Download as pdf or txt
Download as pdf or txt
You are on page 1of 22

Optimized Enhancement of Crude Oil Distillation process

using Artificial Intelligence (AI) techniques


Abhishek S Chaudhary and Vagisha Rastogi, School of Chemical Engineering (SCHEME), VIT
University, Vellore.

ABSTRACT

Crude Oil Distillation is one of the most critical and complex forms of operation that
involve a number of parameters associated with it. Defining the processing conditions is one of
the major challenges faced by an oil and gas industry primarily because of two reasons, first
variation in the quality of crude oil obtained and the second a large and a varied collection of data
obtained during the processes.
The purpose of this research is to provide insights on the various Artificial Intelligence
Methodologies that promises to deliver the most Optimum Production. These methods not only
increase the profit but also produce maximum yield of the required range of distillates. In the oil
and gas sector, majorly Machine Learning (ML) and Artificial Neural Networks (ANN) tend to
show a lot of applications and promising results. The study starts with the characterization of the
physical properties of crude by Ensemble Random Weight Neural Networks. The optimization of
the distillation unit is done using a Support Vector Machine (SVM) algorithm which classifies and
filters the designs based upon column structures and operating conditions and gives the column
performance as the output.
The ANN predicted the crude oil properties with which the SVM model was trained to get
the most optimum operating conditions for the given design of the column. This study can thereby
help the oil and gas industry to optimally design and operate the crude oil distillation process at
minimal cost. The methodologies have improved the efficiency of the Distillation column as per
the crude oil fed without compromising the product quality.

Keywords: Crude Oil Distillation, Ensemble Random Weight Neural Network, Support Vector
Machine (SVM)
INTRODUCTION
Crude oil extraction process is considered to be a very complex operation and has various steps
associated with it. Depending on the region of extraction properties of crude differ widely. The
extraction process is a multistep process and each stage can be split into further steps so as to apply
optimization techniques. Optimizing the layer step by step not only shall improve the operation
efficiency but also shall reduce the costs associated with operation. The entire crude oil extraction
process is divided into 3 processes namely: Exploration of the oil site, Drilling and Production.
Every Oil refinery has a huge historical data in the petroleum industry which makes our process
easier to analyze. But predicting about the future possibilities becomes a cumbersome task as a
result of which decisions made are sometimes uncertain. Scientists and engineers use various
statistical techniques to extract vital information from the data available, but there are a lot of
variables and parameters involved which are mostly non-linearly related and have complex
dependencies. This makes the task of analyzing analytically very challenging. Here is when
Artificial Intelligence comes into rescue
In this paper we shall discuss how Artificial Intelligence plays a distinctive role in analyzing this
data and they also make accurate real-time predictions with good accuracy. Here we are going to
study how Artificial Neural Network and Machine Learning enhance these tasks. These are some
advanced statistical tools that can solve these nonlinear dependencies and give output predictions.
We explain each of these in the following subsections.

Artificial Neural Networks (ANN)


An Artificial Neural Network is a function similar to the biological neurons present in our brain.
In Deep learning ANN is a supervised learning system consisting of a large number of simple
nodes or elements called neurons or perceptron. Each neuron can make a decision which then is
passed on to other neurons. A Neural Network can emulate almost any function, provided given
enough training samples and computing power. The figure here illustrates the structure of the
neural network.
Fig 1: Structure of Artificial Neural Networks

The core functioning of the neural network typically depends on the following aspects:
1. Inputs
It is the source data fed to the ANN, with an aim to make a decision or prediction about the data.
This comprises a set of real values fed into the neurons of the input layer.

2. Training set
A set of inputs for which correct outputs are known that are used to train the ANN.

3. Activation Function
After the input is fed to the neurons, they pass it through the activation function. They are
continuous functions that accept a wide range of values and their nonlinear nature plays a very
crucial role for training the model. Some common functions are sigmoid, hyperbolic Tangent and
ReLu (Rectified Linear Unit).

4. Weight Space
Each neuron is assigned a numeric weight, the weights together with the activation function define
the neuron’s output. In order to generate the most optimal weights ANNs are finely tuned in
repeated cycles.

5. Forward Pass
Inputs are taken from the neurons, pass them forward in the network and allow each neuron to
react to a fraction of the input. The individual neuron generates their output and passes them on to
the next layer.
6. Error Function
This calculates the deviation of the output obtained by the model from the correct output. The main
objective in ANN model training is to minimize the error function and bring output as close as
possible.

