对不起!您没有登录，请先登录论坛.
Copyright 1999, SPE/IADC Drilling Conference
This paper was prepared for presentation at the 1999 SPE/IADC Drilling Conference held in
Amsterdam, Holland, 9–11 March 1999.
This paper was selected for presentation by an SPE/IADC Program Committee followingreview of information contained in an abstract submitted by the author(s). Contents of thepaper, as presented, have not been reviewed by the Society of Petroleum Engineers or theInternational Association of Drilling Contractors and are subject to correction by the author(s).The material, as presented, does not necessarily reflect any position of the SPE or IADC, theirofficers, or members. Papers presented at the SPE/IADC meetings are subject to publicationreview by Editorial Committees of the SPE and IADC. Electronic reproduction, distribution, orstorage of any part of this paper for commercial purposes without the written consent of the
Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted toan abstract of not more than 300 words; illustrations may not be copied. The abstract mustcontain conspicuous acknowledgment of where and by whom the paper was presented. WriteLibrarian, SPE, P.O. Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
Abstract
Future management of drilling operations will face new
hurdles to reduce overall costs, increase performance; and to
do this with possibly less retained experience and
organizational drilling capacity.
Over the past fifteen years huge drilling data
accumulations evolved as part of the drilling process.
Although unused these inert data accumulations offer the new
possibility to create what will be called a Virtual Experience
Simulation (VES) for drilling.
This paper presents a new concept based on heuristics.
Unused data accumulations are processed (or activated) using
a technique cited in the paper. An example, using a 22 well
data accumulation is processed into specific data sets for
geology, tripping, cementing, logging, penetration rate, and
unscheduled events. These new data sets now have a value
added by retaining the field drilling experience and
knowledge.
The last part of the paper explains one type of heuristic
computer simulation engine (using system dynamics) that
gives the user access to this virtual drilling experience and
knowledge.
It is believed the work in this paper offers evidence that the
VES is a new way to retain field drilling experience in a way it
can be learned by others as virtual experience.
Introduction
The fact is: as long as there is an economic motive to explore
and produce oil and gas, drilling costs will be one of the
critical factors that dictate the bottom line economics.
Historically, much of the drilling was driven by
exploration, especially offshore. As companies reported poor
exploration successes over the past decade and moderate
return on investments for development projects a new
industrial dynamic started. This new dynamic will reshape the
oil business possibly as much as the break up of the Standard
Oil monopoly in the early part of the Twentieth Century.
What appears to be emerging is an A&P industry (i.e.
Acquisitions & Production) rather than E&P industry
(Exploration and Production). This doesn’t mean that
exploration will cease but it will not be the main factor for
most upstream oil and gas companies.
The implications of this new industrial direction will have
enormous impacts on all segments of the upstream oil and gas
industry, including drilling.
Historically, company stability and long term ownership of
oil and gas properties had a major influence on how
organizations did their drilling. Whether the drilling
organization was centralized, or decentralized it had one
commonality, retention of how to plan and drill wells for its
organization. This knowledge and experience was retained by
the drilling personnel and management, itself.
During the early 1990’s many companies started to adopt
the concept of outsourcing some of its core needs such as
drilling. This was the beginning of the end of the classical
company experience to have a mastery for drilling its wells.
No longer was the knowledge resident of how to drill a certain
geologic area processed by a certain group.
Later in the 90’s some companies recognized this shift and
its impact and started to take steps to rebuild this deficiency of
experience and capacity. However, because of the oil price
drop of 1998, many companies were forced to merge or be
taken over. The outcome of this restructuring is a massive loss
of personnel both from the oil contractors and service
companies. Again, a major sector of industrial experience will
disappear, including drilling.
The question that begs answering is: how will the drilling
industry cope with another level of lower oil prices, handle
what appears to be another loss of experienced personnel and
the buying and selling of oil and gas properties at an ever
increasing rate?
What if there was a way to capture the knowledge and
experience of drilling a specific field, box it up, and then a
person interested in drilling this field could somehow plug into
this box and learn all this knowledge and experience? At the
present time there is no such way or no such black box.
SPE/IADC 52803
Virtual Experience Simulation for Drilling - The Concept
Keith K. Millheim/University of Oklahoma, USA; Thomas Gaebler/University of Leoben, Austria
2 KEITH K. MILLHEIM, THOMAS GAEBLER SPE/IADC 52803
However, the technology of Heuristics offers the next best
thing: a way to retain knowledge and experience that is
specific to a certain geological and geographical area.
Heuristics also permits the recounting of this knowledge and
experience via what will be called a Virtual Experience
Simulator (VES)
Both heuristics and simulators for drilling are not new.
Back in 1969 Robert Meier et al. wrote a book on the
Simulation in Business and Economics, including one chapter
on Heuristic methods1. This chapter cited the evolving
paradigm of heuristics from the time of Pappus (300 A.D.)
who offered heuristics as the art of solving problems. As
technologies evolved so did the concept of heuristics. Around
the 50’s Thomas Kuehn and Michael Hamburger introduced a
more modern definition:
"We prefer to look at heuristic programming as an
approach to problem solving where the emphasis is on
working towards optimum solution procedures rather than
optimum solutions."
