Eventos Anais de eventos
COBEM 2019
25th International Congress of Mechanical Engineering
DEEP LEARNING BASED INFERENCE SYSTEM FOR REAL-TIME ESTIMATION OF LPG CONTAMINANTS' MOLAR FRACTION
Submission Author:
Jean Mario Moreira de Lima , RN
Co-Authors:
Jean Mario Moreira de Lima, Fábio Meneghetti Ugulino de Araújo
Presenter: Jean Mario Moreira de Lima
doi://10.26678/ABCM.COBEM2019.COB2019-0225
Abstract
The Liquefied Petroleum Gas (LPG) is one of the most lucrative final products of a Natural-Gas Processing Plant. Basically, LPG is composed of propane (C3) and butane (C4), with ethane (C2) and pentane (C5) as contaminants. Measuring and controlling the molar fraction of those contaminants are crucial for product quality monitoring. The molar fractions of the LPG contaminants are measured by gas chromatography or through laboratory analysis, which are high-cost and require large intervals. In this work, a inference system is built to estimate the molar fraction of LPG contaminants in real time. Deep learning models were implemented and their ability for estimation were tested. The results have showed that the deep learning based inference system estimates the molar fraction of LPG contaminants satisfactorily on a minute-by-minute basis. The mean percentage errors of estimation was 0.36% for ethane and 0% for pentane.
Keywords
LPG, Inference System, Deep learning, Convolutional neural networks
