Efficient implementation of envelope analysis on resources limited wireless sensor nodes for accurate bearing fault diagnosis

With the fast development of electronics and wireless communication technologies in recent years, intelligent wireless sensor nodes are becoming increasingly popular in the online machinery condition monitoring systems. They bring a number of benefits, such as reduced investment on the installation and maintenance of expensive communication cables, ease of deployment and upgrading. For the condition monitoring of dynamic signals, distributed computation on wireless sensor nodes is getting popular with wireless sensor nodes becoming more computation powerful and power efficient. As a widely recognised algorithm for bearing fault diagnosis, envelope analysis has been previously proved suitable for being embedded on the wireless sensor nodes to effectively extract fault features from common machinery components such as bearings and gears. As a continuation, this paper studies into several envelope detection methods, including Hilbert transform, spectral correlation, band-pass squared rectifier and short-time RMS. Regarding to the fact that only low frequency components in the bearing envelope is of interest, spectral correlation can be simplified for fast calculation and short-time RMS method can be considered as a simplified band-pass squared rectifier, in which partial aliasing is allowed. Thereafter, spectral correlation and short-time RMS are employed to speed up the calculation of envelope analysis on a wireless sensor node, which thereafter provides the potential to reduce power consumption of wireless sensor nodes. The computation speed comparison shows that the spectral correlation method and short-time RMS can speed up the computation speed by more than two times and five times in comparison with the Hilbert transform method. The simulation study shows that spectral correlation and short-time RMS based methods achieves similar level of accuracy as Hilbert transform. Furthermore, the experimental study shows that spectral correlation and short-time RMS based methods can well reveal the simulated three types of bearing faults while with the computation speed significantly improved.