Offshore pipeline performance evaluation by different artificial neural networks approaches

This paper investigates the upheaval buckling behaviour of offshore pipeline buried in clay soil considering the possible variability in soil, operating condition and pipe properties. A 2-D finite element model of the pipeline–soil system was developed in OpenSEES software to model the upheaval buckling. Further, the uncertainty in the controlling variables was modelled using the optimized Latin Hyper Cube (LHC) sampling technique to draw the samples from appropriate probability distribution. Finally, six different models based on artificial neural networks (ANNs) were developed to predict the performance of offshore pipeline using the simulated upheaval buckling displacement. A total number of 500 data were collected from simulation, randomly divided into 350, 75 and 75 datasets, and were used for training, validating and testing the proposed models, respectively. Comparison between results showed that all models are capable to deliver displacement values very close to the simulated ones. To determine the best performance model, several controlling methods were used and finally one of the models was suggested as the best one. An additional analysis was performed for displacements above 30 mm where the number of achieved data is limited and scattering in data is observed. Analysis of the results illustrated that the models are reliable for predicting displacement values in upper band (above 30 mm) as well.