A Methodology for Benchmarking and Selecting Sequence Optimization Algorithms for Active Debris Removal Mission Planning
Author(s)
Hickey, Alexandra M.
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
The space debris environment grows more concerning every year, and in the last few years, a significant increase in the rate of change of debris has occurred. Increasing amounts of debris threaten existing spacecraft on orbit and increase costs for operators who must analyze a large number of conjunction risks each year. Failure to properly manage the risk to these systems can lead to destruction, as was seen with the Iridium system in 2009, or damage, as has been seen with systems like Sentinel-1A. The growth in the environment has led to increased interest from US and European officials in space debris remediation measures. While governments are investing money in the initial technology development, there is a severe lack of economic and policy infrastructure to support these missions into their commercial phases, making the economic effectiveness of these missions of the utmost importance. One element of creating a cost-effective mission is ensuring that each mission removes the largest number of debris possible.
The number of debris that can be removed is driven by the efficiency of the transfer between any two given pieces of debris targeted and the sequence in which debris is targeted. The latter is referred to as the sequence optimization part of the problem and plays a significant role in the cost-effectiveness of missions. Hundreds of different algorithms have been tested for the sequence optimizer, and significant advances have been made that allow for the optimization of large debris fields. However, there is a lack of synthesizing elements for sequence optimizers, such as common points of comparison for solutions, leading to difficulty in understanding if a new algorithm is an advancement and in what way. Additionally, this makes it difficult for mission planners to understand what algorithms will perform best for their debris field of interest. This work examines the effectiveness of establishing physics-based benchmarks as a solution to this field's current issues with synthesis.
In order to establish benchmarks for ADR mission planning sequence optimization, a series of experiments were completed to determine what factors underlie differences in algorithm performance on the ADR mission planning problem. The first experiment evaluated the impact of orbital element range, clustering, and distribution in debris fields, which were expected to be primary drivers of modality in the fitness function, on differences in algorithm performance. Five different algorithms were used: a genetic algorithm, an ant colony optimization algorithm, a differential evolution algorithm, a beam search, and a greedy search. RAAN and inclination range, along with clustering in all the orbital elements studied, were found to be significant drivers of differences in algorithm performance. The next experiment incorporated size and two additional algorithms, an exhaustive search, and a branch and bound algorithm. This experiment showed that accounting for size drastically changed the results. The primary causes of differences in algorithm performance were found to be size, inclination distribution, RAAN range, inclination range, eccentricity clustering, eccentricity distribution, inclination clustering, and semimajor axis range. The first two experiments created a new understanding for researchers of what aggregate debris field phenomena are predictors of differences in algorithm performance.
The results from the second experiment were then used to establish 19 standardized debris fields, encompassing the interactions and effects found to most drive differences in algorithm performance. Sequence optimization for each of these debris fields was completed using the algorithms from the previous experiments. These results establish an initial baseline of benchmarking results to which new algorithms can be compared. Additionally, for each standardized debris field, the results from the comparison demonstration were analyzed to understand how different algorithm behaviors affected the relative performance of the algorithms. Such comparisons are valuable in allowing researchers to develop new algorithms that will outperform the initial benchmark set and understand existing gaps in the field as more algorithms are benchmarked. Next, three real debris fields, a small debris field at risk of conjunction with the ISS, a midsized field of high-priority rocket bodies, and a large field of debris from the iridium collision, were analyzed and mapped to the standardized debris fields. These standardized debris fields were then used to predict the top performing algorithm for each real debris field. Finally, all the algorithms were tested on the real debris fields. A comparison of the actual and predicted results shows that the standardized debris fields are able to successfully predict the top performing algorithm. The application of the presented methodology for mapping real world debris fields to SDFs allows mission planners a priori knowledge of what algorithms should be used on their debris fields for the first time, enabling better algorithm selection and ultimately lowering per debris removal cost.
Sponsor
Date
2023-12-08
Extent
Resource Type
Text
Resource Subtype
Dissertation