Prediction of natural convection heat transfer in a pipeline bundle

Abstract

The work described in this thesis addresses the issue of flow assurance in hydrocarbon recovery systems. Deposition of wax and formation of hydrates are amongst the critical flow assurance challenges that need to be resolved to avoid serious reduction in hydrocarbon flows emanating from wells and passing through flow lines (which are often sub-sea) to processing facilities. The hydrocarbon streams need to be maintained at temperatures above those at which solids formation occurs. The work described in this thesis is focussed on active heating of the flow lines by arranging them in a tube bundle in which a heat source is provided by a hot fluid (typically water) flowing in a separate tube. The objective of the work was to develop a generic methodology for the prediction of such bundle systems using Computational Fluid Dynamics (CFD) to generate a heat transfer data base and to interpolate this data base using neural network methods. It was convenient to develop this methodology by focussing on a specific geometry (namely a 3 meter long bundle with 4 internal pipes - product, test, heat flow and heat return respectively). Use of this geometry allowed direct validation of the computational method since an experimental investigation of an identical geometry was carried out in a parallel research project. The CFD methodology was first used to investigate the design of the experiments; it was shown that changes were needed in rig flows and in the sensitivity of the temperature measurements to ensure that the experiments complied closely enough with the basic assumption the each of the surfaces was isothermal. Once changes were made, the experimental and computational results were directly compared and satisfactory agreement was observed. For the bundle in a horizontal orientation, a two-dimensional (2D) CFD solution was used and a data base of 1683 solutions (covering the expected range of surface temperatures) was created. A further set of 1053 three-dimensional (3D) cases were calculated for the situation where the bundle was inclined at up to 90o to the horizontal. Both sets of data were fitted by means of neural networks which allowed prediction of bundle behaviour for a wide range of conditions.