科研之友
微信
新浪微博
Facebook
分享链接
摘要
Gear transmissions have been widely used in most of today's manufacturing and production industries; however, they often suffer from deteriorations and damages on gear pairs. Severe damages of the machinery caused by the failures of gears account for 48 %, leading to significant economic losses. Therefore it is crucial to implement fault diagnosis procedure for gearboxes. The gear meshing motion is a kind of typical strong nonlinear movement, and the related vibration signals are the nonlinear mixtures of different kinds of vibration source, leading to great difficulty in the fault feature extraction and fault detection. In order to improve the fault detection of gearboxes, a new method based on the nonlinear fault feature selection and intelligent fault identification is proposed in this work. The blind source separation (BSS) procedure was firstly employed to eliminate the influence of noise signal sources. The useful information related to the fault vibration was hence separated by the independent component analysis (ICA). Then the spectral regression (SR) was used as a nonlinear feature selection technique for the separated vibration sources. Hence, distinct fault features can be obtained. Lastly, the kernel-based extreme learning machine (KELM) was applied for the pattern recognition of single and multiply faults of the gearbox. The fault vibration data acquired from a gearbox fault experimental tester was used to valuate the proposed diagnostic method. The experiment results show that useful fault vibration signals can be separated by the new method, and the fault detection rate of the proposed method is superior to the existing approaches with an increase of 4.4 % or better. Hence, this new development will produce considerable savings by reducing unplanned outages of machinery so a company can get the full benefit from condition monitoring.
