Six interlocking directions, all aimed at reliable real-world diagnostics.
Detecting and classifying mechanical faults in rotating machinery (bearings, gearboxes, planetary roller screws) directly from raw vibration signals using both physics-driven features and deep learning.
Estimating Remaining Useful Life (RUL) and degradation trends of industrial equipment to enable predictive maintenance and reduce unplanned downtime.
Applying CNN, Transformer and transfer-learning models to fault classification, with a focus on explainability and robustness across operating conditions.
Building physics-grounded digital twins (FEA + multibody dynamics) of mechanical systems to generate labeled fault data where real-world failure samples are scarce.
Compressing and deploying diagnostic models onto edge hardware for standalone, real-time inference at the machine level.
Dedicated study of fault mechanisms in gearboxes and planetary roller screws, combining analytical contact mechanics with data-driven detection.
