Crude Fingerprinting and Predictive Analytics - GE Water & Process Technologies

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Crude fingerprinting and predictive

analytics
Predictive modelling technology has been developed to predict the
behavioural characteristics of crude oils and their blends

CONRAD TERAN
GE Water & Process Technologies

I
t is well understood that
crude oil is a complex mix-
ture of a variety of hydro-
carbons and impurities with
varied polarity and polarisabil-
ity, and is prone to cause foul-
ing,1,2 emulsion breaking and
corrosion related issues during
refinery processing. The issues
arising from the processing of
tight oils and other opportu- Figure 1 Same-day deliveries of Eagle Ford crude to a refinery
nity crudes will likely become
more challenging as new crude ficult. This task becomes near GE Water & Process
oils are brought online due to impossible if we are restricted Technologies (W&PT) has
improved and/or more aggres- to the use of traditional crude developed field methods and
sive production methods, trans- assays and other database robust analytics to predict and
port and blending strategies. methods at a time when crude respond to crude processing
The safe and profitable pro- oil names are becoming less issues, irrespective of the source
cessing of crude oil is driven and less relevant in predicting of the crude. These methods and
by the combined effects of best actual behaviour. For example, predictive analytics have been
practices, optimal process condi- Figure 1 shows samples from integrated into a cohesive pre-
tions, reliable equipment, crude seven different crude deliver- dictive modelling technology
oil behavioural characteristics ies to the same refinery on the system called CrudePLUS.
and effective chemical treat- same day, all labelled “Eagle
ment programmes. Any uncon- Ford crude”.3, 4 Key drivers of crude
trolled variation resulting from In every case, incremental processing issues
our inability to predict it, detect variability and uncertainty will Crude oils can be characterised
it or adjust it will result in one or come from the commingling by their structural fractions,
more costly processing issues. of incoming crudes with tank known as saturates (SAT), aro-
Of all these factors, predict- heels, slop and other crudes in matics (ARO), resins (RES), and
ing, managing and controlling the system. The actual charac- asphaltenes (ASP), also com-
the impact of crude oil variabil- teristics of the final charge com- monly known as SARA, which
ity on behavioural characteris- ing out of tankage are typically are determined based on their
tics and associated processing not known for certain in a time polarisability and polarity.
issues is perhaps the most dif- relevant fashion. Typical properties and nature

www.digitalrefining.com/article/1001416 PTQ Q3 2017 1


Instability/incompatibility,
Saturates
Crude oil solids and other contaminants
(Alkanes or paraffins)
Instability / magnify emulsification and
Non polar Incompatibility fouling effects. Figure 3 illus-
Linear, branched, and cyclic
saturated hydrocarbons Solids and other trates the negative impacts of
contaminants these key characteristics on a
Resins
(Maltenes) Emulsification Fouling typical crude unit train.
Second heaviest polar
tendency potential
Natural peptising agents keep
asphaltenes in solution
Saturates, resins, aromatics, asphaltenes Characterisation of crude oils
Soluble in higher MW normal and simulation of processing
alkanes (n-C7)
Insoluble in higher alkanes
issues
Oil storage
Second heaviest fraction in crude oil From a practical standpoint,
Sludging
while the quantification of
Asphaltenes Poor mixing efficiency
Pump transfer limitations SARA fractions and con-
Very high MW, polar
Polycyclic, aromatic and
Higher pressure drop taminants does provide use-
naphthenic ring compounds in ful insight on the directional
colloidal suspension
Soluble in aromatics
behavioural characteristics of
Insoluble in short paraffins (C3-C7)
Heat exchangers a fluid, these parameters by
Carry heteroatoms (S, N, O) and metals Cold fouling themselves in most cases, even
Aromatics
Loss of heat transfer efficiency
when used in conjunction with
Lower raw crude outlet temperature
Non polar Higher pressure drop other physical properties, typ-
Hydrocarbons with one or more ically do not provide a con-
aromatic rings
sistent or expeditious way of
Desalters predicting the key drivers of
Figure 2 SARA classification of crude Lower desalter temperature crude oil processing issues.
oil Poor desalter efficiency The path for the determina-
(water, salt, solids) tion of fluid behavioural char-
Tighter emulsions
of these fractions are summa- Oily brine
acteristics and the design of
rised in Figure 2. Excess water in crude effective mitigation solutions
The complex interactions Atmospheric tower overhead corrosion starts with laboratory analyti-
between SARA fractions and cal testing and simulations that
the presence of solids and characterise the crude sample
Heat exchangers
other contaminants drive a flu- and realistically stress the fluid
id’s behavioural characteristics, Fouling to emulate and quantify its
Loss of heat transfer efficiency
which in turn drive crude pro- Lower PH outlet temperature
potential processing issues.
cessing issues.5,6,7 High pressure drop The data generated from
In the context of CrudePLUS, these lab procedures and the
three crude oil behavioural associated research and devel-
characteristics are the key driv- Furnaces opment, supported by field
ers of crude processing issues: Higher firing rate pilots and validation, are the
• Instability/incompatibility: the Fouling developmental foundation of
stability of a fluid is defined Tube skin temperature increase
CrudePLUS technology.
Lower throughput
by its capacity to maintain Figure 4 illustrates analyti-
asphaltenes and colloidal parti- cal testing and application sim-
cles soluble or dispersed in the Figure 3 Key drivers of crude ulations and their relationship
bulk fluid. Conversely, insta- processing issues to specific fluid behavioural
bility refers to the capacity of characteristics.
a fluid to destabilise itself or fication at typical conditions. Crude oil characterisation
other fluids upon blending. • Fouling potential: defines the includes procedures such as
• Emulsification tendency: tendency of a fluid to foul pre- PIONA and SARA, API gravity,
defines the tendency of a fluid heat exchangers and heaters at TAN, viscosity, sulphur, metals
to emulsify or to resist demulsi- typical conditions. and other industry standards

