Desenvolvimento de um sistema de visão computacional para detecção automatizada de anomalias em equipamentos pressurizados baseado na norma regulamentadora de número 13

Abstract

The evolution of computer vision has democratized access to tools for training and creating Artificial Intelligence models in various areas. However, in the industrial sector, there is a shortage of applied solutions that combine the autonomous identification of failures in boilers and pressure vessels with alignment with Brazilian safety standards for pressurized equipment. To fill this gap, this work presents the development of a modular platform aimed at predictive monitoring of pressurized equipment covered by Regulatory Standard 13 (NR-13). The system integrates object detection and semantic segmentation models, developed with high generalization capacity, to analyze three critical categories of failures: cracks, corrosion and leaks. The tool’s main differentiator is its ability to collect user feedback to continuously improve the accuracy of detection models and mitigate false positives, in addition to automatically generating a structured technical report based on Brazilian regulations for pressure vessels and boilers, which can be downloaded or shared directly by major chat applications such as WhatsApp, Telegram, among others that are installed on the user’s cell phone. This technical document generated by software provides the company with a clear and advance view of the detected anomalies, serving as a basis for predictive maintenance planning and for the targeted activation of a Legally Enabled Professional (PLH), who will carry out the formal inspection and maintenance procedures required by law. After three iterative versions of software, validated by users in the engineering and technology fields, the three models achieved very satisfactory accuracies. Starting with the fourth version, collecting feedback from users made the evolution of model accuracy faster. The technical condition report, with quantification of detected failures and visual recording of anomalies, was assessed as satisfactory by most participants. The proposed architecture demonstrates that it is possible to offer an affordable technological solution for preventive integrity management, reducing the time between the occurrence of failures and corrective intervention, contributing to operational safety in compliance with NR-13.