Frequency Domain Methods in Diagnostics of Rotating Machines
Metsäntähti, Elias (2014)
Metsäntähti, Elias
2014
Automaatiotekniikan koulutusohjelma
Teknisten tieteiden tiedekunta - Faculty of Engineering Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2014-06-04
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tty-201406061301
https://urn.fi/URN:NBN:fi:tty-201406061301
Tiivistelmä
Condition monitoring of machines is very important in maintenance of factories. By monitoring the condition of machines in real time, maintenances can be planned better and unnecessary loss of production and break down of machines can be avoided. Frequency-domain methods provide efficient means for such monitoring. The methods can be used in recognizing periodic signals from data thus providing valuable information of the condition and possible error sources. The aim of the thesis is to review the frequency-domain methods used in literature, and then investigate their performance in recognizing periodic signals from data.
Five frequency-domain methods were chosen from the literature and applied in studies. The chosen methods were autocorrelation, fast Fourier transform (FFT), short time Fourier transform (STFT) and continuous wavelet transform (CWT). The methods were applied to real-world data obtained from an earth crushing facility. Two experiment setups were implemented to compare the different methods with each other.
The results showed that autocorrelation and FFT were very good at finding the periodic signal from the data. The noise and the length of the data affected FFT more than autocorrelation, but the results were still good. The problem with autocorrelation and FFT was that information on the nature of the periodic signal was lost. FFT also had no time information of the periodic signal.
STFT and CWT provided very good information on the nature of the periodic signal. The results showed that CWT was a bit weaker than autocorrelation and FFT at finding the periodic signal, but the noise did not affect the results so much. STFT was always the weakest of the methods at finding the periodic signal, but the results were still good.
The experiments showed that STFT has very good frequency resolution. The frequency resoultion of CWT was worse and the stronger frequencies drowned the other frequencies, but all the frequencies could still be found from the data.
Autocorrelation and FFT are very easy and fast methods to use compared to STFT and CWT, but STFT and CWT provide better information on the periodic content of the data. All the methods have pros and cons. STFT is the best method to use at finding continuous periodic signals from data because the time domain content is preserved in the transformation and the frequency resolution is good. CWT is the best method to use at finding impulsive periodic signals from data because the time domain content is preserved and the impulsive signals are clear and easy to find from the transform. The results of the thesis can be used to select a frequency-domain method for a specific application to perform frequency-domain analysis and condition monitoring.
Five frequency-domain methods were chosen from the literature and applied in studies. The chosen methods were autocorrelation, fast Fourier transform (FFT), short time Fourier transform (STFT) and continuous wavelet transform (CWT). The methods were applied to real-world data obtained from an earth crushing facility. Two experiment setups were implemented to compare the different methods with each other.
The results showed that autocorrelation and FFT were very good at finding the periodic signal from the data. The noise and the length of the data affected FFT more than autocorrelation, but the results were still good. The problem with autocorrelation and FFT was that information on the nature of the periodic signal was lost. FFT also had no time information of the periodic signal.
STFT and CWT provided very good information on the nature of the periodic signal. The results showed that CWT was a bit weaker than autocorrelation and FFT at finding the periodic signal, but the noise did not affect the results so much. STFT was always the weakest of the methods at finding the periodic signal, but the results were still good.
The experiments showed that STFT has very good frequency resolution. The frequency resoultion of CWT was worse and the stronger frequencies drowned the other frequencies, but all the frequencies could still be found from the data.
Autocorrelation and FFT are very easy and fast methods to use compared to STFT and CWT, but STFT and CWT provide better information on the periodic content of the data. All the methods have pros and cons. STFT is the best method to use at finding continuous periodic signals from data because the time domain content is preserved in the transformation and the frequency resolution is good. CWT is the best method to use at finding impulsive periodic signals from data because the time domain content is preserved and the impulsive signals are clear and easy to find from the transform. The results of the thesis can be used to select a frequency-domain method for a specific application to perform frequency-domain analysis and condition monitoring.