The new age of oil and gas industry is being driven by cost effective solutions, aiming to provide cheaper, faster and better products/services. The industry 4.0 brings an opportunity to transform systems and processes to be more efficient, making use of digitalization and new technologies, including the use of artificial intelligence algorithms applied to engineering problems.
In Brazilian offshore fields, the operating conditions for flexible riser applications (deep-water, mean wave frequencies, floating units and corrosive fluids) make the metallic layer’s fatigue failure mode one of the drivers in its design. In a daily basis, nonlinear dynamic finite element analysis uses regular wave scatter diagrams as an equivalent way to model the wave elevation, avoiding the time consuming irregular wave representation. The analysis performed with regular waves are faster but carries conservatisms with it. In a deep-water scenario, the wave height and period ranges of the wave scatter diagram can be refined to improve the fatigue results obtained, leading to a considerable increase in the total amount of wave classes that need to be evaluated.
Great part of the wave classes has a very low participation in the total fatigue damage, spending an unnecessary time to analyze them. Helped by a robust design of simulation experiment (DoSE) and machine learning regressors, a lean representation of the regular wave scatter can be done, where some of them are simulated and the rest of the results can be accurately predicted.
This paper presents the application of supervised learners that are used to predict riser fatigue damage at different riser locations, given partial simulations of a regular wave scatter diagram. The techniques support the strategy to reduce the total amount of fatigue analysis required within a project design phase. The focus stays on the evaluation of the fatigue of metallic layers at two main critical regions, bend stiffener and touch down zone. Hidden patterns inside each scatter diagram are discovered, minimizing the total number of finite element analysis (FEA) required. The amount of the wave class reduction starts from 50% going up to 75%, maintaining a good level of accuracy on the predicted damage values.