In vibration analysis, the aim is to characterize the vibration behavior of measured variables and investigate them for changes over time. Signals can be, for example, electric currents or accelerations. To describe vibration signals, the signal is broken down into its frequency spectrum, which describes how strong the respective frequencies are in the overall signal.
Changes in the known frequency spectrum, such as of an axle bearing, can be an indication of possible wear and tear or a defect. In many cases, the change in vibration frequencies even indicates the type of defect or the main wear component. Machine learning is then used to learn the optimum frequency spectrum of an intact system, which can be used as a reference for a comparison with still unknown spectra of similar systems.
Basically, there are two different approaches: in the first approach, the entire frequency spectrum of a signal is investigated. This shows all the different frequency components that are contained in the signal; however, there is no information about the exact time location of these components. In the second approach, temporally localized oscillating functions, which differ in terms of frequency and time location, are compared with the signal to be analyzed. As opposed to the first approach, the location component remains, and consequently, also signals whose vibration behavior changes over time can be analyzed in a time-based manner.
Change in vibration behavior
In the illustration, the change in the vibration of a mechanical system over time is investigated. At the top, the smoothed acceleration signal is shown as a function of the time. The result of the vibration analysis shows that from about 360s the frequency in the shown period increases continuously, which could be an indication of damage in the investigated component. The normalized spectrum, shown on the right, shows the frequency distribution over the entire period under consideration. The main vibration frequency of the intact component is about 65 Hz. The spectrum of the defective systems differs considerably from this in terms of quality.
Based on this procedure, faults in individual components can be detected at an early stage with vibration analyses. The type of vibration behavior of a machine or a component during a defined processing step can be extremely specific, which means that deviations in the vibration characteristic can provide information about possible damage and wear processes.
in good time
Katana, the data analytics segment of the USU Group, has extensive experience in the application of machine learning methods in the industrial sector. Take advantage of our knowledge and our solutions for building your data-driven business models to reduce your costs and improve your quality and value added.