Bilal Esmael

Automated system for drilling operations classification using statistical features

Operations classification is one of the most needed tasks in the oil & gas industry. It provides the engineers with detailed information about what is happening on the rig site. In this paper we propose an approach to classify drilling operations automatically using machine learning techniques. This approach takes as input the sensors data in a specific time range, and predicts the drilling operation. Our approach is simple but effective, where for each sensor data (channel) a list of statistical features will be extracted, then features selection algorithms will be used to select the most informative features, and finally, a classifier will be trained based on these features. In this paper many feature weighting and selection algorithms were tested to find which statistical measures clearly distinguish between many different rig operations. In addition, many classification techniques were employed to find the best one in terms of accuracy and speed. Experimental evaluation with real data, from four different drilling scenarios, shows that our approach has the ability to extract and select the best features and build accurate classifiers. The performance of the classifiers was evaluated by using the cross-validation method.

Improving time series classification using Hidden Markov Models

Time series data are ubiquitous and being generated at an unprecedented speed and volume in many fields including finance, medicine, oil and gas industry and other business domains. Many techniques have been developed to analyze time series and understand the system that produces them. In this paper we propose a hybrid approach to improve the accuracy of time series classifiers by using Hidden Markov Models (HMM). The proposed approach is based on the principle of learning by mistakes. A HMM model is trained using the confusion matrices which are normally used to measure the classification accuracy. Misclassified samples are the basis of learning process. Our approach improves the classification accuracy by executing a second cycle of classification taking into account the temporal relations in the data. The objective of the proposed approach is to utilize the strengths of Hidden Markov Models (dealing with temporal data) to complement the weaknesses of other classification techniques. Consequently, instead of finding single isolated patterns, we focus on understanding the relationships between these patterns. The proposed approach was evaluated with a case study. The target of the case study was to classify real drilling data generated by rig sensors. Experimental evaluation proves the feasibility and effectiveness of the approach.

A Visual Analytics Approach to Segmenting and Labeling Multivariate Time Series Data

Many natural and industrial processes such as oil well construction are composed of a sequence of recurring activities. Such processes can often be monitored via multiple sensors that record physical measurements over time. Using these measurements, it is sometimes possible to reconstruct the processes by segmenting the respective time series data into intervals that correspond to the constituent activities. While automated algorithms can compute this segmentation rapidly, they cannot always achieve the required accuracy rate e.g. due to process variations that need human judgment to account for. We propose a Visual Analytics approach that intertwines interactive time series visualization with automated algorithms for segmenting and labeling multivariate time series data. Our approach helps domain experts to inspect the results, identify segmentation problems, and correct mislabeled segments accordingly. We demonstrate how our approach is applied in the drilling industry and discuss its applicability to other domains having similar requirements.

Drilling events detection using hybrid intelligent segmentation algorithm

Several sensor measurements are collected from drilling rig during oil well drilling process. These measurements carry information not only about the operational states of the drilling rig but also about all higher level operations and activities performed by drilling crew. Automatic detection and classification of such drilling operations and states is considered as a big challenge in drilling industry. Furthermore, the possibility of detecting such events opens the door to detect and analyze hidden lost time of the drilling process. This paper presents a novel algorithm for drilling time series segmentation using Expectation Maximization and Piecewise Linear Approximation algorithms. The suggested algorithm shows that the incorporation of prior-knowledge about the drilling process is a key step to segment drilling time series successfully. The Expectation Maximization algorithm is used to segment drilling time series based on hook-load sensor measurements. In addition, Piecewise Linear Approximation is hired in our approach to slice standpipe pressure, pump flow rate and rotational speed (RPM) and torque of the top drive motor. Merging the results from both, Expectation Maximization and Piecewise Linear Approximation, gives the suggested algorithm the dynamic ability to detect all drilling events and activities.

Diagnosing drilling problems using visual analytics of sensors measurements

One of the major challenges in the drilling industry is the quick detection of problems that can occur during drilling a deep well due to high cost implications. These problems can occur for various reasons and can exhibit varying symptoms, which make them difficult to identify or prevent automatically. Visual Analytics has emerged as an alternative approach for data analysis. It combines both the computational power of computers and the experience of domain experts to analyze and gain insights into large data. This paper describes a procedure for analyzing and identifying drilling problem using Visual Analytics techniques. It provides results of an elaborated analysis of sensor measurement datasets that contain “Stuck Pipes” situations - one of the most common drilling problems. Statistical features are calculated from the dataset using the sliding window method. We show how visual analysis by means of linked scatter plots enable relating the problem patterns to the computed features and can hence help in identifying “Stuck Pipes” problems.

Intelligent Real-time Drilling Operations Classification Using Trend Analysis of Drilling Rig Sensors Data

