Using Big Data Analysis Tools to Understand Bad Hole Sections on the UK Continental Shelf

MANAGEMENT

Using Big Data Analysis Tools To Understand Bad Hole Sections on the UK Continental Shelf

Joe Johnston, CGG, and Aurelien Guichard, Teradata

Research and analysis show that 90% data is never used alongside seismic data. fields include a retailer analysis of buying of all data in the world today was creat- As the data is growing in volume and patterns and automotive manufacturers ed in the past 2 years alone. Already the variety, it is overwhelming traditional predicting faults and failures. growth of data volume outpaces the con- methods. What business opportunities Large volumes of data are nothing new ventional capacity to analyze and under- are being missed? to the oil and gas industry. The seismic stand, and the trend is only accelerating. business in particular has been success- This trend is also occurring in the oil Big Data fully dealing with rapidly increasing vol- and gas industry with, for example, the The term “big data” refers to more than umes for a long time. For example, work continued growth in seismic channel simply a large volume of different types by seismic equipment manufacturer Ser- counts, integration of multiphysics infor- of data, both structured and unstruc- cel suggests the channel count available mation, logging while drilling, and the tured, with varying degrees of accura- for acquiring a seismic survey has been constant flow of information from “intel- cy. It also includes a suite of applica- steadily increasing by one order of mag- ligent wells” in “digital oil fields.” tions providing solutions and analysis. nitude every 10 years. Current data analysis and interpreta- But big data is really a movement. The big Until now, much of our understand- tion approaches follow well-established data approach is said to be a data-centric ing of reservoirs has come from the and rigid workflows, which study the method adept at uncovering otherwise study of physics-based models rather same small set of relationships between invisible patterns and connections by than directly from data itself. The chal- entities. It is common, for example, to linking disparate data types. It can search lenge we face today is converting ever- use core data in well log analysis or well- and analyze all data and size with great increasing data volumes into models in bore acoustic measurements in surface agility and without regard to user group decreasing time frames, ideally in “real seismic interpretation. However, core or project area. Examples from other time.” The big data approach will instead

Fig. 1—The caliper curve (red in left track) shows an erratic borehole. The rest of the data is poor. The sonic velocity will be wrong for use in seismic analysis; the hydrocarbons and high porosity shown on the right are completely incorrect. Copyright 2015, Society of Petroleum Engineers. Reprinted from the Journal of Petroleum Technology with permission.

60 JPT • OCTOBER 2015 allow us to construct new types of data- driven models to bypass these traditional bottlenecks. It is also expected to lead to different views of standard models, providing new and valuable insights in the process.

Case Study To gain perspective on how a big data approach could be used in the oil and

