---

In order to predict the residual life of offshore platforms and establish efficient schedule for underwater inspection and repair, it is necessary to estimate the fatigue crack growth rate in tubular joints properly. Linear Elastic Fracture Mechanics and Stress Intensity Factor are applicable tools for evaluating growth rate of existing fatigue cracks in offshore tubular joints. In the past several approaches based on Paris crack growth law, have been proposed in this regard. Each of these approaches use different methods for estimating the Stress Intensity Modification Factor (Y). In this research the capability of Artificial Neural Networks for evaluating the fatigue crack growth rate in offshore tubular T-joints under axial loading is investigated, when the crack depth is more than 20% of chord wall thickness. For this condition the crack growth process is highly affected by joint geometry and loading mode. Two types of artificial neural network are developed for predicting the Y factor: Radial Basis Function (RBF) and Multi Layer Perceptron (MLP) networks. The required input data consist of the crack shape and the percentage of crack penetration through thickness. Experimental data from NDE center in University College London are used for training and testing the networks. The results of this research are compared with other existing theoretical and empirical solutions.

---

One of the most important issues in designing coastal and offshore structures is the prediction of wave and current forces on slender cylinders. Such forces are often considered as dominate loadings. Many analytical and empirical methods such as Morison equation have been suggested for estimation of waves and current forces. Such methods, however, have shown inaccuracies in predicting hydrodynamic forces. On the other hand, Artificial Neural Networks (ANNs) have received a great deal of attention in recent years and are being touted as one of the greatest computational tools ever developed. In fact, ANNs are nonlinear systems consisting of a large number of highly interconnected processing units, nodes or artificial neurons, which have the ability of learning. In this research, ANNs have been used to estimate wave and current forces on slender cylinders. Data of 308 experimental specimens have been used for training and testing the networks. Considering the aim of this research for the application of ANNs, these data were consisted of recorded force values in different time series. The supervised learning neural networks models have been used in this research. The results indicate the success of the application of neural networks approach which can efficiently predict waves and current forces on slender cylinders after carrying out appropriate training. Furthermore, the results are within acceptable accuracy in comparison with experimental results and the results obtained from Morison equation.

---

Floating offshore structures, particularly floating oil production, storage and offloading systems (FPSOs) are still in great demand, both in small and large reservoirs, for deployment in deep water. The prediction of such vessels’ responses to her environmental loading over her lifetime is now often undertaken using response-based design methodology, although the approach is still in its early stages of development. Determining the vessel’s responses to hydrodynamic loads induced by long term sea environments is essential for implementing this approach effectively. However, it is often not practical to perform a complete simulation for every 3-hour period of environmental data being considered. Therefore, an Artificial Neural Networks (ANN) modelling technique has been developed for the prediction of FPSO’s responses to arbitrary wind, wave and current loads that alleviates this problem. Comparison of results obtained from a conventional mathematical model with those of the ANN-based technique for the case of a 200,000 tdw tanker demonstrates that the approach can successfully predict the vessel’s responses due to arbitrary loads.

---

Prediction of wave parameters is necessary for many applications in coastal and offshore engineering. In the literature, several approaches have been proposed to wave predictions classified as empirical based, soft-computing based and numerical based approaches. Recently, soft computing techniques such as Artificial Neural Networks (ANNs) have been used to develop wave prediction models. In this work, the performance of regression trees for prediction of wave parameters was investigated. The data set used in this study comprises of wind and wave data gathered in Caspian Sea. Results of regression trees were compared with those of artificial neural networks. Results indicate that error statistics of regression trees and artificial neural networks were nearly similar. In addition, regression trees need lower run-time.

---

Prediction of wave height is of great importance in marine and coastal engineering. In this study, the performances of artificial neural networks (feed forward with back propagation algorithm) for online significant wave heights prediction, in Persian Gulf, were investigated. The data set used in this study comprises wave and wind data gathered from shallow water location in Persian Gulf. Current wind speed (u) and those belonging up to eight previous hours are given as input variables, while the significant wave height with leading time of 1-24 hour are the output parameters. Results show that the artificial neural networks can perform very well in predicting significant wave height, when shorter intervals of predictions (6 hour) were involved. Small interval predictions were made more accurately than the large interval ones. Results of artificial neural networks were compared with those of regression trees. Results indicate that error statistics of neural networks and regression trees were nearly similar

---

Wind waves are one of the important, fundamental and interesting subjects in port and coastal engineering. Thus, within years, different methods such as experimental methods, numerical modeling and soft computing methods have been employed to estimate the wave parameters. In this study, waves height in Anzali port is predicted using soft computing models such as multivariate adaptive regression splines (MARS), regression trees (CART), artificial neural networks (ANNs) and M5' model tree. Among the features of MARS, CART and M5' Models and compared to artificial neural networks, are presentation of regression equations and mathematical relationships which could easily be used to estimate the time series of wave height. The obtained results showed competitive accuracy of regression techniques in estimation of waves height compared to the artificial neural networks.