JAWAHARLAL NEHRU TECHNOLOGICAL UNIVERSITY KAKINADA

UNIVERSITY COLLEGE OF ENGINNEERING(A) KAKINADA

Dept. of Petroleum Engineering and Petrochemical Engineering

Final project review

MODEL BASED OPTIMIZATION OF DRILLING FLUID DENSITY AND
VISCOSITY
By
Chellaboyina SaiKumar(16021A2617)
B.Tech 4th year Petroleum Engineering

Under the guidance and supervision of

Dr.K.Srinivas Rao
Dept of PE&PCE
Department of Petroleum Engineering & Petrochemical Engineering, Kakinada
CONTENTS
 INTRODUCTION

 EXPERIMENTAL DETERMINATION OF DRILLING FLUID DENSITY & VISCOSITY

 INTRODUCTION TO OPTIMIZATION

 WELL HYDRAULICS KERNEL

 PROXY MODEL & DERIVATION

 OPTIMIZATION MODULES

 CASE STUDY

 COST FUCTIONS

 CONCLUSION
INTRODUCTION :
 DEFINITION :
A drilling fluid is a heavy, viscous fluid that is used in oil and gas drilling operations to carry rock
cuttings to the surface and also to lubricate and cool the drill bit.
 FUNCTIONS

a. Control Subsurface Pressure

b. Transport Cuttings

c. Support and Stabilize the Wellbore

d. Support Weight of Tubulars

e. Cool and Lubricate the Bit and Drill String

f. Transmit Hydraulic Horsepower to Bit

g. Provide Medium for Wireline Logging

h. Assist in the Gathering of Subsurface Geological Data and Formation Evaluation

CLASSIFICATION

 RHEOLOGY
EXPERIMENTAL DETERMINATION OF DRLLLING FLUID DENSITY
AND VISCOSITY
 DENSITY
 It is determined by using mud balance

PARAMETER WATERF MUD

SAMPLE(BENT
ONITE)

DENSITY 8.3 lb/gal 8.5 lb/gal

SPECIFIC 1 1.024
GRAVITY

 VISCOSITY
 It is determined by using Fann Viscometer
 S.NO R.P.M DIAL Bingham plastic Power law Herschel
READING model model (T2) Bulkeley

ᵧ T1 model(T3)

1 600 60 1020 17368 30557.78 30583.78

2 300 43 510 8696 21909.24 21935.24

3 200 34 340 5806 18034.48 18060.48

4 100 23 170 2916 12930.32 12956.32

5 6 4 10.2 199.4 3350.59 3376.59

6 3 2 5.1 112.7 2402.30 2428.8

 The values of Y, T1, T2 ,T3, are simulated from matlab

CODE
 Clc graph bw shear rate and shear stress
 Clear all t1 t2 t3

35000
 rpm=[600 300 200 100 6 3];
 d=[60 43 34 23 4 2]; 30000

 mp=d(1)-d(2); 25000

 ty=d(2)-mp;
20000

 n=3.32*(log(d(1)/d(2)));

T1,T2,T3
15000
 k=510*((d(2)/(511^n)));
10000
 fprintf(' rpm deg sr tb tp th \n\n')
 for i=1:6 5000

