Development of signal processing techniques for vibration-based condition monitoring of industrial rotating machines

This dissertation presents innovative signal processing techniques for improving vibration-
based condition monitoring of complex industrial rotating machines. Current methods
often struggle with real-world signals and lack robustness. The study addresses these
limitations by enhancing existing signal processing methods in the literature or proposing
new ones. One of the contributions of this thesis is enhancing signal filtering optimiza-
tion techniques by exploiting the engineering knowledge of the machine. As a result of
the proposed improvement, fault detection is achieved on very complex vibration signals.
Furthermore, condition indicators utilized to assess the health status of rotating machines
are widely discussed. The utilization of several condition indicators recently introduced
to the literature is extensively discussed, enhancements for their effective usage are pro-
posed, and such indicators are merged with signal filtering optimization techniques for
early fault detection. In addition, this study proposes a new framework to generate new
condition indicators that are optimal for early fault detection and their statistical threshold
to alarm the end-user for a potential machine fault. Such a framework enables not only
the generation of novel indicators but also the recovery of the health indicators actively
employed in the field, which explains why they were introduced to the vibration-based
condition monitoring domain in the first place. The study finalizes with a discussion on
how informative two spectral correlation-based indicators in terms of the severity of a
bearing fault in time. The trending ability of two indicators is tested on simulated signals
to explain their performances.