Even though Kuehn and Hamburger successfully proved their
assertion by writing one of the first heuristic simulations: A
Heuristic Program for locating Warehouses1 , heuristic
modeling like artificial intelligence methods never really
caught on. Industry instead, gravitated to the more analytical
approaches for building simulators, especially for drilling.
Simulation and simulators in drilling have received little
attention and importance. Only the well control simulators
have been used with some frequency, and primarily because
most countries require some type of certified well control
training. Other attempts to promote drilling simulation met
with little favor, except for the training of neophytes for
drilling. See references No. 2 to 4 for background on drilling
simulators.
The main reason for this lack of interest is simple.
Experienced drilling personnel found little value added in
using a generic simulation that was either too simple or did not
reflect the real world problems they faced in their own
operations. Also, up until the last few years the computer
capacities needed for advanced simulations, were confined to
the high end Unix based platforms and workstations. This was
okay for the geologist and the reservoir engineer since they
needed to handle large sets of data and sophisticated models
with up to a million grid points. The drilling person on the
other hand, needed portability, simplicity, and speed. Also, the
drilling person did not work with large data sets.
Kuehn and Hamburger’s critical paradigm of heuristics
(Fig. 1) suggests a model of the simulation of Human thought,
Artificial Intelligence, and Heuristic Problem Solving. On its
own this paradigm has limited value as was demonstrated in
the 70’s and 80’s where artificial intelligence methods were
rejected as unrealistic and in many cases, too simple.
Reservoir engineering simulation managed to survive and
grow because the actual simulation had no meaning until the
simulated reservoir was developed from measured data. The
results were then matched with field performance. In essence
what gave significance to reservoir simulation was not the
equations or equation solvers but the data that supported the
uniqueness of the simulation. As 3-D seismic results were
added as a more detailed depiction of the reservoir (the model)
became part of the retention of the reservoir knowledge for a
given field.
Drilling never realized this use of data. There were always
enough experienced people with local drilling knowledge and
experience to essentially function within Kuehn’s and
Hamburger’s heuristic’s paradigm to manifest the planning
and real time drilling of the wells without much data support.
However, because detailed drilling results were kept for most
wells that were drilled (especially from 1980 until the present)
a large data accumulation developed.
It is this data accumulation that allows the concept of
heuristic simulation or a “Virtual Experience Simulation” to
be developed and used. Fig 2 adds the vision of the inert data
accumulation operated on by some method to convert this data
accumulation into retained knowledge and potential learning.
One of the common misconceptions was the idea: "we
have a data base hence the data has value." The recognition
that data is inert until activated is slowly being realized in
other sectors like drilling and production. Reservoir simulation
developed this implicit insight because of the need (value) of
the data for the required end result. Now, the drilling industry
is faced with another challenge: improve drilling performance
and costs without the benefit of localized drilling experience.
This provides the value added driver to convert the inert
drilling data accumulations into activated data sets that retain
knowledge and experience.
The work cited in this paper shows one way to activate an
inert drilling data accumulation for 22 wells to where it
manifests most of the knowledge and experience for drilling
the wells. This is explained in the next section.
The last part of the work then shows the activated data sets
can be operated on by an user via what will be called a Virtual
Experience Simulator (VES). In this work the computer engine
for the VES is a system dynamics model with a special
ergometric design for the 22 well field.
The process of activating an inert drilling data
accumulation to an activated data set.
Various behaviors, events, and situations occur throughout
drilling a sequence of wells. To take advantage of these
“lessons learned” they have to be recognized and kept for
appropriate application. Usually, this knowledge is
accumulated by different individuals, without the use of this
information or “lessons learned” for the rest of the
organization. This is why companies lose performance if the
personnel for one reason or the other leave the organization.
The challenge is to make this job-specific knowledge available
throughout the company to offer a platform where increased
drilling capacity can be obtained. Heuristic simulation is seen
to be the tool to store organizational capacity in a way others
can access it.
SPE/IADC 52803 VIRTUAL EXPERIENCE SIMULATION FOR DRILLING - THE CONCEPT 3
The accumulated drilling data (considered inert or
unusable) could be the daily drilling wires, bit and mud
reports, AFE’s, time vs. depth, electric logs, mud logger data,
well plans, lessons learned reports, and end of well reports.
The following sections discuss how a specific data
accumulation for a sequence of 22 wells was transformed into
a heuristic data set (activated data set) that contains the virtual
experience for drilling these wells.
Description of Data and Data Associated Problems
All data evaluated and processed was taken from 22 wells
drilled for an oil company in one field. Out of these wells four
wells, three deviated wells and one re-entry well, with
different operating procedures are excluded from further
discussion. While 22 wells drilled in one field may seem to be
a substantial number in the oil business, in statistics this
number is relatively small. This fact forced the utilization of
additional data wherever it was possible and legitimate.
