Digital Twin-Based Technical Research on Comprehensive Gear Fault Diagnosis and Structural Performance Evaluation

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Digital Twin Technology

Background:

• Modern industrial gearboxes are critical transmission components, but current monitoring methods suffer from poor data utilization and incomplete evaluation.
• Existing systems lack comprehensive real-time fault diagnosis and performance prediction capabilities, hindering effective predictive maintenance.
• Challenges include limited monitoring, single detection indices, and low data utilization in current gearbox operational assessments.

Purpose of the Study:

• To develop an integrated gearbox monitoring system leveraging digital twin and artificial intelligence for enhanced operational assessment.
• To achieve real-time fault diagnosis, performance prediction, and dynamic visualization of gear health through virtual-real mapping.
• To lay the foundation for advanced predictive maintenance applications in industrial equipment.

Main Methods:

• Developed a five-layer architecture for the digital twin system: functional service, software support, model integration, data-driven, and digital twin layers.
• Utilized HyperMesh and ABAQUS for refined mesh generation and thermal fluid solid coupling simulations to analyze gear stress distribution.
• Implemented a Gaussian process regression (GPR) model for stress prediction and a fault diagnosis algorithm combining wavelet transform and a depth residual shrinkage network (DRSN).

Main Results:

• The integrated system achieved over 99% accuracy in fault diagnosis across various fault types (broken tooth, wear, pitting).
• The stress prediction model demonstrated high accuracy with an average R² of 0.9339 for the driving wheel and 0.9497 for the driven wheel.
• The system successfully supports real-time display of 3D cloud images, visualizing gear stress distribution.

Conclusions:

• The proposed digital twin and AI-based system offers a novel and effective method for intelligent industrial equipment monitoring.
• The system's multi-source data fusion, interaction, and visualization capabilities significantly advance predictive maintenance.
• Future work should focus on improving finite element simulation accuracy and addressing challenges in obtaining actual stress data.