[120] Unified signal and data analysis for integration of condition monitoring and intelligent control
E K Juuso
University of Oulu, Finland
Automatic fault detection with a variety of computational indicators makes it possible to combine reliable condition monitoring with multi-level process control. Various statistical features are widely used in condition monitoring. Current measurements can provide much more and higher-order derivations introduce additional possibilities. Furthermore, combinations of real-order derivatives and generalised real-order norms expand solutions for detecting different signal phenomena. This generalised feature extraction can be tuned to various applications by selecting the sample interval and the orders of derivation and norms. The extensive datasets need edge computing for generating computationally various specific features from the measured signals. The general spectral norms extend this for combining time and frequency domains. The dimensions of all these features are the same as the original signal and all of them can be recursively updated, if necessary, by including new sample intervals of the same size. Non-linear scaling allows different features to be obtained on the same dimensionless scale, which can also be understood on the basis of natural language. The whole analysis can be done by using waveform signals. Non-linear scaling expands dimensionless indices so that scaling takes care of non-linearity processing. In this case, linear interactions are sufficient for the dynamic models used in prognostics and fatigue analysis.
Keywords: advanced signal processing, feature extraction, non-linear systems, condition monitoring, intelligent control.
Keywords: advanced signal processing, feature extraction, non-linear systems, condition monitoring, intelligent control.