2 PTQ Q3 2017 www.digitalrefining.com/article/1001416


Model crudes by type
Oil sample Shale, cond
26%
Blends, refcuts
10%
W&PT proprietary methods and techniques

Conventional
33%
Analytical Turbiscan Static desalter Hot liquid process
simulator simulator

Slop
Oil Instability / Emulsification Fouling 5%
characterisation Incompatibility tendency potential Bit, dilbit, synth
26%

Fluid behavioural characteristics (FBCs) Model crudes by region


USA
44%
Figure 4 Laboratory simulations of crude processing issues
AFR
7%
as well as W&PT’s proprietary W&PT initiated a project to EUR
procedures. On the simulation develop a field deployable 6%
side, W&PT proprietary meth- predictive modelling solu- ME
8%
ods and techniques are imple- tion to deal with the identified
mented utilising a combination limitations. LATAM
21%
of proprietary devices such
as the W&PT Static Desalter Defining the Xs and Ys for the
CAN
Simulator and commercially predictive modelling solution 14%
available devices such as the Any predictive modelling
Turbiscan (Formulaction, Inc.) approach requires a clear defi-
and the Hot Liquid Process nition and understanding of Figure 5 Fingerprinted sources used
Simulator (Alcor by PAC). each of the responses (the Ys) to for CrudePLUS development
Over the years, these sim- be predicted and of the param-
ulations have proven to be eters (the Xs) to be used as methods, techniques and algo-
reproducible and reliable in predictors, where each Y is a rithms were developed to pro-
emulating the processing issues function of the significant Xs cess fingerprint spectrum,
observed in the field, signif- and their transformations. extract the relevant data and
icantly contributing to the For every sample, the appli- assure repeatability and repro-
understanding of the under- cable outputs from the lab ducibility error at or below
lying mechanisms, the devel- simulations (the Ys) are trans- +/-5%. The thousands of pro-
opment of best practices and formed into indexes repre- cessed Xs become the unique
novel chemical mitigation senting the level of severity of identifiers or “fingerprints” of
solutions. the behavioural characteristics the tested oil.
(instability, emulsification and Since solids also play an
From lab simulations to fouling) of a given oil sample. important role in emulsion sta-
predictive models Other Ys targeted for prediction bility and fouling, solids con-
The time required to test crude include a selective number of tent of the oil is another set of
samples and effectively pre- physical properties (SARA frac- Xs that must be considered.
dict and respond to processing tions, API, TAN, viscosity and A new field method to quan-
issues makes laboratory simu- others). tify and classify solids in an oil
lations in most cases impracti- A field deployable infra- sample using a portable heated
cal and very time consuming red spectroscopy device was centrifuge was developed. It is
for day-to-day proactive field selected as the generator of Xs important to understand that
use. More than five years ago, for every fluid sample. Specific this is not a BS&W test. This

