A Physics-Guided Dual-Sensor Framework for Bearing Fault Diagnosis in PMDC Motor Drives

Area of Science:

• Mechanical Engineering
• Condition Monitoring
• Signal Processing

Background:

• Rolling-element bearing faults are a major cause of mechanical failure in rotating machinery.
• Vibration-based diagnostics in Permanent Magnetic DC (PMDC) motors are challenged by variable torque, leading to load-dependent excitation and commutation disturbances.
• These disturbances bias amplitude measurements and decrease the reliability of traditional statistical features used for fault diagnosis.

Purpose of the Study:

• To propose a physics-guided, dual-sensor feature framework, Cross-Reference Energy Attention (CREA), for enhanced three-class bearing state diagnosis in PMDC motors.
• To develop a method that isolates fault-relevant information while suppressing motor-generated disturbances.
• To validate the effectiveness of CREA against conventional methods under variable torque conditions.

Main Methods:

• Implementation of CREA, a dual-sensor framework utilizing a hardware-agnostic, empirically selected mid-frequency carrier band to isolate fault signatures.
• Incorporation of a spatially separated reference sensor to assess signal transmission consistency and suppress local motor noise.
• Experimental validation on a PMDC motor dynamometer with seeded bearing defects, employing GroupKFold cross-validation and per-run normalization.

Main Results:

• Conventional motor-side baseline features showed degraded accuracy (0.495 ± 0.110) under tested conditions.
• The four-feature CREA representation achieved significantly higher accuracy (0.999 ± 0.002).
• Ablation and SHAP analysis confirmed the dominant contribution of carrier-band energy features and the complementary role of cross-sensor metrics.

Conclusions:

• CREA effectively isolates fault-specific vibrations and suppresses noise in PMDC motors operating under variable torque.
• The dual-sensor approach provides robust and highly accurate bearing state diagnosis, outperforming conventional methods.
• CREA's physics-guided design ensures reliable condition monitoring by leveraging structural transmission characteristics.