7. Backward propagation
To get the optimal weights for the neurons, the model performs a backward pass i.e. moving from
the output to the neuron that generated the prediction.
8. Bias
This measures how well the model fits to the training set. Every layer in the ANN has an extra
neuron to fulfil this objective.
9. Hyperparameters
These are the settings which affect the structure or operation of the entire neural network. They
are the number of hidden layers, the activation function and the number of times a training should
be repeated.

Fig 2: Procedure to apply ANN to any problem

Machine Learning techniques


Machine Learning is an application of Artificial Intelligence (AI) that enables a system to
automatically learn and predict without being programmed explicitly. This process of learning
begins with data observation by the computer and looking for patterns. Machine Learning is
categorized into, Supervised and Unsupervised. In supervised ML the learning systems are fed
with large amounts of labelled data. Whereas Unsupervised learning identifies patterns in the data
and then splits that data into categories.
For each category of ML, there are different algorithms:
Supervised Learning: Regression, Decision tree, Random Forest, KNN, Logistic regression.
Unsupervised Learning: Apriori algorithm, K-means.
Apart from these there are some algorithms which can be applied to almost any problem:
1. Linear Regression
2. Logistic Regression
3. Decision Tree
4. Support Vector Machine (SVM)
5. Naive Bayes
6. kNN
7. K-means
8. Random Forest
9. Dimensionality Reduction Algorithms
10. Gradient Boosting Algorithm
a. GBM
b. XGBoost
c. LightGBM
d. CatBoost
In this paper we use the SVM algorithm, as upon doing case study analysis it was foreseen that
there could be a possibility to reduce the search space and this is addressed by SVM easily.

METHODOLOGY
We divide the methodology section into 4 subsections. These are on the basis of the operation done
in the crude oil extraction process, namely:
1. Exploration of the oil site
2. Drilling
3. Production
Each subsection describes the conventional methods and how AI techniques are deployed in the
respective arena.
1. Exploration of Oil Site
The search for an oil site requires a profound knowledge of geography, geology and geophysics as
crude oil is essentially found in certain types of geological structures such as fault raps, salt domes
and anticlines. An area of interest is selected and various geophysical surveys are conducted such
as Magneto metric surveys, Aerial photogrammetric surveys, gravimetric surveys and seismic
surveys. This is essentially done to locate and identify the reservoir rocks.

Reservoir Rock properties estimation


Porosity and permeability are crucial factors in reservoir planning. Nuclear Magnetic Resonance
(NMR) is a tool which is used for well logging. From this technique it is possible to obtain different
characteristics of formation fluids like viscosity, porosity of formation and pore size. Measurement
and evaluation of these properties from NMR log data is a cumbersome and an expensive task.
Developing quantitative correlations for the model in an efficient manner is an objective approach.
Here ANN can affirmatively identify complex and nonlinear systems. In order to obtain an
optimum set of weights for the neurons, the Cuckoo Optimization algorithm is used. As it has
faster convergence and higher accuracy. This algorithm minimizes the MSE criterion that was
encountered in backpropagation of the neural network. The ANN entailed Multilayer Perceptron
(MLP) and a selection of appropriate number of hidden layers and number of neurons in each layer
to any degree random degree of accuracy.
The inputs to the ANN are sonic transit time 𝑫𝑻 , 𝑪𝑵 , electrical resistivity 𝑹𝑻 and bulk density
(RHOB) and the outputs are effective porosity (∅𝒆𝒇𝒇 ) and K.

Fig 3: Artificial Neural Network for Reservoir rock estimation


Prediction of physical properties of Crude Oil
In the oil refining industry for the fast evaluation of crude oil, Carbon residual and asphaltene are
the two significant parameters that are considered for crude oil blending, production planning and
process control of refinery. The correlation between NMR spectrum of the crude oil and its
physical properties has very complicated database models and is sometimes impossible to
characterize using the linear models such as Partial Least Squares (PLS) and Multiple Linear
Regression (MLR). So to overcome this difficulty Ensemble Random Weights Neural Network is
used. This ANN also addresses the issues faced in backpropagation algorithms that are easy to
suffer local optimum. Ensemble ANN based on regularized negative correlation learning is
adopted to avoid over-fitting. The negative correlation learning (NCL) model, the cost function
𝒆𝒎 .

Where, pm is the correlation penalty function, λ is the function parameter and λ ∈ [0,1].

