Identification of fault frequency variation in the envelope spectrum in the vibration-based local damage detection in possible changing load/speed conditions

The problem of local damage diagnosis (based on the detection of impulsive and periodic signals) is discussed. Both features should be checked, as fault frequency must be linked to the true value calculated for a given machine and speed. The precise estimation of the fault frequency is hard due to several factors. If a speed fluctuation exists, it is solved by order analysis. A wider perspective is proposed here, namely, an automatic statistical approach to analyze the distribution of estimated fault frequencies. We propose a procedure to evaluate whether the fault frequency is constant or not. The algorithm uses frequency estimation based on peak detection in the envelope spectrum and statistical testing. We present simulation studies and industrial examples. We have found that if the fault frequency is not constant and its distribution does not follow Gaussian shape with minor variance, then one should use more advanced techniques, e.g. order analysis.

Automated summary

This paper discusses the problem of local damage diagnosis based on the detection of impulsive and periodic signals. Both features need to be checked, as fault frequency should be related to the true value calculated for a given machine and speed. Precisely estimating the fault frequency is challenging due to various factors. A broader perspective is proposed here, introducing an automatic statistical approach to analyze the distribution of estimated fault frequencies. The algorithm uses frequency estimation based on peak detection in the envelope spectrum and statistical testing. Simulation studies and industrial examples are presented, indicating that more advanced techniques, such as order analysis, should be used if the fault frequency is not constant and its distribution does not follow a Gaussian shape with minor variance.