Data mining helps cut cost of gas drilling at BG plc

The Gas Research and Technology Centre - a striking new building located a mile off the M1 motorway near Loughborough - houses the Research & Technology (R&T) operation of BG plc (British Gas). Currently employing some 600 people, R&T brings together the skills and expertise across the gas chain and provides support for the various BG business units. BG Exploration & Production (BG E&P) is presently benefiting from a project that aims to use data mining techniques to achieve significant reductions in the cost of its drilling operations.

The BG E&P drilling department has responsibility for BG-operated wells, in addition to others that are jointly-operated with international partners. Wells are located mostly offshore and are as widespread as Russia, Trinidad, Bulgaria and Tunisia. Drilling is a hugely expensive process, with daily costs for a North Sea operation typically incurring rig costs of around $50,000 per day. Clearly, anything that helps to reduce the time when a drilling rig is not productive has the potential to achieve huge savings.

R&T's Senior Engineer David Reid is part of a five-man project team that is using data mining techniques to help identify the factors that are likely to result in down-time on a rig. Although many problems can occur during a drilling operation, the project has initially focused on "Stuck Pipe" which, as the name implies, involves the drilling pipe becoming stuck while deep in rock below the ocean bed.

Off-shore drilling involves initially sinking a "Conductor" into the sea bed. This is cemented into a large diameter hole in order to seal the well from sea water and to provide a conduit for the drilling pipe work - the drillstring. As the drilling operation continues, the diameter of the hole progressively narrows and is lined with a casing to seal it and to prevent it from collapsing. In the final stages of a drilling operation, the lining may be only 7 inches in diameter. Throughout the operation, mud is pumped through to help clean the drill bit; to balance formation pressure; and to carry rock cuttings to the surface for disposal and to geologists for analysis. The whole drilling process is carefully monitored and surveyed in order to ensure that the well plan is followed. David Reid's project team of engineers is using data mining to identify patterns and trends in well site information that can help to determine the factors most likely to result in what is euphemistically called "non-drilling activity". The project, which currently has an annual budget of around $300,000, is also looking at optimisation of the rate of penetration; optimising the design of the drillstring; and the circumstances surrounding drillstring failures. Data used in the data mining analysis is taken from Daily Reports filed by rig engineers which cover many aspects of the operation - including location of the well; a diary of rig activity; and information about the drilling mud. The drilling of exploration and production wells may result in between 20 and 130 daily reports. The resulting collection of these reports is a valuable source of information - particularly "offset" well data from neighbouring wells and is therefore of enormous potential commercial value.

The project team has had to gather daily report data from two databases: One of which was old and included incomplete or absent data - particularly IADC (International Association of Drilling Contractors) codes. The other database was compiled more recently and included a large amount of additional data about well site geology, drilling costs, etc. This later information has been used as a template for the design of a PC-based FoxPro database and has therefore resulted in a number of gaps from the older data - which needed to be filled manually. David Reid comments on the move to test data mining techniques: "We decided to undertake a feasibility study to see if the data could be mined. Initially, we wanted to look at Stuck Pipe, which is a problem that can occur when drilling beyond the casing in the open hole. Often the problem is a result of mud filter cake building up around the sides of the hole. Even if the pipe can be released, hours drilling time may be lost. If the problem cannot be resolved easily, jarring can release the pipe, although this can take a couple of days of rig time. At worst, it may be necessary to break off the pipe; recover what is possible of the drillstring; cement in the hole; and side track. Using existing data to identify factors likely to result in Stuck Pipe was an attractive option."

The database indicated that there had been sixty recorded occurrences of Stuck Pipe in 170 BG wells. XpertRule Software, a Lancashire-based consultancy and software house with international experience in data mining, was asked to undertake the feasibility study. The company also markets XpertRule Miner - a PC-based software applications that uses "Knowledge Induction" to learn from potentially millions of records in order to generate a graphical "decision tree" that profiles patterns in the data. In this case, not only was the data sample very small, but the data quality was also poor, as David Reid recalls: "Very little of the information from the old database had been transferred automatically and IADC codes were incorrect. We also found that the IADC codes were too vague. Stuck Pipe was simply one of a number of topics listed under a general code that covered Hole Problems. In spite of all this, the results from the feasibility study using XpertRule showed that it was possible to mine the data and to determine trends." Much of the time invested by the project team has concentrated on getting the data in good order. The IADC codes have now been sub-divided to provide BG with their own, more detailed, version of the classification which includes, for example, ten categories under the heading of "Hole Problems". As a consequence, it has been possible to further refine the decision tree produced by XpertRule for Stuck Pipe problems. The results indicate that the length of time the hole has been open; the properties of the drilling mud; and the frequency with which the mud is conditioned all play a significant role in the incidence of Stuck Pipe.

Daily Reports produced on BG E&P drilling rigs now include the new classifications of the IADC codes and provide additional data. This allows the data mining analysis to use the new information to re-run the analysis in order to check and to refine the patterns and trends. The BG R&T project team has purchased XpertRule from XpertRule Software and is now self-sufficient in their use of data mining processes. Data is now routinely conditioned before being included in data tables for data mining analysis.

The analysis is being extended to other non-drilling activities, including drillstring failures. This covers "Twistoffs" (where the pipe shears) and "Washouts" (where the pipe fractures and allows high pressure mud to leak from the inside to the outside of the drill string). World-wide figures indicate that downhole drillstring failures each result in average downtime costs of $500,000 for offshore and $150,000 for onshore operations. The potential for using existing data to avoid costs of this order makes it easy to see why David Reid is confident that the project will be extended to consider other non-drilling activities: "The decision trees can be written up for engineers to use in the field. We are able to provide them with an indication of the sequence of events that can result in drilling problems. It is learning that can be applied to reduce the cost of future wells."