Fault feature extraction method for rolling element bearings based on LMD and MED

Abstract: The vibration signals collected from mechanical systems consist of cyclic impulse response, deterministic component and noise. The rolling bearing's fault features are usually so weak that they are overwhelmed by these components, leading difficulty for fault diagnosis. Compared with the inner race and outer race defects of rolling bearing, recognizing the rolling element defects are much more challenging. Therefore, the key problem of fault diagnosis of rolling bears is to exactly extract the weak fault features from a strong noisy background. In this paper, we developed a method based on the minimum entropy deconvolution (MED) and local mean decomposition (LMD) for diagnosing fault features. First, the LMD was used to decompose the original signals into a set of production functions（PFs） adaptively. Each PF was a product of an amplitude envelope signal and a frequency-modulated signal. By doing so, we aimed to obtaining different components embedded in the original signal. These included the cyclic impulse response, deterministic component and noise. However, the cyclic impulse responses were always submerged by noises and they were helpless to make a decision of fault. The MED filter was adopted to search for an optimum set of filter coefficients that recover the output signal (of an inverse filter) with the maximum value of kurtosis. The MED filter was capable of deconvolution of the periodic impulsive excitations from a mixture of response signals and thus enhanced the impulses arising from spalls and cracks in rolling bearings. The MED filter can also be used to remove most noises. Therefore, the former four PFs were further processed by the MED to enhance the fault impulse information. At last，the signal after processed by the LMD and MED was analyzed by envelop analysis. Through this envelop spectrum，the fault features were ultimately extracted. Experimental investigation of 6205-2RS JEM SKF bearings with rolling element defects was performed. The vibration data were obtained from a test rig for simulating various bearing faults in an electrical engineering lab of the Case Western Reserve University. Single point defects were introduced to the test bearings by the electro-discharge machining with the diameters of 0.177 8 mm. The faulty bearings were installed in the drive end, but the accelerometers were placed at the fan end, so the noise was very strong. Using our LMD-MED method, the fault features were successfully extracted. We concluded based on the experiment that the fep index, which indicates the ratio of the peak value at the fault characteristic frequency versus the mean value of the spectrum in 200 Hz band, was increased by 96.4% compared with the original signal. At the same time, the signal-to-noise ratio (SNR) was raised by approximately 18.3% after the signal processing by the LMD and MED. The experiment results proved that the method was effective to detect and extract the fault features of rolling bearings with strong background noises. Besides, these showed that the method based on the minimum entropy deconvolution and local mean decomposition can also provide a very useful reference for fault diagnosis of rolling bears.