SPE 65618 a New Approach for Drill Bit Selection

Drill bit
of 4
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  Copyright 2000, Society of Petroleum Engineers Inc.This paper was prepared for presentation at the 2000 SPE Eastern Regional Meeting held inMorgantown, West Virginia, 17–19 October 2000.This paper was selected for presentation by an SPE Program Committee following review of information contained in an abstract submitted by the author(s). Contents of the paper, aspresented, have not been reviewed by the Society of Petroleum Engineers and are subject tocorrection by the author(s). The material, as presented, does not necessarily reflect anyposition of the Society of Petroleum Engineers, its officers, or members. Papers presented atSPE meetings are subject to publication review by Editorial Committees of the Society of Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper for commercial purposes without the written consent of the Society of Petroleum Engineers isprohibited. Permission to reproduce in print is restricted to an abstract of not more than 300words; illustrations may not be copied. The abstract must contain conspicuousacknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435. Abstract Presented is a new methodology for selecting rotary drilling bits in an oil or gas well. Currently, bits are selected based onthe performance of similar bits at offset wells. Parametersaffecting a bit performance have a complex pattern. Therelationship between formation properties, drilling fluidcharacteristics, bit design, and operational parameters in these patterns are not easily understood.For a given field, studied were variables such as bit size,weight on bit, rotary speed, pump rate, drilled interval, and bittype. A three-layer artificial neural network was designed andtrained with field data. This method incorporatescomputational intelligence to define the relationship betweenthe variables. Further, it can be used to estimate other drilling parameters. The results indicate that the back propagationachitecture with two hidden slabs is the most effective neuralnetwork design for predicting the optimum bit type.With the given data sets, this new model successfully predicted the bit types for several fields. For different data setsused in this study, the correlation coefficients for the predictedand field used bit types ranged between 0.857 and 0.975. Introduction Drilling engineers deal with many challenges before andduring drilling a new well even in a known area. There aremany parameters related to hardware and daily operations thatare planned and also modified as the drilling progress. Bitselection is one of the important parameters for planning anddesigning a new oil or gas well. The selection of a proper bit isa difficult task since the factors affecting the bit performanceare complex relationships between formation properties, bit  *  Now with Kuwait Drilling Companyhardware design, and operational parameters. Formation properties such as hardness and drillability are based on theinterval drilled and they can not be changed. Additionally,they are not measured prior to drilling and they have to beestimated from geophysical surveys or from offset wellrecords. Several investigators conducted research to estimatethe bit behavior based on operational parameters and datarecorded at offset wells. 1-10  Different models were developed by these researchers with assumptions limiting theapplicability of their model.In general, the data from offset wells are used by theengineer to select a proper bit. The bit type with the highestrate of penetration or minimum cost per foot are the twocommonly used criterias for selecting the bit for the nextinterval. Additional factors such as hydraulics, formationhardness, bit design, and operational parameters are alsoconsidered in the selection process. Due to the number of variables considered, the selection process is a trial and error  procedure. In many cases, this approach can ignore some of the important parameters affecting the bit performance andcan not guarantee the selection of the optimum bit type.When sufficient data exists, the use of neural networks aredemonstrated to identify complex relationships. 