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

Area of Science:

• Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Elevator fault diagnosis is crucial for safe operation but challenged by complex, diverse data.
• Existing intelligent methods struggle to fully utilize elevator fault data characteristics.
• Accurate diagnosis is difficult due to data complexity and limited samples.

Purpose of the Study:

• To propose a novel multi-scale convolutional capsule network for improved elevator fault diagnosis.
• To address challenges of data complexity, insufficient samples, and data imbalance in elevator fault diagnosis.
• To enhance the accuracy and reliability of intelligent elevator fault diagnosis systems.

Main Methods:

• Utilized continuous wavelet transform for extracting image features from vibration signals.
• Developed a deep convolutional generative adversarial network (DCGAN) with spectral normalization and attention for data augmentation.
• Designed a capsule network integrating attention mechanisms with multi-scale convolutional layers for feature extraction and fusion.

Main Results:

• The proposed method effectively generates synthetic image samples to overcome insufficient training data.
• The capsule network successfully extracts and fuses fault features using attention mechanisms.
• Experimental results show superior diagnostic performance compared to existing methods on test datasets.

Conclusions:

• The developed multi-scale convolutional capsule network effectively handles complexity, limited samples, and data imbalance in elevator fault diagnosis.
• The method demonstrates superior diagnostic performance, offering a promising solution for real-world elevator maintenance.
• This approach enhances the reliability and safety of elevator systems through intelligent fault diagnosis.