Comparative study of similarity measures for launch vehicle telemetry data

Post flight analysis of launch vehicles involves comparative study of predicted preflight and flight telemetry time series data which are enormous, high dimensional and nonstationary. This inherent nature of data precludes use of manual methods and calls for data mining techniques for analyzing the similarity between the preflight and flight datasets. Hence different similarity measures like distance measures, correlation coefficients, Dynamic Time Warping (DTW) and Hidden Markov Models (HMM) are applied to the nonstationary time series data of launch vehicles. In the absence of shifts, Euclidean distance was found to be the best similarity measure while v-shift Euclidean distance gave good performance if only vertical shifts were present. Applying HMM after demeaning the data was found to give better performance in detecting similarity even when both horizontal and vertical shifts were present.