Explainable Physics-Informed Deep Learning Framework for Thermoelastic Analysis of Rotating Ti-6Al-4V Disks

Abstract

The integration of artificial intelligence into engineering analysis has enabled the development of intelligent computational frameworks for predicting complex physical phenomena. In this study, a hybrid thermoelastic analysis framework combining analytical modeling, the Chebyshev pseudo-spectral method, and Physics-Informed Neural Networks (PINNs) is proposed for rotating Ti-6Al-4V disks subjected to thermal loading. The disk geometry is defined by an inner radius of 50 mm and an outer radius of 90 mm, while the temperature varies linearly from 40°C to 120°C. The governing thermoelastic equations are solved analytically and numerically using the Chebyshev pseudo-spectral method, providing highly accurate stress and displacement distributions. The generated numerical dataset is subsequently employed to train a PINN model, in which the governing physical equations are incorporated directly into the learning process. Radial stress, tangential stress, and radial displacement are predicted as functions of radial position and temperature. The results demonstrate excellent agreement between analytical, numerical, and PINN predictions, confirming the accuracy and reliability of the proposed framework. Increasing temperature levels are observed to significantly affect both stress distributions and radial displacement throughout the disk. Compared with conventional numerical approaches, the developed PINN model provides accurate predictions with reduced computational effort. The proposed methodology contributes to the integration of physics-based modeling and artificial intelligence for advanced thermoelastic analysis and offers potential applications in digital twin systems, structural health monitoring, and predictive maintenance of rotating mechanical components operating under thermomechanical environments.