Fault Detection Method for Port Equipment Using Modified Recurrent Neural Network

Smart port construction drives optimal resource allocation and operational efficiency through the deep integration of new-generation information technologies with port operations. However, its intelligent evolution poses significant challenges to traditional maintenance paradigms. The current manual operation pattern, characterized by high workload and human-induced risks, urgently requires automated fault detection technologies to enhance safety and efficiency in port equipment maintenance. This paper addresses two critical challenges in port equipment fault diagnosis: the trade-off between high-precision fault detection and efficient diagnostic output, and the dataset class imbalance caused by infrequent equipment failures. An intelligent fault diagnosis method based on an improved recurrent neural network (AE-RNN) is proposed, achieving performance breakthroughs through a three-stage architecture: 1) a single-layer LSTM-Dropout module captures long-term temporal dependencies in acoustic emission signals while mitigating overfitting; 2) a single-layer GRU-Dropout structure efficiently extracts critical feature information; 3) the Focal Loss function dynamically adjusts sample weights to alleviate class imbalance interference during model training. Experimental results on a self-built acoustic emission dataset demonstrate that the AE-RNN model achieves <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">${9 3. 0 4 \%}$</tex> classification accuracy, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">${7 3. 8 1 \%}$</tex> recall, and 82.34 % F1-score. Quantitative comparisons validate the method's effectiveness in port equipment structural fault diagnosis, providing a novel technical foundation for intelligent port equipment maintenance.