Towards anomally detection using stationary and non-stationary signal analysis

Abstract

This paper focuses on demonstration of an enhanced model for investigating data signals features, i.e., whether the given signal has stationary or non-stationary features. The accurate detection of the features of signals is crucial for the right directions towards methodology of further preprocessing to perform data analysis of the data signal, specifically in the tasks of finding anomalies in the given signal and big data environment. A problem often encountered is the exact determination of the occurrence of stationary or non-stationary data signal features in data processing. Within this research paper, the mathematical foundations of data signal processing are described. Based on the mathematical model of the input signal processing, an improved workflow using the enhanced statistical KPSS test and autocorrelation function (graphical) analysis is demonstrated here, to confirm the accuracy and usability of selected methodology. The alternative approach described here leads to a much lower computational effort and the achievement of accurate identification of signal features in big data environment for possible deployment of A.I. or machine learning anomaly detection pipeline. The obtained dataset and model are based on the real environment and measured signals in the production process of machine tools in company Tajmac-ZPS Zlin.