Fluid-Induced Vibration and Buckling of Pipes on Elastic Foundations: A Physics-Informed Neural Networks Approach

This study presents an analysis of transverse vibration behavior of a fluid-conveying pipe mounted on an elastic foundation, incorporating both classical analytical techniques and modern physics-informed neural network (PINN) methodologies. A partial differential equation (PDE) architecture is developed to approximate the solution by embedding the physics PDE, initial, and boundary conditions directly into the loss function of a deep neural network. A one-dimensional fourth-order PDE is employed to model governing transverse displacement derived from Euler–Bernoulli beam theory, with additional terms representing fluid inertia, flow-induced excitation, and stochastic force modelled as Gaussian white noise. The governing PDE is decomposed via separation of variables into spatial and temporal components, and modal analysis is employed to determine the natural frequencies and mode shapes under free–free boundary conditions. The influence of varying flow velocities and excitation frequencies on critical buckling behavior and mode shape deformation is analyzed. The network is trained using the Resilient Backpropagation (RProp) optimizer. A preliminary validation study is presented in which a baseline PINN is benchmarked against analytical modal solutions for a fluid-conveying pipe on an elastic foundation under deterministic excitation. The results demonstrate the capability of PINNs to accurately capture complex vibrational phenomena, offering a robust framework for data-driven modelling of fluid–structure interactions in engineering applications.