(183p) Prototype Study for Monitoring Flare Performance

Flares are a key source of volatile organic chemicals and particulate matter in petrochemical plants. Automating the measurement of these flare emissions offers plants the ability to reduce the dependence on human operators and enhance environmental compliance efforts. In industrial settings, plant operators often neglect the existing flare automation because it is just as easy to control flaring behavior through visual confirmation; integrating this data into automated flare control systems will allow for better flare control. In this paper, a soft sensor is developed that can predict emissions using this visual data with still images. A natural gas burner’s carbon monoxide and methane emissions are predicted using this deep neural network.