Development of a system based on multimodal transformers for diagnosing events on offshore oil and gas extraction platforms.

• Grant number: 25/09605-0
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: July 01, 2025
• End date: June 30, 2026
• Field of knowledge: Engineering - Electrical Engineering - Industrial Electronics, Electronic Systems and Controls
• Principal Investigator: Eduardo Aoun Tannuri
• Grantee: Guilherme Luiz Müller Machado Filho
• Host Institution: Escola Politécnica (EP). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Company: Universidade de São Paulo (USP). Escola Politécnica (EP)
• Associated research grant: 22/03698-8 - OTIC Offshore Technology Innovation Centre, AP.PCPE

Abstract

Offshore oil platforms are equipped with numerous sensors to monitor whether the oil extraction systems are functioning correctly. In the event of a failure, the responsible operators must follow three key steps:1 - Timely detection of the failure2 - Diagnosis of its causes3 - Taking appropriate actions to restore all platform processes to normal operation [1].However, with technological advancements, offshore platforms now have an increasing number of sensors, which makes it more difficult for a team of technicians to monitor all system data in real time. As a result, several issues may arise, such as:1 - Late detection of the failure2 - Incorrect diagnosis of the failure3 - Delay in determining the root causes4 - Delay in identifying the most appropriate control actionAn example of an accident caused, for instance, by late failure detection was the explosion of the P-36 Platform in 2001, which resulted in 11 deaths and an estimated loss of USD 500 million [2].Therefore, to prevent financial losses and human casualties in the offshore platform sector, it is necessary to develop a tool capable of handling large volumes of data and making accurate inferences from them within a short time frame. A promising technology with this capability is artificial intelligence, specifically models based on multimodal transformer architecture [3], which are capable of handling various data formats such as images, tables, and text.Some of the current applications of multimodal transformers include medical diagnosis based on imaging exams, object detection in photographs, and audio-to-text transcription. Given that sensor data from offshore platforms are often represented as graphs and numbers (i.e., two different data formats), these models are promising for fault detection and diagnosis in such systems.Thus, the objective of this work is to investigate the application of multimodal transformer models in offshore platform environments, as these systems can process and analyze large amounts of data and make inferences from them much faster than a human could.These models will be used to develop a fault detection and diagnosis system for offshore oil platforms. The system will be capable of generating alerts in the event of abnormal events and can also be consulted by the responsible personnel to provide information about the causes of potential anomalies and the best way to address them.Currently, the application of artificial intelligence models on offshore oil platforms has shown promising results, particularly with models that process time series data to detect abnormal events [1]. This work aims to contribute to these advancements by using more robust neural network models capable of interpreting more complex data.1 - VARGAS, R. E. V. et al. A realistic and public dataset with rare undesirable real events in oil wells. Journal of Petroleum Science and Engineering, v. 181, p. 106223, 2019.2 - AGÊNCIA NACIONAL DO PETRÓLEO, GÁS NATURAL E BIOCOMBUSTÍVEIS (ANP); DIRETORIA DE PORTOS E COSTAS (DPC). Análise do acidente com a plataforma P-36: relatório da Comissão de Investigação ANP/DPC. Rio de Janeiro: ANP, jul. 2001.3 - XU, Peng; ZHU, Xiatian; CLIFTON, David A. Multimodal learning with transformers: a survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, jun. 2022. (AU)