75 / 2025-04-01 16:40:16

Multi-Sensor Fault Detection Based on Wavelet Time-Frequency Maps

Fault Detection,Time-frequency Maps,Wavelet Transform,Two-dimensional Convolutional Neural Network

Most frequency-domain-based fault detection methods only extract partial data features from multi-sensor data, such as frequency-domain characteristics or the coupling relationships between sensors. In this paper, a fault detection method based on wavelet time-frequency maps is proposed to enhance detection accuracy. To begin with, wavelet transform is applied to perform time-frequency analysis on time-series data, generating time-frequency maps that capture the time-frequency properties of different frequency components. Then, the time-frequency maps are sliced along the time axis to refine the capture of local time-frequency features, and the covariance matrix of each slice is computed to extract the correlation characteristics between sensors. Finally, a two-dimensional convolutional neural network (2D-CNN) is constructed to achieve fault detection using the covariance matrices as input. Simulations on the tennessee eastman process (TEP) dataset demonstrate that the proposed method effectively extracts both time-frequency features and inter-sensor coupling relationships, achieving high accuracy and robustness in multi-sensor fault detection.