2. Drilling
We jump to drilling after finding a suitable location for an oil field. It can be done by drilling a
long hole using an oil rig. It is a process which uses a drill bit to make a hole, can be of different
materials, polycrystalline diamond is mostly preferred. After setting up an oil rig we pass the
drilling fluid to break down the reservoir rock in order to extract the oil content present in the
pores. Drilling fluid plays a major role in maintaining hydrostatic pressure to prevent the entering
of formation fluids into the well bore, it also keeps the drill bit cool and clean to make it work
efficiently. Drilling fluid ranges from incompressible fluids such as brine and water to
compressible fluids such as air.
Parameters affecting Drilling fluid performance
Some factors affecting drilling fluid performance are -
1. Differential Sticking
Many drilling programs run by the engineers shows the permeable formation which leads
to differential sticking. Prevention can be done by maintaining all the mud parameters
within the conditions of the drilling program, particularly it is important:- HTHP and low-
pressure fluid loss are be maintained inside specification (these values ought to be
measured three instances in keeping with day on the minimal) Even as keeping consistency
with suspension and hole cleansing requirements, the gel strengths are maintained on the
minimal fee possible.

2. The change of drilling fluid viscosity


Increment in the viscosity of the drilling fluid, increases the mud resistance and in turn it
will show an effect in the cutting and lifting performances. The condition may be worsened
by the increment of ultra-fine drill solids in the drilling fluid which reduces the viscosity
at the fixed mud weight.
A rotational viscometer is used to measure the viscosity at different shear rates in order to
determine the rheology model coefficient.

PV (cps) = Ɵ600 – Ɵ300


YP (lb/100 sq ft) = Ɵ300 – PV
where,
PV - Plastic Viscosity
YP - Yield Point

3. The change of mud pH


Either pH can be calculated by using Sorensen scale or by using the formula
pH = -log10[H].
pH of the drilling fluid should be maintained to prevent the contamination by substances
which later on affects the solubility and effectiveness of the chemical additives. Companies
should check pH levels of the drilling mud which will save their efforts and money in the
long run.

4. Corrosion or fatigue of the drill string


Drill pipe failures result from the metal fatigue which originates due to the result of
fluctuating stresses with maximum values less than the tensile strength of the material.
Estimation of Collective Fatigue wear in the Drill String during well construction: -
By knowing the bending stress at a definite point along the drill string, we can determine
the maximum number of rotations that cause drill pipe failure.

where,
ni & N - Actual and Maximum number of cycles at a specific bending stress
Di - Collective fatigue wear

The actual number of cycles at a specific bending stress is calculated by using the
formula given below

where,
𝑁06 - Rotary speed in RPM
𝑉𝑚𝑒𝑥 - Drilling Rate in m/hr

5. Thermal stability of the drilling fluid


Instability of the drilling fluid occurs due to the high temperature in geothermal and deep
wells. High Temperature causes the change in the physical and chemical properties which
are the main causes of the problems such as gelatin and formation damage occur.

6. The change of drilling fluid density


Reduction of drilling fluid density below the pore pressure gradient creates an unfavorable
stress distinction that could permit formation fluid to go into the wellbore - below
maximum occasions an unwanted situation. Density or mud weight is the mass according
to unit volume. In an oil field, it’s far measured with a mud stability and is most customarily
said in kilos according to gallon (lb/gal or ppg); specific gravity or SG (g/ml); kilograms
according to cubic meter (kg/cu m); or kilos according to the cubic feet(lb/cu ft). Density
is used to decide the hydrostatic stress of the mud column and also can be measured and
expressed as a gradient such as pounds per square inch per thousand feet (pai/1,000 ft).

Optimization techniques for minimizing Drilling costs


Drilling is a huge task which involves labor, money and time. The main objective is to reduce the
cost of a drilling well using optimization techniques which includes the consideration of seven
variables such as mud properties, hydraulics, bit type, weight on bit, rotary speed these five
variables are controllable ones and rest two are unaltered ones: formation to be drilled and depth
in conjunction with mathematical models selected. The most concerning variables which are taken
into account is - weight on bit and rotary speed and other remaining controllable variables are
assumed to be already optimized.