www.digitalrefining.com/article/1001416 PTQ Q3 2017 3


new method requires a fraction out first having to predict phys-
of the typical three hours per ical properties.
W&PT hybrid
sample that it takes to run a fil- monitoring
terable solids test. Training, validation and
Advanced
Figure 5 summarises the num- chemometrics predictive error
bers, types and sources of oils Genetic Models were trained using
that were used in the develop- algorithms
70-80% of the available exper-
ment of CrudePLUS. Machine learning
algorithms imental data for training pur-
W&PT proprietary poses and applying various
Modelling approach error-in-variable (EIV) cross validation methods on
algorithms
The under-determined nature the remaining data to test and
of the modelling problem, the tune model performance and to
complexity of the Ys to be pre- Figure 6 W&PT advanced hybrid assure the robustness (long term
dicted and the generalised modelling predictive power) of the mod-
predictive performance require- els. Prediction errors on valida-
ments presented unique and rithm, to guarantee predictive tion sets range from +/-10% to
difficult challenges to the model consistency and robustness. +/-15% for predictions of fluid
development process. SVR was coupled to proprie- behavioural characteristics and
A hybrid modelling approach tary kernel functions to deal from +/-5% to 25% for basic
combining advanced chemo- with non-linearities. physical properties (API, TAN,
metrics, machine learning, • Genetic algorithms for non- viscosity, SARA).
genetic algorithms and W&PT linear solution searches and One key aspect of
proprietary methods was archi- complex classification. CrudePLUS technology is the
tected to deal with identified • Proprietary error-in-variable inclusion of algorithms and
and latent modelling challenges (EIV) algorithms to deal with protocols for calibration and
(see Figure 6). noise in predictors (Xs) and standardisation capable of
The methods and algo- responses (Ys). detecting instrument drift, mal-
rithms included in the hybrid Figure 7 depicts the major functions and/or fluids with
approach include: steps in the modelling process. anomalous markers to prevent
• Continuum regression and The fluid fingerprint is pro- false predictions. All anomalies
dimension reduction methods cessed by the hybrid model- are analysed and acted upon
such as partial least squares ling algorithms, extracting from using rigorous protocols that
(PLS), principal component the spectrum the latent crude may ultimately result in model
regression (PCR) and correlated markers associated with the retraining.
component regression (CCR). fluid behavioural characteristics Model robustness is assured
• Support vector regression (instability, emulsification and by testing every available oil.
(SVR), a machine-learning algo- fouling) being predicted with- Models are retrained if pre-
dictions for new oils deviate
Models X’s from established error thresh-
oil fingerprints olds, if they are deemed to con-
and solids loading
tribute additional information
Latent Ys Lab Lab crude Basic Ys to improve overall prediction
simulations oil charact. errors, or if new algorithms
WPT hybrid modelling algorithms or methods are developed to
improve overall model errors.
Latent crude markers (λL) Basic crude markers (λB)
The CrudePLUS system
Predictive models f(λL) f(λB) Predictive models
CrudePLUS implementation,
Instability Emulsif. Fouling Minimise API, TAN, VIS, γSARA, others
predictive analysis, maintenance and
error
management is executed by
W&PT personnel with refinery
Figure 7 Major steps in the modelling process personnel.

