Modelling the Corrosion Rate of Buried Pipes Using Modified Artificial Neural Network (MANN) Coupled with Monte Carlo Simulation

International audience Several researchers have developed models to predict corrosion rate in buried structures with good results using methods such as artificial neural networks, Monte Carlo to mention a few. This paper presents a novel approach to predicting corrosion rate in buried structures using Monte Carlo and a modified artificial neural network (MANN). Monte Carlo Simulation is used in this paper to estimate the probability of occurrence of uniform corrosion in buried steel pipes in different soil locations using the fixed walk method. The central limit theory and law of large numbers were utilised to reduce errors. While Modified Artificial Neural Networks was used to establish the relationships. Data used for this study were obtained from weight loss study of Nickel electroplated and non electroplatedAISI-1051 samples. The parameters measured were soil pH, soil temperature and atmospheric temperature of different soils taken from an oil producing site in Delta State, Nigeria. These parameters were assumed normally distributed. Corrosion Penetration Rate (CPR) was calculated for each scenario using the weight loss method. Relating these input parameters to the calculated CPR was possible via the employment of Modified Artificial Neural Networks, which makes use of the least square polynomial regression equations instead of the neuron box to avoid complexities in neuron number determination. Weights and Scaling factors were set appropriately to allow for easy convergence. Third degree polynomial equations were derived from the MANN and used as inputs for the MCS. Results from the MCS showed 80% correlation with data used and a reliable estimate for the CPR was achieved using this MCS-MANN approach. The CPR of the Nickel Electroplated Sample was less than zero which was in good agreement with the weight loss data. It is observed that the CPR values for Non-electroplated samples were 60% of the time higher than the values for electroplated samples.