Transfer learning enhanced physics-informed neural network for buried pipeline deformation analysis under permanent ground deformation

Pipelines are essential infrastructure networks, critical for transporting materials over long distances. However, buried pipelines are susceptible to permanent ground deformation (PGD) which can introduce substantial strain, potentially leading to structural issues such as bending, wrinkling, or cracking. This paper proposes a deep learning framework based on physics-informed neural networks (PINNs), enhanced with transfer learning (TL), to analyze buried pipeline deformation under PGD. By incorporating the residuals of governing physical equations into the loss function, the PINN model integrates the underlying physical laws directly into the deep neural network. The PINN model demonstrates superior generalization and parameter inversion capabilities by integrating physical mechanisms, reducing the dependence on large datasets, compared to traditional deep neural networks (DNNs). Using TL, the model is efficiently extended to new scenarios by partially re-training the network, enhancing computational efficiency. The effectiveness of the proposed model is demonstrated through three scenarios, using only a small amount of training data. A parametric study further shows that the TL-enhanced PINN model offers an efficient surrogate model to reproduce the influence of key parameters on pipeline deformation under PGD. The TL-enhanced PINN framework provides a robust and efficient tool for accurate pipeline deformation prediction, even with limited data, making it suitable for practical engineering applications.