4 PTQ Q3 2017 www.digitalrefining.com/article/1001416


The technology was commer-
cially launched in mid-2016
incorporating the instability/
incompatibility and the fouling
potential models, after success-
ful completion of multiple lab-
oratory and field validations.
Validation of the emulsification
tendency model was recently
completed and has now been
integrated into the CrudePLUS
system. This section includes a
detailed description of all appli- Figure 8 RIX interpretation guide
cable elements and capabilities.
CrudePLUS utilises in-the- crude names are not needed for abilise other fluids in a blend
field rapid response analytical analysis. (incompatibility), relative to a
testing via an oil fingerprint- Integral to CrudePLUS is a stable and proprietary bench-
ing and solids testing system W&PT proprietary, centrifuge mark. It classifies the tested
coupled to proprietary pre- based solids loading and clas- fluid into one of five severity
dictive modelling software, to sification field test with a typi- regions across a range from 0 to
accurately and reliably predict cal testing capacity of four fluid 10 (stable to severely unstable).
the instability/incompatibility, samples (100 mL each) per hour Severity regions provide a gen-
emulsification tendency, foul- per centrifuge. eral indication of the potential or
ing potential and basic phys- The system also incorporates preponderant locations where
ical properties of crude oils, a powerful blend simulator and processing issues may manifest
slop oils, other refinery flu- a constrained optimiser, which themselves (see Figure 8).
ids and their blends. Integral provide rapid blend assessment
to CrudePLUS are the site spe- and optimisation capabilities Crude precipitant index (CPI)
cific mitigation actions and using the actual fingerprints provides a measure of the rel-
strategies, including blend- and solids loading of the indi- ative potential precipitant
ing order (when blending two vidual fluids without the need amount upon destabilisa-
or more crudes) and, when to physically create and then tion of asphaltenes and other
deemed necessary, recom- test each of the blends being related colloids in the fluid.
mendations of crude stabi- assessed. CPI is a function of RIX and the
liser, emulsion breaker, and/ The major elements of level of available precipitants
or antifoulant dosage ranges CrudePLUS are: (asphaltenes and other col-
necessary to minimise crude loids); as such it does not have
processing issues that result in The instability/incompatibil- a defined upper limit. Two flu-
poor desalter performance, slop ity model utilises a fluid’s fin- ids with the same RIX value
generation, poor brine quality gerprint latent markers to may have different CPI values –
and fouling. determine its relative insta- in those cases, the one with the
The fingerprint of a fluid is bility index (RIX), crude pre- higher CPI is deemed the most
generated in less than five min- cipitant index (CPI), blending unstable. CPI and RIX are used
utes from a couple of drops of a order (when two or more flu- in the model to drive crude sta-
well-mixed representative sam- ids are blended), crude stabi- biliser dosage range and injec-
ple of a single fluid or blend liser dosage range and injection tion location (cargo/tank or a
to be tested. Unlike other test location. specific crude or component in
methods, the oil fingerprinter a blend).
does not require solvent dilu- Relative instability index (RIX) is
tion, which changes the native a key parameter that quantifies Blending order: the order in
properties of the oil. The his- the potential capacity of a fluid which two or more fluids are
torical characteristics based on to self-destabilise or to dest- blended has a direct impact on

www.digitalrefining.com/article/1001416 PTQ Q3 2017 5


severe). Each severity regime
provides an indication of the
potential or preponderant loca-
tions where fouling issues may
manifest themselves. An impor-
tant outcome of our research
has been the identification of
different fouling mechanisms
for fluids falling within two
specific FPX regions. As Figure
9 shows, fluids with FPX<3 fol-
Figure 9 FPX interpretation guide low conventional fouling mech-
anisms, and fluids with FPX>3
the actual degree of incompati- Emulsion breaker dosage range: follow unconventional and
bility and precipitant amount. CrudePLUS automatically more complex mechanisms.
When two or more fluids are establishes the recommended It is important to note that at
blended, CrudePLUS will pro- dosage range of a reference RIX values greater or equal to
vide a recommended order for emulsion breaker (EB) required six, unconventional and severe
blending those fluids. for mitigation. Major adjust- fouling will be preponderant
ments and product change rec- in the atmospheric heaters, par-
Crude stabiliser dosage range ommendations are consulted ticularly if tight oils are being
and injection location: crude sta- with a product applications processed.
bilisers are designed to improve specialist, taking into consid-
the stability of asphaltenes in eration all relevant parameters Antifoulant dosage range: the
the bulk hydrocarbon phase, as and unit operations. antifoulant (AF) product rec-
well as helping in the de-oiling ommended for mitigation must
of solids. W&PT’s crude stabi- The fouling potential model uti- be specific to the preponder-
lisers can prevent agglomera- lises the fluid’s RIX, CPI, solids ant fouling mechanism iden-
tion of large, relatively polar loading and other fluid mark- tified. For fluids following
molecular units that can stabi- ers generated from the finger- conventional fouling mecha-
lise desalter emulsions and foul print to determine the fouling nisms (FPX<3), CrudePLUS
heat exchangers. CrudePLUS potential index of the fluid and automatically establishes the
automatically establishes the the recommended antifoulant most appropriate AF dosage
recommended dosage range of dosage range required for mit- range, based on the most effec-
our top of the line crude stabi- igation. Poor desalter opera- tive AF product formulation.
liser and the fluid in the blend tions can result in poor solids For unconventional fouling
where it should be injected. removal efficiency and will mechanism, a customised AF
directly impact fouling poten- programme must be designed
The emulsification tendency tial. The CrudePLUS fouling by a W&PT product applica-
model utilises the fluid’s RIX, potential model has been built tions specialist, based on all rel-
CPI, solids loading and other with desalted crude. Therefore evant CrudePLUS parameters
fingerprint latent markers if the tested fluid is raw crude, and unit operations.
to determine its emulsifica- SRE is used to adjust sol- It is important to note that
tion tendency index (ETX) and ids loading levels to match severity regions and poten-
emulsion breaker dosage range. actual desalter solids removal tial and preponderant locations
efficiency. for all the models provide the
Emulsification tendency index means for pinpointing the most
(ETX) quantifies the fluid’s ten- Fouling potential index (FPX) effective chemical programme
dency to emulsify and classi- quantifies fluid fouling poten- and dosage range required to
fies it into one of five severity tial and classifies it into one mitigate the issues identified
regions, across a range from 0 of five severity regions across by the corresponding model.
to 10 (low to severe). a range from 0 to 10 (low to The location drives the type

