An Investigation of Time Series Embeddings and Topological Data Analysis for Fault Analysis

Topological data analysis (TDA) is an important part of a data scientist's tool box; through the extraction of topological information, TDA provides an automated means of feature engineering. Thus, TDA may be an appropriate tool for fault analysis of mechanical systems, where the methodologies have traditionally relied on expert intuition and signals processing techniques. However, appropriate embedding of the data must be defined before TDA can be applied. In this paper, we investigate several different embeddings of time series data for the application of TDA for fault analysis, and show that features engineered by TDA improve fault classification accuracy.