The rotation speed of a rotating machine frequently fluctuates when approaching the machine’s working speed. Current non-stationary signal processing methods hardly detect the rotation speed fluctuations. To tackle the shortage of existing instantaneous frequency evaluation methods, this paper adopts a fast path optimization algorithm for instantaneous frequencies of for non-stationary signals, which is verified with simulation signals and actually measured signals. On this basis the short-time order tracking principle of the parameterized time-frequency analysis theory is elaborated in this paper. Based on the instantaneous frequency variation function of the estimated vibration signals, this method can construct corresponding rotation operators to rotate the time-frequency characteristics of the rotation speed fluctuation signals. Moreover, it can obtain the short-time order tracking spectrum of each time-frequency window which reserves all instantaneous information of the rotation speed fluctuation signals. Experiment results show that this method can accurately extract the characteristics of typical failures of rotating machines under the working conditions of fluctuating rotation speeds, providing a decent precision and reliability.