WO2018029454A1

WO2018029454A1 - Method and system for analysing drilling data

Info

Publication number: WO2018029454A1
Application number: PCT/GB2017/052322
Authority: WO
Inventors: Pablo Vicente Albert MAESTRO; Thibault DAOULAS; Frode Sørmo; Krishna Srinivasan
Original assignee: Datacloud International Inc.
Priority date: 2016-08-08
Filing date: 2017-08-07
Publication date: 2018-02-15
Also published as: GB2552939A

Abstract

There is provided a method of analysing drilling data from a drilling operation in real-time, the method comprising a processor: receiving a real-time data stream of drilling data, wherein the drilling data is received from at least one sensor at the drilling operation and the drilling data is time-indexed and/or depth indexed; processing the drilling data using complex event processing, wherein the complex event processing comprises implementing one or more declarative statements on the received drilling data; and generating one or more real-time outputs in dependence on the processed drilling data, which one or more outputs aid the drilling operation. There is also provided a system for controlling a drilling operation, the system comprising: a cloud based server system configured to receive an input parameter from at least one sensor at a drilling operation, wherein the cloud based server system is remote from the drilling operation, wherein the cloud based server system comprises a processor for analysing real-time drilling data configured to perform the above method.

Description

METHOD AND SYSTEM FOR ANALYSING DRILLING DATA

FIELD The present invention relates to the drilling of oil wells. More particularly, embodiments of the invention relate to the monitoring and control processes during a drilling operation.

BACKGROUND

According to known techniques for drilling an oil well, a number of sensors are provided on the drilling rig, near the drill bit, and along a drill string that measure the hookload, torque, rpm, temperature, pressure and other relevant parameters during a drilling operation. This drilling data is provided to an operator at the surface wellhead who is controlling the drilling operation. The operator may change the drilling process given the information that is received from the sensors.

A problem with this approach is that it is expensive to employ operators who are experts at making drilling decisions as workers at a wellhead, especially when the wellhead is in a remote and inhospitable location. A solution to this problem is for the drilling data to be transmitted by satellite communication to a remote database. The remote database then provides information in real-time to skilled operators who are in a remote location from the wellhead. The skilled operators can monitor the real-time data of the drilling operation remotely and then call the operators who are actually at the wellhead if they foresee any potential problems or think that the efficiency of the drilling operation could be improved. This avoids the requirement of experts being present at an actual wellhead. However, a problem still exists in that it is inefficient for a skilled operator to be monitoring the drilling data and relaying advice to the operator at the wellhead.

It has therefore been suggested that modelling techniques may be used to facilitate the role of the skilled operator. Currently proposed approaches for detecting drilling problems automatically are machine learning techniques, such as neural networks, that learn to recognize patterns in the drilling data, statistical approaches such as Bayesian Networks, that can be manually built by an expert operator, trained or a combination of these two, and physical models that seek to simulate the physical environment of the drilling operation.

Machine learning techniques are typically divided into unsupervised and supervised approaches.

Unsupervised approaches use data to automatically build a model of the process, for instance by forming a "normality model" that can characterize new data in terms of how normal it is compared to already observed data, or clustering algorithms that can recognize that the pattern of data fit into one or more states (for instance corresponding to drilling activities such as drilling, tripping, circulating, etc.).

Supervised machine learning methods provide data that is labelled, typically manually by an expert operator, and attempts to form a model that is able to provide labelling for new data that is provided. For instance, by labelling data where a problem such as stuck pipe or tool damage occurs, the system can be trained to recognise or predict such events on data it has not yet seen.

The limitations of unsupervised machine learning to many problems is that it is difficult to direct the algorithm to the exact pattern of interest. This problem is solved by applying supervised machine learning, where the expert operator focuses the learning by providing the labelled training data. However, the more complex and subtle the pattern to be learned, the more data must be supplied. In particular, it is difficult to learn patterns where the variance over time is important. This means that for many problems in drilling, the amount of manually labelled data required to train a model is too great to be practical.

Another problem is that machine learning methods tend to express learned models as 'black boxes' in the sense that end users, and even expert operators in the field, cannot decipher how and why the model provided a particular answer. This can be acceptable in many domains (for instance, controlling the autofocus of a compact camera), but when dealing with expert users in high- impact situations, such as medical diagnosis or potentially dangerous industrial processes such as oil well drilling, a deep understanding of the decision process is necessary in order for the operators to trust the system.

A third limitation with machine learning methods is that in drilling and many other industrial processes, the physics governing the behaviour of the parameters is very well known, and is a source of knowledge for a human operator who typically draws on this heavily when interpreting the data. This knowledge is not available to the machine learning methods that have only the data to draw from. Hence, machine learning techniques are forced to learn these laws of applied physics from the data itself, which means that the machine learning techniques spend their learning potential - the amount of model complexity it is possible for a machine learning to learn from a given set of data - on quite basic physical relationships. The result of this is often underwhelming in practice, as the physics is well understood, and the machine learner fails to provide insights at the expected level.

The use of techniques based on Bayesian Networks allows a combination of learning from data and modelling knowledge about the physical systems. A main limitation with these techniques is that they require a high level of expertise in Bayesian modelling to create and maintain the models to an acceptable accuracy. In particular, it is difficult to translate expertise from petroleum engineering and drilling experts, which are used to thinking in terms of physics equations and heuristic rules, to a fundamentally probabilistic model.

Physical models seek to simulate the physical system of the drilling rig, string and the earth in order to be able to "measure" in the model what happens under different input scenarios, thus allowing it to predict the consequence of such input from the operators. This has the advantage that the causal chain between the inputs and consequence in the model is explicitly modelled, and can as such serve as an explanation for operators of why the model predicts that a certain outcome will occur under a certain circumstance. However, simplifications must be made to the model in order for it to be able to run in near real time, and these simplifications can introduce errors in the model. Modelling the complete causal chain is also a much more difficult problem than simply understanding what kind of inputs are associated with what kind of outputs, as statistical and machine learning models tend to do.

SUMMARY OF INVENTION

The present invention improves on known methods and apparatuses for analysing real-time drilling data.

In accordance with a first aspect of the present invention, there is provided a method of analysing drilling data from a drilling operation in real-time, the method comprising a processor: receiving a real-time data stream of drilling data, wherein the drilling data is time-indexed and received from at least one sensor at the drilling operation; processing the drilling data using complex event processing, wherein the complex event processing comprises implementing one or more declarative statements on the received drilling data; and generating one or more real-time time-indexed outputs in dependence on the processed drilling data, wherein the one or more outputs aid the drilling operation.

One advantage of the system of embodiments of the present invention is that it is highly scalable and a plurality of drilling rigs can be simultaneously supported.

