Real Time Determination of Rheological Properties of Spud Drilling Fluids Using a Hybrid Artificial Intelligence Technique

Abstract

Abstract Measuring the rheological properties during the drilling operation is time-consuming and usually, these properties are measured twice a day. The rheological properties play a key factor in controlling the drilling operation. The knowledge of these properties are very important for hydraulic calculations which are required for hole cleaning optimization, surge and swab pressure calculations and others. Wrong estimation of these properties may lead to big disaster during the drilling operation such as pipe sticking, loss of circulation, and/or well control issues. The aforementioned problems will increase the non-productive time and hence the overall cost of the drilling operations. Artificial intelligence techniques, once they are optimized, can be used to predict the rheological properties in a real time. The main idea of this paper is to use the frequent measurements of the mud density, Marsh funnel viscosity and solid percent, which are measured every 15-20 minutes and build new artificial intelligent models for plastic viscosity, yield point, apparent viscosity, and flow behavior index. Different AI techniques were evaluated such as; artificial neural network, support vector machine (SVM) and adaptive-network-based fuzzy inference system (ANFIS). The model which yielded the highest accuracy (lower average absolute error (AAPE) and highest correlation coefficient (R)) was used to develop new empirical correlations for each rheological properties. For the first time, the artificial intelligence techniques were combined with the self-adaptive differential evolution algorithm to optimize the best combination of the AI parameters. The results obtained showed that ANN is the best AI technique to predict the rheological properties from the mud density, Marsh funnel viscosity, and solid percent. It is very important to combine the self-adaptive differential evolution with the artificial neural network to predict the rheological properties with high accuracy (AAPE less than 5% and R greater than 95%). The ANN black box was converted to a white box by extracting the weights and biases of the optimized SaDe-ANN model for each rheological parameter and a new empirical correlation was developed. The developed technique will help the drilling engineer to predict the rheological properties every 15 to 20 minutes and this will help in hole cleaning optimization and avoid most of the drilling problems such as pipe sticking and loss of circulation. The developed correlation can be used without the need for the ANN model and can apply using any software. No additional equipment or especial software is required for applying the new method.