Development of a Brush Wear State Diagnosis Framework for PMDC Motors via Digital-Twin-based Virtual Data Fusion

Key Points

• The study aims to improve the diagnosis of brush wear in PMDC motors by integrating real vibration data with virtual data from a digital twin.
• Developed a framework for data fusion integrating real and digital twin data.
• Evaluated the similarity of vibration data using DTW-based coefficients.
• Determined the necessary amount of virtual data automatically using simulator reliability.
• Classified the brush wear states using models such as XGBoost, 1D-ResNet, TCN, and TST.
• Achieved over 97% accuracy across all models, with XGBoost reaching 99.4% accuracy.
• Confirmed improvement in diagnosis accuracy using the data fusion framework compared to using real data alone.