Multidirectional odd-weighted transform: Algorithm and applications

Du, Shuaicong

doi:10.1016/j.eswa.2026.133758

Multidirectional odd-weighted transform: Algorithm and applications

2026-08-01

0

PRE

AI

DOI：10.1016/j.eswa.2026.133758

Original

Original request for help

Save

Abstract

En 中文

Accurate characterization of multimodal signals containing both transient and harmonic components remains a key challenge in condition monitoring. Classical single-direction time-frequency (TF) analysis (TFA) methods cannot effectively process multimodal signals; mode-division-based TFA methods suffer from poor resolution and mode misclassification. Hence, a new TFA technique termed multidirectional odd-weighted transform (MOWT) is developed. Differing from existing mode-division-based TFA methods with a divide and conquer strategy, the MOWT aims to generate highly readable TF representations (TFRs) by exploiting the geometric characteristics of the energy distribution within the odd-weighted TFR to adaptively extract TF coefficients along TF ridges in the rough TFR. Specifically, it is implemented by adopting partial derivatives of the original TFR and an eight-direction selective positioning algorithm to identify ridge points, and mapping ridge points onto the original plane by a ridge reassignment operator. Moreover, the mask-connected recognition algorithm is proposed to extract TF ridges. Both simulated and experimental signals validated that the MOWT exhibits superior performance, indicating its suitability for nondestructive testing and fault diagnosis.

Keywords:

Time-frequency analysis

Multimodal signal

Odd-weighted

Eight-direction selective positioning algorithm

Mask-connected recognition algorithm

Fault diagnosis

Multidirectional odd-weighted transform: Algorithm and applications

Abstract

Journal

Organization

Cited Papers