Accelerated Fatigue Reliability Analysis of Stiffened Sections Using Deep Learning
Abstract
Fatigue is one of the main failure mechanisms in structures subjected to fluctuating loads such as bridges and ships. If inadequately designed for such loads, fatigue can be detrimental to the safety of the structure. When fatigue cracks reach a certain size, sudden fracture failure or yielding of the reduced section can occur. Accordingly, quantifying the critical crack size is essential for determining the reliability of fatigue critical structures under growing cracks. Failure Assessment Diagrams (FADs) can be used to determine the critical crack size or whether the state of the crack is acceptable or not at a particular instant in time. Due to the presence of uncertainties in loads, material properties and crack growth behavior, probabilistic analysis is essential to understand the fatigue performance of the structure over its service life. A time dependent reliability profile for the structure can be established to help schedule maintenance and repair activities. However, probabilistic analysis of crack growth under complex geometrical and loading conditions can be very expensive computationally. Deep learning is a useful tool that is used in this study to curtail this lengthy process by establishing multi-variate non-linear approximations for complex fatigue crack growth profiles. This study proposes a framework for establishing the fatigue reliability profiles of stiffened panels under uncertainty. Monte Carlo simulation is used to draw samples from relevant probabilistic parameters and establish the time dependent reliability profile of the structure under propagating cracks. Deep learning is adopted to improve the computational efficiency of the probabilistic analysis in establishing the probabilistic crack growth profiles. The proposed framework is illustrated on a bridge with stiffened tub girders subjected to fatigue loading.
Collections
- OSU Theses [15752]