A Novel Rheological Hierarchy Optimization Methodology with Artificial Intelligence and Inverse Technique Improves Spacer Design

Abstract

The quality of zonal isolation and well integrity are two main objectives for a successful cementing job. These objectives require proper placement of cement in the annular depth interval of interest. Cement placement, in turn, is dependent on effective drilling mud removal. A spacer fluid is designed to aid displacement of the drilling fluid and to minimize cement contamination. It takes into consideration not only wettability tests and compatibility results between spacer/mud and spacer/cement, but also rheological properties to improve friction pressure hierarchy. The rheological properties vary depending on flow rate, polymer concentration, spacer density and temperature. Trial-and-error lab tests and re-designs are needed to establish a spacer to meet those requirements. To optimize the costly process, and to expedite the job design, it is necessary to propose a smart rheological hierarchy optimization methodology that can predict the spacer rheologies and polymer concentrations by leveraging information from known data. In this paper, a novel rheological hierarchy optimization methodology with artificial intelligence and inverse technique is introduced to improve spacer design. First, sufficient existing rheological property data (FANN readings) as a function of spacer density, temperature and polymer concentration are used to train the machine learning algorithm. Second, depending on the selected mud and cement, density of the spacer and the bottom-hole temperature, the required spacer properties are predicted by the trained machine learning algorithm for each polymer concentration. Third, given the wanted rheological properties, the best solution can be found by the inverse optimization algorithm with the least- square error in the range of polymer concentration. The proposed methodology is applied to a new class of spacer system, which is a premium, water-based spacer, designed to effectively displace the drilling fluid in the annulus. Hundreds of rheological properties are recorded with various combinations of density, temperature and polymer concentration. The data are used in training the nonlinear machine learning algorithm. In real cement-job design, with the required FANN readings of the spacer and FANN readings of mud and cement, the algorithm finds an optimized solution. The FANN readings predicted by the machine learning algorithm are compared in the inverse search step. The calculated polymer concentration and FANN readings are confirmed by laboratory tests with minor variations.

Prediction of Total Organic Carbon in Organic-Rich Shale Rocks Using Thermal Neutron Parameters

“…However, they incur prohibitively high costs and cannot be used conventionally, hence requiring new robust and more convenient methods for TOC quantification . Artificial intelligence (AI) has been considerably used in the last few years in oil and gas research, and much work has been made on the prediction of TOC based on core and well log data. ,− In most cases, AI methods are not globally applicable due to the heterogeneity of shales, which indicates the cruciality of studying the nature of the targeted fields and picking proper logs for the model. , Huang et al demonstrated one of the early applications of AI in predicting TOC using only three conventional (gamma ray, resistivity, and sonic) logs as an input and a pseudo-TOC log calculated from an empirical correlation following the Passey et al approaches . Subsequently, conventional logs have been fed as inputs, that is, gamma ray log, density log, acoustic log, deep and medium resistivity logs, and porosity log in addition to uranium (U), thorium (Th), and potassium (K) contents. − X-ray fluorescence elements’ data (such as copper and nickel) and thermal neutron porosity as well as conventional log combinations displayed correlation with the TOC in the investigated environment − and were proofing evidence of AI reliability in predicting TOC.…”

“… 7 Artificial intelligence (AI) has been considerably used in the last few years in oil and gas research, and much work has been made on the prediction of TOC based on core and well log data. 15 , 35 − 41 In most cases, AI methods are not globally applicable due to the heterogeneity of shales, which indicates the cruciality of studying the nature of the targeted fields and picking proper logs for the model. 1 , 33 Huang et al 39 demonstrated one of the early applications of AI in predicting TOC using only three conventional (gamma ray, resistivity, and sonic) logs as an input and a pseudo-TOC log calculated from an empirical correlation following the Passey et al 32 approaches.…”