Accurate prediction of liquid holdup associated with multiphase flow is a critical element in the design and operation of modern production systems. This prediction is made difficult by the complexity of the distribution of the phases and the wide range of fluid properties encountered in production operations. Consequently, the performance of existing correlations is often inadequate in terms of desired accuracy and range of application. This investigation focuses on the development of a neural network model, a relatively new approach that has been successfully applied to a variety of complex engineering problems. 2292 data sets from five independent sources were used to develop a neural network model for predicting liquid holdup in two-phase flow at all inclinations from upward(+90 degrees) to downward(-90 degrees) flow. A three-layer backpropagation neural network has utilized. Seven parameters including inclination from horizontal, pipe diameter, gas and liquid superficial velocity, liquid viscosity, density and surface tension are used as inputs to the network. A detailed comparison with some empirical correlations which are applicable for whole range of inclinations reveals that the developed model provides better accuracy and predicts liquid holdup in terms of the lowest absolute average percent error (15.31), the lowest standard deviation (30.15) and the highest correlation coefficient (0.9962).
Multiphase flow is defined as the concurrent flow of two or more phases, liquid, solid or gas, where the motion influences the interface between the phases. The prediction of liquid holdup in pipeline is very important to the petroleum industry. Liquid holdup, which is defined as the fraction of pipe occupied by liquid, must be predicted to properly design separation equipment and slug catchers in pipeline operations. Many correlations have been published for predicting this important parameter. The commonly used correlations are those of Eaton et al.(1), Guzhov et al. (2), Beggs and Brill(3), Minami and Brill(4), Gregory et al. (5), Mukherjee and Brill(6), Hughmark and Pressburg(7), Hughmark(8), Abdul-Majeed(9) (10), Xiao et al. (11), Baker et al. (12) and Gomez et al. (13). Some are very general while others only apply to a narrow range of conditions. Each of those correlations was developed for a specific orientation (vertical, horizontal or inclined upward or downward) and can't predict liquid holdup for other orientations correctly.
Many of these approaches begin with a prediction of flow pattern, with each flow pattern having an associated method of predicting liquid holdup. The liquid holdup prediction is used to determine a two-phase friction factor from which a pressure gradient is calculated. One of the problems with this approach is that it is dependent on the accuracy of flow pattern predictions and is subject to discontinuities in predictions made across flow pattern transition boundaries. (14) Comparative studies(15) (16) have shown that these models perform inconsistency as flow conditions change. This limitation makes difficult the task of selecting the most appropriate flow correlation.
Artificial neural networks (ANN) are parallel-distributed information processing models that can recognize highly complex patterns within available data. The recent development and success of applying ANNs to solve various difficult engineering problems has drawn the attention to its potential applications in the petroleum industry.
