Efficient time simulation regression procedure for predicting offshore structural responses

Abstract

The assessment of accurate hydrodynamic loads on structures due to the extreme environmental loadings is the primary concern in the design of offshore platforms. The wind-induced waves are normally the most potential loadings with nonlinear behaviour, contributing to a complicated solution. The conventional Monte Carlo Time Simulation (MCTS) method is required for an accurate analysis of wave loads without offering any approximation error. The MCTS technique is considered a realistic and versatile approach because it can cover all sorts of nonlinearities to evaluate the offshore structural responses. However, this conventional technique is very computationally demanding, as reliable results require a large number of simulations due to unavoidable excessive sampling variability. Past studies showed that an Efficient Time Simulation (ETS) method offered a more effective result without scarifying accuracy. Nevertheless, the ETS method is limited to specific sea state conditions, in which the level of accuracy decreased with the presence of the wave current. Therefore, this study aims to improve the ETS method by taking advantage of their excellent correlation between extreme surface elevation and corresponding structural responses. Hence, an extended version of the ETS method is introduced. A novel model is developed based on regression algorithms and known as an ETS-Regression (ETS-Reg) procedure contributing a simplified method for the direct calculation of the wave-induced loads. Two ETS-Reg models were developed based on different input variables with similar output variables. In model development, the first relationship-based model was developed based on the surface elevation (input) and nonlinear responses (output), defined as the ETS-RegSE model. The second model was an improved version of the ETS-RegSE model, the linearised responses (input) with their corresponding nonlinear responses, known as the ETS-RegLR model. In short-term analysis, these models will be tested by three sea state conditions and three different wave-induced currents. The probability distribution of the 100-year extreme response values from the ETS-Reg models have been compared with corresponding distributions of 100-year response values from the MCTS procedure to examine the accuracy and the efficiency of the developed technique. As a result, for the short-term analysis, the ETS-RegLR model delivered an excellent accuracy in the range of 93% to 99% in predicting 100-year responses compared with the benchmark value using the MCTS method for all cases of wave conditions. Meanwhile, the ETS-RegSE model varies between 20% and 96%. Remarkably, the efficiency level achieved by the ETS-RegLR model was in the range of 43 to 51 times more efficient than the MCTS method in terms of variance ratio of sampling variability, whereas, the ETS-RegSE model was in the range of 22 to 42 times. The same inference appeared for long-term analysis since both the ETS-Reg models' accuracies were closely matched to the previous short-term analysis. The ETS-RegLR model’s accuracy was 95% to 99%, whereas, 89% to 96% by the ETS-RegSE model. Overall, the ETS-Reg models can lead to better performance without extensive simulations, which the models require less computationally demanding processes and time. Thus, these innovative models are proposed as an alternative technique for frequency domain in the probabilistic assessment in the oil and gas industry.