Hybrid Learning Model for Valve Stiction Detection Based on Dynamic Time Warping and Convolutional Neural Network

In industrial production, changes in valve stiction significantly affect safety and efficiency. Traditional detection methods are inefficient and face challenges in real-time monitoring. With advances in artificial intelligence, deep learning-based automated detection methods have gained significant attention. This paper proposes an ensemble learning model based on Residual Network9(ResNet9), incorporating Dynamic Time Warping (DTW) and 2D image mapping techniques to enhance data quality and feature representation. We also integrate the Squeeze-and-Excitation (SE) module to enhance the representational power of the convolutional network, and introduce the novel ECANeXt attention module for more efficient information processing across multiple time scales. Experimental results on the International Stiction Data Base (ISDB) demonstrate that the proposed method outperforms traditional models in terms of accuracy and robustness, offering an effective solution for valve stiction detection and intelligent industrial inspection.