Detection of oilwell drilling operations is an important step for drilling process optimization. If drilling operations are classified accurately, detailed performance reports not only on drilling crews but also on drilling rigs can be produced. Using such reports, the management can evaluate the drilling work more precisely from performance point of view. Mud-logging systems of modern drilling rigs provide numerous sensors data. Those sensors measurements are considered as indicators to monitor different states of drilling process. Usually real-time measurements of the following sensors data are available as surface measurements: hookload, block position, flow rates, pump pressure, borehole and bit depth, RPM, torque, rate of penetration and weight on bit. In this work, collected sensors measurements from mud-logging systems are used to detect different drilling operations. Detailed data analysis shows that the surface sensors measurements can be considered as a main source of information about drilling operations. For this purpose, a mathematical model based on polynomials approximation is constructed to interpolate sensors data measurements. Discrete polynomial moments are used as a tool to extract specific features (moments) from drilling sensors data. Then we use these moments for each drilling operation as pattern descriptor to classify similar operations in drilling time series. The extracted polynomial moments describe trends of sensors data and behavior of rig’s sub-systems (Rotation System, Circulation System, and Hoisting System). Furthermore, this paper suggests a method on how to build patterns base and how to recognize and classify drilling operations once sensors data received from mud-logging system. Drilling experts compare the results to manually classified operations and the results show high accuracy. Introduction Improving performance of drilling process is a big challenge in nowadays drilling industry. To improve drilling performance, we need first to measure it. Performance measurement means determining quantitative values or weights that describe each drilling operation and complete drilling process as resultant. For example, duration of each drilling operation can be considered as a useful measure. Also number of drilling operations and distributions of those operations over different well drilling phases are important measures of drilling performance. Automatic detection and recognition of drilling operation is the first step towards drilling performance measurement and improvement (G. Thonhauser, Mathis, et al. 2006). Automatic detecting and recognition of drilling operations gives flexibility in monitoring and recognizing drilling process. Monitoring drilling operations helps rig operator in finding limitations and shorts in performance of either drilling crews or rig equipment or even both. Then saving potentials are accurately estimated through training plans for drilling crews and/or replacement parts for rig’s equipment. Furthermore, evaluation of drilling performance supports the drilling crews in their drilling tasks with consideration to safety and technical limits of their drilling equipment (G. Thonhauser, Mathis, et al. 2006). Through drilling process, a huge amount of data in form of sensors measurements is produced over time. This data contains not only readings of sensors but also information about each drilling operation i.e. start, end, and behavior of each equipment. Drilling operations such as formations drilling, making connection for new drillstand, breaking connection, pulling out of hole, running in hole, and cleaning hole are carefully chosen as basic drilling operations performed by drilling crew. Each of those drilling operation has a specific pattern in rig sensors measurements. Detecting drilling operations patterns in sensors data supports rig operators in finding out the state of drilling rig instantly. At the end of the day, it gives detailed information on rig state over any span of time. So rig’s operator can easily observe operating time of drilling rig and how the actual performance matches with the pre-defined well plan (G. Thonhauser, Wallnoefer, et al. 2006). Rig Sensors Systems Drilling rig performs its functionality in drilling boreholes through collaboration of three main sub-systems: Rotary System, Circulation System, and Hoisting System. Rotary System is the system that turns the drillstring. Top drive as type of rotary system which consists of one or more motors (electric or hydraulic) connected with appropriate gearing to a short section of pipe called a quill, that in turn may be screwed into a saver sub or the drillstring itself. Also rotary table another type of rotary system and it consists of revolving or spinning section of the drillfloor that provides power to turn the drillstring in a clockwise direction (as viewed from above). The rotary motion and power are transmitted through the kelly bushing and the kelly to the drillstring. RPM and torque sensors measure the revolution per min and torque of rotation at the surface (Florence et al. 2010). Circulation system is defined as the complete, circuitous path that the drilling fluid travels. Starting at the main rig pumps, major components include surface piping, the standpipe, the kelly hose (rotary), the kelly, the drillpipe, drill collars, bit nozzles, the various annular geometries of the openhole and casing strings, the bell nipple, the flowline, the mud-cleaning

Distributed recognition system for drilling events detection and classification

Several sensor measurements collected from drilling rig during oil well drilling process. These measurements carry information not only about operational states of drilling rig but also about all high-level operations and activities performed by drilling crew. The work presented in this paper shed the light on analysis of hidden lost time in drilling process through automatic detection and classification of drilling operations. This paper develops a novel algorithm for detecting drilling events and operations in sensor data of drilling rig. Expectation Maximization EM and Piecewise Linear Approximation PLA algorithms applied for detecting drilling events. The Expectation Maximization algorithm performs high-level segmentation on hook-load sensor data. In addition, Piecewise Linear Approximation algorithm slices standpipe pressure; pump flow rate; rotational speed and torque of top drive motor into labeled segments low-level segmentation. Merging results from both Expectation Maximization and Piecewise Linear Approximation gives the suggested algorithm ability to detect all drilling events and activities performed by drilling rig and crew. Moreover, this paper shows the usage of discrete orthonormal basis functions Gram basis as a tool to classify drilling operations from detected segments in drilling time series. The classification process performed in cooperation with the concept of Patterns Templates Base. The optimal polynomial degree to represent drilling operations has been concluded through analysis of polynomial spectrum of each drilling operation.

A hybrid multiple classifier system for recognizing usual and unusual drilling events

Up to very recently, the applications of machine learning in the oil & gas industry were limited to using a single machine learning technique to solve problems in-hand. As the complexity of the demanded tasks being increased, the single techniques proved insufficient. This gave rise to intelligent systems that are hybrids of several machine learning techniques to solve the most challenging problems. In this paper we propose a hybrid multiple classifier approach for recognizing usual and unusual drilling events. We suggest using two different information sources namely: (1) real time data collected by sensors located around the drilling rig, and (2) daily morning reports written by drilling personnel to describe the drilling process. Text mining techniques were used to analysis the daily morning reports and to extract textual features that include keywords and phrases, whereas data mining techniques were used to analysis the sensors data and extracting statistical features. Three base classifiers were trained and combined in one ensemble to obtain better predictive performance. Experimental evaluation with real data and reports shows that the ensemble outperforms the base classifiers in every experiment, and the average classification accuracy is about 90% for usual events, and about 75% for unusual events.