Abstract

Most existing studies on vibration-based fault detection for planetary gears were developed and tested under constant speed conditions. Recently, some methods were developed to consider the variability of the rotating speed; however, these methods have limitations. Specifically, these methods are applicable only for small fluctuations of speed, or the methods require additional angular information as an input. This paper thus proposes a new method, the positive energy residual (PER) method, for fault detection of planetary gears. PER does not require the assumption of only small fluctuations of speed, nor does it need angular information. The proposed PER algorithm is based on two techniques, the wavelet transform (WT) and the Gaussian process (GP), which are used to remove the variability of the signals while extracting the faulty signals. Further, a fault feature is presented that is able to effectively quantify the characteristics of faulty signals. The performance of the proposed method is demonstrated using two case studies: vibration signals from a simulation model and vibration signals from a real test-bed. A comparison study with other methods, WT and energy residual (ER), is also presented to clarify the performance of the proposed PER algorithm. From the results, we conclude that the proposed PER method is capable of detecting faults of a planetary gear under variable speed conditions, while showing better performance than the two other methods.