gas industry, we selected a real business (ft/hr) erage Rate of Penetration case as a test bed. We thought it was Av important to look for specific answers using data from very different disciplines and sources. Drilling a well is a com- Average Weight on Bit (klb) plex operation with significant risk, both economic and health, safety, and envi- Fig. 2—The visualization of multidimensional data presents a challenge. In ronment-related. Typical modern drill- the trial, the size of the circle shows the variation from the ideal case and the colors represent the formations encountered. The Punt formation (brown) ing operations generate vast quantities of shows an enlarged borehole. data and metadata. In this study, we chose to seek the trends and correlations that could all referred to the same kind of measure- the borehole quality by a single well or by improve drilling results using predictive ment from different sources). We also formation or by formation and geograph- relationships to indicate drilling efficien- encountered and solved issues with mis- ical location or any combination. cy and high-risk situations. We aimed to labeled formation tops, missing or mis- An example of the type of discovery is find ways to save on drilling time and, placed data columns, and widely differ- illustrated in Figs. 3 and 4. Plots of WOB hence, cost, along with enhancing safety ing scales and granularity. vs. torque (Fig. 3) and of WOB vs. ROP aspects. We also noted that poor bore- show an anomalous set of bad data points hole conditions make the subsequent Results associated with an increase in WOB. This logging operations difficult and often Once we had thoroughly prepared and is found to be a single well. The two plots render the acquired data useless (Fig. 1). loaded all the many types of input data, suggest a problem with the well. The vari- Our study made use of data relating to we began to form connections between ous reports and plots associated with this approximately 350 wells from across the drilling observations, measurements, well were checked. UK North Sea, provided by CGG as offi- and the subsurface environment. Our The answer was found in the log plot cial UK Continental Shelf data release objective was to try and link drilling (Fig. 4), which shows a dramatic decrease agents on behalf of the UK Department of parameters to wellbore condition. In in the caliper. The device was appar- Energy & Climate Change. The Teradata this study, we specifically considered ently closed because the logging tool Aster platform was used for data loading, weight on bit (WOB), rate of penetration was becoming stuck owing to the bore- quality control, and analysis. (ROP), torque, and caliper, with the latter hole conditions. A significant increase The combined input were drilling parameter normalized using the bit size in the tension on the logging cable was parameters, well logs, geological for- to define the differential caliper, which reported, a potentially dangerous situ- mation information, and well locations/ can be used as an indicator of hole condi- ation. An additional wiper trip, repre- deviations. It took the form of approxi- tion and flagging of “bad holes.” senting an extra cost and time, was made mately 20,000 files in a range of com- An essential concept of this type of to recondition the hole before running mon formats including LAS, TXT, CSV, analytics is that there is no a priori data casing to eliminate potential problems XLS, and PDF. The location enabled a model or correlations and algorithms with that operation. All this was noted in geospatial analysis while the stratigra- between the data types. All data, whether the reports. phy allowed an exploration of the verti- sampled every 6 in. as in a well log or at In addition, the logs themselves cal dimension. random intervals such as formation tops, were affected by the sticking tool, with We needed our data management is treated in the same way. This allows spikes and straight lines. This effect expertise to address various challeng- for rapid associations between dispa- was not noted in any report but would es encountered with the input data. We rate data types not normally connected have been discovered in the subsequent were careful to apply business rules to (Fig. 2). interpretations. data preparation and quality control. We The results were surprising and instruc- The ease and speed with which this was resolved inconsistent well log mnemon- tive. It was possible to visualize where uncovered using big data analysis con- ics (for example, CAL, CALI, CAL1, and C1 changes in drilling parameters affected trasts with the traditional approach of a

JPT • OCTOBER 2015 61 tions. Unexpected correlations were exhibited and differences were discovered both spatially across areas and ver- tically through formations. The correlations allowed predictive statistics to be computed. This can provide the drillers with indications of difficult areas. A set of operational parameters can be calculat- ed enabling them to drill future wells

rque (kft lb) with fewer problems. In addition, inno- To vative quality control techniques applied on disparate data types were developed, which saved specialist time. Our experience suggests that oil and gas data are highly suited to big data analysis. However, expertise is required

Weight on Bit (klbf) to properly prepare and control the quality of the input. The preparation of data using big data techniques allows for a Fig. 3—In a single formation called Humber, the colors show whether the DCAL number of quality control steps to be rap- value is good (green) or bad (orange), based on some modifiable criteria. The idly performed. circled cluster of points represents a single well. This “exception” uncovered It is better to focus any analyses on with the big data approach revealed an anomaly. specific questions because it allows the user to obtain quick answers rather than detailed examination of well reports and hole in one place are not necessarily cor- fuzzy generalizations. It is no longer a other documents. In the case of multiwell rect in other places. question of whether big data has arrived, phenomena, this task is extremely time but how the oil and gas industry will consuming and prone to various types Conclusion use them and what insights and break- of errors. We successfully deployed big data tech- throughs they will provide. JPT The example is one of many to come niques on the diverse data used in this out of this study. Other examples show study and used advanced visualization variations on a regional basis in the same capabilities to display multiple types formation. This suggests that drilling of data. Multivariate analysis was per- parameters giving a perfectly good bore- formed on the data without preconcep- Joe Johnston is chief petrophysicist with CGG. He has worked in the industry Curves straight lining—bad data for more than 40 years, 5 of them with CGG, in a number of positions around the world. He is the author of a number of papers on petrophysical topics and holds a BSc degree in physics from Imperial Spikes in data due to caliper College London. being closed due to high-tension overpulls Aurelien Guichard is a senior consultant at Teradata where he focuses on uncovering business value from data and analytics in oil and gas. Previously, he worked for Schlumberger in business consulting, reservoir engineering, well testing, and seismic. He holds a master’s degree in ocean engineering from Texas Bad data A&M University, a master’s degree in Fig. 4—The log shown is from the anomalous well discovered in Fig. 3. The civil engineering from Ecole Speciale des caliper curve (red in left track) goes sharply to the left, indicating a dramatic Travaux Publics in Paris, and an MBA from reduction in apparent hole size. The logging problems are shown on the Tulane University. other tracks.

62 JPT • OCTOBER 2015