A multi-scale convolution capsule network with data augmentation and attention mechanisms for elevator fault diagnosis

Elevator fault diagnosis plays a vital role in supporting maintenance practices that ensure safe operation. However, because of the complexity and diversity of elevator fault data, existing intelligent fault diagnosis methods are often struggle to fully exploit the data's characteristics, making accurate fault diagnosis challenging. To address this problem, a multi-scale convolutional capsule network with data augmentation and attention mechanisms is proposed for elevator fault diagnosis. Firstly, a continuous wavelet transform method is used to extract image features from the original vibration signals of elevators. Subsequently, a novel model is introduced to optimize the discriminator and generator of deep convolutional generative adversarial networks. By employing spectral normalization and space-time attention mechanisms, the model generates additional synthetic image samples to mitigate the issue of insufficient training data. Furthermore, a capsule network is designed to integrate attention mechanisms into multi-scale convolutional layers, enabling the extraction and fusion of fault features. Finally, the effectiveness of this method is verified through experiments and compared with existing intelligent elevator fault diagnosis methods. The results demonstrate that the proposed method effectively addresses the complexity, limited training samples, and imbalance of elevator data in real-world environments. Moreover, its diagnostic performance on the test dataset is superior to that of other commonly used methods.