6 PTQ Q3 2017 www.digitalrefining.com/article/1001416


CrudePLUS software Refinery site

Constrained Blending Fingerprint Oil WPT oil fingerprint


optimiser simulator library fingerprint A couple of drops,
<5 min per sample

WPT oil
Latent crude Processing Basic crude Workstation fingerprinter
markers algorithms markers

WPT
WPT solids test
Instability Emulsification Fouling Physical personnel 100 ml fluid,
model tendency model potential model properties model 4 samples/cent/hr
(RIX) (ETX) (FPX) (PPROPS)
Index severity Index severity Index severity API
Problem location Problem location Problem location TAN
WPT
Blending order EB dosage range AF dosage range VIS solids sample
CS dosage range YSARA 100 ml fluid,
4 samples/cent/hr
Injection location YCII
Others
Predictive analytics

Solids loading

Value calculator, process monitoring and modelling tools

Figure 10 CrudePLUS system

of chemistry and the severity mitigation of crude oil process- to confirm BCOP and support/
drives the dosage range. ing issues. validate the effect of mitigation
The implementation can typ- actions.
CrudePLUS system integration: ically be completed in two to
As Figure 10 shows, the predic- three months and includes the Value generation
tive modelling analytics have following activities: While the paths, phases and
been bundled into a software • Tank farm mapping and strategies to ultimately mitigate
package at the centre of a port- understanding of blending and capture maximum value
able and robust system, which operations and practices from all identified opportuni-
includes the oil fingerprinter, • Crude unit survey and best ties will vary from refinery to
the heated centrifuge and other practices analysis refinery, there are three possi-
support tools, with W&PT • Best practices gap reduc- ble, not mutually exclusive sce-
personnel executing imple- tion and mitigation actions narios under which a refiner
mentation, analysis, system recommendations may decide to use CrudePLUS
maintenance, day-to-day service • Establish sampling plan (loca- to maximise value generation.
and management, process mod- tion, frequency, and so on) Working towards and attain-
elling and expert support. • Commissioning test equip- ing the point of maximum
ment (fingerprinter, heated cen- value requires the engagement
CrudePLUS system trifuge, PC, and so on) and data and support of other refinery
implementation management/communication groups beyond operations and
includes a series of orchestrated tools process engineering, such as
activities designed to effectively • Standard reports setup economics planning and sched-
deploy the technology and ulti- • Initial analysis and value map uling and crude purchasing.
mately maximise value gener- to estimate the baseline cost of
ation opportunities that result processing issues (BCOP) Scenario 1: reactive mode
from the continuous detection, • Empirical process modelling Under this scenario, CrudePLUS
quantification and cost effective (preheat, furnaces, for instance) is used as a monitoring (pro-

