Approaches for Maximum Deformation Prediction Using Computational Intelligence

Resumo

Oil and gas pipelines, although very efficient, are subject to several defects, such as dents, which are deformations that in turn compromise  their integrity and safety. These failures require fast and accurate assessment, following codes of standards such as ASME B31.8 (2018), which recommends complex analyses, such as the Finite Element Method (FEM). However, FEM has a high computational cost, which limits its application in situations that require agility. Therefore, there is a need for alternative methods, in order to overcome the limitations of FEM, generating more agile and accurate analyses. Machine Learning (ML) algorithms show themselves as solid options in cases like this due to their ability to predict deformations, with lower computational cost. This approach is crucial to avoid catastrophic failures, reduce environmental risks and improve preventive maintenance, since the safety and structural integrity of oil and gas pipelines are critical to avoid environmental and economic accidents. Find the best ML model is important to improve the prediction, but the search for the hyperparameters is a difficult task.  Apply exhaustive search techniques such as Grid Search and its variations, Random Search, Halving Grid Search and Halving Random Search is an option. These methods will allow exhaustive testing with predictive modelssuch as Decision Tree (DT), Random Forest (RF), Multi-Layer Perceptron (MLP), Gradient Boosting (GB), Gaussian Process Regressor (GPR), in order to ensure the best configuration for each scenario. Based on related studies, it is expected to identify the most efficient model (accuracy and time) thus allowing rapid and accurate diagnoses of critical deformations, greatly reducing computational and operational costs, offering technical decisions that align with industry demands for safer and more agile methods.