CONDITION MONITORING

 

 

BEARING FAULT DETECTION USING VIBRATION SIGNALS

 

Bearing fault detection using vibration measurements is presented in Figure 1. The spectral amplitude (envelope) corresponding to the fault frequency of the bearing is monitored. Two methods have been applied to process of vibration signals; high-pass filter (TRADITIONAL) and Kanflex method (ADVANCED). Using the Kanflex method a bearing failure is detected clearly earlier than using the traditional method, so maintenance operations can be performed well in advance before a possible machine breakdown.
FIGURE 1
Responsive image

 

 

 

BEARING FAULT SIZE ESTIMATION USING VIBRATION SIGNALS

 

A supervised machine learning method is used to assess the severity of a bearing failure. The best features (indicators) are calculated from the vibration signals measured from bearings to describe the severity of a bearing failure. The machine learning method is "taught" by using the best features calculated from the vibration signals and the actual directly measured bearing fault sizes corresponding to the vibration measurements. Figure 2 shows the estimated bearing fault sizes for four (unknown) bearings. The shades of red illustrate the severity of the bearing failure from small (light red) to significant (dark red). The estimation error of the bearing fault sizes is about 6% in these cases.
FIGURE 2
Responsive image

 

ESTIMATION OF BEARINGS REMAINING USEFUL LIFE

 

After a bearing failure is detected, an estimation of the remaining useful life is performed. From the previously measured (history) bearing vibration signals, the best possible characteristic (index) is calculated to describe the degradation. A sampling algorithm based on the MCMC method is used for the remaining useful life estimation. Figure 3 shows a forecast of the remaining service life of a damaged bearing. In the picture, a dashed turquoise vertical line depicts the last vibration measurement from which the forecast is seen.
FIGURE 3
Responsive image