Abstract
Flexible pipelines are widely used in the offshore oil and gas projects. Comparing to steel pipes, flexible risers consist of multiple components layers, with each layer serving designated functions. For example, the helical tensile layers are used to work against axial tension from static pipe weight and amplification from the dynamic motions. As the risers are constantly experiencing stress variations due to different loadings, the fatigue performance of the tensile layers is essential to the design and operation of the flexible pipe system. Global and local fatigue analyses are usually performed in the design stage to estimate the fatigue life of tensile layers. Various parameters like vessel characteristic, pressure and temperature, environmental loadings affect the riser long-term performance. Also, multiple locations like top end fitting, bend stiffener region, touch-down zone are all potential fatigue hot-spots. To evaluate the fatigue damage at these locations, the industry standard practice is to run fatigue analysis in time domain with irregular wave. This approach is however time and resource consuming, especially when the designers have to consider thousands of load cases with long duration for numerous combinations of sea-states and operating conditions.
Baker Hughes developed a frequency domain technique for fatigue assessment on tensile layers of flexible risers. The approach builds transfer functions based on chosen time domain simulation and predict the total stress spectrum for other cases. Dirlik’s approximation method is used to estimate the fatigue damage based on the stress spectrum. Comparing to time domain approach, this approach is considerably more efficient in implementation. Since frequency domain technique simplifies the riser response as a linear system, the discrepancy must be addressed by calibration factors. To improve accuracy, more transfer functions are required to be built prior to the damage calculation.
In this paper, Baker Hughes introduces artificial neural network (ANN) technique to the developed frequency domain fatigue damage methodology. The ANN model is built between the wave spectrum and stress spectrum. Unlike the general AI technique, the type of neuron at each layer is carefully selected to represent the dynamic behavior of the riser system to improve the model efficiency. The ANN model is trained based on selected sea states. The trained model is used to predict riser damage and compared to that calculated from the time domain approach.
