Abstract
Wavelet transforms are capable of separating the raw vibro-acoustic signals into different frequency and time bands. They have exhibited potentials in the detection of fault related impulsive signals by using their multi-resolution time-frequency analyses. To ensure the design of wavelet transforms is simple and the processing is not time intensive, discrete type of wavelet transforms (DWTs) become popular as they are composed of low-pass and high-pass digital filters only, making them easier to implement and processing faster. Recently, a number of publications have applied the similar type of DWTs commonly used for data compression (dyadic type of DWTs) in vibration based machine fault diagnosis. However, the results are not satisfactory. The main reasons are the poor resolution provided by DWTs and the inappropriate design of digital filters causing undesirable frequency aliasings. Without taking care of these problems, they may lead to false alarms in fault diagnosis. In this paper, we present a new family of DWTs, which mainly consists of a series of Butterworth filter banks. They are capable of providing sufficient resolutions in different time and frequency ranges, and minimizing the effect of frequency aliasing. The results have shown that the new types of DWTs are promising in solving the problems and tailor-made for machine fault diagnosis. With the help of the new DWTs, the faults that exhibit non-linear and non-stationary signals can be detected easier and the diagnosis becomes more reliable.
