Predicting the casing failure in pre-salt wellbores using neural networks
Anna Luiza de Castro Santos1; Fernanda Lins Gonçalves Pereira1; Deane Roehl1
1 PUC-Rio
doi:10.20906/CPS/CILAMCE2017-0148
Resumo
An integrity analysis on steel casings of pre-salt wellbores is a key factor to indicate failure in a reservoir system since it is mainly surrounded by salt rocks. Halite has an elasto-viscoplastic time-dependent behavior. In cases where there is no cement in the annular, as the salt expands, it can reach the casing in the lifetime of the wellbore. This contact is not impeditive to the structural integrity of the system. Yet it can generate plastic zones, which indicate the casing failure. The salt reaches the casing initially in some points before contacting the entire surface. It creates a non-uniform loading on the steel surface, leading to localized points with high stress values. Eventually, once the salt reaches all the points on the casing surface, it is possible that an analysis at the end of the well lifetime does not show some plastic stress zones due to the relief of the stresses. Nevertheless, it is considered in this work that the material has reached the failure condition once the stresses reach the yield surface. The numerical model proposed in this study has a 100% fault in cementation (no cement in the annular), considering a plane-strain analysis in a quarter symmetric representation. The analyses were carried out through a Finite Element Method by ABAQUS® software. For the constitutive model of the Halite, it was implemented the Double-Mechanism model to represent the creep effects in the salt rock. This study objective is to present an equation that receives a few parameters of the problem as incognitos and returns the time in which occurs the first plastic stress zone in the whole casing as a result. An artificial neural network is created. It is capable of generalizing the results through learning by experience, based on a few number of samples, which the entry parameters and the answer are known. After successful training, with the activation functions and the weight and biases vectors, the final equation with a prompt answer is generated, based on the variation of just physical parameters,
Palavras-chave: Artificial neural network; Casing failure; Pre-salt wellbore; Finite element method; Integrity analysis