 sh=1.7*rpm(i); 0
-200 0 200 400 600 800 1000 1200
 tb=ty+mp*sh;
-5000
SHEAR RATE
 tp=k*(sh^n);
 th=ty+(k*(sh^n));
 format shortg
 disp([rpm(i) d(i) sh tb tp th])
 end
 INTRODUCTION TO OPTIMIZATION
 Drilling automation is an active area within the oil and gas industry, driven by the need to improve well
construction quality, safety and cost efficiency. The drilling fluid or mud system is also a vital
component of every drilling operation and accounts for about 10% of total well costs. Drilling fluid
management is a critical component of process safety and cost effective drilling operations.
 Drilling fluids are characterized by properties such as density, rheology, gel strength, oil water ratio,
water-phase salinity, and electrical stability, to name the most important ones . The downhole pressure
ensuring borehole stability is mainly controlled by the density of a drilling fluid. Too low, mud weight
can cause compressive shear failure of the borehole and formation fluid inflow whereas too high mud
weight can cause tensile failure of the formation, leading to losses.
 The rheology has a major influence on the efficiency of cuttings transportation from the drill bit to the
surface. Drilling fluids should be shear thinning and form a gel when no shear is applied. Too little
viscosity at low shear rates can lead to barite sag or insufficient carrying capacity in static periods.
Excessive viscosity at high shear rates might result in excessive downhole pressures due to poor
cuttings removal.
 To determine the appropriate ranges of drilling fluid properties, the cuttings concentration and pressure
distribution along the entire wellbore are simulated using a well hydraulics software, or often referred
to as the well hydraulics kernel. The cuttings concentration everywhere along the wellbore should be as
low as possible and the simulated pressure distribution should fall within the expected fracture and pore
pressure gradient. Those gradients are determined prior to the simulation by a borehole stability
software. This software uses knowledge gained from offset wells and data acquired with geological
studies.
 Here J(X) is the overall objective function that can be decomposed as:
 Ji(fi(X)) : Objective function i associated with an output fi(X) of the kernel,
for example the hole cleaning index or the equivalent circulating density.
 fi(X) :Output i of interest of the well hydraulics kernel such as hole
cleaning index or equivalent circulating density, represented as unknown
non-linear function of input variables of the kernel.
 X : Vector containing the values of all the kernel its input parameters.
 U and L : Superscripts denoting the upper and lower bound respectively.
 Firstly, a well hydraulics kernel that predicts the down hole pressure and hole cleaning should be
available. Secondly, given the advised parameter ranges of drilling fluid properties, drilling
parameter simulations following a systematic experimental design procedure are run. Thirdly, proxy
models are derived to approximate the input output relationships given the simulation results.
Finally, numerical optimization techniques are used to find the optimal drilling fluid properties given
an objective function, the proxy models, and constraints

The Work Flow of Study

WELL HYDRAULICS KERNEL
 The goal of this work is to develop an optimization module with the objective of maximizing hole
cleaning and/or minimizing pressure subject to holecleaning, pressure, and other constraints. Shell has
developed an in-house software, the Well Hydraulics kernel, which is a simulation tool for predicting
the downhole pressure and various quantifications of hole cleaning capabilities. To simulate the
downhole pressure and hole cleaning indicators as a function of depth, the kernel divides the spatial
domain of the wellbore in O(103) grid cells and solves the Navier Stokes equation. The calculations
performed by the kernel (typically one second per calculation) have been used by well engineers as a
planning and real-time monitoring tool to optimize drilling parameters.

 The input data to the kernel covers both well design parameters as well as operational parameters.
Well design parameters include among others the open hole diameter, the casing inner diameter, the
true vertical depth and the measures of the drill string. Operational parameters include among others
cuttings density, cuttings size, flow rate, rate of penetration and drilling fluid properties.
 The primary outputs of the kernel are pressure and cuttings concentration, both simulated as a
function of depth. The kernel also calculates several quantities related to those two. These include the
frictional and static pressure, the equivalent circulating density, the suspended cuttings concentration,
the bed height, the open flow area, and the hole cleaning index.
 OUTPUT PARAMETERS OF INTEREST
 The to-be-developed optimization module will make use of two outputs of the kernel, i.e., the
equivalent circulating density and the hole cleaning index. The equivalent circulating density ρEC
[kg/m3] is directly related to the total pressure ptotal and defined as

 The hole cleaning index, IHC [-], is defined as the ratio of the minimum drill string cross sectional
open area (often the bit open area) Aopen [m2], and the annular cross sectional area occupied by
cuttings Ac [m2], with a maximum value of 5.00.