www.digitalrefining.com/article/1001416 PTQ Q3 2017 7


cessing issue detection) tool and other recommendations Case studies
and advisor. The value gen- – such as blending, segrega- The case studies presented here
erated from this mode can tion, rerouting of streams, best present experiences and find-
only be realised if specific practices, and so on – that will ings gathered from the various
actions are taken to mitigate incrementally minimise the pilots that ran over periods of
the processing issues identi- overall cost of crude processing time ranging from six months
fied by CrudePLUS and control issues as a first step, followed to one year and other ongo-
chemical injection (if any) based by cost effective chemical miti- ing applications. In every case,
on the model’s feed-forward gation when needed. a BCOP was quantified using
recommendations and process CrudePLUS tools, actual W&PT
feedback. Scenario 3: strategic mode process service experience, and
Under this scenario, CrudePLUS direct interaction and feedback
Scenario 2: proactive mode is used as a crude purchas- from operations, process and
In this scenario, the refinery uses ing decision enhancement tool. crude planning at each of the
CrudePLUS as a monitoring, While this scenario may not be sites.
simulation, optimisation, advice easily attainable by many refin- Specific mitigation strate-
and control tool. The blending ers, it is the one with the highest gies and projected reductions
plan is simulated in advance by value generation opportunity, in BCOP were made based on
testing actual tank or crude sam- as it tackles the potential pro- the learnings from CrudePLUS
ples in the plan prior to actual cessing issues associated with implementation, expert analy-
scheduling. The blending plan specific opportunity crudes or sis and site specific refinery eco-
is then simulated using the blends prior to actual crude pur- nomics. In sites where specific
blend simulator to assess mul- chase. Many dumb-bell crudes mitigation recommendations
tiple blend scenarios using the are created and sold matching were executed, empirical pro-
saved test results of the individ- the API, TAN and sulphur spec- cess models were built and val-
ual crudes in the blend in con- ifications of an actual crude oil idated using historical process
junction with the constrained and will eventually end up in data to normalise for process
optimiser to find the blending the refinery crude system under conditions variability and to
combination that in combina- that name, but in many cases categorically quantify, prove or
tion with chemical mitigation with significantly different yield disprove the effects of executed
(if needed) minimises process- and fluid behaviour characteris- mitigation actions.
ing issues, subject to a set of con- tics, among other issues, making
straints provided and adjusted the names of the crudes in these Case study 1
by the crude planner. cases totally useless for predic- A North American refinery
The objective here is not to tive purposes. processing a complex blend of
provide a trivial solution by rec- Having the capability to rate crude oils and purchased inter-
ommending exclusion of the and discriminate crude oils mediate oils was experiencing
most problematic crudes from using CrudePLUS provides the significant processing issues
the blend, nor is it to automat- opportunity to enhance crude due to crude hot preheat and
ically attempt to solve all pro- purchasing decisions that can coker heater fouling. The diet
cessing issues by chemical positively affect the bottom to the crude train included for-
mitigation. The objectives are: to line (for example, driving dis- eign and domestic heavy sour
inform the refiner ahead of time counts). Of course, deciding to crudes and slop. The diet to the
of potential processing issues purchase, process and mitigate coker train included foreign
associated with a blend in the a crude deemed problematic heavy sour crudes, purchased
plan; to recommend viable miti- by CrudePLUS will continue to resids, decant oil, heavy cycle
gation alternatives that will ulti- be a viable option, but one that oils, and pipeline transmix.
mately minimise the overall cost if taken will now have defined Following CrudePLUS imple-
of processing; and to monitor ahead of time the potential mentation protocols, the opera-
advice and control. incremental processing issues tion was baselined to quantify
Any viable mitigation will and costs associated with that crude processing issues, fol-
include adjustments, changes decision. lowed by fingerprinting of all

