Physics Informed Gaussian Process for Bolt Tension Estimation

Bolted joints are fundamental components in many engineering applications. Therefore, the need for monitoring their tension over their life span is an essential for securing their integrity. Modelling of the dynamics of a bolt has shown success through Euler-Bernoulli beam theory where a relationship of boundary conditions and tension allows determining the changes in the modal parameters. However, the widespread adoption of this approach has faces challenges, as obtaining high-fidelity data for bolts under all tension phases is often unfeasible in practice, particularly for those in low tension. Nevertheless, merging data and prior physics knowledge can provide practical constraints for bolt tension estimation in areas lacking observational data. This study establishes its foundation by developing a stochastic model for bolt tension estimation through the integration of bolt data-driven and physics-based predictions using Gaussian Process Regression (GPR). The core concept of this approach involves predicting data observations through a stochastic simulation using a physics-based model, particularly in scenarios where observational data is absent. The proposed methodology is validated with experimental data to critically evaluate its performance.