Selection of an Optimum Drilling Fluid Model to Enhance Mud Hydraulic System Using Neural Networks in Iraqi Oil Field

In drilling processes, the rheological properties pointed to the nature of the run-off and the composition of the drilling mud. Drilling mud performance can be assessed for solving the problems of the hole cleaning, fluid management, and hydraulics controls. The rheology factors are typically termed through the following parameters: Yield Point (Yp) and Plastic Viscosity (μp). The relation of (YP/ μp) is used for measuring of levelling for flow. High YP/ μp percentages are responsible for well cuttings transportation through laminar flow. The adequate values of (YP/ μp) are between 0 to 1 for the rheological models which used in drilling. This is what appeared in most of the models that were used in this study. The pressure loss is a gathering of numerous issues for example rheology of mud), flow regime and the well geometry. An artificial neural network (ANN) that used in this effort is an accurate or computational model stimulated by using JMP software. The aim of this study is to find out the effect of rheological models on the hydraulic system and to use the artificial neural network to simulate the parameters that were used as emotional parameters and then find an equation containing the parameters μp, Yp and P Yp/ μp to calculate the pressure losses in a hydraulic system. Data for 7 intermediate casing wells with 12.25" hole size and 9 5/8 " intermediate casing size are taken from the southern Iraq field used for the above purpose. Then compare the result with common equations used to calculate pressure losses in a hydraulic system. Also, we calculate the optimum flow by the maximum impact force method and then offset in Equation obtained by (Joint Marketing Program) JMP software. Finally, the equation that was found to calculate pressure losses instead of using common hydraulic equations with long calculations gave very close results with less calculation.


Introduction
Drilling mud has a crucial part in the drilling processes for petrol wells. The Water base (WB) mud are desirable drilling fluid because of its character, the cheaper price, ecologically friendly, and equipped to justifying the well control difficulties [1]. A representative WB drilling mud comprises of water by way of a base and some additives to achieve aimed roles, for instance rheology controlling additive and density control additives [2]. The supreme public way for assessing and enhance the performance of mud is by inspecting their rheological properties for instance yield point and plastic viscosity [3]. Throughout circulations of the mud inside the well, the friction between mud and the wall of the annulus and drill pipe reason to pressure loss [4]. The launch of the JMP program was the beginning of the nineties, and then it developed little by little until it became as good as it is now to take benefit of the graphical border presented by the its operating systems [5]. This program has since been expressively revised and made obtainable also designed for the Windows system [6]. JMP program is used for applications for instance, quality regulator, and engineering, strategy of experiments, in addition to investigation in science and engineering. In supreme cases of an Artificial Neural Networks (ANN) is an updated system which make modifications in its construction based on exterior or interior instruction which flows inside the network throughout the learning stage information [7] , [8]. Current neural networks should be nonlinear numerical information demonstrating tools [9]. Nevertheless, at the moment, an excessive deal of determinations is attentive on the improvement of artificial neural networks for requests for instance information compression and optimization [10]. Artificial neural network contains of an organized collection of artificial neurons using a connectionist method for computation [11]. The main drilling problems for instance fluid loss, torque, wellbore strengthening, drag, well control, carrying capacity, and stuck pipe outcome from the inappropriate corresponding of drilling mud properties. Those problems happen because of differences in pressure, and temperature that have an excessive influence on the mud rheological properties. Mud properties may be improved for the effective drilling process [12]. The main goal of this study to identifying the effect of rheological parameters influence on the total pressure losses in the hydraulic system using experimental results such as the viscosity and yield point for the better understanding of the rheological model and pressure losses. This study is based on that the rheology model for pressure loss  prediction can be investigated to the desired level in an experimental laboratory facility and 111 drilling field data, which can be applied to reduce drilling problems in wells. This study presents a simplified procedure for selecting the rheological model of 25 samples which best fits the properties of a given hydraulic fluid to represent the shear-stress, shear-rate relationship for a given fluid. Throughout this particular study, an Artificial Neural Network model was implemented through the fitting tool of (Joint Marketing Program) JMP software. The study assumes that the model obtained by ANN technique which gives the lowest Absolute Average Percentage Error (AAPE) between the measured and calculated total pressure losses is the best one for a hydraulic system calculation. The results are of great importance for achieving the correct pressure drop and hydraulics calculations., in general, the best prediction of total pressure losses for the mud samples and drilling field data considered (AAPE = 8.7%) and (R=91.3%). The study also included optimum flowrate calculation for better hydraulic system.

