed. Specifically, an automatic scheduling procedure was used to create an overall plan for operation of a model that assumed closed circulation. The tanks, processors, and module subsystems were controlled by multi-agent reinforcement learning (MARL) in a decentralized autonomous system. Currently, the calculations are complete for the simplest system, that with two tanks. These results indicate that MARL requires preliminary learning, particularly for emergency events, but otherwise, all learning can be done by MARL on-line.

Considerations of Material Circulation in CEEF Based on the Recent Operation Strategy

In Closed Ecology Experiment Facilities (CEEF), with integrating the Closed Plantation Experiment Facilities (CPEF) and the Closed Animal Breading & Habitation Facilities (CABHF), closed habitation experiments without material exchange with the outside will be conducted after the 2005 fiscal year. Cultivation experiments of approximately 30 crops and the integrating test of the material circulation system required for the closed habitation experiments have been performed since fiscal year 2000. Using data reported in these experiments, material circulation in CEEF is simulated based on the recent operation strategy, and the storage capacity needed for the buffer of an air processing subsystem was estimated. In order for two humans to dwell for more than 120 days, the storage capacities of the carbon dioxide tank, the oxygen tank, and the waste gas tank in CPEF, and the carbon dioxide tank and the oxygen tank in CABHF are 820 g, 2830 g, 4425 g, 1780 g, and 1792 g, respectively. This implies the storage capacities needed under the best conditions. It is confirmed important to set the closing period of material circulation as the longest cultivation period of the crops, and to limit the processing amount of the wet oxidation processor to the quantity of waste product for one day.

System Structure Modification in Advanced Life Support System using Autonomous Control Method

An Advanced Life Support System (ALS) recycles and circulates materials within a living environment, and will eventually make it possible to sustain life in outer space. This study addresses the system subset that will recycle the elements of carbon, hydrogen and oxygen, with planned functions that include waste recycling and food production. We have previously proposed a procedure for such a system that combines automatic generation of scheduling and multi-agent reinforcement learning (MARL), based on a hierarchical control method. This paper reports the application of this procedure during a material circulation simulation that includes modifications to the system. Specifically, a model with two installed tanks each for oxygen (O2) and carbon dioxide (CO2) is expanded by the installation of three additional tanks after scheduling calculations have been performed, thus providing a total of five tanks for each gas, at which point the calculation is repeated. It was confirmed that the increase in tanks caused no disturbances in the scheduling-based upper control layer, or any problems in the MARL-based lower layer.

Autonomous Control Method for Material Circulation in Advanced Life Support System

An Advanced Life Support System (ALS) conducts recycling and circulation of materials within the system, and is the component ensuring the survival of living beings during missions. The present research examines schemes for recycling the elements carbon, hydrogen and oxygen inside a system to process wastes and produce food. The circulations of materials become very complicated in this kind of ALS, so it is difficult to manually draw up comprehensive control principles. It is also difficult to handle emergency conditions. Therefore, it would be preferable to carry out environmental control automatically, and to provide for autonomous handling when an emergency arises. We applied methods for automatically creating schedules in an attempt to enable the system to continue degenerate operation automatically during emergencies; one control method employed multi-agent reinforcement learning, and the other employed Lagrangian decomposition and coordination method. The two procedures were found to offer complementary advantages and weaknesses. A system was then proposed under which the two procedures would compensate for each other’s weaknesses.