The complex event processing is based on a number of declarative statements that, unlike prior art techniques, do not automatically learn from data, but embody a heuristic that is already understood or can be easily explained to an expert in oil well drilling. The declarative statements can be combined into complex reasoning chains in such a way that both the individual steps and overall reasoning chain can be directly understood by operators. As such, the 'black box' scenarios of the prior art are advantageously avoided. The declarative statements have in common with stasticial and machine learning methods that they do not alone or in combination form complete physical simulation models of the full causal chain between inputs from the operator and measured sensor data. Rather, they represent heuristics that extract information of particular interest to detect symptoms, predict problems or provide higher- order indicators.

For example, in order to detect the risk of stuck pipe due to cuttings accumulations, a machine learning method such as a neural network may use the sensor measurements directly, and based on a set of example of stuck pipe produce a model that associates certain progression of such measurements with the increased risk of stuck pipe. The neural network may then provide an output representing the risk of going stuck at any moment on a scale from 0 (no risk) to 1 (certain to go stuck). A physical model might approach the same problem by modelling the wellbore, drill pipe, drill bit and drilling fluid system in order to estimate how much cuttings is produced from the drill bit. Then, using a fluid model to determine the type of flow along the drill string as well as the well trajectory, estimate how much of the cuttings are carried to the surface. Warnings can be provided if more cuttings are produced than carried to the surface, or if the estimated rotational friction (torque) or along the drill pipe (drag) is higher than expected in the circumstances.

The complex event processing approach according to embodiments provides an alternative solution to the same problem by encoding similar heuristics as used by a skilled operator, such as recognizing when friction tests are performed during connections (measuring the hookload when pulling out and then in at constant speed, or rotating when not on bottom) and looking at trends in friction development that suggest that cuttings have accumulated. The advantage of these declarative statements over machine learning and statistical models is that they are highly structured in a logical manner. They can therefore be easily understood and are clear, as well as being conceptually much closer to the language and processes used by experts in the field. The advantage over physical models is that they are much less complex to create and execute, as they do not attempt to model the complete causal chain. This allows them to run comfortably in real time. The declarative statements operate on time-indexed data. This means that it is possible to express conditions such as "if the current pressure is more than 3 standard deviations above the average pressure seen in the last 20 minutes, then trigger a Pressure Spike alert" as well as "if the variance of RPM and WOB is less than 1 standard deviation of the last 10 seconds, then we are in 'Stable Drilling' state". It is then possible to build further declarative statements that build on these basic declarative statements, just as in traditional expert systems: "if we are in 'Stable Drilling' and 'Pressure Spike' alert is triggered, trigger 'Unexpected Pressure Spike'" and so on.

Each declarative statement produces at least one new output stream with a value for every time point, t. Although individual declarative statements can individually provide useful outputs, an important feature is that the output value of each declarative statement again can be used by more other declarative statements so that more complex statements can be constructed. Advantageously, with this approach complex statements can be built with refined data, so that it is accurate, and that have a high information content. For example, a complex statement may be built from hundreds of declarative statements in order to achieve a high predictive capability for a desired outcome.

Embodiments provide a wellhead operator with recommendation(s) based on the real-time outputs so that the operator knows what subsequent action to take. The recommendation(s) are automatically sent to a computer at the wellhead and displayed to an operator. The operator can then make changes to how the drilling operation is controlled in response to the feedback.

For example, an operator may receive messages that are green, yellow and red. A green message tells the operator that they are controlling the drilling operation well and should make no changes, a yellow message may provide a warning to an operator and recommend a change and a red message may order an operator to change something. The messages may be automatically displayed on a display of the operator's console/computer system and are therefore received by the operator substantially in real-time. In addition, the messages may be sent by, for example, email, SMS or instant messaging services so that they are delivered to a plurality of people, e.g. office workers, managers, etc.

The real-time data stream of drilling data may be depth-indexed instead of time- indexed. Preferably, the processor is remote from the drilling operation and/or the processor is a processor of a cloud based server system. Preferably, the one or more declarative statements in the complex event processing comprises applying a time window to the received drilling data such that the one or more outputs are only generated in dependence on drilling data received in that time window. Preferably, the method further comprises buffering the drilling data at least over the time window. The time window is preferably a predetermined and moving time period immediately preceding real-time. The time window may be a time period of two hours or less, preferably twenty minutes or less, more preferably five minutes or less, and yet more preferably 10 seconds or less.

Preferably, the one or more declarative statements in the complex event processing comprises filtering the drilling data over the time window such that the one or more outputs are dependent only on the filtered drilling data. The filtering may be in dependence on one or more conditions. A condition could be that rotation per minute is higher than 25. Alternatively or additionally, a condition could be that bit depth equals hole depth. Alternatively or additionally, a condition is that the time window is shortened immediately preceding suspension of a drilling operation. Alternatively or additionally, a condition is that the time window is shortened immediately following suspension of a drilling operation. It is particularly advantageous in these examples to shorten the time window in order to remove transient effects in the drilling operation. Preferably, the one or more declarative statements in the complex event processing comprise applying one or more functions to the filtered drilling data. The one or more function(s) may be an averaging function or a linear regression. In some examples, the one or more declarative statements in the complex event processing may comprise detecting a dysfunction in a drilling operation. The one or more generated outputs may comprise providing a recommendation for correcting the detected dysfunction in the drilling operation. The one or more declarative statements in the processing may further comprise determining whether the recommendation for correcting the detected dysfunction in the drilling operation has been implemented. Finally, the one or more declarative statements in the processing may further comprise, if the recommendation has been implemented, determining whether the recommendation has corrected the detected dysfunction in the drilling operation.

Preferably, the data stream of time-indexed real-time drilling data comprises one or more input parameters selected from the following: hole depth, bit depth, block position, standpipe pressure, differential pressure, mud flow in, rate of penetration, hook load, weight on bit, inclination and autodriller settings.

Preferably, the one or more generated outputs is one or more of the following: average rate of penetration, drilling activity, mechanical specific energy (MSE), MSE erraticity, differential pressure, differential pressure erraticity, pressure spike event, stationary drilling state, ROP erraticity, WOB erraticity and AD Dysfunction.

In a particularly preferred example, the real-time data stream of drilling data comprises rate of penetration, bit depth and hole depth, a declarative statement in the processing comprises a) applying a time window to the received drilling data such that the one or more outputs are only generated in dependence on drilling data received in that time window, the time window being a predetermined and moving time period of up to two hours immediately preceding real-time (such as ten minutes preceding real-time), b) filtering the drilling data over the time window such that the one or more outputs are dependent only on the filtered drilling data, wherein the filtering is in dependence on a condition that bit depth equals hole depth, and c) applying one or more functions to the filtered drilling data, wherein the one or more functions is an averaging function, and the one or more generated outputs comprises average rate of penetration. In some example, the method may further comprise storing the received drilling data in order to generate a historical database. The method may further comprise storing past data from another database in the historical database.

