Parameter-Adaptive Feature Mode Decomposition Method for Compound Fault Feature Extraction in Rotating Machinery

Rotating machinery is a critical and easily damaged component in large-scale equipment. The coupling between the various components of the system often leads to compound failures, which can severely affect the safe operation of the equipment. Consequently, the prompt and accurate detection of fault information is essential for ensuring the safety and reliability of mechanical equipment. Recently proposed feature mode decomposition (FMD) tailored for mechanical fault feature extraction has received wide attention due to its superior filtering performance. However, the FMD is sensitive to input control parameters and the period estimation of the signal in this algorithm is also susceptible to noise interference, which will bring limitations to the practical compound fault diagnosis applications. To solve the above issues, a parameter-adaptive FMD (PAFMD) is proposed in this article. First, a spectral trend band division method is employed to initialize the finite impulse response (FIR) filter bank, enabling the adaptive acquisition of accurate filter coefficients. Then, a period estimation method based on an autocorrelation signal-to-noise ratio (ASNR) is proposed to determine the fault period accurately, thus iteratively guiding the filter to realize the accurate extraction of fault features. Finally, the decomposition modes containing fault information are automatically selected based on the period similarity of the fault signal, and a filter length selection strategy from coarse to fine is proposed to determine the optimal filter length to obtain the best fault feature modes. Both simulation and experimental validations demonstrate that the proposed PAFMD method can accurately and efficiently extract the multi-fault feature information with higher robustness and noise immunity.