Since the location of the field and the wells have to remain
confidential a detailed description of the evaluated subject
wells is not possible. The wells are therefore put in a sequence
from 1 to 22, though not in historical order. They were drilled
between 1988 and 1997 (Fig. 3).
Seventeen wells were drilled with contractor “A”, two
wells with contractor “B”, and one well with contractors “C”
to “F”, thus providing insufficient data to compare the
contractor’s performance.
Almost all the data included in the computer model was
taken from the morning reports, either directly, or where
possible, from the electronically stored database. Morning
reports were generally filled out each day by the tool-pusher.
They summarize all data and actions gathered in a twenty-four
hour period. Typical entries are the depth, the footage drilled,
bit type, number and size, the weight-on-bit, the revolutions of
the drillstring, pump pressures, mud volumes, properties and
materials, including the operations performed.
It is the structure of the morning reports which is
responsible for part of the inaccuracy. The minimum reporting
interval for the different actions is for a half an hour of
operations. This time sometimes does not reflect the actual
event time. That means, that procedures or operations done,
which took the rig crew only a few minutes, are either not
resolved at all, or overemphasized in the reports, thus lost for
an detailed analysis. This is the main hurdle an analyst has to
face when evaluating morning reports.
Further data was extracted by using master logs (provided
by the logging company), bit records (by the drilling
contractor), weekly lithology report (by the geologist), and
final reports (by the drilling supervisor).
Geology
Although the overall data analysis includes only 18 vertical
wells drilled in the specific field, three deviated wells are also
included wherever possible, e.g. in the geological analysis. As
the spatial distribution of the well is neglected, due to a very
homogenous lithology, the use of the deviated wells increases
the size of the sample. Consultation with the current drilling
supervisor in charge exposed the hypothesis, that the lately
occurred improvement of drilling performance might have
been experienced, due to the west-to-east movement. Yet this
could not be proved to be due to geological differences.
The sequence of geological layers that have been drilled
through in the respective field consists of maximum 19 layers
drilled to a total depth of approximately 12,500 ft. Most of the
wells reached layer 19. It can be seen, that all wells show the
same sequence of layers with no faults or missing layers. It is
therefore most likely, that future drilling activities in this field
will not change the picture.
For a statistical distribution of the formation sequence the
layer thickness is the only necessary distinction. Due to
proprietary aspects, the layers will be numbered from top to
reservoir bearing layers from layer 7 to layer 19. The other
layers were omitted:
Layer 1-6 were skipped because the simulation starts with
drilling after the 13 3/8 in. surface pipe has been set, which is
in layer 7.
Layer 19 is the lowest reservoir bearing layer, and drilling
was stopped before drilling the entire layer. A computation of
the layer thickness was therefore not possible.
From the analysis of the layers a histogram of layer
thickness for layer 15 was developed. See Fig. 4.
It should be noted that this is the unique outcome for the
particular geographical and geological feature to this VESD
and that in another area a geology would yield different results
and significance levels, thus making a new evaluation
necessary.
Tripping
In generic drilling simulators the calculation of the tripping
rates is usually done by a constant factor for running in the
hole and pulling the drillstring out of the hole5. Based on the
available data of the 18 wells under investigation a different
approach was chosen by calculating the tripping rates as a
function of depth. Thus, all tripping times were collected,
statistically evaluated, and two tripping functions computed to
be implemented into the drilling model.
A major problem was encountered during the data
analysis: For most of the tripping processes the tool handling
prior or after tripping is not explicitly stated in the morning
reports, and therefore the duration of a particular action was
questionable. To illustrate the problems mentioned above,
three entries out of the morning reports are compared with the
corresponding entries in the database.
A regular tripping tour for changing a dull bit is recorded
in the morning report of well No. 3 at 8,261 ft given by Table
1. Which resulted in an according database entry given by
Table 2.
Although a coarse example, it illustrates the problems the
authors were confronted with similar occurrences throughout
the data analysis. (1) In this case the operational code is not
analyzed correctly by the data entry person. Out of four hours
the laying down the shock sub only a half an hour of the total
4 KEITH K. MILLHEIM, THOMAS GAEBLER SPE/IADC 52803
period was used. Therefore the operational code for tripping
out would have been a better choice. (2) The make up of the
new bit is included in the running in the hole. (3) Additionally
simple input errors occur. In this example, the time spent on
blowout preventer (BOP) and valve testing is wrongly
digitized, by using four instead of two hours for BOP testing.
To generate the tripping times as function of total depth
drilled, all available tripping data was collected and
subsequently sorted in increasing order. With this data two
scattered plots, one for tripping in and one for tripping out of
the hole, were generated with the help of the statistical
evaluation software “Origin 5.0”6. Based on these plots, a
second order polynomial curve fitting calculation was
performed for each data set. The fitting function is given by
Eq. 1.