MODEL EQUATIONS -
Galle and Woods made an assumption which applies to any optimization techniques based on their
equations -

where,
𝐶𝑓 - the formation drill ability factor
𝐴𝑓 - the formation abrasiveness factor
S - the drilling-fluid factor
The above three formulas represent drilling rate, dulling rate and bearing life respectively.
We will be using three optimization methods: Point Optimization, Interval Optimization and
Multi-Interval Optimization.
The First Method is used to minimize the cost per foot drilled during a bit run. The Second Method
minimizes the cost of a selected interval, and the third method minimizes the cost over a series of
intervals.

Artificial neural network model for predicting drill cuttings settling velocity
Nature Inspired Algorithm (ANN) is used to develop a model for estimating cuttings settling
velocity. It can only be evaluated on the basis of some properties relating to the cuttings, the mud
and the wellbore. These properties include: the cuttings sphericity, its diameter, and density as well
as the mud's density, its viscosity and flow rate. Cuttings Transport is an art and science of taking
out drill cuttings from the well. To start cutting transport, mud is circulated to serve as an expulsion
agent.
Artificial neural network is a mathematical model consisting of several input signals with synaptic
weights and a new term bias is added to the total function in order to raise or lower the input which
is received by the activation function.

where,
x - input vector
𝑤𝑗𝑖 - connection weight from the ith neuron in the input layer to the jth neuron in the hidden layer
𝑤𝑘𝑗 - connection weight from the jth neuron in the hidden layer to the kth neuron in the output
layer
𝑏𝑘 - bias of the kth output neuron
𝑓ℎ & 𝑓𝑜 - transfer functions for the hidden and output neuron respectively.
The optimal architecture of the ANN Network for the five inputs is explained below
Fig 4: Artificial Neural Network Diagram for Drill Cutting Velocity

INPUT PARAMETERS - Cuttings sphericity or shape, cuttings density, cuttings diameter, mud
viscosity and mud density are the inputs fed into the neural network.
OUTPUT PARAMETER - It should be predicted and it depends on the input parameters. The
values of this parameter are a set of real values and ranges between 0.001 and 0.78 m/s.
The ANN Model generated by applying The Levenberg Marquardt algorithm which is stated below

where,
j - number of input variables = 5
b1 - bias of the hidden layer
b2 - bias of the output layer
This equation represents the ANN model correlating five parameters and the settling velocity in
MATLAB. Here, ‘purelin’ and ‘tansig’ are the input functions to calculate output from its input
parameters and the network is simulated to predict the cutting settling velocity after training the
neural networks with LM algorithm.
In order to quantify the values that how well the model prediction values match the actual values
which will be evaluated by using four performance metrics such as: mean squared error (MSE),
root mean square error (RMSE), sum of squares error (SSE) and goodness of fit (R2).
STATISTICAL EQUATIONS -

where,
𝑃𝑖 - predicted outputs of the ANN model
𝐴𝑖 - corresponding actual/target values
A - the mean of the actual values
n - number of data points.

3. Production of Oil and Distillation operation


Crude oil distillation is an energy and capital-intensive process. In the oil refineries crude oil
distillation starts with distillation in the Atmospheric Distillation Unit (ADU). Crude oil feedstock
undergoes preprocess treatments and is then converted into different fractions. The plant consists
of a distillation unit, a heat recovery network and furnace where the feedstock is heated and
partially vaporized.

Fig 5: Crude Oil Distillation unit flow Diagram


Optimization of the ADU column
Distillation columns are made up of multiple structures each of them transferring heat and mass.
The distillation unit is equipped with side-strippers and pump-arounds which are used to remove
the light components and provide internal reflux. The distillation process is strongly associated
with the heat recovery through pump-arounds, product coolers and condensers. Thereby, changes
in the design and operation of the ADU affect the design and operation of the heat exchanger and
furnace.
For the overall optimization of the unit, structural variables and operating conditions are
considered. The structural variables include the location of the feed tray, side-stripper draw
streams, number of pump arounds and number of trays in each section of the column. Operation
control variables include feedstock inlet temperature, temperature drops along the column, reflux-
ratio and stripping steam flow rates.

