Bearing Fault Diagnosis Based on Stacking Ensemble Learning for Fusion of Spatiotemporal Features

To address the limitations of existing rolling bearing fault diagnosis techniques in terms of diagnostic accuracy, generalization capability, and performance improvement, this paper proposes a method based on Stacking ensemble learning that integrates spatiotemporal features. This method employs Kernel Principal Component Analysis (KPCA) to extract spatial features and Long Short-Term Memory (LSTM) networks to extract temporal features from the signals, providing a comprehensive representation of their spatiotemporal properties. Subsequently, Convolutional Neural Networks (CNN) and Support Vector Machines (SVM) are used to establish classification models based on the extracted features. Finally, the Stacking ensemble learning algorithm integrates these models to deeply explore the spatiotemporal correlations within the data, leveraging the strengths of each model to achieve efficient fault diagnosis of bearing signals. Experimental results demonstrate that this method significantly improves diagnostic accuracy compared to traditional approaches and other existing models, proving its effectiveness and superiority in rolling bearing fault diagnosis.