Additionally or alternatively, the method may further comprises obtaining a reference set of historical data that is measured from an actual drilling operation from a historical database, and generating one or more outputs in dependence on a direct comparison of the drilling data and the reference set of historical data.

Preferably, the method further comprises processing the drilling data by one or more neural networks, wherein the one or more real-time outputs for aiding the drilling operation are dependent on output(s) of the one or more neural networks.

Preferably, the output(s) of the one or more neural networks are dependent on the real-time data stream of drilling data.

Preferably, the output(s) of the one or more neural networks are dependent on past drilling data.

Preferably, the generated one or more real-time outputs for aiding the drilling operation are prioritised in dependence on the output(s) of the one or more neural networks.

Preferably, the processing of the drilling data using complex event processing is dependent on the output(s) of the one or more neural networks.

Additionally or alternatively, the method may further comprise performing filtering and complex event processing on the obtained reference set of historical data, wherein the complex event processing comprises implementing one or more declarative statements on the obtained reference set of historical data. Preferably, the one or more generated outputs are dependent on a direct comparison of the processed drilling data and the processed historical data. Preferably, the one or more generated outputs comprise the obtained reference set of historical data and/or a recommendation generated in dependence on the obtained reference set of historical data. Preferably, the obtained reference set of historical data is obtained in dependence on one or more contextual factors that include: geological formation of oil well, oil field, geological location, oil well profile, bit size of drill, bottom hole assembly make-up, drilling fluid properties and equipment installed on rig. In some example, the method may further comprise selecting a subset of the reference set of historical data that correlates to a desired performance improvement.

Preferably, the desired performance improvement is a desired output and preferably the desired output is a desired drilling rate of penetration.

Preferably, the method further comprises comparing the input parameter(s) associated with a current drilling operation and the input parameter(s) associated with desired performance improvement

Preferably, generating one or more outputs is further in dependence on the comparison of input parameters.

In a preferred example, the method further comprises presenting the one or more outputs to an operator of a drilling operation.

Additionally or alternatively, the method may further comprise a step of automatically controlling a drilling operation using the output of the declarative statements in the complex event processing.

In accordance with a second aspect of the present invention, there is provided a processor for analysing real-time drilling data configured to perform any one of the above methods. In accordance with a third aspect of the present invention, there is provided a system for controlling a drilling operation, the system comprising: a cloud based server system configured to receive an input parameter from at least one sensor at a drilling operation, wherein the cloud based server system is remote from the drilling operation, wherein the cloud based server system comprises a processor for analysing real-time drilling data configured to perform any one of the above methods.

Preferably, the system further comprises a buffer for storing time-indexed real- time drilling data.

Preferably, the system further comprises a remote database for storing historical data. The remote database that has received the real-time drilling data communicates the drilling data in real-time to the cloud based server system. The cloud based server system that preferably includes or is communication with a database of historical and other information performs an analysis of the drilling data and automatically generates advice for operators at a wellhead. Preferably, the system further comprising a drilling operation that, in use, generates drilling data that is said received input of cloud based server system.

In accordance with a fourth aspect of the present invention, there is provided a method for operating a drilling operation, the method comprising: generating a real-time data stream of drilling data at a drilling site, wherein the drilling data is time-indexed; sending the drilling data from the drilling site to a cloud based server system; performing the method of any one of claims 1 to 41 at the cloud based server system; and sending a recommendation, that is one more outputs generated by any one of the above methods, to an operator of the drilling operation. LIST OF FIGURES

Certain preferred embodiments of the present invention will now be described by way of example only with reference to the accompanying drawings, in which: Figure 1 shows a schematic representation of a system in accordance with an embodiment of the present invention;

Figure 2 is a graph showing oscillation in control parameters by an autodriller; Figure 3 is a graph showing no oscillation in control parameters by an autodriller; Figure 4 is a graph showing oscillation in control parameters by an autodriller together with other activities;

Figure 5 is a graph showing a complex event processing example in accordance with an embodiment of the present invention;

Figure 6a is a diagram showing a first workflow for monitoring autodriller dysfunction in accordance with an embodiment of the present invention;

Figure 6b is a diagram showing a second workflow for monitoring autodriller dysfunction in accordance with an embodiment of the present invention

Figure 7 is a graph showing depth of drilling over time for various oilwells;

Figure 8 is an example is a driller's console with benchmark intervals for rotations per minute, differential pressure and weight on bit according to an embodiment of the invention;

Figure 9 is a flow chart showing an embodiment of the method of the present invention; DESCRIPTION OF EMBODIMENTS

Figure 1 is a schematic diagram that shows the basic configuration of a system for monitoring, aiding and/or controlling a drilling operation according to embodiments. The system comprises a cloud based server system 102 configured to receive an input parameter from at least one sensor at a drilling operation 103, wherein the cloud based server system 102 is remote from the drilling operation 103, and the cloud based server system comprises a processing device for analysing real-time drilling data that is received from the drilling operation. The cloud based server system communicates with the drilling operation over network 101 , that may include satellite communications.

Figures 2, 3 and 4 illustrate situations in which streaming analytics, hereinafter used interchangeably with complex event processing (CEP), that is used to analyse real-time drilling data according to the techniques of embodiments. Streaming analytics, or CEP, is a methodology for performing pattern matching, signal processing and event detection on streams of time-indexed data. The data is time-indexed because a timestamp is attached to each set of measurements. Drilling data can for instance be produced by sensors measuring temperature, pressure, rotational speed (rpm) and/or rotational force (torque) in a drilling process or similar. While individual measurements of these sensors can be valuable, e.g. providing an alarm if the temperature or pressure exceeds a danger threshold, much more information can be gained by looking at how drilling data changes over time.

Although in embodiments drilling data may be received from a single sensor, embodiments include monitoring the correlation of two or more sensors over time. An example of this is oscillation in control parameters by an autodriller. An autodriller is an automated control loop system that controls input variables to a drilling process, such as weight on bit (WOB) and the rate that drilling fluid ("mud") is pumped through the inner pipe, out through nozzles on the bit and back up through the annulus between the pipe and the hole walls. The operator of the autodriller can set targets for input variable such as RPM, but also for indirectly controlled variable such as the stand pipe pressure, which is affected by the drilling fluid flow rate. However, it is possible to set targets that in some given situation are mutually exclusive. As higher WOB increases the friction, it tends to affect the flow rate of the drilling fluid, which again changes the pressure. Thus, if both a WOB and a pressure target are set, the autodriller may start oscillating between fulfilling the WOB and the pressure goals. These oscillations can mean inefficient drilling and can damage downhole equipment.

