A Diagnostics Model for Detecting Leak Severity in a Regenerable CO2 Removal System

Human spaceflight beyond Earth orbit will experience high latency in communication and data transmission requiring autonomous deep space habitats that can keep the crew alive when present and keep the habitat "alive" when not. To achieve this goal, the autonomous agent must be both self-aware and self-sufficient. A self-aware system able to perform advanced diagnostics and prognostics of possible failures will be crucial towards enabling autonomy. Keeping the crew alive will demand a robust Environmental Control and Life Support System (ECLSS), the health of which can be sensed in real-time (self-aware) and appropriate corrective action taken when something's wrong (self-sufficient). To investigate the feasibility for autonomous control of ECLSS, a case study utilized a machine learning-based diagnostics model for a leaky regenerable CO2 removal system. A zeolite 13X sorbent bed processed simulated cabin atmosphere flows laden with elevated levels of CO2. Experiments were conducted at nominal operating conditions as well as with faults to generate a diverse set of data for training the model. For this paper, a leak was introduced into the CO2 removal bed. We present the experimental data, describe model development for diagnostics, and then discuss its validation and performance. This paper will further pose a design framework for self-aware ECLSS that utilizes machine learning-based algorithms.