Tien Vuong Nguyen

The increasing complexity of mechanical systems creates challenges in modeling, monitoring, and predicting their behavior using traditional methods. Physics-based models require extensive expertise, time-consuming calibration, and often fail to capture real-world complexities. Engineers need practical solutions that can learn from data and provide actionable insights without requiring deep theoretical knowledge or manual parameter tuning for each specific application. 

This research applies machine learning to solve challenging problems in mechanical engineering. The approach demonstrates that neural networks can learn complex mechanical behaviors directly from data, eliminating the need for intricate mathematical formulations. Two distinct applications illustrate this capability: first, predicting friction forces in mechanical systems where traditional models struggle with accuracy; second, estimating clearance in mechanical joints using sensor data, something previously impossible to measure during operation.

Both leverage time-series neural networks to transform raw measurements into useful predictions. The benefits are transformative: engineers can deploy models without specialized knowledge in mathematical modeling, systems can be monitored in real-time for early detection of degradation, maintenance becomes predictive rather than reactive, and development time is dramatically reduced. This data-driven approach democratizes advanced modeling capabilities, making sophisticated analysis accessible to practitioners without requiring years of expertise in theoretical mechanics.

Unlike traditional physics-based models requiring expert calibration, vibration monitoring systems with limited predictive power, or commercial solutions offering only basic anomaly detection, this machine learning approach learns directly from data with minimal domain expertise. It adapts automatically to different systems and conditions, providing accurate predictions across diverse applications while remaining practical and accessible for real-world engineering use.