In order to detect such patterns, it is necessary to look at measurements from a number of sensors at once, and it is necessary to look at them over time. In Figure 2, a simple example of oscillating WOB and pressure (y-axis) and time (x- axis) is shown. Here, we can see that the pressure and WOB are out of phase, which may suggest that the autodriller is oscillating between fulfilling each one. Note that if they were in phase then that would not be a sign of oscillating between the parameters. In Figure 3, only the pressure changes over time while WOB is steady. This is not a sign of oscillation, as the autodriller is responding on one control parameter (pressure) to changing conditions. In actual measured drilling data, the measured parameters are not continuous as shown in Figures 2 and 3. During drilling operations, drilling is interrupted by other activities, such as connecting a new pipe to the string, or circulating mud to surface in order to clean the hole. During these activities, drilling is suspended, and there are transient periods before and after such activities as well. Figure 4 shows oscillations with two discontinuities, i.e. breaks in the drill for other activities. The discontinuities are shown in Figure 4 as non-drilling periods and the WOB and pressure falls during these periods. In order to detect the primary pattern, embodiments use streaming analytics to filter out these discontinuities. Figure 5 shows an example of how streaming analytics is applied to a method for analysing real-time drilling data according to embodiments. A declarative statement is created that takes the linear regression of the pressure over a five minutes window. However, a set of conditions are applied so that a measure of pressure at time t is accepted only if the conditions are valid for that time t.

The conditions are:

At t, RPM must be higher than 25.

At t, the Bit Depth and Hole Depth must be the same (i.e. the bit must be at the bottom of the well).

We see pressure, RPM, hole depth and bit depth plotted on a time chart. The time window ensures that only received data from the last five minutes is used. In the time window, the above conditions are applied so that only two short intervals of pressure data are accepted. A linear regression is then produced from these values and provided as the result of the declarative statement at time to. This declarative statement produces a regression value at every time where there is data in accordance with the above conditions and it therefore creates a time series of data that can be used in other declarative statements. According to embodiments, a set of declarative statements are created for monitoring autodriller, AD, dysfunction in drilling. An AD Dysfunction Alerting System according to embodiments consists of two workflows, one for detecting dysfunctions in the autodriller tool and giving recommendation(s) to mitigate any detected dysfunctions, and one for monitoring manual intervention on the tool, quantifying the resulting improvement and sending further recommendation(s) if necessary.

Figure 6A shows the first workflow that is constructed from a set of declarative statements designed to detect underlying symptoms of autodriller dysfunction from drilling data. The input parameters may be one or more of the following: hole depth, bit depth, block position, standpipe pressure, differential pressure, mud flow in, rate of penetration, hook load, weight on bit, inclination and autodriller settings.

Advantageously, an operator can see how each of the separate dysfunctions are detected, and which declarative statements are used for each detection. Although not shown in Figure 6A, an operator can also see how each declarative statement is implemented, i.e. its inputs, the applied filtering and conditions, the applied functions and the output generated. This is a significant advantage over neural networks or Bayesian network techniques for analysing data in which the processes performed by the analysis are not transparent to operators. The analysis just provides an output from a 'black box'. Another advantage of the construction of an analysis process from a plurality of declarative statements according to embodiments is that an analysis process can be easily designed to achieve a desired result, or goal. For example, in addition to detecting AD dysfunction, embodiments include implementing a plurality of other analysis processes in parallel. These may include analysis processes for detecting if current best practice requirements are being followed and advising an operator if changes to a drilling operation should be made in order to comply with the current best practices. A best practice may be, for example, keeping an RPM in a particular range when the temperature pressure and other conditions are at certain levels. This requirement can be easily defined and implemented in a declarative statement. If the best practice requirements changes, the declarative statement can be easily changed in a corresponding manner. This flexible control over analysis processes is not possible with known techniques such as neural networks or Bayesian network techniques that cannot flexibly adapt to such changes in desired results.

A particularly preferred feature of embodiments is that processing is performed that removes transient effects from the drilling data and this increases the accuracy of the determined characteristics of the data, such as its variance.

The inventors have realised that transient effects, in particular of the hook load measurement, occur whenever drilling is started and stopped/paused. The transient effects are shown in Figure 4 as occurring in the time periods immediately before and immediately after non-drilling time. Advantageously, the processing according to embodiments detects whenever a drilling operation is stopped, which often occurs a number of times during a drilling operation either as part of a planned process or due a fault occurring. When such a suspension of the drilling operation is detected, embodiments apply a filtering to the drilling data immediately preceding the suspension of the drilling operation so as to filter out transient effects in measured parameters that are caused by the suspension. Similarly, when such a starting or restarting of the drilling operation is detected, embodiments apply a filtering to the drilling data immediately following the starting or restarting of the drilling operation so as to filter out transient effects in measured parameters that are caused by the starting or restarting.

The effect of the filtering is to shorten the time window over which data is analysed. However, by using the data in the shortened time window accuracy is improved as transient effects are mitigated. The filtering may shorten the time window by a predetermined length of time, such at 5 seconds, or an analysis may be performed that detects the transient effects and either shortens the time window when the transient effects have fallen below an acceptable level or performs other processing to mitigate the detected transient effects. Figure 6B shows the second workflow for detecting manual changes to the autodriller once a change recommendation is sent, detecting the impact driven by human intervention and sending additional recommendation(s) if the autodriller dysfunction has not been solved.

An example of how the second workflow is implemented according to an embodiment is provided below.

After a dysfunction alert or recommendation is sent to an operator, a time window of 90 seconds starts where the following AD settings are monitored by declarative statements operating on the received drilling data:

AD Differential Pressure Setpoint

AD Differential Pressure Limit

- AD WOB Setpoint

AD WOB Limit

AD ROP Setpoint

If the value of any AD setting is differs by a predetermined amount over the time window (e.g. 5% relative difference for all non ROP settings, 10% for ROP setting), then the change is stored, a message reporting the change is sent, and a 4-min window opens where oscillations are tracked and compared to the 4 minutes prior to the alert. If a new setting change is detected within 10 sec, the 4-min monitoring window is reset.

If no change is detected 90 seconds prior to the alert and within 90 seconds after the alert was sent, an alert is sent to the operator that informs the operator that no response to the previous alert has been detected. The second alert may state: 'No action recognized for sent AD Dysfunction alert'.

After 4 min elapsed without AD setting change, oscillations before the alert and after the AD setting change are compared based on their frequency and amplitude. A score is attributed to the reduction in oscillations which is given as a percentage of improvement. If the improvement is greater than a predetermined amount, e.g. 10%, a message stating 'AD Dysfunction resolved' along with the percentage of improvement is sent to the operator. If the improvement is less than the predetermined amount, the sent message to the operator is: 'AD Dysfunction Unresolved - Try Again' is sent, and a new 90 sec window monitoring for AD setting change is opened again.

If, during the time period following an action being taken of after 1 min and before 4 min, drilling is interrupted, an extrapolation of the quantity of oscillations in the current window is made and a comparison is made using the extrapolation to calculate the improvement which can then be reported immediately.

