The sparsogram: A new and effective method for extracting bearing fault features

Rolling element bearing is the most frequently failed component in rotary machine. Its failure could cause unexpected machine breakdown. This paper presents a novel fault diagnostic method called sparsogram that can enable early bearing fault detection in a prompt manner. The main concept of sparsogram is derived from the sparsity measurement commonly used for analyzing ultrasonic signals. Sparsogram is capable of detecting high resonant frequency bands that magnify the bearing faulty signals. By using sparsogram, the fault related resonant frequency bands can be determined and used to reveal the temporal waveform contained in each band. Envelope analysis is then applied to convert faulty signals from high frequency band to low frequency band. Finally, power spectrum is employed to display these low frequency signals at their respective frequency spectrum. From the spectrum, the bearing characteristic frequencies related to different types of faults can be easily can be detected and identified. To verify the effectiveness of sparsogram, three different types of simulated signal and a real bearing faulty signal collected from industrial machine were tested. The results show that the sparsogram has good abilities in detecting the health status of bearings, and if faults occurred, determining the causes of faults. © 2011 IEEE.