ABSTRACT

This study evaluated a multi-objective flow modelling of pipeline under spatially varying corrosion defect using Artificial Neural Network (ANN)-trained particle swarm optimization (PSO). The objectives were to develop several stochastic corrosion predictive models from corrosion parameters using a metaheuristic algorithm; to spatially predict internal corrosion of pipelines using the developed models; to develop a simulation code for modelling of similar corrosion rates based on the driving parameters; and to establish the relationship between corrosion rates and the driving parameters via extensive sensitivity analysis. Data comprising corrosion rates and defect depth from influencing parameters such as crude temperature, partial pressure, hydrogen sulphide concentration, pH value, flow velocity, water content, pipe diameter, pipe length, and age of pipeline in years was used as obtained in the results. To cater for variation in different combinations of the driving parameters, seven models were developed using comparatively different parametric grouping. The first model predicted the corrosion rate as a function of crude temperature, partial pressure, hydrogen sulphide concentration, pH value, flow velocity, water content, pipe diameter, pipe length, and age of pipeline in years; the third model predicted corrosion rate as a function of fluid velocity, and partial pressure while thefourth model considered the effective prediction of corrosion rate as a function of fluid velocity, fluid partial pressure, hydrogen sulphide concentration, and pH value.A feed-forward ANN topology was used to construct the corresponding models with an avalanche of hidden neurons in one hidden layer which communicates with the input and corrosion rates. Furthermore, the PSO algorithm was applied to train the parameters of ANN network using relevant number of particle size at optimum conditions for the models with 500 iterations across all models while validating model performance with relevant data. Results from values obtained from the first model and field data have shown relatively close proximity between corrosion rates. Parameters such as crude oil temperature, partial pressure, hydrogen sulphide concentration, pH value, flow velocity, water content, pipe diameter, pipe length, and age of pipeline in years, although having varying degrees of relationship with corrosion rate have been modelled with PSO and the results are noteworthy. Very strong correlation up to 0.84211 was obtained from the first model. Consequently, correlation between data and model values of up to 0.90976 was obtained from the second model. Consequently, defect depth from the location can be closely estimated on provision of the driving parameters like fluid velocity, fluid partial pressure, hydrogen sulphide concentration, and pH value, in that order in the developed MATLAB code. This model is a good tool for the prediction of corrosion of crude oil in internal pipeline.