Related Experiment Video
Updated: Sep 19, 2025
07:34
A Simple Stimulatory Device for Evoking Point-like Tactile Stimuli: A Searchlight for LFP to Spike Transitions
Published on: March 25, 2014
10.0K
Motor fault diagnosis method based on spiking convolutional neural network with multi-scale decomposition local
Gongping Wu1, Zhiwen Huang1, Zhuo Long1
1College of Electrical and Information Engineering, Changsha University of Science and Technology, Changsha, Hunan 410114, PR China.
ISA Transactions
|June 3, 2025
Summary
This study introduces an advanced motor fault diagnosis method using a Spiking Convolutional Neural Network (SCNN) with multi-scale decomposition. The approach achieves high accuracy (up to 99.49%) in identifying motor faults with reduced computational cost.
Area of Science:
- Engineering
- Artificial Intelligence
- Signal Processing
Background:
- Traditional neural networks struggle with temporal feature extraction in motor fault diagnosis.
- Accurate motor fault diagnosis is crucial for manufacturing system reliability.
Purpose of the Study:
- To propose a novel motor fault diagnosis method using Spiking Convolutional Neural Network (SCNN) with multi-scale decomposition.
- To enhance feature representation and model generalization for improved diagnostic accuracy.
Main Methods:
- Utilized Discrete Wavelet Transform (DWT) for multi-scale decomposition of motor fault signals.
- Employed Gaussian population encoding for time spike generation and Spiking Convolutional Neural Network (SCNN).
- Integrated Batch Normalization Through Time (BNTT) and surrogate gradient method for stable and efficient training.
Main Results:
- Achieved high test accuracy: 99.49% on Induction Motor Data Sets (IMDS) and 96.31% on Case Western Reserve University (CWRU) datasets.
- Demonstrated effective extraction of local features at various scales (frequency and time).
- Showcased reduced computational cost compared to traditional methods.
Conclusions:
- The proposed SCNN-based method significantly improves motor fault diagnosis accuracy and generalization.
- The integration of DWT, Gaussian encoding, BNTT, and surrogate gradients offers a robust and efficient solution.
- This method provides a reliable and computationally inexpensive approach for motor fault diagnosis in manufacturing systems.