Accordingly, embodiments include the real-time analysis of drilling data to detect problems, or determine improvements that can be made, and proving recommendation(s) to operators. Embodiments also include the analysis of drilling data subsequent to recommendation(s) being sent to determine both if the recommendations have been implemented and, if they have, if the recommendations were effective at improving the drilling operation.

As described above, embodiments use CEP processes to generate recommendations for operators of drilling operations. An advantage of the advice generated through CEP processes is that a human operator can analyse the advice and obtain an understanding of how the advice has been generated. Embodiments also include using neural network techniques, in addition to CEP processes, to provide a system that has combined advantages of both CEP and neural network processes.

Neural networks, or artificial neural networks, use real-time and historical/past data to generate logical and/or quasi-physical models of relationships between parameters in the data. They are able to automatically learn from the data and, by detecting patterns in the data, predict potential future events. Neural networks are therefore capable of automatically and very quickly determining if a high risk situation is potentially going to occur. A limitation of CEP processes is that, in some circumstances, it is necessary for a clear data trend, that has already been confirmed as correct by a human, to be detected in order for the occurrence of a potentially high risk situation to be detected.

Embodiments include performing CEP processes on real-time data to generate real-time outputs for advising an operator as described above. However, in addition to the CEP processes, real-time and/or historical data is also used as inputs to one or more neural network models. The neural network models are trained with historical data until appropriate confidence levels on their pattern recognition capabilities are obtained. The training of the neural networks is preferably by supervised learning. The output of the neural networks are risk levels of potentially dangerous or high consequence situations occurring. The risk levels are determined by the pattern recognition performed by the neural networks. The risk levels are then used to enhance the advice provided to operators. For example, the neural networks may have detected that a high risk situation may occur in advance of the CEP processes and alert the operator of the potential high risk situation. Alternatively, or in addition, the risk levels may be used to prioritise the CEP processes that are performed so that the CEP processes are directed towards confirming whether or not the potential high risk situation may occur in order to corroborate the detection by the neural networks. Alternatively, or in addition, the risk factors may be used to prioritise the advice provided to operators. For example, the CEP processes may have detected five potential high risk situations that may occur but not identified any one of these situations as being significantly more likely to occur than the others. However, pattern recognition performed by the neural networks may have detected that one of the situations is a lot more likely to occur than the others and therefore given this situation a higher risk level. Embodiments use the risk levels generated by the neural networks to alert the operator that that a specific one of the high risk situations is more likely to occur. Advantageously, the situation has been analysed by CEP processes and the CEP analysis is available for inspection, analysis and understanding by a human operator. The neural networks have just identified the CEP analysis of a specific situation as more important than the other potential situations that the CEP analysis determined could occur. Neural network techniques are therefore used to enhance the CEP processes.

Embodiments also include using benchmarking to generate advice for operators of drilling operations. Embodiments advantageously find relevant drilling data within a historical database of drilling data, process the found drilling data using the streaming analytic techniques of embodiments and then use the data to provide advice based on the historical data to operators. Benchmarking in the context of oil well drilling, is the process of using data from similar drilling operations in the past to provide a benchmark for how quickly it should be possible to drill, typically expressed in ROP (Rate of Penetration). In addition to the benchmarking information itself, information about how this was achieved, for instance by providing information about what Weight on Bit (WOB) was used, the RPM and mud flow settings. If an autodriller is used, the set- points for the autodriller can also be relevant.

For information like this to be transferable from a past drilling operation to a new one, the context of the drilling must be comparable. For instance, a drilling operation drilling the initial vertical hole with a large bit will not be comparable to the end of a long horizontal section at the end of a well that is drilled with a small bit in another geological formation.

The contextual factors that can impact the relevance of well data from the past include (but not limited to):

The geological formation.

The bit size

The bottom hole assembly make-up

· The drilling fluid

The equipment installed on the rig

The known approach to benchmarking requires building a model of the drilling operation. The various modelling methods include machine learning (e.g. neural networks, support vector machines and similar), statistics (e.g. regression models, Bayesian models) or physical modelling. The models capture how the different contextual factors affect what kind of ROP can be achieved, and sometimes what kind of parameters would achieve that ROP. However, known approaches suffer from two main shortcomings. First, the systems are a 'black- box' from the perspective of an end user such as a drilling engineer or driller. Since it is based on a generalized model learned or tuned over many wells, there is no transparency as to how that advice was reached. Physical models based on simulations of the wellbore tend to do a little better as they are able to ground their reasoning in physical explanations that may be more accessible to end users, but they still require deep expertise to interpret those explanations. It is also very difficult for expert users to determine when the advice from the system should be followed as there may be a difference in circumstances not accounted for by the model.

The second problem is that it is difficult to develop a truly global model, as new equipment, drilling fluids and configurations are brought to market continuously. The relationship between these are also very complex and hard to model generally, so a model tends to behave unexpectedly when facing new conditions and therefore requires updates.

The above problems with known techniques make it difficult to build trust in the models among end users, and they tend not to be used in practice. Embodiments provide a new approach to benchmarking. A single, or small, set of wells from historical drilling data is identified for use as a reference to the well of a current drilling operation. Big Data technologies are used to automatically compile the reference set and identify one or reference wells from that data. The current well is tracked against the reference wells in real-time, the reference set being updated when the circumstances of the current drilling operation change. Advantageously, the data used for advising operators is fine tuned as well as being closely based on data from actual drilling operations, rather than the 'black box' of model based techniques. Benchmarking according to embodiments uses a database of historical drilling data from previous well drilling operations. The data in the database is structured in a particular way that allows comparisons across wells to take place. The database of embodiments is built from two sources of data. These are:

1. Time-indexed sensor data from current drilling operations

2. Manually entered or imported data form other drilling operations, such as data on rigs, wells, runs and geological formations.

The time-indexed data is provided either in real-time as a well is drilled through API standards such as WITSO or WITSML, or as historical data logs through databases or files. This data consists of time-stamped measurements from sensors on the rig such as block position (BPOS), rotations per minute of the drill string (RPM), drilling fluid flow rate (MFI), stand pipe pressure (SPP) and so on. In addition, data from sensors in the hole near the bit may be provided, such as ECD, gamma and inclination.

The system of embodiments can receive this data from live wells in real-time, and in doing so compare it to the historical data in its database to provide benchmarking data in the way described below. In addition, data can be populated from historical sources such as log files, databases or similar. This data may be first streamed through the system data to correctly standardized and index the data in the database so that the data can be used as reference wells for benchmarking.

