Robustness analysis of the FFT-based segmentation, feature selection and machine fault identification algorithm

This paper aims to experimentally investigate the robustness of a recently developed fast Fourier transform (FFT)-based segmentation, feature selection and fault identification algorithm for the development of practical maintenance applications for high-speed rotating machinery using the acoustic emission (AE) technique. 50 experiments are carried out for five machine health conditions: healthy, with a compressor air leak and with three different bearing outer race defects, in order to evaluate the performance of the algorithm using an industrial air blower system. The sensitivity analysis introduced in this study investigates the effect of changing the sample time length, changing the position of the data window (sliding window) and varying the rotational speed on the certainty of fault identification results. Disturbance and measurement noise are also considered. Moreover, the ability of the algorithm to identify degradation outside of the datasets for which it was trained is investigated. The results show that the fault identification is impervious to changes in these parameters and that the algorithm demonstrates an ability to perform during machine degradation. However, a number of the addressed parameters adversely affect the level of confidence in the fault identification results, which increases the potential for a false diagnosis.