11-13  Theincreased availability of down hole measuring tools resulted indata sets with many recorded variables related to the drilling process. As a result, different neural networks can be designedfor a region or a field to predict unknown parameters. Approach In this study, a new methodology is introduced to select rotarydrilling bits. This approach uses a three-layer feed forwardneural network to select bit types. Several neural network models were used to determine the complex relationship between formation and bit properties together with operating parameters.The data sets used in this study were obtained fromdifferent fields in the Middle East. The data were screened andthe reaming and coring operations were excluded from thestudy. First data (K-1) consisted of approximately 2000 sets of recorded field information with 277 different bit types fromone region. The input data used for the first neural network design were consisted of bit size, total nozzle area of the bit,depth the bit was pulled, drilled interval length, rate of  penetration, maximum and minimum weight on bit (WOB) SPE 65618 A New Approach for Drill Bit Selection H.I. Bilgesu, SPE; A.F. AL-Rashidi * ; K. Aminian, SPE; S. Ameri, SPE, West Virginia University  2H.I. BILGESU, A.F. AL-RASHIDI, K. AMINIAN, S. AMERISPE 65618 and rotary speeds (RPM), and the mud circulation rate. Bitsizes included in the K-1 data set ranged between 4½-in. and28-in. The WOB and RPM data included minimum andmaximum values for each bit resulting in two separate inputsfor the neural network. The range of data used in data set K-1are listed in Table 1. In the first neural network, data set K-1with 10 input parameters was used to predict the bit type asoutput for the next drilling interval.A larger data set from different regions in the Middle Eastwas used to develop the second neural network. The seconddata set (K-2) consisted of 489 different bit types andvariables recorded were the same as data set K-1. Bit sizesranged between 4¼-in. and 36-in. for the second data set andthe range of data for other parameters used in this study arelisted in Table 2. Similar to the first neural network design, 10variables were used as input and the output from the modelwas the bit type.The third data set (K-3) contains more than 2,000 records.The variables recorded for the data set consisted of bit size,total nozzle flow area, depth the bit pulled out, drilled intervallength, penetration rate, WOB, RPM, and mud flow rate. Theranges covered by each recorded variable are listed in Table 3.In the design of the third neural network, eight variables wereused as input.Several neural networks were developed to predict the bittype to use in the new drilling interval. Different neuralnetwork design architectures were used to select bit types. Thedesign structure with two hidden slabs with differentactivation functions and a jump connection was found to bemore effective in predicting bit types. A typical neuralnetwork architecture used in the final design for all data sets isshown in Figure 1. Discussion of Results The neural networks developed in this study used 90% of thedata after selecting the 10% of data as the test set. The 90% of the data set were used to train the network and the test set wasused to verify the neural network design. The networks weresubjected to test sets for the first time. The predicted bit typeswere compared with the bits used in the field for threedifferent data sets and the results are shown on linear scale plots. For a perfect match, all points should lie on the straightline.Figure 2 compares the predicted bit types with the selected bit types for data set K-1. The runs conducted with this model produced linear correlation coefficient (r) values ranging between 0.960 and 0.975 and the results with the highestcorrelation coefficient is shown in Figure 2. The learning rateand the momentum used in the final design were 0.05 and 0.5,respectively.The results from data set K-2 are shown in Figure 3. Thelearning rate and the momentum values for the final designwere both 0.1. The highest correlation coeefficient obtainedfor the second network was 0.857.The results from the third neural network are presented inFigure 4. For this design, fewer bit types were availablecompared to data sets K-1 and K-2. The learning rate and themomentum used in the final design were 0.05 and 0.5,respectively. The linear correlation coefficient for the thirdmodel was 0.877. Conclusions 1.   A new methodology was introduced to select bits for thenext bit run or for new wells.2.   The neural networks developed in this study successfullyselected the bits for new sections and they can be used toenhance the planning process for a new well.3.   A three layer neural network structure using two slabswith different activation functions and a jump connectionfor the hidden layer was found to be the most effectivedesign in predicting the bit types.4.   