The process for processing data into the database is similar both for real-time and historical wells, as real-time wells are stored in the database so they can be used as reference wells for future drilling operations. The only difference is that for historical wells, data is streamed through the analysis system at a higher rate than real-time (so that it take much less than two weeks to import two weeks of data), and that advice is normally not produced for the historical wells as they have already been drilled and advice cannot change the past. However, advice may be produced for historical wells in order for testing and validation purposes. The steps taken when streaming data into the database according to embodiments are: 1. Parameter Name Standardization: Different names are used for the same parameters across the oil & gas industry. These are automatically translated to standard names so that all wells in the database use the same names for the same parameters. 2. Unit Standardization: Different units of measurement can be used in different countries, territories, operators and rigs. This process automatically standardizes the units to a common, Sl-based format.

3. Streaming Analytics Analysis: The data is enriched by analysing the data to provide higher level data, such as indexes ("MSE", ... ), recognition of drilling activities ("drilling", "sliding", "tripping in" etc), events of interest ("pressure spike", "overpuH", "Erratic ROP"). These are used to filter data for the benchmark slices, e.g. by activity. 4. Time- and depth based summarization. This calculates descriptive statistics for each parameter over certain time- and depth intervals as well as other intervals (per run, per section, per well).

5. Database storage. The standardized, analysed data is stored directly in a database, along with the time- and depth based summarization. The data is indexed both on time and (measured) depth so that it can be retrieved quickly.

In Figure 7, shows a time-depth chart that shows a comparison of four wells. The wells are plotted on a chart where the X-axis represents the time since drilling started, and the Y-axis represents the depth drilled.

The line for each well comprises sections of drilling, during which the depth of the well increases over time, and periods of time where non-drilling operations are performed, during which the hole depth is not increasing (i.e. is 'flat'). A 'flat' period can either represent an operation that is planned and required, e.g. setting casing, cementing, or a planned BHA or bit change, or it can be due to an unplanned event, such as equipment failure, early bit dulling or a stuck pipe (called Non-Productive Time, NPT). The amount of 'flat' periods combined with the speed of drilling during drilling periods determines the time it takes to reach the target depth of the well. The ideal well would be a well where planned flat time is minimized, i.e. no NPT is encountered, and the rate of drilling is as high as possible. Specific details of the benchmarking technique of the embodiments are set out below in four steps:

1 ) Identify a reference set of wells that are similar enough to the current well of a drilling operation for experience from them to be potentially relevant. For example, this can be wells drilled in the same formation with similar equipment.

2) Identify a subset of the reference wells, referred to as the benchmark wells, that are considered 'best of breed' for the current circumstances. Typically, these are wells with no, or low, NPT that were drilled quickly.

3) Identify "benchmark slices" that are relevant data from the benchmark wells, i.e. identify data from a short period of time for each well that is relevant to benchmark against in the current well under the current circumstances. Typically, this is data from the same formation and a similar location on the well profile, performing the same type of drilling activity.

4) Presenting the "benchmark slices" to the end user.

Below, is the description of each of these steps of embodiments in more detail. They assume the existence of a database of historical data about wells and an approach to capturing real-time data that is detailed later in this document.

1 Identify the reference well set This step identifies a set of wells similar enough to the current well that they may be used for comparison. This set will not only include 'good' wells, but all wells, or a representation of the population. This allows this set to be used as a reference set to compare how the current well is drilled not only to the 'best' wells, but also to see how it does in the population as a whole. In order for a well to be included in the reference set, it preferably: is drilled through the same formations, which typically mean they are drilled in the same field (or a field that is deemed to be very similar).

· has a similar well profile, i.e. it has the same number of sections, and the casing points for these sections are similar.

use the same size bit in the different sections, and have comparable bit types.

have a similar BHA make-up.

be drilled by the same or similarly equipped rig.

have similar drilling fluid properties.

the geographical location of the hole may also be used so that closer wells are preferred to wells that are further away. In the system of embodiments, the reference set can be determined either by providing this context information to the system and having it calculate a reference set, or by explicitly labelling each well with a label that identifies its reference set. A reference can be calculated because each well in the database has a data structure providing information about all the above elements. The data structure may be filled in manually or provided through third party integrations. Before the drilling of a new well is started, the same data structure is filled out for that well. The reference set is calculated by performing a comparison of the data structure of the new well with each well in the database before and/or during the drilling operation.

For each such comparison, each element in the data structure is compared to see if they are compatible. This calculation is specific for each element and depends on the detail of information available. For instance, if a complete 3- dimensional well path is available for all wells in the database as well as planned path for a new well, it is possible to use this data to calculate how close the planned trajectory is to each well in the database, providing a similarity score between the new well and each well in the database. However, if such detailed trajectory data is not available, it may be sufficient to provide a template label for each type of trajectory. In this case, the system would require that a past well have the same template as the new well for it to be included in the reference set. The result of each comparison is a compatibility value between 0 and 1 , where 1 means completely compatible and 0 is completely incompatible.

The well is included in the reference set if and only if the smallest compatibility value of the elements exceeds a threshold, and the average compatibility value exceeds a second threshold. Once this is performed for each well, a compatibility set for the new well is defined.

An alternative approach is to use a manual label attached to each well to establish reference sets. This approach is particularly useful in situations where contextual data is not formalized in a way easily accessible by computer systems, but where a human operator can determine what wells belong together in reference sets.

2. Identify Benchmark Wells This step identifies a subset of the reference wells that serve as positive examples, best of breed or benchmark wells. These are typically wells that are drilled quickly, andwithout problems leading to NPT. In theory, the benchmark wells could simply be picked by measuring the time from drilling begins until the target depth is reached, picking the five (or other number) quickest. However, it is not unusual for individual wells to have individual sections that are good, while other sections may have had problems that reduce the performance of the well as a whole. For that reason, embodiments include generating synthetic 'best wells' that combine the best sections from different wells. As such, one approach of embodiments is to only use the performance of the current well section when evaluating if a well is included in the benchmark set. This however, means that the benchmark set is not the same for each section of the well, but must be updated for each section. An implicit assumption in this process is that there is minimal dependency between well sections in the sense that actions taken in an earlier section have little impact on the performance of later sections. If it was, for instance, possible to take shortcuts or perform short-term optimizations that could lead to problems and slowdowns in later sections, they could not be de-coupled. Because each section is cased and cemented and no longer exposed to the formation after a section is complete, this is generally accepted to be the case, but there are exceptions to this rule. For instance, in some cases, the details of the placement of the kick-off point of a new section makes a big difference for how that section can be drilled. Some of these factors are possible to incorporate into the algorithm of embodiments. However the main advantage of the approach of embodiments is that advice is generated and provided from specific past wells. It is therefore possible for a human experts to validate the advice by looking into the history of those wells. This is not possible when the operator is provided with no more than the output of a 'black box' model.

There are special circumstances when embodiments generate a special benchmark set that does not necessarily use the best section as a whole. Specifically if a well section has an NPT event that necessities tripping out and changing the bit, the performance of the bit after tripping in may be better than reference wells where the trip did not happen as the bit becomes duller over time and loses performance. In these situations, the system may include well sections where a bit change had occurred at a similar depth to provide more accurate reference data, if such data exists. Similarly, if a sidetrack is necessary, data from sidetracks from the same well section and similar depths are preferred as reference data if they are available.