 In the field the hole cleaning index is the standard parameter to quantify the goodness of the
achieved hole cleaning. Validation of the IHC with a large number of North Sea wells showed that
low IHC values (0 to 0.75) indicate that hole cleaning problems are likely and proper hole cleaning
procedures should be followed. Intermediate risk exists when the IHC is in the range of 0.75 to 1.25.
Larger IHC’s, i.e. IHC > 1.50, correspond to low risk and good hole cleaning. As an example,
below Figures show a typical equivalent circulating density and hole cleaning index profile for a
certain set of input parameters as simulated by the hydraulics kernel.
 SENSITIVITY ANALYSIS OF KERNAL

 The behaviour and sensitivity of the kernel its output on the selected parameters can be visualized by
varying them one by one over a certain range. For this purpose, each individual input parameter is
varied by ±20% around its nominal value and the simulated equivalent circulating density at the bit
and average hole cleaning index as simulated along the borehole are plotted in below Figure .
 Visualization of the behavior and sensitivity of a) the equivalent circulating density at the bit and
b) the average hole cleaning index on applying a ±20% variation on one input parameter at a time.
 Varying the parameter values over a larger range for example 50% does show less smooth behaviour .
This demonstrates already the non-linear character of the kernel. A more complex proxy model would be
required to approximate the kernel its behavior over more extended ranges
 Conclusion
In this work the hydraulics kernel is used as the foundation for developing the envisioned automatic optimization
module. This module will make use of the simulated pressure profile and cuttings concentration profile via two
related quantities, namely the equivalent circulating density and the hole cleaning index. The most significant
input variables for these quantities are identified, and a one dimensional sensitivity analysis of them within a
restricted range showed smooth relations having limited curvature.
PROXY MODEL&DERIVATION

 In order to solve this non-linear optimization problem, parametrization of f1(X) and f2(X) is needed. An
analytical expression for ρEC does exist , and it partially captures the input/output structure of the kernel.
However, both the ρEC as well as the IHC output are the result of numerical simulation. Therefore, a set of
math functions is needed that accurately approximates f1 and f2 , denoted as f^1 and f^2 . These functions
will be called proxy models and will later serve as objective functions and/or constraints in the optimization
module.
Based on the two acquired data sets, a variety of proxy models can be built. The models considered in
this study are
1. The equivalent circulating density at the bit
2. The hole cleaning index at the most critical point along the well bore, that is, the worst hole cleaning
index encountered
3. The total hole cleaning index
 Equivalent circulating density at the bit
It is sufficient to only build a proxy for the equivalent circulating density at the bit because the cased
sections of a well are less vulnerable to too low or too high pressures compared to the open hole section.
In addition, ρEC varies relatively slowly with depth
 Worst hole cleaning index
Hole cleaning index values below IHC,min = 1.5 mean that there is increased likelihood of drill pipe
sticking problems due to cuttings compaction behind the drill bit when pulling out of hole. Ideally one
wants to have IHC > 1.5 along the entire well.
 Total hole cleaning index
Instead of only including one value of a (discretized) hole cleaning index profile, this proxy model
includes them all by building a model based on the sum of all ND values making up one profile. This
results in an overall optimized hole cleaning, instead of optimization at only one point along the wellbore.
Figure10 ; Here, the worst hole cleaning index occurs near the start (~1700m) and end (~4000m) of the
highly deviated section.
This suggests that the shape of the hole cleaning index profile is more or less independent on the
operational parameters. This allows us to state that maximizing the hole cleaning index at the most
critical point along the wellbore – i.e. the worst hole cleaning index – will likely result in a maximized
hole cleaning index along the entire wellbore. Hence, building a proxy model for the most critical depth
is considered sufficient.
 Therefor the proxy model has to meet several criteria. Those will be introduced in the next Section
based on a case study. Proxy models are built for ρEC,bit, IHC,crit, and IHC,total based on 161 samples
taken from the subspace of flow rate Q, mud weight ρdf, plastic viscosity µpv, and yield point µyp,
varied over a range of ±20% with respect to their nominal values as indicated in Table.
 The subspace used in this case study spans four dimensions ranging ±20% around the nominal
parameter values.
Parameter Minimum value Nominal value Maximum value