Fig 6: ANN to optimize the SVM processed data space

Existing design methods involve rigorous simulation models which require a trial and error before
the design is finalized. Even though these models are versatile and provide accurate estimates of
the performance of the column, it requires accurate and conditionally well initial points to converge
to a feasible solution. The tray-by-tray column model optimization is computationally intensive.
To overcome this limitation surrogate distillation models based on ANNs are constructed where
in the feasibility constraints are defined using a supervised Machine Learning algorithm known as
Support Vector Machine (SVM). In this model, inputs are given by the column structure and
operating conditions. While the outputs are related to the column performance. The SVM filters
out the infeasible design from the search matrix, thus reducing computational time and faster
convergence. The SVM implemented a genetic algorithm to search for column structure variables
and operating conditions. SVM defines an optimal hyperplane that separates two or more classes
of data points. Hyperplanes are boundaries that are used to classify the data points. In a given
dataset the optimal hyperplane is the one with the maximum distance to the nearest data point in a
n-dimension Euclidean space.
The SVM classifier function
y(x) = sign(w . x + b)
Corresponds to +1 and -1, where +1 denotes feasibility and -1 denotes infeasibility. This is used
to classify the new samples, the bias (b) and the weight vector (w). The objective functions of the
SVM are stated below

Here, n is the number of training data points; are the slack variables; C is a constant to scale
constraint violations.
The data for the surrogate model is generated using multiple rigorous simulation. For each set of
input data simulation is carried out on HYSYS to obtain the set of output variables. The input
variables including both structural and operational are adjusted to improve the column
performance. The output variables represent product flow rates, product quality (T95 and ASTM
T5), Boiling temperatures of each product), enthalpy changes, diameters of each column and target
temperatures of stream. The data obtained from simulation is sent to SVM to eliminate any
infeasible solutions. The processed data is fed to the ANN to correlate the independent variable to
each group of dependent variables. The ANN comprises feedforward Multilayer Perceptron (MLP)
with one input layer, one hidden layer and one output layer. Hidden layer uses sigmoid and output
layer uses ReLU as an activation function. Each hidden layer and output layer consist of 10
neurons. The coefficient of determination, which has a typical range of 0 to 1, is used to judge the
performance.
The operating cost of the crude oil distillation is majorly by the cost of furnace heating, using; fuel
oil of natural gas as fuel, the cost of steam for stripping indirect heating and the cost of cold utilities.
The optimization framework integrates the ANN, the SVM and the costs into a unified framework.
A genetic algorithm is deployed to search the cost-effective design. Every iteration the algorithm
proposes a column structure and its operating conditions, which is later checked by the SVM to
confirm the feasibility of the design. When the SVM confirms the feasibility of the design the
inputs are then fed to the neural network to predict its performance. Inequality constraints are
applied to check the product qualities and flow rates specification.

RESULTS
1.Reservoir Rock properties estimation

The ∅𝑒𝑓𝑓 as predicted by ANN-Cuckoo Algorithm are plotted as shown below for the Training data and
Test data. Performance of the modelling for porosity (K) is also made.

Fig 7: Data response validation from the ANN-Cuckoo Algorithm

ANN- Cuckoo predicted the performance of K-modelling accurately as the 𝑹𝟐 coefficient was close to 1
and the MSE value was much lower.
2. Prediction of physical properties of Crude Oil
The optimum weights of ERNN are determined by exhaustive linear search. And the number of hidden
layer nodes is mostly between 100-300. The regularization coefficient and penalty term coefficient were
searched in the range [0, 1] and [0, 0.6] respectively. Results of the ERNN based model of carbon residual
and asphalt is shown in the figure. It can be observed that the prediction errors of the ERNN are smaller
than that of the other involved data-based models on the whole. Due to the non-linear approximation ability
of ERNN the regularization term is introduced in the ensemble.

Fig 8: Data response validation from the ANN-Cuckoo Algorithm

3. Optimization techniques for minimizing Drilling costs

Fig 9. Scatter plots of ANN model developed with five input parameters.
Fig 10. Comparison of the experimental data with the ANN model predictions.

The above figure shows the good match of the model output from ANN with the experimental
data.
TABLE 1: Statistical Validness of the model

The closer the MSE, RMSE, SSE, MAPE and MAE is to zero, the better the model is said to be.
4. Optimization of the ADU column
A prototype of the actual refinery process was entered into the simulation environment and the
raw crude temperature, pressure and mass flow rate were input. The table below contains all the
parameters of the stream summarized.
TABLE 2: Process Stream Parameters

The ANN model was able to adequately represent the complex ADU process due the non-linearity structure
of MLP. The figures below represent the correlations between the input and output variables.

Fig 11: Correlation between predicted values and simulated values


The constraints of the problem as discussed in the methodology are summarized in the tabular format
below
TABLE 3: Constraints of Operating variable before Optimization process

The optimum Operating Conditions as predicted by the ANN were obtained on the basis of the structure
and design variables as the input. The ANN was able to predict the optimum operating conditions at
which the ADU can be operated without compromising the products quality.

