Rolling Element Bearing Fault Diagnosis Based on Improved l₁ Trend Filtering

In order to remove the noise in the vibration signals of fault bearings, improved l₁ trend filtering is exploited. In this method, regularization parameter is used to control the performance of the filtering method, and is experimentally determined through the feature information of raw signals. It is inconvenient to real applications. In this case, the suited parameter is selected based on the linear relation between it and its maximum. By analyzing the two vibrations measured from the bearing with an outer race fault and an inner race fault respectively, the results demonstrate that the proposed method is effective and robust to diagnose the two bearing faults. At the same time, empirical mode decomposition is adopted for further comparisons. It shows that the improved approach has better performance in the fault feature extractions.