Previsão de integridade em tubulações industriais através de redes neurais: uma abordagem abrangente

Abstract

This article evaluates the integrity of industrial drainage pipes, aiming at operational reliability in complex systems. In the face of diversity of materials and fluids used, the use of Artificial Neural Networks (ANNs) emerges as a promising approach. Authors such as Elsayed and Lafraia highlight the relevance of reliability in industrial contexts. The methodology adopted it comprised the normalization of data collected from the pipes, pre-processing and exploratory analysis. The Multi-Layer Perceptron (MLP), validation Stratified K-Fold crossover and L2 regularization were strategies chosen for the model construction. The results were consistent, with accuracy test averages ranging from 83.77% to 91.54%. The predictions confirmed trends identified in the exploratory analysis, highlighting the influence of the material and type of fluid in the predictions. The synergy between technical knowledge and AI has proven to be a valuable resource to improve reliability management in environments complex industrial. The conclusions indicate that the application of ANNs in the evaluation of pipeline integrity offers the potential to improve decision-making, optimize maintenance and extend the useful life of these systems, impacting positively the industry as a whole.