This paper describes an original approach to reduce drastically the analysis time needed for fatigue design of structures by using machine learning techniques. The approach is applied to the spectral fatigue analysis of various structural details on a converted FPSO hull, where design iterations are usually time consuming. For the structural detail example used in the present study, numerical results show that by including the right inputs to the machine learning algorithm, the predicted fatigue life could compare well with the spectral fatigue analysis output with a score of up to 0.997. For the critical elements with high fatigue damage, the predicted fatigue life is found up to 2.5 times of the actual value. Overall an estimated 5.5 hours (out of 6 hours) are saved for one iteration of spectral fatigue analysis.
Ocean developments for oil and gas and renewables involve site specific floaters designs. For such assets, it is required to perform detailed structural analysis at the design stage to consider all the specificities of the structure and the environments in the design and make sure the floater can operate safely throughout its design life. For fatigue design, the state of the art approach for such floaters is spectral fatigue analysis.
The calculations of the fatigue damage using a full spectral direct calculation approach is labor intensive, especially when design iterations are needed. The complete assessment procedure comprises hydrodynamic analysis to compute wave-induced loads; structural analysis for both global coarse mesh and local fine mesh finite element models; statistical analysis to calculate the short-term stress response based on the environmental conditions on site; and lastly fatigue strength evaluation using the applicable fatigue S-N curve. This analysis approach can be rather time consuming in cases where many structural details are assessed or where multiple iterations are needed to reach a satisfactory design.
