A Multi-Task Causal Knowledge Fault Diagnosis Method for PMSM-ITSF Based on Meta-Learning.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Robotics

Background:

• Diagnosing inter-turn short circuit faults in industrial robot permanent magnet synchronous motors is challenging due to limited and sparse fault data.
• Existing methods struggle with accurate fault degree evaluation, location locking, and cause tracking, leading to potential misdiagnoses.

Purpose of the Study:

• To propose a novel multi-task causal knowledge fault diagnosis method for inter-turn short circuits in permanent magnet synchronous motors based on meta-learning.
• To address the limitations of small and sparse fault sample data in achieving quick and accurate fault diagnosis.

Main Methods:

• Investigated motor parameter variations under inter-turn short circuit faults and selected characteristic quantities.
• Utilized Simulink, Simplorer, and Maxwell for comprehensive simulations to generate labeled fault data.
• Developed a meta-learning network for multi-task synchronous diagnosis of fault degree and position.
• Constructed a Neo4j database incorporating causal knowledge of motor inter-turn short circuit faults.

Main Results:

• Achieved multi-task synchronous diagnosis of fault degree and position.
• Demonstrated accurate diagnosis of fault location, degree, and cause under varying voltage unbalance.
• Attained high diagnostic accuracy: 99.75 ± 0.25% for fault degree and 99.45 ± 0.21% for fault location and degree.

Conclusions:

• The proposed meta-learning based multi-task causal knowledge method effectively diagnoses inter-turn short circuit faults in permanent magnet synchronous motors.
• This approach significantly enhances diagnostic accuracy and reliability for industrial robot motor fault detection.