Deep Learning Approach for Vibration Signals Applications.

Area of Science:

• Engineering
• Machine Learning
• Signal Processing

Background:

• Vibration signal analysis is crucial for industrial machinery health monitoring.
• Convolutional Neural Networks (CNNs) offer powerful tools for complex signal pattern recognition.
• Existing methods may lack efficiency or accuracy in specific diagnostic tasks.

Purpose of the Study:

• To investigate the application of Convolutional Neural Networks (CNNs) for vibration signal analysis.
• To explore both one-dimensional CNNs (1DCNN) and two-dimensional CNNs (2DCNN) for regression and classification tasks.
• To optimize CNN structures for improved performance in industrial diagnostics.

Main Methods:

• Utilized 1DCNN and 2DCNN models for vibration signal analysis.
• Applied CNNs to regression (surface roughness estimation) and classification (fault diagnosis, tool wear detection).
• Employed techniques like Uniform Experimental Design (UED), neural networks, multiple regression, and particle swarm optimization for CNN structure optimization.

Main Results:

• Demonstrated the effectiveness of 1DCNN for machining surface roughness estimation.
• Successfully applied CNNs for bearing fault diagnosis and tool wear classification using vibration signals.
• Optimized CNN structures showed enhanced performance in experimental validation.

Conclusions:

• CNNs provide a robust and effective framework for analyzing vibration signals in industrial applications.
• The proposed optimization methods significantly improve CNN performance for diagnostic tasks.
• This study validates the practical applicability and effectiveness of CNNs in machinery health monitoring.