8 PTQ Q3 2017 www.digitalrefining.com/article/1001416


Figure 11 Summary of instability/incompatibility model results

streams and tanks over a period Working with crude plan- Case study 2
of several months. BCOP ning and operations, and sup- A North American refin-
revealed significant annual pro- ported by CrudePLUS tests and ery operating two large crude
cessing costs totalling around simulations, the recommenda- units was experiencing signif-
$25 million. Of this total, 77% tion was to move the transmix icant economic penalties due
was attributable to crude pre- to the crude train and the move to critical heat transfer equip-
heat and coker heater fouling. was made immediately after. In ment fouling in one of the
The balance was attributed to addition, several other mitiga- units. Preheat train and atmos-
slop generation reprocessing, tion strategies were proposed, pheric heater fouling in Unit A
90% of which was coming from targeting net savings in pro- was costing approximately $30
coker operations. cessing costs of $10 million/y. million/y.
Although crude preheat foul- To capture the impact of this The refinery processes a
ing was very significant and and other recommendations, variable mix of West Texas
over time a throughput lim- crude preheat and coker heater Intermediate, West Texas Sour,
iting issue, the biggest chal- empirical process models were tight oils, Canadian crudes,
lenge had been trying to built using actual process previ- four different offshore crudes,
identify suspected incompat- ous and current runs data. and slop oil. At the tank farm,
ibility issues that were caus- The change increased coker crudes are blended and segre-
ing the coker heater skin heater run length by 40 days, a gated by sulphur content into
temperatures to episodically statistically significant increase, three different groups of tanks
and uncontrollably increase, without a statistically signifi- designated as light, medium
causing on the average three cant impact on crude preheat and high sulphur.
shutdowns per year to decoke fouling rate. The increase in Unit A processes predomi-
the heater. Increasing coker run coker run length resulted in a nantly light sulphur oils, with
lengths was the main priority. net reduction in the cost of pro- the remainder of its diet from
Predictive analytics clearly cessing of nearly $1.4 million/y the medium sulphur tank.
indicated that the purchased and without additional chemis- These two feeds are blended
pipeline transmix was the most try. Incremental improvements to meet specified sulphur and
unstable stream. Figure 11 sum- in run length would require gravity targets for the unit feed.
marises the instability/incom- implementation of additional Tight oils represent 27-35% of
patibility potential of refinery and more complex strategies the total charge to Unit A.
streams and exploratory blends focused on operational prac- In contrast, Unit B processes
of transmix with various tices and process realignments, mostly medium sulphur crude
streams. as well as chemical mitigation. oil, supplemented with a nota-

www.digitalrefining.com/article/1001416 PTQ Q3 2017 9


Figure 12 Unit B relative instability and fouling potential indexes

ble amount of high sulphur each of the units. Following Figure 13 shows the analytics
crude oil from western Canada. CrudePLUS protocols, each for Unit A crude charge sam-
CrudePLUS was used to tank and total charge to each ples, indicating high levels of
measure the impact dissimilar unit was fingerprinted over instability and fouling poten-
crude slates impart on each unit several months. tial, consistent with actual unit
and determine potential issues Predictive analytics for Unit performance (high to severe
of various blends. Having two B crude charge samples (see fouling in preheat and heaters).
units with dissimilar process- Figure 12) indicated low lev- Recommended mitigation
ing issues also allowed us to els of instability and fouling, strategies included changes to
determine, over a period of sev- matching the actual trouble free blend order, blend ratio adjust-
eral months, if the technology performance of Unit B. The rec- ments and chemical treatment,
was capable of correctly pre- ommendation for this unit was and projected a net reduc-
dicting the processing issues to make no changes and con- tion in processing costs of $12
and their magnitude in tinue monitoring. million/y.

