Recognition of Bearing Type by Spectrum Analysis

Abstract

The methods of bearing condition analysis advanced much since the processor technologies had enormously raised the computational power. Both smaller and larger platforms have gained in computational speed. The manufacturers of portable vibration measurement devices have got the opportunity to apply such efficient but heavy tools as digital filters, convolution, fast Fourier transforms with better resolution, wavelet transforms, pattern recognition or even neural networks with 2D- and 3D-representation of data. But even then obtaining of good results is limited with multiple conditions as shown in [1]. Having solved the problem with CPU power it may seem that there are no limitations to use the hard methods. But, the electronic component’s lifetime still has the limits: read/write operations and etc. Although these limits are high, it is always good to prolong the lifetime of entire device or reduce its power consumption, or increase the diagnostic speed by using a less demanding algorithm at least at some parts of diagnostic procedure. The laboratories of bearing analysis which are commonly the part of bearing production enterprises have the best opportunities to test the bearings. The type is known, the dimensions are known. The test SW is tuned to the bearing type and the detailed diagnosis gives the best outcome. It is unlike in field service where the bearing type is unknown. Knowing it in the field would help to tune the vibrometer (as in laboratory) and then to switch on the heavy diagnostic methods. The purpose of this work was to create a prototype of software and the bearing model. The model would have the features enough to be determined by the frequency analysis of signal mix and recognized as a certain bearing type.