Peridynamics for damage prediction in ships and offshore structures

  • Cong Nguyen Tien

Student thesis: Doctoral Thesis

Abstract

Ships and offshore structures can experience damages due to many reasons such as collisions, groundings, explosions, corrosion, fatigue, overloading, or extreme conditions, etc. To date, the prediction of progressive damages in these structures is a challenging research area. The classical continuum mechanics uses partial differential equations which become invalid in the presence of discontinuities. By contrast, the recently introduced nonlocal peridynamics (PD) theory uses integrodifferential equations that are valid in both continuous and discontinuous models. Therefore, the peridynamics theory is highly suitable for predicting crack initiation and crack growth. In this thesis, progressive damages in ship and offshore structures are predicted by using peridynamics. To do that, first, novel PD models for predicting linear elastic deformations of 3D beam structures and 3D shell structures are developed. The deformations of 3D beams and 3D shell structures predicted by using the developed PD beam and shell models agree very well with the FEA results with less than 3% relative errors. It is also found that the developed PD beam and shell models are suitable for predicting progressive brittle damages in ship and offshore structures. The PD shell model can also predict the ultimate bending moment of a ship with only 0.102% difference from the experimental result.;Second, novel nonlinear PD models for predicting damages in one-dimensional (1D), two-dimensional (2D), and three-dimensional (3D) structures, 3D beam structures, and plates subjected to large deformations are developed. The large deformations structures predicted by using the developed nonlinear PD models agree very well with the FEA results with maximum 5% relative errors. The developed nonlinear PD models show a capability to predict progressive damages for many complex problems. The damage patterns captured by the nonlinear PD models agree very well with the experimental results in the literature. Third, a novel energy-based PD model for fatigue cracking is also developed. Instead of using the cyclic bond strain range for PD fatigue equations available in the literature, the energy-based PD fatigue model proposes a definition of the cyclic bond energy release rate range and use this term for PD fatigue equations. The fatigue life of the structure predicted by the energy-based PD fatigue model is 4.108% different from the experimental results while the predicted fatigue crack growth, �{9D}{91}{9E} �{88}{92} �{9D}{91}{81} curve agrees very well with experimental results. The energy-based PD fatigue model can be more suitable for beam and shell structures since in these structures, the bond energy release rate is unique although the bond strain consists of in-plane, shear, and bending components.;Finally, to reduce the computational cost for PD simulations, novel 1D and 2D peridynamic-based machine learning models for damage prediction are developed. The relations between displacements of a material point and the displacements of its family members as well as the externally applied forces are obtained by using linear regression. The machine learning models can easily be coupled with the PD models. Specifically, the PD model is used for the regions that are near crack surfaces or near boundary areas. Meanwhile, the ML model is used for the remaining regions to reduce the computational cost. Like the traditional PD model it is found that the coupled PD-ML model is also suitable for damage prediction. The crack patterns predicted by using the coupled PD-ML model agree very well with experimental results in many complex problems. Therefore, the hybrid approach of coupling ML with PD can be a potential approach for future research to reduce the computational cost for PD simulations while the capability of PD
Date of Award2 Feb 2021
Original languageEnglish
Awarding Institution
  • University Of Strathclyde
SponsorsUniversity of Strathclyde
SupervisorSelda Oterkus (Supervisor) & Evangelos Boulougouris (Supervisor)

Cite this

'