TABLE 4: Optimized variable values

CONCLUSION
Through this paper we fulfilled the ways in which the crude oil distillation process can be
optimized using Artificial Intelligence techniques such as Artificial Neural Networks: MLP,
ERNN and also using SVM Machine Learning Algorithm. Suitable dependent and independent
variables were identified and their feasibility was checked to obtain the most optimum model.
ANN – Cuckoo Algorithm was used to obtain the Reservoir Rock properties and it predicted the
results with high degree of accuracy. Drilling properties and the parameters that influence the
drilling process were discussed. ANN was used to predict and validate the cutting settling velocity
and it indeed gave a good fit to that of the experimented data.
The ANN predicted the crude oil properties with which the SVM model was trained to get the
most optimum operating conditions for the given design of the column. This study can thereby
help the oil and gas industry to optimally design and operate the crude oil distillation process at
minimal cost. The methodologies have improved the efficiency of the Distillation column as per
the crude oil fed without compromising the product quality.

The methodology discussed here can be used for any Oil refinery plant and suitable inputs can be
given to the models. Using the optimum conditions obtained the overall running costs of the unit
can suitably be lowered.

REFERENCES

[1] Zargar, Ghasem, et al. “Reservoir Rock Properties Estimation Based on Conventional
and NMR Log Data Using ANN-Cuckoo: A Case Study in One of Super Fields in Iran Southwest.”
Petroleum, vol. 6, no. 3, 2020, pp. 304–10. Crossref, doi:10.1016/j.petlm.2019.12.002.

[2] Wilson, David C., and Ramon G. Bentsen. “Optimization Techniques for Minimizing
Drilling Costs.” Fall Meeting of the Society of Petroleum Engineers of AIME, 1972. Crossref,
doi:10.2118/3983-ms.

[3] K. Aminian, et al., Prediction of flow units and permeability using artificial neural
networks, SPE Western Regional/AAPG Pacific Section Joint Meeting, Long Beach, California,
2003, pp. 19–24 (May).

[4] Hansen, L. K. and Salamon, P. (1990). Neural network ensembles. Pattern Analysis &
Machine Intelligence IEEE Transactions on, 12(10), 993-1001.

[5] Lu, Jun, et al. “Prediction of Physical Properties of Crude Oil Based on Ensemble
Random Weights Neural Network.” IFAC-PapersOnLine, vol. 51, no. 18, 2018, pp. 655–60.
Crossref, doi:10.1016/j.ifacol.2018.09.349.

[6] Yang, H.-S., & Kim, N.-S. (1996, January 1). Determination of Rock Properties By
Accelerated Neural Network. American Rock Mechanics Association.
[7] Ibrahim, Dauda, et al. “Optimization-Based Design of Crude Oil Distillation Units
Using Surrogate Column Models and a Support Vector Machine.” Chemical Engineering Research
and Design, vol. 134, 2018, pp. 212–25. Crossref, doi:10.1016/j.cherd.2018.03.006.

[8] Fadhil Ahmed, Duraid, and Ali Hussein Khalaf. “Artificial Neural Networks Controller
for Crude Oil Distillation Column of Baiji Refinery.” Journal of Chemical Engineering & Process
Technology, vol. 07, no. 01, 2015. Crossref, doi:10.4172/2157-7048.1000272.

[9] A. Zakerian, S. Sarafraz, A. Tabzar, N. Hemmati, S.R. Shadizadeh, Numerical


modeling and simulation of drilling cutting transport in horizontal wells, J.Pet.Explore. Prod.
Technol. 8 (2) (2018) 455–474, https://doi.org/10.1007/ s13202-018-0435-6

[10] Al-Azani, K., Elkatatny, S., Abdulraheem, A., Mahmoud, M., & Al-Shehri, D. (2018,
August 16). Real Time Prediction of the Rheological Properties of Oil-Based Drilling Fluids Using
Artificial Neural Networks. Society of Petroleum Engineers. doi:10.2118/192199-MS.

[11] Ibrahim, D., Jobson, M., Guillén-Gosálbez, G., 2017b. Optimization-Based Design of
Crude Oil Distillation Units Using Rigorous Simulation Models. Ind. Eng. Chem. Res. 56, 6728–
6740.

You might also like