3. Identify benchmark slices. A benchmark slice is an interval of data from a benchmark well that is most relevant to use for reference in the current situation in the current well. This interval is typically defined according to embodiments on time or depth measured along the wellbore (measured depth).

The sensor data provided as a basis for the benchmark data may be time series data supplied at 1 to 10 second intervals, where each data point will have measurements for a set of surface and/or downhole data sensors. Vibration, measurement noise and micro physical properties of the geological formation and drilling process cause some level of variance between each measurement, it is preferable to average out these measurements over an interval of time or depth. This requires the following steps:

1. From the time and depth in the current well, identify an interval in time and depth for each well in the benchmark set. The approach of embodiments is as follows: a. Identify the length of the ideal interval. This is a parameter that may change from field to field, and are generally measured in stands (a stand is usually 90 ft length of pipe), with the minimum interval a single stand and the maximum covering all stands drilled in the formation/section. b. Identify a corresponding point in the benchmark well corresponding to the current well. For drilling operations, this would be when the same depth as measured along the hole (measured depth). For example, if a well is currently drilled at 6200 ft, the corresponding point in benchmark wells would also be at 6200 ft. Since each depth is only drilled once, this also translates to a unique time point in each well. c. The benchmark slice is established by identifying the stands around the depth point, so that it covers the number of stands specified while keeping the corresponding point as close to the centre of the interval as possible. For instance, if the corresponding point is at 6200 ft, which is in a stand that goes from 6135 to 6225 ft, and the length of the ideal interval is 9 stands, the interval would go from 5775 ft (6135 ft, the start of the stand the corresponding point is in, minus 4 stands at 90 ft) to 6585 ft (the end of the stand the corresponding point is in, plus 4 stands at 90 ft). d. If the benchmark slice is not all in the same formation that the current well is being drilled in then it is shifted up or down until it is fully in the same formation as the current well. If the formation is not large enough to allow the full interval to exist (e.g., if the interval should be nine stands, which is 900 ft but the formation is only 200 ft deep), the interval is reduced so that it does not include data from another formation. This may shift the benchmark slice so much that it does not include the depth from the current well. For instance, if a formation shift occurs at 6100 ft in a benchmark well, but at 6200 ft has not yet occurred in a current well, the benchmark slice would include an interval ending at 6100 ft. 2. Apply a filter to the interval identified to remove data associated with activities that are not relevant to the current activity in the active well. For instance, if a drilling operation is currently drilling with a downhole, mud motor only ("sliding"), data from drilling using a top drive, connecting new pipe, circulating and other activities are not relevant in order to provide benchmark information.

3. Combine the benchmark data for those presentations that require a compact form. The interval will typically contain thousands of rows of data for each well, which makes it impractical to convey the complete dataset to an end user. The goal of benchmarking is also to establish the macro picture, making local variations less important. However, some idea of the distribution of values is important, so a compaction here includes descriptive statistics of the distribution shape (variance, central moments, Px-scores) in addition to the average values.

Note that while embodiments include the above calculations being performed on demand in real-time during a drilling operation of a current well, embodiments preferably perform these calculations ahead of time by calculating them for every stand in the benchmark well and for every relevant drilling activity and storing them in a depth-index database table.

4. Present "benchmark slices" to the user

There are three user groups that have different requirements attached to viewing the benchmark slice data. These are the driller, the engineer and the manager.

The driller controls a single drilling operation in real-time. The driller is tasked with taking the right decisions moment to moment, to drill as quickly and safely as possible and to follow best practice. The driller needs short, concise messages in text, continuous indexes on dials and/or information as red/yellow/green type traffic lights. A driller's console according to embodiments preferably has virtual dials or number displays that show the current values for parameters such as ROP, WOB, RPM, MFI, SPP etc. These allow secondary values to be visible as background intervals or similar, as well as an area for short textual updates. In Figure 8, a simplified example of driller's console according to an embodiment with Benchmark intervals or RPM, Diff. Pressure and WOB is shown.

In this illustration, a single benchmark well is selected that is the closest analogy to the current well, and an interval is shown for each of ROP, WOB and SPP. These intervals are taken from the benchmark slice of data, with the interval defined as one standard deviation around the mean. The textual window shows the name, depth and formation of the benchmark well these values are from.

The text message can change to show direct advice if the driller deviates from the benchmark values over a period time, e.g. to "Experience on well X suggest that ROP may safely increased by 10% by increasing WOB 8%."

The engineer is not in moment to moment in control of the operation but is close to the operation and has a deep understanding of the physics of drilling the well. The engineer performs more in-depth analysis when required, but is not necessarily watching the well continuously. Embodiments provide engineers with the data that they need and also alerts to notify them if a well is needs their attention, and the tools to dig deep into the problem if it does.

Engineers typically use standard personal computers and are accustomed to viewing drilling data on time and depth charts. For the engineer, more transparency on exactly what wells are in the benchmark set is required, as well as what the current benchmark slice is. It is also necessary for them to be able to see what the state of this was previously, e.g. what the benchmark wells and benchmark slices were two hours ago.

Embodiments provide the engineer with a user interface that shows information from all reference wells rather than just the benchmark wells, as it is helpful and interesting to determine the difference between successful and unsuccessful wells at certain points along the well path. For instance, a scatter plot showing ROP on the X-axis and WOB on the Y-axis, where each dot on the plot is the average for a reference well for a current depth, with the colour of the dot corresponding to the outcome of that run.

The manager is responsible for drilling operations overall. The manager typically wants the high level picture, such as how current wells are performing relative to past benchmark wells overall, KPIs that measure performance trends over time and how well staff adhere to best practise.

For the real-time view of the current situation, embodiments present the manager with a time/depth chart where the time dimension is not absolute time, but relative time starting from the time drilling starts. Assuming that all current wells in a single chart share a set of reference wells, a line shows the calculated synthetic best well, real best well, and the mean (P50) well for comparison.

The metrics and KPIs include:

Time to drill a well (with trend lines to show if this is increasing or decreasing) Variance in time to drill each well (with trend lines)

Amount of time spent during drilling operations spent inside the recommended benchmark settings. Embodiments also include using neural networks in addition to the above- described benchmarking techniques. The use of neural networks enhances the above-described techniques. For example, using corresponding techniques to those described earlier in the present document, the outputs from the neural networks can be used to prioritise the above-described processes and advice provided to operators.

Figure 9 shows a flowchart of a method according to an embodiment.

In step 901 , the process begins.