2
TripRate = A + B1 * Depth + B2 * Depth .......................(1)
and resulted in the parameters listed in Table 3.
By use of these parameters in Eq. 1 the tripping rates are
calculated in feet per hour. This is shown by Fig. 5.
Comparing the tripping curves in Fig. 5 it can be seen, that
tripping into the hole takes longer than tripping out of the hole.
This is in contrast to what is measured during a typical roundtrip
in other areas. Smaller tripping-in rates therefore have to
be seen as a local peculiarity experienced in the field under
investigation, and will probably be different at other locations.
Phenomena like this prove the need of heuristically simulated
field experience in order to capture location-related
characteristics.
Logging, Setting Casing, and Cementing
Opposite to defining the different tripping rates, where the
available data generated difficulties, the challenge for
analyzing the casing setting process and subsequent cementing
was the procedure itself. By comparing all 18 wells a generic
casing setting procedure had to be determined.
Based on this procedure the means and standard deviations
of each task, i.e. reaming, logging, rigging up the casing tongs,
running casing, cementing, etc. were calculated and
implemented into the activated data set. Again, this approach
differs from generic drilling simulators, where, for example,
the casing running time and waiting on cement (WOC) are
constant factors5 (e.g. 30min/100ft for running in the casing
and 24 hours for WOC).
As all eighteen wells are drilled in one particular field with
very homogenous lithology the casing program is the same for
all wells.
Rate of Penetration
If there are no major complications, the actual on bottom
drilling process makes up about fifty percent of time spent on
this geographical location. (Table 4). Therefore, the rate of
penetration has a significant impact on the overall time for
drilling and completing a well.
The most important factors affecting the penetration rate
(ROP) have been identified and studied. They are the bit type,
the formation characteristics, drilling fluid properties, and
operating conditions. A considerable amount of research has
been accomplished to study these effects on the rate of
penetration. In most of the experimental work, however, the
effect of a single variable was studied while holding all other
influencing variables constant. The references No. 7 to 12
explain why no explicit penetration rate models are possible.
Therefore, another approach was taken using the accumulation
of drilling data for the 22 wells.
The basic principles are to take an isolated look at each
individual layer and to keep the factors, which influence ROP
due to lithology, depth, hydraulics, etc. constant. The first step
in the statistical evaluation was the digitalization of the data
according to the specific layers taken from the masterlogs. The
masterlogs were recorded by a service company and contain
information, like mud properties, ROP (given in minutes
needed to drill 5 ft), lithology, hydrocarbon gas content, a
geological description. Out of 18 vertical wells, only 12
different masterlogs were available to the authors. However,
this yielded more than 30,000 data points for the ROP study.
Prior taking any operating conditions into consideration an
evaluation of the overall drilling performance, neglecting all
parameters affecting ROP, had to be done because (1) many
variables are not known, e.g. rock hardness, or rock
abrasiveness, (2) some variables do not offer enough data for a
statistical meaningful analysis, e.g. mud properties, pump
pressures, and (3) the combination of variables is in many
cases too complex to give reliable results, or are not
measurable like dynamic bit wear. It was assumed, that with a
given equipment and a particular experience, the rig crew
always tried to accomplish the best performance within the
observable data and the knowledge (learning) for various bit
performances (trial and error).
Generic drilling simulators, must account for the effects of,
e.g. bit selection, bit wear, hydraulics, mud weight, rock
strength, etc. to compute the rate of penetration. The heuristic
concept allows all the assumptions. This can be done because
heuristics reflect the level of decision making for running bits
and controlling performance which incorporates these
variable.
After evaluating the overall drilling performance it is
necessary to predict the influence of the operating variables
WOB and RPM on the drilling performance. This will give the
user of the drilling simulator the opportunity to set his/her
decisions and compare different effects. One way to do this is
by a technique the authors call “Isomeric Mapping”.
Based on the need to give the user of the drilling simulator
the flexibility of choosing WOB and RPM the idea of
interpreting the ROP as heights in topographical maps was
formulated. This was never been done before.
In the following the software program “Surfer V6.02”13,
was used to generate the isomeric maps and the threedimensional
models. Prior to the map generation the data used
had to be slightly modified. Due to obvious reasons no data is
available with WOB=0. Therefore, to establish the WOB=0
boundary at the x-axis zero values for the WOB/RPM data
pairs were added to the existing data sets.
SPE/IADC 52803 VIRTUAL EXPERIENCE SIMULATION FOR DRILLING - THE CONCEPT 5
Using the data “Surfer” computes following isomeric two
dimensional map (Fig. 6). Regions clearly showing better
performance than their surroundings are highlighted, and
ranked as well.
Region No. 1, respectively the WOB/RPM combination,
apparently shows a much better performance at 17,5klbs WOB
and 65 RPM. Additionally three other areas (#2, #3, and #4)
with drilling performance above the average can be identified.
