Human Locomotion Monitoring in Space Flight: Retrospective Nonparametric Changepoint Detection Methods

Abstract

The paper is dedicated to the analysis of medico-biological data obtained during locomotor testing of astronauts. Accurate data interpretation plays a crucial role in locomotion system monitoring, prophylaxis of long-duration spaceflight negative effects and thus in the development of an autonomous medical support system for deep space expeditions. During the locomotor testing the astronaut changes motion modes in accordance with the prescribed training protocol while running on the treadmill, and data such as speed, support pressure, heart rate frequency, etc., are collected simultaneously. The astronaut may follow either an individual protocol developed by specialists or perform his personal protocol at every fourth day of the micro cycle. Our task is to identify unknown motion modes by the means of a posteriori time series segmentation and, specifically, in the presence of various transitional processes as well as signal loss periods. The presence of tricky profiles does not allow for preliminary hypotheses about the distribution pattern of the dataset under study. The article consists of two parts. Firstly, it provides a detailed overview of several modern retrospective (offline) nonparametric multiple changepoint detection methods in multidimensional time series. A change point means an abrupt change in the probability properties of the observed series occurring at an unknown time instant. When describing the algorithms, emphasis is placed on statistics as a measure of data homogeneity, numerical methods for solving optimization problems, and model selection methods. Secondly, the real speed profiles resulting from locomotor testing have been handled through the mentioned algorithms. The validation was performed on three characteristic experimental data samples, allowing for an assessment of the prospects of applying the described methods to the entire dataset.