Robust wheel wear detection for solid carbide grinding under strong noise interference: A new approach based on improved ensemble noise-reconstructed empirical mode decomposition

Abstract

Listen

Translate article

Wheel wear detection is of great importance to achieve the desired surface quality and productivity in solid carbide grinding, while the Hilbert–Huang transform composed of empirical mode decomposition (EMD) and Hilbert transform is widely utilized in the research of grinding wear detection. However, previous studies suffer from the following drawbacks: (1) the intrinsic mode functions (IMFs) decomposed by the EMD are prone to mode mixing, hampering the extraction of weak wear features; (2) the heavy noises which could submerge the weak fault features are not considered, whereas collecting signals with a low signal-to-noise ratio (SNR) is ineluctable in practical production. Although the ensemble EMD (EEMD) is proposed to handle the first drawback, the additional Gaussian white noise can aggravate the noise interference. In this paper, an improved ensemble noise-reconstructed EMD (IENEMD) is proposed to tackle these drawbacks. The inherent noise hidden in raw signals is extracted by a newly developed noise estimation algorithm and leveraged in the IENEMD to avoid mode mixing and denoise the inherent noise itself. Both repeatable numerical simulations and practical grinding experiments are studied to validate the reliability and feasibility of the proposed IENEMD, while contrastive analyses are performed with the EMD, EEMD, and ENEMD. The results indicate that the proposed IENEMD has superior abilities of mode mixing elimination and feature extraction in processing signals with low SNR. Moreover, based on the combination of the IENEMD, the Hilbert transform, and the one-dimensional convolutional neural networks, a robust and highly accurate wheel wear detection model is established. Results indicate the model can act as a promising tool for the wear detection of grinding wheels.