Areas that were not covered by data before have been
successfully interpolated. Transforming the isomeric map into
a three-dimensional surface (Fig. 7) these areas can be seen
even better. The fall off of the performance towards higher
WOB also can be seen better.
Again it should be emphasized, that Fig. 6 and Fig. 7 are
solely valid for layer 15. The isomeric maps of all other layers,
make up the complete activated penetration rate data set.
Unscheduled Events – Trouble Time
It is outlined in the previous chapters that the statistical
approach of a data-based drilling module strongly relies on
probabilities of different actions rather than on rigid
deterministic algorithms used in conventional simulators.
Whereas regular drilling actions, like tripping, change mud,
drilling, etc. can be simulated by using empirical relationships,
the time spent on unscheduled events is subjected to a
statistical evaluation. Such an evaluation has two steps: (1)
description of major unforeseen events, and (2) adding these
probability-based events to the drilling model will result in
different “virtual” wells when simulated. This concept
therefore makes it necessary to evaluate what happened during
drilling the data-set wells, and what are their associated
probabilities.
The description of unscheduled events can probably unveil
best the differences of a heuristic drilling simulator in
comparison to a generic simulator for drilling. While a generic
simulator will, most likely, not have built in any unscheduled
events because of lack of data for such a calculation, a
heuristic simulator will strongly rely on exactly these data.
And whereas generic simulators are best used for teaching
certain operations encountered during drilling a well, heuristic
simulators should be used to capture local circumstances and
peculiarities. Thus, facilitating an organization to memorize
and present the “lessons learned” throughout the company.
During drilling the 18 wells the drilling company
experienced severe fluid losses in almost every well. In 6 wells
they encountered some well control problems. Hence special
attention was given during the development of VESD for these
two occurrences.
At some particular depths the drilling contractors had to
fight severe drilling fluid losses due to fractured rocks. These
losses varied between only a few barrels per hour (bbl/h) to
total losses with no mud returns. The required responses for
the occurrences cause different actions. To cure the losses in
some cases the rig crew mixed and pumped additional mud, in
other situations they continued drilling, replacing the mud with
water. If both of these two measures did not prevent further
losses, one or more cement plugs were cemented to restore a
trouble free mud cycle.
In order to get a statistical meaningful set of data, which
are an input prerequisite for the subsequent simulation, the
data taken from the morning reports had to be processed. By
doing this, a distinction could be made whether the cement
plug technique worked or not.
Other events like stuck pipe, well control, fishing jobs, and
rig repairs were converted into probabilistic “activated data
sets” from the data occurring in the records. Note, that again
these data are unique and not generic. But it is this uniqueness
that causes the once inert accumulated drilling data to become
activated useful retention of experience ready to be accessed
by an interested user.
The next section presents the way the activated data sets
for geological layers, tripping, logging, running casing, rate of
penetration, and unscheduled events can be accessed as virtual
experience and learned by a user.
The Development of the Virtual Experience
Computer Model
Based on the activated data sets a computer model is
developed. The purpose of this “heuristic engine” is to present
the user an interactive environment to gain insights of a certain
domain, and test different scenarios. The purpose built
software PowerSim Constructor 2.5d14 was used to develop
the heuristic engine to transfer the knowledge stored in the
activated data sets. The PowerSim software is a system
dynamics modeling language, and provides an easy-to-use
programming environment.
System dynamics is concerned with creating models or
representations of real world systems of all kinds and studying
their behavior. The method provides a distinctive set of easily
usable tools (centered on a generic set of building blocks). The
formulation of these universal tools makes system dynamics
widely applicable. See references Nos. 15 to 19 for suggested
reading.
The Drilling Model – Generic Part
The design steps and design considerations throughout the
development of the Virtual Experience Simulator for Drilling
is basically divided into the description of the generic part, and
the heuristic part of the drilling model. The generic part of the
drilling simulator is mainly represented by a procedural course
of events which makes up the basic drilling. It is the skeleton
for the subsequent heuristic part, where field-specific data is
being implemented into the VESD.
The actual number of core processes modeled in this
simulation is based on the already discussed necessity for
statistical meaningful data. And this is dictated by the available
information. For this work the major data sets support five basic
processes encountered during the drilling of a well: (1) Actual
drilling, (2) Setting casing, (3) Setting lost circulation plugs, (4)
Logging, and (5) Coring. These five processes are depicted in
Fig. 8. However, theses five activities account for more than 90%
of the actual time spent while on location.
6 KEITH K. MILLHEIM, THOMAS GAEBLER SPE/IADC 52803
The flow diagram in Fig. 8 can be interpreted as follows:
The five core processes are represented in five columns,
whereas the user’s input and the subsequent heuristic
calculations are elements of the individual columns. The user’s
inputs are marked as shaded boxes, and the probabilistically
based heuristics are unshaded. Similar to a conventional
software program flow diagram, the rhombs are being used
whenever decisions are being made, either by the user or the
drilling model.