Figure 13 Unit A relative instability and fouling potential indexes

10 PTQ Q3 2017 www.digitalrefining.com/article/1001416


Case study 3
A US refinery processing a
variable crude diet was expe-
riencing hot preheat heat trans-
fer losses, which resulted in
crude throughput limitations
and crude throughput loss
while cleaning exchangers. The
crude diet was primarily com-
posed of Alaska North Slope
and Bakken, in addition to vari-
ous concentrations of Canadian Figure 14 Average crude slates and CrudePLUS analytics during a run
crudes (Cold Lake, Alberta
Western Blend, Lloyd Kerrobert ant CrudePLUS analytics for mean, upper and lower statis-
and Bow South). three consecutive periods dur- tical process control limits for
The refinery expressed a need ing a run. the untreated system. As can
to better manage the crude To determine the effective- be seen, the corrected U (rep-
fouling impact with both the ness of the treatment pro- resented by the purple trend)
existing and future planned grammes, a heat transfer starts to move out of control (a
crude purchasing regime. The coefficient model was built good thing in this case) shortly
goal would be to minimise the around a critical desalted after the antifoulant was intro-
risk of not capturing the gain in crude/vacuum bottoms set duced into the system and con-
profitability. of bundles, to track the per- tinues almost unchanged after
Following CrudePLUS pro- formance over time, incor- stabiliser treatment was initi-
tocols, testing was performed porating all critical predictor ated, clearly and statistically
on the component crudes and variables affecting heat transfer demonstrating the efficacy
crude blends at the refinery. performance. of the programs, even under
The ultimate mitigation strat- The cumulative throughput increasing instability and foul-
egy implemented was the addi- reduction due to heat transfer ing conditions.
tion of an antifoulant to the losses was estimated at $11.6 The simplest measure of suc-
crude unit preheat train and million/y. Figure 15 shows a cess was the reduction in the
treatment of problematic crude statistical control chart depict- frequency of bundle cleanings,
blends with crude stabiliser at ing the behaviour of the U which went from six cleanings
the tank farm. Figure 14 under a normalised set of oper- to zero between turnarounds.
summarises the aver- ating conditions. The three lines The actual net reduction in
age crude slates and result- sloping down represent the refinery processing costs was

Figure 15 Heat transfer U coefficient model

www.digitalrefining.com/article/1001416 PTQ Q3 2017 10


estimated at $5.7 million/y. their blends. Blending order of challenges of tight/shale oil refining,
components in a blend, as well Processing Shale Feedstocks, 2014,
Conclusions as crude stabiliser dosage range 4 Dion M, Challenges and Solutions for
W&PTs’ predictive modelling and injection location, emulsion Processing Opportunity Crudes, AM-14-
13, AFPM Annual Meeting, 2014.
technology, CrudePLUS, has breaker and antifoulant dos-
5 Park S J, Mansoori G A, Energy Sources,
been developed to reliably pre- age ranges are automatically
1988, 10, 109.
dict the behavioural charac- generated based on predictive 6 Speight J, Ch 1: Composition of
teristics of crude oils and their analytics. Petroleum in Petroleum Chemistry and
blends, their potential impact Refining, Taylor & Francis,1998.
on the process and recommend CrudePLUS is a mark of General Electric 7 Carbognani L, Izquierdo A, Asphaltenes
the viable alternatives availa- Company. and Asphalts in Dev. Petrol. Sci., 4 (Part
ble for mitigation. The technol- B), 2000, 335–362, 2.
ogy is totally agnostic to crude Acknowledgments Conrad K Teran is a Senior Technical
name, type or source, utilis- I would like to thank the team working in Manager of Refining Technology with
ing actual fluid analytics to the development, validation and support GE Water & Process Technologies, The
drive model predictions, with- of the predictive modelling technology Woodlands, Texas. He holds a BS degree
out using any type of histori- presented in this article and those who in chemical engineering from Brigham
cal, experiential or crude assay have contributed to the completion of Young University, a master of engineering
database. the article: Nimeshkumar Patel, Naveen in chemical engineering from Lamar
CrudePLUS is at the centre Agrawal, Vijaysai P, Manish Joshi, University and is a Registered Professional
of a service providing a porta- Sudhanshu Kashyap, Greg Yu, Karl Engineer in the State of Texas.
Kuklenz, Kurt Ginsel, Gabriel Castillo,
ble solution that utilises in the
Mike Dion, Jeff Zurlo, Steve Bakas.
field rapid response analytical
testing via an oil fingerprinting References
and solids testing system cou-
1 Watkinson A P, Wilson D I, Chemical LINKS
reaction fouling: a review, Experimental
pled to proprietary predictive
Thermal and Fluid Science, 1997, 14,
modelling software, to predict 361–374.
More articles from the following
categories:
the instability/incompatibility, 2 Watkinson A P, Deposition from crude Corrosion and Fouling Control
emulsification tendency, foul- oils in heat exchangers, Heat Transfer Heavy and Sour Feedstocks
ing potential and basic physi- Engineering, 2007, 28, 177-184, doi: Instrumentation, Automation and
cal properties of crude oils, slop 10.1080/01457630601064413. Process Control
oils, other refinery fluids and 3 Benoit B, Zurlo J, Overcoming the

12 PTQ Q3 2017 www.digitalrefining.com/article/1001416

You might also like