Crack Size Identification for Bearings Using an Adaptive Digital Twin

Area of Science:

• Mechanical Engineering
• Data Science
• Signal Processing

Background:

• Bearing health monitoring is crucial for industrial machinery.
• Accurate fault diagnosis and crack size identification are essential for predictive maintenance.
• Existing methods often struggle with complexity and uncertainty in vibration signals.

Purpose of the Study:

• To develop an adaptive digital twin (ADT) algorithm for bearing fault diagnosis and crack size identification.
• To enhance the reliability and accuracy of fault detection using a hybrid modeling approach.
• To validate the proposed ADT technique on a benchmark dataset under various operating conditions.

Main Methods:

• Developed an ADT integrating mathematical and data-driven techniques for normal vibration signal modeling.
• Employed Gaussian process regression, enhanced by a Laguerre filter and fuzzy logic.
• Utilized an adaptive observer, combined with a proportional-integral observer and Lyapunov-based algorithm for signal estimation.
• Applied Support Vector Machine (SVM) for fault classification and crack size identification.

Main Results:

• The proposed ADT achieved an average fault diagnosis accuracy of 95.75%.
• Average crack size identification accuracies reached 97.33% for roller faults, and 98.33% for inner and outer race faults.
• The technique demonstrated robustness across diverse torque loads and motor speeds.

Conclusions:

• The developed adaptive digital twin algorithm offers a reliable and accurate solution for bearing fault diagnosis and crack size identification.
• The hybrid approach effectively handles signal complexity and uncertainty, outperforming traditional methods.
• This research contributes a robust framework for enhancing the operational safety and efficiency of rotating machinery.