Method for detection and automatic identification of damage to rolling bearings

An method for detection and automatic identification of damage to rolling bearings, in which vibrations of a bearing in operation, measured as an analogue timewave signal, are converted into digital data and prefiltered, wherein the detection of shock pulses are performed with use of a binary vector (p) that is determined in the following actions:

• selecting a first and second adjacent vector intervals in the prefiltered digital timewave data, said first adjacent vector intervals being a reference interval and said second adjacent vector interval being a test interval, filtering the signal in said test interval and said reference interval multiple times whereupon after each filtering a shock pulse likelihood ratio is calculated as a ratio of mean square values of filtered signal present in said test interval to the mean square values of filtered signal present in said reference interval, comparing said likelihood ratios with a preset threshold value and the probability of shock pulse occurrence in the first test interval is estimated, recording information about shock pulse occurrence in said test interval as initial elements of binary vector (p), wherein if said likelihood ratio is greater than said preset threshold value, the elements of the binary vector (p) take the value 1 and if said likelihood ratio is less than said preset threshold value, the elements of the binary vector (p) take the value 0, and thereafter determining all remaining elements of the binary vector (p) for all successive test and reference intervals that are selected iteratively with an offset in a specific direction, wherein the identification of the defect in the bearing or bearings is done automatically for the binary vector (p) determined for all examined said test intervals and said reference intervals of timewave (X) of the processed signal and the set of known defect frequencies of examined bearing or bearings ((Ff)=[ff1, ff1, . . . ffK]), by means of the following actions:
• for each successive, known defect frequency (ffK) from a set (Ft) of the examined bearing, a test frequency (νfKε└(1−Pg)ffK,(1+Pg)ffK┘) that belongs to a neighbourhood (Pg) of the frequency (ffK) is determined, where (Pg) is a configuration parameter denoting the searching frequency range related to known defect frequency and next, using maximum likelihood mathematical principles, for each test frequency (νfK) the maximum likelihood function (J) in time domain is determined using the binary vector (p), checking which test frequency (νfK) correlates to a maximum likelihood function value (JA), calculating a ratio (FA) of the maximum value (JA) to a noise level (nA) and comparing said ratio (FA) with a reset threshold parameter (SA):
• generating information about examined bearing defect with specific frequency that belongs to set (Ft) and corresponds to test (νfK) if the ratio (FA) is greater than the preset threshold parameter (SA), generating information that no specific defect occurred in the examined bearing or bearings if the ratio (FA) is less than or equal to the preset threshold (SA).