The Ergometric Design
When starting the simulator a front-end with two major
windows shows up. The first window is the Drilling
Commands Window, shown by Fig. 9, the second is the
Simulator Output Window. The Drilling Commands Window
is the control panel of the simulator where the user can set the
parameters and manage the needed actions.
At the beginning of the simulation, with the bit above the
rotary table, the user has to decide upon a bit type and
diameter he/she is going to use, the drilling assembly, and the
operating parameters WOB and RPM. After the drilling
command “Trip In” is switched on the “Pause/Resume
Simulation” button starts the tripping in procedure. In the
Output Window (Fig. 10) the increasing tool face depth can be
read off. Once, at bottomhole depth the user has to circulate
out (“Wash Down”) the settled cuttings to commence drilling.
The actual drilling progress can be observed in the Output
Window, where the rate of penetration, the mudlosses, and the
bottomhole depth are displayed.
With time the bit wears and the user has to pull the bit out
of the borehole, change the bit, and trip in again. Unless no
unscheduled events occur, this procedure goes on until the first
(9 5/8”) casing setting depth is reached. At that depth the user
has to trip out and set the casing. The actions needed to be
done pop up automatically as computed by the drilling model,
e.g. there might be a wipertrip requested or not. After the
casing is set the regular drilling process (with a smaller bit
diameter) continues.
By enlarging the small menu bars (Fig. 11) below the
Output Window the user has access to more information. The
Time vs. Depth Window depicts the bottomhole depth over
time, and shows the layer tops of the different formations. In
the VESD Model Window the user can browse through the
entire drilling, and watch the change of variables throughout a
simulation run.
If messages of unscheduled events show up throughout the
simulation the user is asked to respond, either by setting a specific
action, or if no action is required, to notice the estimated duration.
These messages can be mudlosses, rig defects, well control
problems, required leak off tests, a twist off, etc.
Whenever the user decides to set a cement plug because of
serious mudlosses, he/she has to pull the drillstring out of the
hole and start the setting plug procedure. By doing so the
drilling model calculates the success rate, whether the plug
does cure the mudlosses problem or not. It is the up to the user
to decide to continue drilling, or to set another plug in case the
first plug did not cure the losses.
When the initially calculated total depth is reached, the
simulator computes the probability of hitting an economically
viable formation. Based on the results the user is informed
whether the hole is plugged and abandoned or completed.
Results - One Set of Simulation Runs
To verify the model results from the simulator were compared
to real well data. Prior to verification thirty five runs were
used to debug and test the model. After testing the robustness
of the model the runs were focused on the total drilling time
from drilling the 13,375” casing setting cement to total depth.
One simulation run takes the experienced user about thirty to
forty five minutes. This depends on the scope of the
simulation, i.e. whether comprehensive notes are taken or not.
The bit wear of the individual bits could not be derived
from the given data sets. Therefore, the majority of the runs
were done to calibrate the model’s wear factors. Out of a
series of fifteen runs the last three runs were evaluated.
Using these bit wear factors a total drilling time (starting
with the 9 5/8” casing section to completion) of fifty to ninety
days is computed. This is shown by Table 5.
The total bits consumed, ranged in the simulation from
twenty-one bit to thirty-five bits per well. This is
approximately what was used in the field, with a range from
seventeen to thirty-two bits per well.
Minor drilling mudlosses were experienced in all simulated
wells. Most of the time the mudlosses were within eight to
twelve barrels per hour. Run #1 and run #2 show total
mudlosses, which consequently resulted in one “kick”, and
one successful set cement plug in run #3, and two kicks in run
#2.
Stuck drillstring problems were simulated in runs #1, and
run #3 and could be solved within half an hour and twenty
four hours. A twisted off drillstring in the last depth interval of
run #1 accounted for more than twenty-six hours of trouble
time. The simulated percentages of trouble time spent during
drilling ranging from 3.1 %to 6.3% lie within the actual data
percentages (Table 6).
The time vs. depth curve of run #3 is given by Fig. 12,
where the time read off has to be multiplied with the time
lapse factor nine, which results in 1,602 hours or sixty seven
days of total drilling time.
The above comparison of simulated well data with actual
well data proves the reliability of the heuristic modeling
approach. However, with more data attributes and better well
information accuracy a more “real world” model of the drilling
experience could have been replicated. This should certainly
be a subject of further investigation.
The authors introduced several people to the simulation
model, gave them a short introduction. After about twenty
minutes everyone of the “scholars” were able to run a
simulation on their own.
SPE/IADC 52803 VIRTUAL EXPERIENCE SIMULATION FOR DRILLING - THE CONCEPT 7
Conclusions
This paper presents the overall heuristic concept. Methods
have been introduced to show how experience and knowledge
gained in a certain domain, i.e. drilling a well, can be captured
and retained, as well as being used as a tool to transfer
learning. This is accomplished by introducing the concept of
the virtual experience simulation.
