An Expert System for Rotating Machine Fault Detection Using Vibration Signal Analysis.

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Preventive maintenance in industrial enterprises relies on early fault detection to prevent downtime and ensure safety.
• Vibration signal analysis is crucial for assessing the health and condition of rotating machines.
• Existing methods require robust signal processing and feature extraction for accurate fault diagnosis.

Purpose of the Study:

• To develop and validate a fault detection system for rotating machines using vibration signal analysis.
• To investigate the effectiveness of empirical mode decomposition for signal conditioning.
• To identify optimal feature combinations and classifiers for high-accuracy fault classification.

Main Methods:

• Acquisition of 3D vibration signals from large induction motors in healthy and faulty states.
• Signal denoising and conditioning using empirical mode decomposition (EMD).
• Multi-domain feature extraction (temporal and spectral) and classification using Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Decision Trees, and Linear Discriminant Analysis (LDA).

Main Results:

• Empirical mode decomposition effectively conditioned the vibration signals.
• A hybrid combination of temporal and spectral features yielded the most discriminant information.
• Support Vector Machines with a Gaussian kernel achieved the highest performance: 98.2% accuracy, 96.6% sensitivity, 100% specificity, and 1.8% error rate.

Conclusions:

• The proposed fault detection system, utilizing vibration signal analysis and SVM classification, is highly effective for rotating machinery.
• The integration of EMD for signal conditioning and hybrid feature extraction significantly improves fault detection accuracy.
• This approach offers a reliable solution for preventive maintenance, enhancing industrial equipment reliability and safety.