The ranges of data used in the development of neuralnetworks define the applicable boundaries of the modelsdesigned in this study. For other ranges new neuralnetworks need to be developed.5.   The correlation coefficients between neural network  predicted bit types and bit types selected in the field were between 0.857 and 0.975 for the data sets used in thisstudy. References 1.   Falconer, I.G., Burgess, T.M., and Sheppard, M.C.: “SeparatingBit and Lithology Effects from Drilling Mechanics Data,” paper SPE 17191 presented at the 1988 IADC/SPE Drilling Conference,Dallas, TX, Februrary 28 – March 2.2.   Fear, M.J.: “An Expert System for Drill Bit Selection” paper SPE27470 presented at the 1994 IADC/SPE Drilling Conference,Dallas, TX, February 15 – 18.3.   Fear, M.J.: “How to Improve Rate of Penetration in FieldOperations” paper SPE 35107 presented at the 1996 IADC/SPEDrilling Conference, New Orleans, LA, March 12 – 15.4.   Burgess, T.M., and Lesso, Jr., W.G.: “Measuring Bit Wear of Milled Tooth Bits Using MWD Torque and Weight-on-Bit” paper SPE 13475 presented at the 1985 SPE/IADC Drilling Conference, New Orleans, LA, March 6-8.5.   Xu, H., Tochikawa, T., and Hatakemaya, T.: “A Practical Methodfor Modeling Bit Performance Using Mud Logging Data,” paper SPE 37583 presented at the 1997 SPE/IADC Drilling Conference,Amsterdam, The Netherlands, March 4-6.6.   Perrin, V.P., Mensa-Wilmot, G., and Alexander, W.L.: “DrillingIndex – A New Approach to Bit Performance Evaluation,” paper SPE 37595 presented at the 1997 SPE/IADC Drilling Conference,Amsterdam, The Netherlands, March 4-6.7.   Fear, M.J., Thorogood, J.L., Whelehan, O.P., and Williamson,H.S.: “Optimization of Rock-Bit Life Based on Bearing FailureCriteria,” SPEDE (June 1992) 163.8.   Fay, H.: “Practical Evaluation of Rock-Bit Wear During Drilling,”SPEDC, (June 1993) 99.9.   Hemphill, T., and Clark, R.K.: “Effects of PDC-Bit Selection andMud Chemistry on Drilling Rates in Shale,” SPEDC, (September 1994) 176.10.   Simmons, E.L.: “A Technique for Accurate Bit Programming andDrilling Performance Optimization,” paper IADC/SPE 14784 presented at the 1986 IADC/SPE Drilling Conference, Dallas,TX, September 21-24.  SPE 65618A NEW APPROACH FOR DRILL BIT SELECTION3 11.   Arehart, R.A.: “Drill Bit Diagnosis Using Neural Networks,” paper SPE 19558 presented at the 1989 SPE Annual TechnicalConference and Exhibition, San Antonio, TX, October 8-11.12.   Bilgesu, H.I., Tetrick, L.T., Altmis, U., Mohaghegh, S., andAmeri, S.: “A New Approach for Prediction of Rate of Penetration (ROP) Values” SPE paper 39231, presented at the1997 SPE Eastern Regional Meeting, Lexington, KY, October 22-24.13.   Bilgesu, H.I., Altmis, U., Mohaghegh, S., Ameri, S., andAminian, K.: “A New Approach to Predict Bit Life Based onTooth and Bearing Failure” SPE paper 51082, presented at the1998 SPE Eastern Regional Meeting, Pittsburgh, PA, November 19-21. TABLE 1 – RANGE OF PARAMETERS FOR DATASET K-1ParameterRange Bit size, in.4.5 – 28Total nozzle area ,in 2 0.1534 – 1.491Depth out, ft127 – 16,463Drilled interval, ft100 – 5,105Rate of penetration, ft/hr1 – 188Minimum WOB, 1000 lbs5 – 85Maximum WOB, 1000 lbs5 – 95Minimum RPM, rev/min25 – 180Maximum RPM, rev/min30 – 180Mud circulation rate, gal/min73 – 1,200 TABLE 2 – RANGE OF PARAMETERS FOR DATASET K-2ParameterRange Bit size, in.4.5 – 36Total nozzle area ,in 2 0.049 – 7.517Depth out, ft104 – 20,774Drilled interval, ft100 – 8,672Rate of penetration, ft/hr1 – 488Minimum WOB, 1000 lbs1 – 80Maximum WOB, 1000 lbs2 – 100Minimum RPM, rev/min15 – 259Maximum RPM, rev/min15 – 259Mud circulation rate, gal/min23 – 1,900 TABLE 3 – RANGE OF PARAMETERS FOR DATASET K-3ParameterRange Bit size, in.6.125 – 28Total nozzle area ,in 2 0.173 – 2.506Depth out, ft104 – 17,892Drilled interval, ft100 – 5,833Rate of penetration, ft/hr2.32 – 448WOB, 1000 lbs2 – 83RPM, rev/min11 – 315Mud circulation rate, gal/min10 – 1,297  4H.I. BILGESU, A.F. AL-RASHIDI, K. AMINIAN, S. AMERISPE 65618 Figure 1. Neural network structure used in the bit selection. 0100200300400500600 NN PredictedBitType 0 100 200 300 400 500 600Selected Bit Type Figure 2. Comparison of neural network selected bit types withbits used in the field for data set K-1. 0100200300400500 NN PredictedBitType 0 100 200 300 400 500Selected Bit Type Figure 3. Comparison of neural network selected bit types withbits used in the field for data set K-2. 010002000300040005000600070008000 NN PredictedBitType 0 2000 4000 6000 8000Selected Bit Type Figure 4. Comparison of neural network selected bit types withbits used in the field data set K-3.
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!