Based on the actual data of 22 wells drilled in a specific
geographical and geological environment a simulator is
developed which offers a special ergometric design that is
compatible with specific drilling data sets.
The major difference of the virtual experience simulation
(also called heuristic simulation) to conventional drilling
simulators is the fact that the heuristic simulation is developed
around data sets of actual wells. Heuristic simulation is the
bridge between the knowledge contained in activated data sets
and the ability to quickly learn the previous gained insights
and experiences. This approach has major advantages. The
VES is not subjected to the formulation of analytical equations
to compute predictable results. In fact, this is why many
previously built predictive simulators failed to establish broad
acceptance. The development of heuristic simulators is
founded on satisfactory “real world” data accumulations,
which can only be obtained if there is a serious commitment of
all persons involved in the data acquisition and data
management process. This includes the definition of the
different data accumulations attributes, the data acquisition,
the data management, and finally the statistical data evaluation
A spin-off of this work is the formulation of the rate of
penetration based on isomeric maps in dependency of the
operating values weight-on-bit and revolutions of the drillstring
Testing the model against actual data shows that the
implemented concepts yield reliable results. With the given
data sets more than three quarters of actual time spent for
drilling the wells can be reproduced, with all major unforeseen
events included.
It is believed this work establishes the potential to take
“inert” unused drilling data and create an activated drilling
data set which contains the majority of the experience and
knowledge for drilling in a specific area. Furthermore, by use
of “a heuristic computer engine” (system dynamics based) this
potential knowledge and experience can become virtual
experience and knowledge, thus transferring knowledge.
Similar VES can be made for any geological domain or for
specific types of drilling like slim hole wells, extended reach
wells, underbalanced drilling, etc.
References
1. Meier, Robert C., Newell, William T., Pazer, Harald L., (1969):
“Simulation in Business and Economics”, Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey, U.S.A.
2. Millheim, Keith K. (1982): “The Role of the Simulator in Drilling
Operations”, paper SPE 11170 presented at the 1982 SPE
Annual Technical Conference, New Orleans, LA, U.S.A., Sept.
26-29
3. Millheim, Keith K. (1983): “An Engineering Simulator for
Drilling, Part I”, paper SPE 12075 presented at the 1983 SPE
Annual Technical Conference, San Francisco, CA, U.S.A., Oct.
5-8
4. Millheim, Keith K. (1983): “An Engineering Simulator for
Drilling, Part II”, paper SPE 12075 presented at the 1983 SPE
Annual Technical Conference, San Francisco, CA, U.S.A., Oct.
5-8
5. Cooper,George A. (1996): “PayZone – Operator’s Manual”,
Department of Material Science and Mineral engineering,
University of California, Berkeley, CA 94720
6. Microcal Software, Inc. (1997), “Origin 5.0 – User’s Manual”,
Northampton, MA, U.SA.
7. Bourgoyne, Adam T. Jr., et al. (1991): “Applied Drilling
Engineering”, Society of Petroleum Engineers, Richardson, TX
8. Warren, T. M. (1979): “Drilling Model for Soft-Formation Bits”,
paper SPE 8438 submitted to the 1979 SPE Annual Technical
Conference, Las Vegas, Nevada, U.S.A., September 23-26
9. Warren, T.M. (1983): “Penetration Rate Performance of Roller
Cone Bits”, paper SPE 13259 submitted to the 1983 SPE Annual
Technical Conference, Houston, TX, U.S.A., October 5-8
10. Bingham, M. G. (1965): “A New Approach to Interpreting Rock
Drillability”, reprint from the Oil and Gas Journal Series,
Petroleum Publishing Co., Apr. 1965
11. Maurer, W. C. (1962): “The Perfect Cleaning Theory of Rotary
Drilling”, Journal of Petroleum Technology, Nov. 1992
12. Onyia, Ernest. C. (1983): “Geological Drilling Log (GDL) – A
Computer Database System for Drilling Simulation”, paper SPE
13113 submitted to the 1983 SPE Annual Technical Conference,
Houston, TX, U.S.A., Oct. 5-8
13. Surfer (Win 32), V6.04, (1996), “User Manual”, Surface
Mapping System, Golden Software, Inc., Golden, CO, U.S.A.
14. PowerSim 2.5d (1996): “Reference Guide”, Powersim Press,
Powersim AS, Promenaden, Knarvik Senter, 5100 Isdalsto,
Norway
15. Checkland, P.B. (1987): “The Application of System Thinking in
Real-World Problem-Situations: The Emergence of Soft-Systems
Methodology”, in M.C. Jackson and P. Keys (eds) New
Directory in Management Science, Gower, Aldershot, England
16. Forrester, Jay W. (1961): “Industrial Dynamics”, MIT Press,
Cambridge, MA, U.S.A.
17. Forrester, Jay W. (1968): “Principles of Systems”, MIT Press,
Cambridge, MA, U.S.A.
