Detection of locally stationary segments in time series : algorithms and applications

In many applications it is required to segment a time series into its locally stationary parts. Two applications are presented: As a first example consider online monitoring of a BTA Deep-Hole-Drilling process. Here chattering and spiralling of the drilling tool should be avoided by process control. A second example is the analysis of vocal sound signals. It may be required to analyze only specific tones instead of a whole song. Three new algorithms are introduced in this paper, all based on the theory of Dahlhaus (1997) and the analysis of the spectrum of time series, but with different methods to distinguish locally stationary parts of the signals.