Company: GreyLogix Brasil

Technologies:

• Python;
• Deep Learning;
• Transfer Learning;
• MATLAB;
• Keras; imageai; opencv

I've recently developed a PoC for a big O&G company in order to address a common issue in offshore oil rigs.
Gas flares are equipment that burn natural gas when a processing unit is not capable of processing all the gas coming from a well. Methane and other gases present in natural gas is a lot worse than CO2 in terms of greenhouse effects. So the gas must be burned in order to convert it to CO2. Monitoring of flares in offshore oil rigs is done by having a person periodicaly check whether the flare is lit or not. This is unacceptable in a context where companies must have control of their greenhouse gases emissions.

So we made a vision system based on neural networks and a camera that would check it 24/7. In this first PoC, the system only checks whether the flare is on or off and estimates gas emissions based on that. In the future, the vision system should be able to assess flame quality (e.g. whether there is smoke or not, or estimates of how much of the gas is actually being burned), and take that into account in emission reports.