18. Wolstenholme, Eric. F. (1990): “A System Dynamics Approach”,
John Wiley & Sons Ltd., Baffins Lane, Chichester, West Sussex
PO19 1DU, England
19. PowerSim 2.5d (1996): “Reference Guide”, Powersim Press,
Powersim AS, Promenaden, Knarvik Senter, 5100 Isdalsto,
Norway
8 KEITH K. MILLHEIM, THOMAS GAEBLER SPE/IADC 52803
Time Operations
07:00 – 11:00 Dropped Totco. POH for bit change. Shock sub leaking oil, laid out same.
11:00 – 13:00 Made up BOP test plug. Tested 5” pipe rams, kill & choke valves and manifold to
3,000psi. Tested Hydrill to 1,500psi. Held all tests for 10 minutes o.k.
13:00 – 15:00 Made up bit No.7 and back-up shock sub. RIH @ 13 3/8” Csg. Shoe
Table 1, Example of actual data recorded in the morning reports
Description Duration [h] OpsCode
dropped Totco, TOH, lay down shock sub(leaking oil) 4 O
BOP, kill-& choke line tested 4 BOP
picked up new bit & back up shock sub, TIH (2773ft) 2 TI
Table 2, Data base entry equivalent to the of morning report example
Parameter for the Trip Rate Calculation
A B1 B2
Trip In -266.00 0.49 -2.87e-5
Trip Out 584.86 0.36 -2.28e-5
Table 3, Parameters used for the tripping rate estimation
Summary of Timing [%]
Activity Well 2 Well 3 Well 5 Well 7 Well 9 Well 21
Drilling 50.3 54.3 41.4 47.5 52.9 48.8
Roundtrips 20.3 12.9 17.4 25.1 15.7 18.1
Coring 3.7 3.5 3.7 4.9 2.6 7.5
Conditioning Mud 3.1 2.2 2.8 3.3 2.9 2.1
Setting & Cementing Casing 3.9 6.7 3.4 5.1 4.6 4.6
Logging 2.9 4.6 3.6 3.8 3.7 2.9
Table 4, Time distribution of activities on location
Total Drilling Time [days]/(Well No.) Run [days]
82 (2) 33 (3) 71 (4) 81 (5) 79 (7) 73 (9) #1 86
68 (10) 74 (11) 71 (12) 66 (13) 84 (14) 73 (15) #2 67
74 (16) 89 (18) 60 (19) 54 (20) 88 (21) 74 (22) #3 54
Average Time: 72 69
Table 5, Comparison total drilling time for 13 3/8” casing section to completion: actual data – simulation
Trouble Time [%]/(Well No.) Runs [%]
4.1 (2) 9.4 (3) 2.3 (4) 20.7 (5) 3.9 (7) 0.7 (9) #1 5,1
5.8 (10) N/A (11) 12.4 (12) 6.8 (13) 16.6 (14) 1.4 (15) #2 4,2
8.0 (16) N/A (18) 12.7 (19) 5.9 (20) 9.0 (21) 7.5 (22) #3 9,9
Table 6, Comparison trouble time: actual data - simulation
SPE/IADC 52803 VIRTUAL EXPERIENCE SIMULATION FOR DRILLING - THE CONCEPT 9
Fig. 1, Heuristic Triangle
Fig. 2, Merging Inert Data Set with the Heuristic Triangle
0
1
2
3
Wells per year
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
Wells drilled
Fig. 3, Time Distribution of Wells Drilled
Fig. 4, Histogram of layer thickness for Layer 15
10 KEITH K. MILLHEIM, THOMAS GAEBLER SPE/IADC 52803
Fig. 5, Trip rate derived from actual well data
Fig. 6, ROP surface map for layer 15 in m/h
Fig. 7, 3-dimensional surface map of layer 15
SPE/IADC 52803 VIRTUAL EXPERIENCE SIMULATION FOR DRILLING - THE CONCEPT 11
no
yes
no no no
yes
no no
yes
yes
no
yes yes
yes
no
VESD - Flow Diagram
OEDP Core Bit / Core Assembly Bit / BHA Run Log Set Casing
SET PLUG CORE DRILL LOG SET CASING
Trip In Rate
Set Plug Core Drill Set Casing Cement
TIME LOG TIME CAS
TIME PLUG ROP CORE ROP Drill TIME CEM
Drill
Success
TO for Drill
req.
length
Losses
Core
req. TO for CORE TO Rate
TO for SET CAS
Wiper Trip
Cas.Set
Depth TO for SET PLUG
LCM
Well
Control
TIME TASK L.O.T.
Drill
Twist off
Fig. 8, VESD Flow diagram
Fig. 9, Drilling Commands Window
12 KEITH K. MILLHEIM, THOMAS GAEBLER SPE/IADC 52803
Fig. 10, Output Window
Fig.11, Menu bars of the Time vs. Depth Window and the VESD Model Window
Fig. 12, Time vs. depth curve of run #3
[ 此贴被superswpu在2007-05-19 22:52重新编辑 ]