Q [lit/min] 760 950 1140

ρdf [kg/m3] 992 1240 1488

µpv [deg 20 25 30

µyp [deg] 13.6 17 20.4

Table :the subspace used in case study spans four dimensions ranging 20% around nominal parameter values
 For each of the 161 samples the kernel simulates an equivalent circulating density profile, ρEC,
and a hole cleaning index profile,

Based on Figure b the IHC,crit proxy model is built for the measured depth of 4303m. Among all
profiles the hole cleaning index has its lowest value most frequently at this depth, i.e. 101 times.
Conclusions
 A workflow to obtain expressions for the proxy models f^1(X) and f^2(X) has been identified. Using
this workflow the input/output structure of the hydraulics kernel can accurately be captured within a
given subspace.
 To achieve sufficient accuracy a minimum number of samples should be taken from the kernel, at least
twenty times the number of parameters included in the proxy model.
 One of the limitations of the proxy models is that they cannot be accurately built over large ranges and
simultaneously include a large amount of parameters.
OPTIMIZATION MODULES
 The workflow of proxy model derivation based on polynomial regression models is given in earlier
sildes . It was shown that for the drilling parameter range of interest, the proxy models for IHC and ρEC
are capturing the original IHC and ρEC as simulated by the well hydraulics kernel quite accurately.
These models can thus be used in the optimization module. The optimization module will take the
proxy models as its objective function(s) and/or constraint(s).
 Two optimization problems are considered: The first one is with the objective function of maximizing
the worst hole cleaning index IHC,crit and the second one considers maximization of the total hole
cleaning index IHC,total. Given the proxy models for IHC,crit = f^c1(X) , IHC,total = f^t1(X) , and
ρEC,bit = f2(X) , the optimization problems can be formulated as follows
 The optimization problems are non-linear, constrained optimization problems. To solve them, the
Matlab optimization function fmincon will be used with the sequential quadratic programming
algorithm. This steepest descent method is one of the most effective algorithms to solve non-linear
constrained optimization problem

CASE STUDY
 The well considered, Well II, is an S type production well with a measured depth of 5845m. The drilling
fluid program specifies the constraints for the drilling fluid properties

PARAMETER LOWER BOUND UPPER BOUND

Mud weight ρdf [sg] 1.24 1.24

Plastic viscosity µpv [deg] As low as possible

Yield point µyp [deg] 16.9 20.6

6 rpm Fann reading µr6 [deg] 7.5 11.2

Table; Targeted ranges of the drilling fluid properties as taken from the drilling fluid program
 PARAMETR FIELD VALUE

Avg Low High

Mud weight ρdf [sg] 1.24 1.24 1.24

Plastic viscosity µpv [deg] 25 23 27

Yield point µyp [deg] 17 17 17

6 rpm Fann reading µr6 [deg] 11 10 11

Flow rate Q [lit/min] 950 350 950

Rate of penetration Rp [m/h] 17.5 15 20

Rotary speed vrot [rpm 65 50 80

Table ; Average field values of mud properties and other operational parameters which can be included in
the proxy models. The lowest and highest value as observed while drilling are mentioned
 As required, the parameter values as used in the field do fall within the ranges mentioned in this table.
Given the average values, the kernel simulates circulating density at the bit of 1385 kg/m3. The
corresponding hole cleaning index profile has a minimum value of 1.19 and an average value of 3.01
(Figure).

Figure; Simulated hole cleaning index profile based on the average parameter values as used in the field.
 In the field the flow rate is the parameter which can relatively quickly varied. Therefor it is of interest to also include
the flow rate as a parameter in the optimization module. Here a flow rate variation of ±150lit/min is considered. Below
Table summarizes the parameter ranges of interest. PARAMETR LOWER UPPER
Plastic viscosity µpv [deg] 23 27
Yield point µyp [deg] 16.9 20.6
6 rpm Fann reading µr6 [deg] 7.5 11.2
Flow rate Q [lit/min] 800 1100
Rate of penetration Rp [m/h] 15 20
Rotary speed vrot [rpm 50 80
 PARAMETR Optimized value
Module 1 Module 2
Plastic viscosity µpv [deg] 23 23
Yield point µyp [deg] 16.9 16.9
6 rpm Fann reading µr6 [deg] 7.5 7.5
Flow rate Q [lit/min] 1067 1069
Rate of penetration Rp [m/h] 15 15
Rotary speed vrot [rpm 80 80