Methodology
This study will be conducted with the input data sets of μp, Yp, (Yp /μp), Total Flow Area (TFA) and all drilling data provided. To maximize the accuracy and reliability of the model, the variables that been selected based on the importance toward the Pressure losses. The selection will be based on the R-Square (R) for each parameter. The variables with high R-square(R) will then selected. In order to minimize the randomness and increase the relevancy or the model, the data is then being simplified before being evaluated in Artificial Neural Network. Artificial Neural Network will be trained for the input of the selected variable with simplified data and the output of the target data, in this study, Pressure losses. With trainings and validations, the simulation for sample data will then be evaluated in order to find the matching between predicted and actual. An error analysis is then being observed in order to find the relevancy of the matching, and as shown in Figure (

Data Preparation
Data used in program has been obtained from seven wells in x-oil field in south of Iraq and included flowrate, plastic viscosity, yield point and total flowrate as input data while the output data include total pressure losses in the hydraulics, In the critical operations, the ECD is used to control the formation pressure and prevent kicks without fracturing the drilled formations.
When the mud pumps are switched off, the reduction of ECD may result in underbalanced conditions which require good knowledge of the ECD to avoid any drilling problems. At the same time, it is not possible to increase the mud weight due to fracture pressure limitations. the higher plastic viscosity generates higher resistance in mud which in turns will affect cutting lifting performance and increase pressure losses. The situation may be worsened by the increase of ultra-fine drill solids in the drilling fluid which causes incremental trend of plastic viscosity at constant mud weigh. The pressure losses & ECD have been calculated by using Excel program after analysis lab data and determine the type of model Table (2) illustrate the data used:

Results and Discussion:
The number of neurons in the input and output layers are fixed and can be determined from the number of input and output parameters. For example, in case I the number of output neurons is one, which are total pressure losses. The numbers of input neurons are five, representing the input parameters which were found to be most effective for predicting total pressure losses.

Network Architecture Design & Analysis Of Data Base:
The neural network model was tested to predict total pressure losses in a new well for the same formation. Although tested for the same formation, the new well test was carried out in a 59 effects on error value but an increasing in number of hidden layers has no effect on decreasing error and performance of network. In this study activation function and different number of neurons has been used to achieve this job and considered the fastest method for training moderate-sized feed-forward neural networks reach to several hundred weights.

Prediction of Pressure Losses Equation Based on Ann:
The results are extracted and discussed here from training and testing regression with BPNN technique. The BPNN has been implemented for carried out non-linear regression to obtain predicted pressure losses

Calculation of Optimum Flow Rate:
Depending on the data in Tables (3) and (4) and equations (2 and 3), the optimum flowrate is founded by using maximum impact force method.   Where P b = bit pressure psi, Q= flowrate gpm,ρ= mud density ppg ,AT=nozzle area in 2 ,P c = circulating pressure psi,P s = stand pipe pressure psi

Calculatingn -Index:
The optimal bit pressure drop is associated to the factor "n" that is a distinguishing of a specific system. The slope of the pressure loss curvature for the whole system (Pcirc), not including of the bit, designed on a log-log chart. The entire pressure losses of the system, that should be corresponding to standpipe pressure (Psurf), possibly will also be designed by way of the sum of the bit pressure loss (Pbit), and also the circulating pressure loss of the system (Pcirc). It The result of tests shows that the samples (1,2,3,4,5,6,8,11,12,13,14,15,16,18,22,23,24,25) are represent power low model but samples (7,9,10,19,20,21) are represent Bingham model. Figure (10) shows the relationship of viscosity, yield point and pressure loss, as it is clear that increasing viscosity and yield point leads to an increase in pressure loss. Because this leads to fatigue of the mud pump in order to pump high viscosity drilling fluids, and thus leads to additional pressure loss. Figure (11) shows the relationship between the equivalent density and pressure loss, as it is clear that increasing the equivalent density leads to an increase in   3-In order to set up model the number of output neurons is one target, which is the total pressure losses. The numbers of input neurons are five, representing the input parameters which were found to be most effective for predicting total pressure losses. The effective parameters that have been choose is μp, YP, YP/ μp, TFA, Q as input data.

4-
The neural network model was tested to predict total pressure losses in a new well for the same casing sections; analysis in an altered formation has been done for enhancing the simplification of the developed model.