Experimental and machine learning study of thermal conductivity of cement composites for geothermal wells

Abstract

Geothermal energy is the thermal energy present within the sub-surface of the earth. The efficiency of any geothermal energy system depends on net heat obtained at the surface during the production stage. However, thermal energy can be easily dissipated during the production stage of a geothermal well. In deep geothermal wells, these losses can be caused by the exchange of heat between the coolant and heated water, through the movement of the hotter fluid in the annulus, or through the convectional exchange of heat with the surrounding formations having low temperatures. The proper selection of cement material is crucial for a geothermal well as it provides a medium for heat transfer between the well and formation. On one hand, a low thermal conductive cement for the upper sections of a well can help to limit the heat transfer between the relatively cooler formation and the produced heating fluid. On the other hand, high thermal conductive coefficient cement can be preferred for the bottom sections of a geothermal well to avoid thermal loading and micro-crack development. A highly conductive cement in lower sections of the well will also support greater heat absorption from the formation to the working fluid.This experimental study measures and analyzes the thermal conductivity of class G cement along with different additives such as Fly Ash (FA), Micro cellulose (MC), Sand, Titanium Oxide (TiO2), and Gilsonite for up to 45 days of curing time. The samples are cured in both dry and wet conditions to observe the effect of the curing environment on the thermal conductivity of cement. Moreover, the Ultrasonic Pulse Velocity (UPV) is used for indirect measurement of compressive strength. It was noted that samples cured in the dry condition showed lower thermal conductivity and higher transit time irrespective of the additives and curing age.To expand the understanding of cement behavior with influencing factors such as additive type, curing time, and curing environment, clustering-based machine learning was implemented on the experimental database. Clustering, an unsupervised machine learning algorithm is immensely helpful in discovering hidden patterns and trends in the dataset. Four distinct clusters of cement behavior were observed and characterized by various influencing factors.