ADVANCED SYSTEM FOR AUTOMATICALLY DETECTING FAULTS OCCURRING IN BEARINGS

Maintenance is essential in all kinds of machinery. In order to prevent long-term breakdown or catastrophic failure, faults should be detected at their incipient stage. Bearing faults are the most frequent cause of failures in machinery. Most of the faults occurring in bearings will introduce impact vibration signals when bearings are rotating. Such impacts behave like damped oscillations and can be detectable by using accelerometers. In the past, maintenance staff assessed a machines' condition by measuring the increase of either the overall vibration level or the magnitudes of characteristic frequencies generated by roller bearings. To date, many staffs still adopt these methods for vibration-based machine fault diagnosis. However, these methods are not effective. Bearing vibrations displayed in a frequency spectrum are often overwhelmed by higher vibrations generated by larger components located adjacent to the inspected bearing. Faulty bearing information could be difficult to reveal. Hence, maintenance staff fail to detect bearing faults at their earlier stages prior to a fatal breakdown of bearing. Research has been conducted to invent effective techniques so that such anomalous impacts can be more easily detected either in its time or frequency domains. For example, the ever-popular Fast Fourier transform, the Wavelet transform, the Hilbert-Huang transform etc. are aimed to detect bearing faults through the use of either conventional frequency spectrum or advanced decomposed waveforms that may exhibit instantaneous changes in temporal and frequency characteristics. Meanwhile, while these techniques have proven to be successful in various degrees of fault detection, the industry is reluctant to adopt these techniques as they require intensive analysis and experts to interpret the results. The purpose of this chapter is to introduce an automatic and effective but simpler-to-use system for bearing fault diagnosis that the industry can apply to its equipment without the need of hiring experts. The system uses the reassignment wavelet transform to decompose the captured vibration signals into a number of waveforms at different frequency bands. Based on the calculated values of kurtosis and root-mean-square (RMS) for each decomposed band, the system will automatically select the most suitable frequency band for detecting whether impact occurs. If the appearance of impacts is confirmed, then the system will automatically extract the characteristics of impacts for further fault classification. The effectiveness of the system was verified by both simulated data and real machine generated vibrations. Moreover, a comparison study was conducted to show the results generated by several popular techniques used in this system. The system has been designed and implemented as a series of virtual instruments with man-machine interface. The process of developing the system as well as the required data acquisition and sensors are briefly described in the chapter. By incorporating this system in bearing maintenance practice, the industry may enjoy not only the convenience in automating the process of bearing fault detection, but also the enhanced accuracy in determining the possibility of bearing failure which may eventually lead to the complete breakdown of the inspected machine.