Multi-Sensor Vibration Signal Based Three-Stage Fault Prediction for Rotating Mechanical Equipment

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Predictive Maintenance

Background:

• Reducing maintenance costs and preventing accidents in rotating machinery requires effective fault prediction.
• Current methods focus on fault diagnosis and remaining useful life (RUL) prediction, lacking specific health and fault type identification.
• Predicting degradation periods and fault types in advance is crucial for proactive maintenance.

Purpose of the Study:

• To develop a novel three-stage method for simultaneous identification of degradation periods and fault types in rotating mechanical equipment.
• To provide a comprehensive fault prediction solution that goes beyond RUL estimation.
• To enhance the accuracy and specificity of fault prediction for industrial machinery.

Main Methods:

• A hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) network extracts spatiotemporal features from multi-sensor vibration signals.
• An attention-bidirectional (Bi)-LSTM network is employed as a regression model to predict future feature trends.
• A Support Vector Classification (SVC) model identifies specific degradation periods and fault types using predicted features.

Main Results:

• The proposed method successfully identifies degradation periods and fault types simultaneously.
• The three-stage approach demonstrates effectiveness in comprehensive fault prediction.
• Validation on the NSF I/UCR Center for Intelligent Maintenance Systems (IMS) dataset confirms the method's feasibility and efficiency.

Conclusions:

• The novel three-stage fault prediction method offers a significant advancement in monitoring rotating mechanical equipment.
• This approach enables more precise maintenance planning by identifying specific fault types and degradation stages.
• The study highlights the potential of integrating deep learning models for robust predictive maintenance solutions.