Multiscale Entropy Constrained Decomposition for Time-Frequency Enhancement in Industrial Machinery Vibration Signals

The diagnostics of rotational machinery using vibration signal is limited due to the presence of noise and components overlap in time-frequency (TF) domain representation. Traditional signal decomposition algorithms, like empirical mode decomposition (EMD) and variational mode decomposition (VMD), would often extract redundant or noise-dominated modes, leading to loss of the diagnostic accuracy. In this paper, a new entropy constrained mode decomposition framework is introduced to improve the clarity of signals, by using the multiscale permutation entropy (MPE) calculated on the Teager-Kaiser energy series (TKES) of each mode. Unlike fixed filtering or postprocessing approaches, this approach quantifies the structural complexity of the decomposed modes at various time scales. Modes with low entropy are regarded as fault-related, representing deterministic transients, while high-entropy modes are discarded as noise or irrelevant information. The chosen modes are then reconstructed to create the denoised signal which in turn is analyzed through a sharpened TF representation for improved resolution and interpretability. Experimental results obtained on a synthetic ball bearing fault vibration signal demonstrate that the proposed approach is capable of achieving better transient localization as opposed to baseline decomposition methods. The proposed method provides a data-based approach to mode selection that offers robust fault diagnosis.