Bearing fault diagnosis via a parameter-optimized feature mode decomposition

Because actual vibration signal collected from mechanical equipment (e.g., wind turbines and high-speed trains) are strongly non-stationary and have low signal-to-noise ratios, which indicates that it is arduous to excavate beneficial fault characteristics from collected vibration data via traditional detection methods, such as Fourier transform and envelope demodulation. Feature mode decomposition (FMD) is a recently proposed new signal analysis approach that has been smoothly applied to mechanical fault diagnosis. Nevertheless, input parameters (i.e., the mode number K and filtering length L) must be defined artificially when FMD is introduced for vibration signal analysis. That is, the decomposition performance of FMD is easily affected by its parameter setting. To handle this, this study proposes a parameter-optimized feature mode decomposition (POFMD) for bearing fault diagnosis. Firstly, a new index termed signal cycle kurtosis-to-noise ratio (SCKNR) is designed as an objective function of FMD in adaptive parameter searching, which can take the impact intensity and noise immunity of the signal into consideration simultaneously. Subsequently, particle swarm optimization (PSO) based on SCKNR is adopted to automatically select the optimal combination parameters (i.e., the mode number K and filtering length L) of FMD. Meanwhile, POFMD is used to separate the collected data into a series of mode components. Next, main mode component is selected based on maximum criterion of feature energy rate of squared envelope spectrum (SES-FER). Ultimately, squared envelope spectrum of main mode component is computed to achieve fault identification of wind turbines and high-speed train bearings. A study of simulated signals and two cases validate the availability of our presented approach. Furthermore, our presented POFMD is more efficient in extracting fault characteristics than the well-versed variational mode decomposition (VMD), symplectic geometry mode decomposition (SGMD) and spectral kurtosis (SK).