A CEEMDAN-CNN-BiLSTM-Based Method for Monitoring Wear Particles With Electrostatic Sensors

Electrostatic sensors play a crucial role in wear detection; however, they are highly susceptible to environmental interference. To address signal recognition issues under such conditions, a complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN)-convolutional neural network (CNN)-bidirectional long short-term memory (BiLSTM)-based method for oil debris electrostatic sensor particle recognition is proposed. To address noise interference and improve feature extraction, this study uses CEEMDAN to reduce mode mixing and enhance target signals. Subsequently, a CNN is used to extract local features, combined with a BiLSTM network to capture long-term dependencies in time series, thereby improving signal classification accuracy. Experimental results indicate that the CEEMDAN-CNN-BiLSTM method exhibits superior recognition capabilities in both simulated and real experimental data, achieving a classification accuracy of 99.1%, surpassing methods like empirical mode decomposition (EMD)-CNN-long short-term memory (LSTM) and CEEMDAN-CNN-LSTM. Moreover, it maintains strong robustness under challenging conditions such as high temperatures and vibrations, with a classification accuracy of 98.3%.