Adaptive Gaussian process regression as an alternative to FEM for prediction of stress intensity factor to assess fatigue degradation in offshore pipeline

The linear elastic fracture mechanics (LEFM) based remnant fatigue life (RFL) assessment of offshore pipeline is used to determine the inspection interval frequency of the aforementioned asset. One of the vital factors of the LEFM approach that determines the accuracy of the RFL estimate (and in turn of the inspection interval frequency) is the Stress Intensity Factor (SIF), which must be evaluated as accurately as possible. For simple crack geometries numerous closed-form equations available in various handbooks and industrial standards provide accurate SIF results. However, it is a common industry practice to utilize finite element method (FEM) for evaluating the SIF for the intricate crack geometries and the complex loading conditions. Although, FEM is known for its accurate SIF calculation, but due to its high computational expense and time-consumption, cycle-by-cycle SIF evaluation (required for the LEFM based RFL assessment) makes the aforementioned method quite laborious. Furthermore, using FEM to evaluate SIF for thousands of pipeline location (undergoing fatigue degradation) on an offshore platform seems to be impractical. Thus, in this manuscript authors have proposed a computationally inexpensive adaptive Gaussian process regression model (AGPRM) which may be utilized as an alternative to FEM for prediction of SIF to assess fatigue degradation in offshore pipeline. The training and testing data for AGPRM consists of 105 and 50 data points (load (L), crack depth (a), half-crack length (c) and SIF values), respectively. Latin Hypercube Sampling (LHS) is used to generate (L, a and c) values while SIF values are evaluated using FEM by carefully accounting for the discretization error emanating due to the finite mesh size in the FEM simulation. After the GPRM has been adaptively trained, it is used to predict the response of the 50 data points. On comparing the values of the SIF (obtained by AGPRM) with the SIF values obtained from FEM, the average residual percentage between the two is found to be 1.76%, thus indicating a good agreement between the AGPRM and FEM model. Furthermore, the time required to predict the SIF of 50 test points is reduced from 50 min (for FEM) to 12 s with the help of the proposed AGPRM, thus making RFL assessment less laborious and time consuming. - Highlights: • Adaptive Gaussian process regression model (AGPRM) is proposed as an alternative to FEM for prediction of SIF of a growing crack. • SIF values predicted by AGPRM are in good agreement with the SIF values obtained from FEM simulations. • Time required to predict the SIF values using the proposed model is substantially lower than FEM, making fatigue life assessment less time consuming.