Table ; Optimized parameter values after optimizing for all seven parameter

Figure ; The hole cleaning index profiles obtained after optimizing using the parameter values of above Table.
 PARAMETER ORIGINAL Module 1 Module 2

ρEC,bit [kg/m3] 1385 1385 1385

IHC,total [-] 1.08*10^3 1.18*10^3 1.18*10^3

IHC,ave [-] 3.01 3.31 3.31

IHC,crit [-] 1.19 1.55 1.55

DMD,crit [m] 4.3*10^3 3.6*10^3 3.6*10^3

Table; Quantifying the improvement made by optimizing six parameters

 An improvement of the worst hole cleaning index from 1.19 to 1.55 is obtained for both optimization
modules. Based on the average hole cleaning index, an improvement of 10% is made. This value honors the
constraint set in for the hole cleaning index. Also the two constraints set for the equivalent circulating
density at the bit are honored.
 COST FUNCTIONS
 Both preferred IHC values as well as preferred ρEC,bit values are valuated with a 2% savings on the total
well construction costs. The least preferred values come at a 2% cost. This is based on the average cost of
drilling non-productive time caused by hole cleaning issues or borehole instabilities. Summing the
proposed cost functions then results in a new objective function J(X). This function will be based on
both the proxy model and the IHCcrict(X) or the IHCtotal(X) proxy model, instead of on one of the
latter two only.
 Hole cleaning index
 The optimization modules introduced ,assume that an equal benefit is achieved with an increase of the
hole cleaning index from 3 till 4, as with, let’s say, from 0.5 to 1.5. This is a simplification of reality
given that larger IHC values, i.e. > 1.5, all correspond to low risk and good hole cleaning practices (Sub-
section 2.1.3). This makes an improvement of the worst hole cleaning index from 0.5 to 1.5 much more
significant compared to an improvement from 3 to 4. To value this difference a cost function should be
introduced for the hole cleaning index. This can be done by writing the savings.

 It is noted that in the latter case, a conversion is required from the total hole cleaning index value to a
representative hole cleaning index value between 0 and 5.
 The non-differentiable function of above Equation can be approximate by

 This function will be part of the new objective function. The above two equations are graphically shown in
below

Figure ; Cost function for the hole cleaning index IHC.

 The graph for J1(X) shows that a hole cleaning index of zero comes at a loss of 2% compared to the ‘base
case’, when an hole cleaning index of 1.5 is achieved. For a hole cleaning index of zero, two percent more time
is required to deliver the well due to required additional circulating time. For improvements of the hole
cleaning index above 3 savings of 2% are made, meaning no additional value is created by increasing the hole
cleaning index from 3 to 5.
 Equivalent circulating density
 The optimization modules also assume an equal cost of having a downhole pressure anywhere between
the pore pressure and the fracture gradient. The optimization module ended up a few times at the
fracture gradient (see for example Table 4-2 and Table 5-9). In the field one wants to refrain from
approaching this gradient. Neither one wants to approach the pore pressure gradient. The equivalent
circulating density at the bit should preferably be located as far away from both gradients as possible,
while being shifted more towards the pore pressure gradient. To value this difference a cost function
should also be introduced for the equivalent circulating density. This can be done by writing the savings
J2(X) for example as

 Equation of J2(X) is graphically shown in Figure.

