Updating subsystem-level fault-symptom relationships for Temperature and Humidity Control Systems with redundant functions

Abstract

As we aim for deep space exploration, supporting vital systems, such as the Temperature and Humidity Control System (THCS) in the Environmental Control and Life Support System (ECLSS), through timely onboard fault detection and diagnosis becomes paramount for mission success. Many existing fault diagnosis approaches assume that the function that models the relationship between faults and associated symptoms (fault-symptom relationships) will remain constant throughout the THCS’ lifetime. Therefore, many of these diagnosis methods are not robust enough to automatically account for changes in fault-symptom relationships as a result of changes in the habitat (e.g., system reconfiguration). The work highlighted here is on (i) surveying existing work on adaptable fault diagnosis methods and (ii) showcasing a real-life case study, in which we identified the need for an automatically adaptable fault diagnosis method. The case study focuses on a reconfigured terrestrial THCS analog, the Heating, Ventilation, and Air Conditioning (HVAC) system, where the original fault-symptom relationship is revealed to be no longer accurate. We then apply current adaptable fault-symptom relationship generation methods, such as Model-Based Dependability Analysis (MBDA) methods and data-driven causal discovery methods. Through this analysis, we detail our procedure in (i) identifying relevant fault-free system information, such as redundancy, to revise fault-symptom relationships used in fault diagnosis and (ii) evaluating the fault diagnosis performance in a THCS with the original and revised fault-symptom relationship. Our contribution lies in identifying the shortcomings of current methods and pinpointing future steps in creating an adaptable fault diagnosis framework. We found that although the MBDA methods can automatically generate fault-symptom relationships given system flow information and fault mode of components, they also required manual revision of the aforementioned information to create fault-symptom relationships that reflect redundancies. On the other hand, we concluded that the causal discovery methods can detect fault-free system information, such as redundancies, that may help us revise fault-symptom relationships, but suspect variables that contribute to redundancies may have to be hand-picked.