Abstract:The local defect in rotating machine always gives rise to repetitive transients in the collected vibration signal. However, the transient signature is prone to be contaminated by strong background noises, thus it is a challenging task to detect the weak transients for machine fault diagnosis. In this paper, a novel adaptive tunable Q-factor wavelet transform (TQWT) filter based feature extraction method is proposed to detect repetitive transients. The emerging TQWT possesses distinct advantages over the classical constant-Q wavelet transforms, whose Q-factor can be tuned to match the oscillatory behavior of different signals, but the parameter selection for TQWT heavily relies on prior knowledge. Within our adaptive TQWT filter algorithm, the automatic optimization techniques for three TQWT parameters are implemented to achieve an optimal TQWT basis that matches the transient components. Specifically, the decomposition level is selected according to a center frequency ratio based stopping criterion, and the Q-factor and redundancy are optimized based on the minimum energy-weighted normalized wavelet entropy. Then, the adaptive TQWT decomposition can be achieved in a sparse way and result in subband signals at various wavelet scales. Further, the optimum subband signal which carries transient feature information, is identified using a normalized energy to bandwidth ratio index. Finally, the single branch reconstruction signal from the optimum subband is obtained with transient signatures via inverse TQWT, and the frequency of repetitive transients is detected using Hilbert envelope demodulation. It has been verified via numerical simulation that the proposed adaptive TQWT filter based feature extraction method can adaptively select TQWT parameters and the optimum subband for repetitive transient detection without prior knowledge. The proposed method is also applied to faulty bearing vibration signals and its effectiveness is validated.

Trend extraction based on variable time window length median filter

Trend feature extraction method based on important points in time series

Real-time Algorithm for Trend Analysis of Dynamic Data Streams

A novel piecewise linear segmentation for time series

Trade Filtering Method for Trend Following Strategy Based on LSTM-extracted Feature and Machine Learning

Integrating Data-Driven Segmentation, Local Feature Extraction and Fisher Kernel Encoding to Improve Time Series Classification

Research on the Fastest Detection Method for Weak Trends under Noise Interference

Variable Step Algorithm for Sub-Trend Sequence Searching

Wasserstein total variation filtering

A Real-Time Computer Network Trend Analysis Algorithm Based on Dynamic Data Stream in the Context of Big Data

Process Trends Analysis Via Wavelet-Domain Hidden Markov Models

A Hybrid Online Series Pattern Detection Algorithm

RobustTrend: A Huber Loss with a Combined First and Second Order Difference Regularization for Time Series Trend Filtering

An Advanced Segmental Semi-Markov Model Based Online Series Pattern Detection

Main Trend Extraction Based on Irregular Sampling Estimation and Its Application in Storage Volume of Internet Data Center

Adaptive TQWT Filter Based Feature Extraction Method and Its Application to Detection of Repetitive Transients

A novel signal feature extraction technology based on empirical wavelet transform and reverse dispersion entropy

On the Spectral Properties of Matrices Associated with Trend Filters

Pipeline signal feature extraction with improved VMD and multi-feature fusion

Outlier Mining for Multivariate Time Series Based on Sliding Window

W-FENet: Wavelet-based Fourier-Enhanced Network Model Decomposition for Multivariate Long-Term Time-Series Forecasting