 Figure ; Cost function for the normalized equivalent circulating density at the bit ρNbit EC

 It shows that a money loss of 2% is associated with the downhole pressure when ρEC,bit is too low
or too high. Ideally the downhole pressure is located around bit ρNbit EC= 0.37, which saves 2% of
time.
CONCLUSION

 The objective of this project is the development of an optimization module, based on an existing hydraulics
kernel, that systematically derives the optimal drilling fluid properties and operational settings, in the sense
that they maximize the hole cleaning index while maintaining borehole stability. The developed tool could
help engineers plan their drilling fluid programs more scientifically, i.e., compared to the current practice.
 A set of optimization modules is devised that requires users to only define the ranges of interest for the input
parameters to be optimized, and fixed parameters such as the well geometry, and the pore pressure and
fracture gradient. The procedure will:
 automatically derive proxy models for the hole cleaning index and the equivalent circulating density that are
used in an optimization framework;
 check the derived models for their accuracy based on four proposed criteria;
 run the selected optimization module giving the optimum mud properties and/or operational parameters as
an output.
 Second-order polynomial regression models proved to be adequate as a basis for the proxy models since
they capture the input/output structure of the hydraulics kernel sufficiently within a given subspace.
 Three optimization modules are considered is this work taking as objective function:
 The worst hole cleaning index encountered along a profile.
 The summed hole cleaning index profile.
 A cost function containing both the hole cleaning index as well as the equivalent circulating density.
 BIBLOGRAPHY
 [1] Thesis of Roel F.H. Roijmans at “ shell “,January 6, 2017.
 [2] “BaraLogix™ Real-Time & Automation Services,” Halliburton, [Online]. Available:
http://www.halliburton.com/en-US/ps/baroid/real-time-services/baralogix.page.
 [3] B. Bloys, N. Davis, B. Smolen, L. Bailey, O. Houwen, P. Reid, J. Sherwood, L. Fraser and M. Hodder,
“Designing and Managing Drilling Fluid,” Oilfield Review, pp. 33-43, April 1994.
 [4] T. Schuit, “Thesis: Control of drilling fluid properties; Density and Viscosity,” Shell, Rijswijk, 2015.
 [5] R. Nafikov and M. Glomstad, “Automatic Mud Mixing,” in SPE-163473-MS, Amsterdam, 2013.
 [6] J. Matheus, M. Ignova and P. Hornblower, “A Hybrid Approach to Closed-loop Directional Drilling
Control using Rotary Steerable Systems,” Elsevier, vol. 45, no. 8, pp. 84-89, 2012.
 [7] H. Melgares, W. Grace, F. Gonzalez, C. Alric, J. Palacio and G. Akinniranye, “Remote Automated
Directional Drilling Through Rotary Steerable Systems,” in SPE-119761-MS, Amsterdam, 2009.
 [8] T. Vromen, N. v. d. Wouw, A. Doris, P. Astrid and H. Nijmeijer, “Observer-based outputfeedback
control to eliminate torsional drillstring vibrations,” Proceedings of the IEEE Conference on Decision and
Control, pp. 872877, Febr 2015.
 [9] W. Aldred, J. Bourque, M. Mannering, C. Chapman, B. d. Castel, R. Hansen, G. Downton, R. Harmer,
I. Falconer, F. Florence, E. G. Zurita, C. Nieto, R. Stauder and M. Zamora, “Drilling Automation,” Oilfield
Review, vol. 24, no. 2, pp. 18-27, 2012.
 [10] Drilling Fluids Engineering Manual - Chapter 5, M-I, 1998.
 [11] Wells Distance Learning Package, Volume 1, Section 4, Part 1, The Hague, The Netherlands:
Shell International Exploration and Production B.V., 2005.
 [12] A. Chan, “The STABOR User Companion,” Shell International Exploration and Production
Inc., Houston, 2010.
 [13] J. Wentink, “WellHydraulics NextGeneration Verification and Acceptance Report,” Shell
Global Solutions International B.V., Rijswijk, 2014.
 [14] J. Dudley, J. Nicholson, J.C.Woodhouse, M. Albrecht, S. Zeilinger, E. v. Oort and A. Tutuncu,
“Borehole stability analysis prospect princess, Offshore Gulf of Mexico,” Shell Technology EP,
Houston, 2000.
 [15] A. Tutuncu, J. Nicholson, S. Zeilinger, E. V. Oort and Q. Sharif, “Borehole Stability A9alysis
Prospect Auger, Offshore Gulf of Mexico,” Shell Technology EP, Houston, 2001.
THANK YOU