Wavelets for spindle health diagnosis

Abstract

Industrial development and customer demands have increased the need to look for high-quality products at low cost and, at the same time, ensure safety during manufacturing. As a result, rotary machinery and its components have become increasingly complex, making their repairs more expensive. Therefore, many efforts must be focused in preventing breakdowns in machines, for which real-time fault diagnosis and prognosis are mandatory. Considering that the element most prone to failure in a machining center is the spindle, and with it its bearing system, the diagnosis of failures of these elements is of paramount importance. To ensure the safe operation of the bearing, some methods of fault detection have been developed based on different techniques. One of the most commonly used is vibration analysis. There are several difficulties when dealing with analyzing vibration signals, they are complex and non-stationary signals with a large amount of noise. Conventional analysis have not been able to solve this problem, thus, alternative methods such as Wavelet Transform have been gaining ground. The following research is focused in detecting bearing faults, as well as the main shaft faults, which eventually also lead to bearing damage, by using wavelets. Different signals, presenting distinct bearing fault conditions, of different data sets are evaluated for validating the proposed methodology. An exhaustive analysis has been developed for selecting the best parameters of this methodology. As results, an improvement around 20% in magnitude of bearing fault frequency peaks was found, compared to the traditional methodology. The proposal of giving more weight to high energy components allows increasing these fault frequencies, as well as reducing low frequency noise. This provides a great advantage in pursuit of an automatic fault detection. An industrial approach was also validated, by proving that the proposed methodology is more immune to noise. Even though, the magnitudes of the bearing fault peaks are diminished by noise, a comparison between the proposal and the traditional methodology reveal an increase of approximately 70% of those magnitudes. Demonstrating that the fault information is barely attenuated by noise. Also, an early diagnosis was proved, which could benefit future studies of fault prognosis. Finally, the filtering property of wavelet decomposition is exploited to limit the frequencies of the signal to few harmonics of the shaft speed. This with the aim of restricting the spectrum for detecting other faults, that mainly affect the spindle shaft, which are diagnosed by analyzing speed harmonics and subharmonics. Thus, a complete methodology is proposed to deal with the main spindle faults.