A Statistical Analysis and Predictive Modeling of Safing Events for Interplanetary Spacecraft

Author(s)
Pujari, Swapnil R.
Editor(s)
Associated Organization(s)
Organizational Unit
Daniel Guggenheim School of Aerospace Engineering
The Daniel Guggenheim School of Aeronautics was established in 1931, with a name change in 1962 to the School of Aerospace Engineering
Supplementary to:
Abstract
Unexpected spacecraft failures and anomalies may prompt autonomous on-board systems to change a spacecraft’s state to a ‘safe mode’ in order to isolate and resolve the problem. Future interplanetary missions such as Psyche and the proposed Next Mars Orbiter mission concept, plan to use solar electric propulsion on-board. Continuous operation of the thrusters is necessary in order to achieve their mission objectives. The mo tivation for this paper stems from a need to better predict safing events based on various mission factors such as mission class, destination, duration, etc. Modeling spacecraft inoperability due to a spacecraft entering safe mode is imperative in order to appropriately allocate spacecraft margins and shape design & operations requirements. This paper contributes to the area of safing events by further analyzing trends and dependencies within the available data subsets, and develops predictive models of frequency and recovery times of safing events for interplane tary spacecraft missions. First, the full safing event dataset is split into multiple subsets based on various mission classifiers. By employing the Chi Squared hypothesis test, the degree of dependency between classifiers is assessed. A parametric analysis is conducted using a single and mixture of two Weibull distributions. The optimal parameters that would best fit the full dataset and subsets are computed by a maximization likelihood algorithm. The mean square error and Akaike Information Criteria represent goodness-of-fit criteria for the computed distributions; insight into any inherent bi-modal behavior is identified through these criteria. A supervised learning algorithm is utilized in captur ing and understanding relationships between input and output variables, and utilizing these to predict unknown outcomes. For the safing event database, two Gaussian process models are trained, tested, and deployed: one for time-between-events and the other for recovery durations. By incorporating these Gaussian Process models into a mission simulation framework, a Monte Carlo simulation of the likelihood of inoperability rates is conducted to robustly predict safing events. A greater understanding of the safing event dataset through statistical & parametric analyses, and the development of a Gaussian Process model for predictions enables interplanetary mission planners to make more informed decisions during spacecraft development
Sponsor
Date
2018-04-27
Extent
Resource Type
Text
Resource Subtype
Masters Project
Rights Statement
Unless otherwise noted, all materials are protected under U.S. Copyright Law and all rights are reserved