Local Optimal Scaling Chirplet Transform for Processing Nonstationary Mechanical Vibration Signals

Vibration signals collected from complex rotating machines often contain close-spaced or nonproportional instantaneous frequencies (IFs), including crossed IFs, and current time–frequency analysis (TFA) methods should be improved or are difficult to characterize the above IFs and detect mechanical faults with high time–frequency resolution. To tackle the above challenge, a TFA algorithm, termed local optimal scaling chirplet transform (CT) (LOSCT), is proposed. First, based on the scaling-basis CT (SBCT), the scaling chirplet basis is introduced to calculate various time–frequency representations (TFRs); then, Rényi entropy-based local optimal theory is constructed to capture local optimal TFRs, and finally, the local maximum extraction criterion is defined to calculate ideal time–frequency amplitudes on IF curves from the optimal TFR. The primary contribution is that the LOSCT can process nonstationary signals, whose IF are nonproportional or close-spaced, with high time–frequency concentration and detect mechanical faults. The LOSCT is verified by two simulated signals, whose IF curves are close-spaced or crossed, respectively. A comparative analysis with current TFA algorithms is used to evaluate the superiority of the developed technique. Finally, the engineering applications for processing mechanical vibration signals, i.e., fault bearing and planetary gearbox signals, are discussed.