O uso de NLP e data science na indústria de óleo e gás - um case deaquisição de conhecimento sobre integridade e confiabilidade naárea submarina

A system was developed to improve the analysis of integrity and reliability in subsea equipment, dealing with the complexity of unstructured data of around 120,000 PDF documents. This system consists of two services: an Extractor, which uses Machine Learning (ML) and regular expressions to mine data from inspection reports, and an Interface for viewing and analyzing structured data. Identification of relevant sources and system requirements was obtained in meetings with subsea experts, and, through sample inspection reports, used to train the Extractor. The methodology involved Support Vector Machines (SVM) for classifying table types and regex for extraction. Difficulties arose due to the diversity of reporting structures, leading to a hybrid approach with ML for context identification. The resulting web interface offers features such as advanced search, visualization of extracted data, statistical analysis of reliability and analytical dashboards. The software contributed significantly to the efficiency of underwater analysis activities, promoting digital transformation in the area by providing a valuable historical base of structured data to support decision making.