276 / 2025-06-15 22:29:42

Prediction of Wear in Rotary Dynamic-Seal Friction Pairs Based on Acoustic Emission Technology

Rotary dynamic seal,Friction pair wear,Acoustic emission technology,Bayesian optimization,Deep learning model

To address the complexity of the dynamic wear behavior in rotary dynamic-seal friction pairs and the challenges of real-time monitoring, an intelligent prediction method based on acoustic emission (AE) technology and a Bayesian optimization (BO)-enhanced deep learning model is proposed in this study. An AE friction experimental platform was established to capture dynamic AE signals from M106K-WC friction pairs, and a high-signal-to-noise ratio dataset was constructed using data augmentation and normalization techniques. A prediction model named BO-CNN-BiLSTM-AM was developed, integrating a one-dimensional convolutional neural networks (CNNs), bidirectional long short-term memory (BiLSTM) networks, and a multi-head attention mechanism (AM). Bayesian optimization was employed to adaptively adjust the hyperparameters (e.g., convolutional kernel size, BiLSTM layers, and number of attention heads), significantly improving the feature extraction efficiency and prediction accuracy. Experimental results demonstrated that the model achieved a mean absolute error below 5% and a coefficient of determination of 0.972 on the test set, with prediction errors below 3% in the moderate wear stages. Compared to traditional models (CNN, CNN-BiLSTM, and CNN-BiLSTM-AM), the proposed model effectively addressed the noise interference and nonlinear temporal modeling challenges through multi-modal feature fusion and dynamic parameter tuning. This study pioneered the application of AE technology to monitor wear in rotary dynamic seals, resolving composite burst-type (material wear) and continuous-type (end-face vibration) signal features, and it provides a data-driven solution for the high-precision, low-latency intelligent maintenance of industrial equipment.