Before running out of fuel, the car's dashboard flashes red, warning that the tank is low. The driver doesn't need to know how many kilometers they've driven since refueling, because the system indicates the right time to stop at a gas station. This is the same principle behind two computer programs developed at the Faculty of Chemical Engineering (FEQ) at Unicamp: translating data – far more complex than that of a car – into information ready for use by engineers and managers. Instead of warning about the risk of running out of fuel, the Unicamp software estimates the probability of explosions and oil and gas leaks, contributing to the prevention of operational failures in industrial plants.
The technologies, named Fuzzy BowTie and Artificial Neural Network for Leak Detection (AN24LD), were created from a research, development, and innovation (R&D&I) project coordinated by Unicamp, with the participation of the Unicamp Development Foundation (Funcamp) and financed with resources from the National Petroleum Agency (ANP). The intellectual property generated in the research was protected with a strategy from the Inova Unicamp Innovation Agency and licensed, on an exclusive basis, to Shape Digital, a technology company operating in the oil and gas sector. Modec, a Japanese multinational that builds and operates floating oil and gas platforms, joined the research project as a partner and will enable testing of the technologies with real data from its operations in Brazil.
Following licensing, in a new phase of the research partnership with industry, the companies conduct complementary tests and developments to incorporate the computer programs into commercial systems. A ship off the coast of São Paulo was chosen to receive the experimental application of one of the software programs.
“We are monitoring the tests with real data that are being carried out at this offshore oil production unit,” says Sávio Souza Venâncio Vianna, professor at FEQ Unicamp and coordinator of the research project that generated the two software programs registered by Unicamp.
From research to licensing.
The programs were born from an industry demand for tools capable of optimizing safety and risk management in highly complex sectors, such as oil and gas exploration and production. The priority was to create an effective leak prevention system and present a new BowTie approach (a butterfly-shaped risk analysis diagram, hence the name).
To meet the challenge, Professor Vianna combined his expertise in automation and risk analysis with Professor Flávio Vasconcelos da Silva's knowledge of artificial intelligence. The pair combined two advanced techniques: fuzzy logic , which mimics human reasoning, and neural networks, inspired by brain learning.
Together with a team of five students, the researchers developed Fuzzy BowTie and AN24LD. To transform academic research into industry innovation, the first step was to ensure the protection of the technology. Before any public disclosure, the inventors reported the invention to Inova Unicamp to protect the intellectual property of the computer programs. This led to licensing agreements, aimed at raising the technology maturity level (TRL) of the tools, and the publication of scientific papers on the results achieved.
“We were the first group in the world to publish about this highly innovative technology, even when the TRL wasn't so high. This caught the industry's attention and, today, we see our work becoming a benchmark,” summarizes Vianna, regarding the project that combined process safety and artificial intelligence.
How do technologies work?
The AN24LD program focuses on the instantaneous detection of leaks. The technology integrates neural networks with long-term recurrent memory (capable of identifying complex patterns in large volumes of sequential data) with computational fluid dynamics (CFD) simulations, which reproduce the behavior of fluids in three dimensions. This combination allows for precise analysis and identifies irregular flows and potential failures, preventing losses and increasing operational safety.
“Imagine an oil production platform 330 meters long by 50 meters wide. How do you quickly locate a leak on such a large platform? With this program, identification becomes more precise,” describes Vianna. To achieve this result, the team trained the network with 2,6 million data points generated from leak scenario simulations. “We don't need to have all leak scenarios to train the network. We train with what already exists, with what has already happened, and also with what we simulate, and thus we can expand protection to possible future scenarios,” says Vasconcelos da Silva regarding the generalization capacity of neural networks.
The Fuzzy BowTie program is an advanced accident prediction system capable of analyzing risk scenarios and quantifying the frequency and consequences of critical events. The tool stands out for creating quantitative BowTie diagrams in an intuitive graphical interface. "This tool is very visual: you look at it and you can understand the system," says Vianna.
Risk calculation combines two methods: Boolean logic, for objective decisions and binary answers (yes or no), and fuzzy inference, which incorporates uncertainties for more realistic and flexible assessments. "Fuzzy inference offers a holistic and quantitative view of the risk of a given system, from a mathematical framework, based on fuzzy logic," says Vasconcelos da Silva.
In practice, the program prioritizes the weight of each threat on the final outcome, showing, for example, that a failure can account for 80% of the risk of a specific event occurring. The sooner a leak is detected, the less product is released and the less severe the accident's consequences; and the sooner a threat is identified and weighed within the system, the easier it is to decide where to invest in security, the researchers explain.
To illustrate the importance of this practical translation, Professor Vianna uses another analogy: “If I tell someone that the risk of an airplane accident is 1,38 x 10⁻⁸ , the data may mean nothing to them. Now, if I say that this is a chance in tens of millions or that the risk is low, the understanding is different. A complex number, in safety systems, only makes sense if it is translated into practical language that anyone can understand.”
From nuclear risk analysis to the chemical industry.
According to the researchers, the BowTie technique, originating from the nuclear and aerospace industries, was already known in the chemical industry, but applied in a predominantly visual and qualitative way. The tool licensed by Unicamp adds fuzzy logic to prioritize the threats that lead to an accident.
Fuzzy BowTie has already been incorporated as a module within a system that Shape uses internally, which facilitates adoption by operations teams, according to the researchers. AN24LD is in a more advanced stage of validation, being trained with typical data from an offshore oil production unit. Field tests are expected to continue throughout 2026 and will be monitored by researchers for teaching, research, and tool improvement purposes.
The same logic could be applied to biological risks in hospitals, the food industry, or the automotive industry. Companies interested in licensing technologies developed at the University can consult Unicamp's Technology Portfolio and contact the Business and Innovation Coordination (CNI) of Inova Unicamp, which is responsible for negotiating licensing agreements, both paid and unpaid, and research partnerships with companies.
Inventors Award 2026
In its 19th edition, the Unicamp Inventors Award, organized by Inova Unicamp, recognizes and values inventors who have excelled in the transfer of technologies from the University and in the creation of academic spin-off companies.
Award-Winning Inventors
Sávio Souza Venâncio Vianna, Flávio Vasconcelos da Silva, Felipe Matheus Mota Sousa, André Zamith Selvaggio, Raphael Issamu Tsukada, Raphael Santana Almeida, and Vitor Augusto Oliveira da Silva were awarded prizes in the Licensed Intellectual Property category in 2026.
Check out the complete list of all award winners on the Unicamp Inventors Award website.
Article originally published on the Inova Unicamp website.
Cover photo:
