Sensitivity analysis and faults diagnosis using artificial neural networks in natural gas TEG-dehydration plants

Abstract

In this work, a typical process for natural gas dehydration using triethylene glycol (TEG) as a desiccant is simulated using a steady state flowsheet simulator (Aspen Plus). The flowsheet includes all major units in a typical dehydration facility, that is: absorption column, flash unit, heat exchangers, regenerator, stripper, and reboiler. The base case operating conditions are taken to resemble field data from one of the existing TEG-dehydration units operating in United Arab Emirates (UAE). Using Aspen Plus, the flowsheet is then used to study the effects of different input parameters and operating conditions of the absorption column, the stripper and the overall plant, on BTEX emission, volatile organic components (VOCs) emission, TEG losses and water content (dew point) of the dehumidified natural gas. Contactor performance has been found to be most sensitive to disturbances in operating pressure and wet gas flow rate, whereas flow rate of stripping gas and temperature of inlet solvent have the major impact on the stripper performance. The potential of artificial neural network (ANN) to detect and diagnose process faults in the dehydration plant has also been explored. ANN successfully detects the disturbance severity levels in the input variables considered for the contactor. In particular, abnormal levels of BTEX concentrations in the rich solvent (exiting the contactor) are shown to precisely indicate the severity levels in the input variables. Faults in the stripper–regenerator unit have been perfectly predicted by the ANN for two symptoms (TEG emissions and BTEX emissions in vents) and to a lesser extent for faults in VOCs emissions. The best ANN prediction is obtained for the overall plant where the ANN simulates the imposed disturbances for three severity levels of imposed malfunctions for all symptoms considered.