An alternative method of monitoring rolling element bearing condition is the ‘shock pulse method’ (SPM). This method depends upon the signal input to the measuring instrument being passed through a narrow-band filter whose center frequency corresponds to the accelerometer natural frequency (of the order of 20-50 kHz). As part of the bearing collides with a defect, a shock wave is transmitted through the bearing and machine casting thus exciting the accelerometer by means of a pulse input. The accelerometer output is a damped transient waveform whose frequency is much higher than the frequencies characteristic of specific bearing faults, and whose magnitude is dependent upon the magnitude of the defect in the bearing. Again this method relies heavily upon establishing reliable baseline data and on monitoring changes in output level rather than on absolute values. SPM cannot indicate the cause of the vibration levels, but certainly it is a less expensive technique.

In acoustic flaw detection (incipient flaw detection or IFD) the band from 80 kHz to 120 kHz is responsive to bearing defects and being used in most operational applications. This choice is highly empirical, however, and other bands are being used. Following pre-amplification and band-pass filtering to eliminate as much noise and extraneous information as possible, the acoustic flaw detection signal is conditioned to produce three outputs which can be measured as numerical values on a meter. First, the RMS amplitude within the band-pass is determined and displayed as a measure of overall condition. Next, the energy content of the spikes or pulses in the high frequency signal above some arbitrary and automatically set multiple, usually 2-4 times the RMS level, is detected and displayed as indicator of discrete flaws. This measurement can be thought of as the energy content above some fixed crest factor. Finally, the rate at which the signal crosses a given threshold is counted and displayed. Because large signals will produce more crossings per event or pulse, the count rate measurement can be a second approximation of the energy or severity of the defect. Acoustic flaw detection thus provides three measurements with which to judge overall condition as well as permits recognition and evolution of local defects. As indicators of bearing condition, field experience ranks the RMS (root mean square) measurement first, followed by the SAT with count rate a distant third.

A review of vibration and acoustic measurement methods for the detection of defects in rolling element bearings was presented by Tandon and Choudhury (1999). Detection of both localized and distributed categories of defect was considered. An explanation for the vibration and noise generation in bearings was given. Vibration measurement in both time and frequency domains along with signal processing techniques such as the high-frequency resonance technique were covered. Other acoustic measurement techniques such as sound pressure, sound intensity and acoustic emission were reviewed. Recent trends in research on the detection of defects in bearings, such as the wavelet transform method and automated data processing, were also been included.

While operating a rolling bearing with local faults an impulse is created, the high-frequency shock vibration is then generated and the amplitude of vibration is modulated by the pulse force. The envelope analysis method provides an important and effective approach to analyse the fault signals of high-frequency impact vibration, it has been applied to the fault diagnosis of rolling bearings successfully (Radcliff, 1990; Randall, 1986; Brown, 1989). However, in the traditional envelope analysis method, the fault is identified through the peak value of envelope spectrum. Thus, this traditional method has two aspects of disadvantages. On the one hand, FFT method is widely used in the spectrum analysis of envelope signals; however, it could only give the global energy-frequency distributions and fail to reflect the details of a signal. So it is hard to analyse a signal effectively when the fault signal is weaker than the interfering signal (Ho and Rand, 2000; Randall, 1997; McFadden, Smith, 1984). At the same time, it is easy to diffuse and truncate the signal’s energy as FFT regards harmonic signals as basic components, which will lead to energy leakage and cause lower accuracy. On the other hand, the central frequency of the filter is determined with experience while forming an envelope signal, which will make great subjective influence on the results (McFadden, Smith, 1984; Tse, et al., 2001; Randall and Gao, 1996).

Generally speaking, the process of the rolling bearing fault diagnosis consists of three steps: (1) the collection of the rolling bearing fault vibration signals; (2) the extraction of the fault features; (3) condition identification and fault diagnosis. How to extract the fault features and identify the condition from the rolling bearing vibration signals are the key steps in the fault diagnosis of rolling bearings. As the fault vibration signals of rolling bearings are non-stationary, it is a crux how to obtain feature vectors from them for the fault diagnosis. The traditional diagnosis techniques perform this from the waveforms of the fault vibration signals in the time or frequency domain, and then construct the criterion functions to identify the working condition of rolling bearings. However, because the non-linear factors such as loads, clearance, friction, stiffness and-so-on, have distinct influence on the vibration signals due to the complexity of the construct and working condition of rolling bearings, it is very difficult to make an accurate evaluation on the working condition of rolling bearings only through the analysis in time or frequency domain (Sandy, 1988; Li and Wu, 1989). Features are those parameters derived from the measured data that robustly indicate the presence of the rolling bearing faults. The main feature extraction methods include: time-domain methods, frequency-domain methods, and time–frequency methods. Time domain methods such as peak amplitude, root-mean-square amplitude, crest factor, kurtosis and shock pulse counting have been successfully applied to rolling bearings (Ma and Li, 1993; Martin, 1989; Volker and Martin, 1986). Frequency domain methods applied to rolling bearing fault diagnosis include Fourier spectra time waveform, cepstrum analysis, sum and difference frequencies analysis and the envelope spectra technique (Ma and Li, 1993; Radcliff, et al., 1990; Peter, 2001; Zheng and Wang, 2001). A comparative study of various feature extraction methods that fall into the time-domain and frequency-domain methods is presented in  (Elbestawi, and Tait, 1986). As the rolling bearing vibration signals possess non-stationary characteristic, time–frequency methods is effective to extract the feature of the original data. The wavelet transform has been applied to feature extraction for rolling bearing vibration signals and efficient results have been obtained (Li and Ma, 1997; Geng and Qu, 1994; Lin and Qu, 2000). The mathematical model need to be established or the fault mechanism of the rolling bearing vibration system need to be studied before the feature extraction in above-mentioned methods. For example, in the envelope spectra technique, the centre frequency and bandwidth of the band-pass filter must be determined correctly while forming envelope signal and the fault characteristic frequency of the rolling bearing must be computed (Randal, 1986; Randal and Gao, 1996). However, for a complex rolling bearing vibration system, the related parameters and the mathematical model are difficult to be determined. In many cases, these parameters (such as the centre frequency of the band-pass filter in envelope spectra technique) are determined with experience, which will make great subjective influence on results.

The application of wavelets has emerged in the context of damage detection, and an excellent review of this is given in (Staszewski, 1998). Mori et al. (1996) have predicted the spalling on the ball bearing by applying discrete wavelet transform to vibration signals. Jing and Qu (2000) proposed a de-noising method based on Morlet wavelets for feature extraction and they applied it to inner race fault detection of the rolling bearing. The previous works (Jing and Qu, 2000; Mori et al., 1996; Li and Jun, 1992) dealt with the detection of one fault in a bearing using wavelet transform. In the study of Jing and Qu (2000), the diagnosis of single and multiple ball bearing race faults was investigated using discrete wavelet transform. Bearing race faults were detected by using discrete wavelet transform (DWT) by Prabhakar et al. (2002). Vibration signals from ball bearings having single and multiple point defects on inner race, outer race and the combination faults were considered for analysis. The impulses in vibration signals due to bearing faults were prominent in wavelet decompositions. It was found that the impulses appear periodically with a time period corresponding to characteristic defect frequencies. It was shown that DWT could be used as an effective tool for detecting single and multiple faults in ball bearings