In step 903, a real-time data stream of drilling data is received, wherein the drilling data is received from at least one sensor at the drilling operation and the drilling data is time-indexed and/or depth indexed In step 905, the drilling data in processed using complex event processing, wherein the complex event processing comprises implementing one or more declarative statements on the received drilling data.

In step 907, one or more real-time outputs are generated in dependence on the processed drilling data, wherein the one or more outputs aid the drilling operation.

In step 909, the process ends. The flow charts and descriptions thereof herein should not be understood to prescribe a fixed order of performing the method steps described therein. Rather, the method steps may be performed in any order that is practicable. Although the present invention has been described in connection with specific exemplary embodiments, it should be understood that various changes, substitutions, and alterations apparent to those skilled in the art can be made to the disclosed embodiments without departing from the spirit and scope of the invention as set forth in the appended claims. In particular, the specific examples discussed above relate to the analysis of drilling data from an oil well drilling operation. Embodiments include the method being used for other drilling operations, such as the drilling for water or any other resource. Embodiments also include applying the above-described techniques in the mining industry and embodiments include all of the references to drilling operations throughout the present document alternatively being references to mining operations. For mining operations, measurement of parameters such as the hardness of the rock can be used to determine topographical information. The processes may be performed on real-time data only, on real-time and historical data or on historical data only.

Embodiments also include the use of the disclosed techniques in streaming analytics (or complex event processing) and benchmarking in other applications, such as generating advice for medical practitioners during an operation and generating financial advice from a real-time analysis of financial data.

Methods and processes described herein can be embodied as code (e.g., software code) and/or data. Such code and data can be stored on one or more computer-readable media, which may include any device or medium that can store code and/or data for use by a computer system. When a computer system reads and executes the code and/or data stored on a computer-readable medium, the computer system performs the methods and processes embodied as data structures and code stored within the computer-readable storage medium. In certain embodiments, one or more of the steps of the methods and processes described herein can be performed by a processor (e.g., a processor of a computer system or data storage system). It should be appreciated by those skilled in the art that computer-readable media include removable and non-removable structures/devices that can be used for storage of information, such as computer-readable instructions, data structures, program modules, and other data used by a computing system/environment. A computer-readable medium includes, but is not limited to, volatile memory such as random access memories (RAM, DRAM, SRAM); and non-volatile memory such as flash memory, various read-only-memories (ROM, PROM, EPROM, EEPROM), magnetic and ferromagnetic/ferroelectric memories (MRAM, FeRAM), and magnetic and optical storage devices (hard drives, magnetic tape, CDs, DVDs); network devices; or other media now known or later developed that is capable of storing computer-readable information/data. Computer-readable media should not be construed or interpreted to include any propagating signals.

The following is a clause indicating a preferred aspect according to the present disclosure.

1. A method of processing drilling data from a drilling operation in real-time, the method comprising a processor:

receiving a real-time data stream of drilling data, wherein the drilling data is time-indexed and received from at least one sensor at the drilling operation;

processing the drilling data using complex event processing, wherein the complex event processing comprises implementing one or more declarative statements on the received drilling data;

obtaining a historical drilling data, that is measured from one or more actual drilling operations, from a historical database; and

generating one or more outputs in dependence on a comparison of the processed drilling data and the historical drilling data.

Claims

A method of analysing drilling data from a drilling operation in real-time, the method comprising a processor:

receiving a real-time data stream of drilling data, wherein the drilling data is received from at least one sensor at the drilling operation and the drilling data is time-indexed and/or depth indexed;

processing the drilling data using complex event processing, wherein the complex event processing comprises implementing one or more declarative statements on the received drilling data; and

generating one or more real-time outputs in dependence on the processed drilling data, wherein the one or more outputs aid the drilling operation.

A method according to claim 1 , wherein the real-time data stream of drilling data is only time-indexed.

A method according to claim 2, wherein the processor is remote from the drilling operation.

4. A method according to claim 3, wherein the processor is a processor of a cloud based server system.

5. A method according to any preceding claim, wherein implementing the one or more declarative statements in the complex event processing comprises applying a time window to the received drilling data such that the one or more outputs are only generated in dependence on drilling data received in that time window.

6. A method according to claim 5, wherein the method further comprises buffering the drilling data at least over the time window.

7. A method according to claim 5 or 6, wherein the time window is a predetermined and moving time period immediately preceding real-time.

8. A method according to any one of claims 5 to 7, wherein the time window is a time period of two hours or less, preferably twenty minutes or less, more preferably five minutes or less, and yet more preferably 10 seconds or less.

9. A method according to any one of claims 5 to 8, wherein implementing the one or more declarative statements in the complex event processing comprises filtering the drilling data over the time window such that the one or more outputs are dependent only on the filtered drilling data.

10. A method according to claim 9, wherein the filtering is in dependence on one or more conditions.

1 1. A method according to claim 10, wherein a condition is that bit depth equals hole depth.

12. A method according to claim 10 or 11 , wherein a condition is that rotation per minute is higher than 25.IFSI]

13. A method according to any one of claims 10 to 12, the method comprising detecting the suspension of a drilling operation; and applying a filtering to drilling data immediately preceding the suspension of the drilling operation so as to filter out transient effects.

14. A method according to any one of claims 10 to 13, the method comprising detecting the starting, or restarting, of a drilling operation; and applying a filtering to drilling data immediately following the starting, or restarting, of the drilling operation so as to filter out transient effects.

15. A method according to any one of claims 9 to 14, wherein implementing the one or more declarative statements in the complex event processing comprises applying one or more functions to the filtered drilling data.

16. A method according to claim 15, wherein the one or more function(s) is an averaging function or a linear regression.

17. A method according to any preceding claim, wherein the one or more declarative statements in the complex event processing detect a dysfunction in a drilling operation.

18. A method according to claim 17, wherein the one or more generated outputs comprise providing a recommendation for correcting the detected dysfunction in the drilling operation.

19. A method according to claim 18, wherein the one or more declarative statements in the processing further determine whether the recommendation for correcting the detected dysfunction in the drilling operation has been implemented.

20. A method according to claim 19, wherein, if it is determined that the recommendation has been implemented, the one or more declarative statements in the processing further determine whether the recommendation has corrected the detected dysfunction in the drilling operation.

21. A method according to any preceding claim, wherein the data stream of time-indexed real-time drilling data comprises one or more input parameters selected from the following: hole depth, bit depth, block position, standpipe pressure, differential pressure, mud flow in, rate of penetration, hook load, weight on bit, inclination and autodriller settings.

22. A method according to any preceding claim, wherein the one or more generated outputs includes one or more of average rate of penetration, drilling activity, mechanical specific energy (MSE), MSE erraticity, differential pressure, differential pressure erraticity, pressure spike event, stationary drilling state, ROP erraticity, WOB erraticity and AD Dysfunction.

23. A method according to any preceding claim, wherein the real-time data stream of drilling data comprises rate of penetration, bit depth and hole depth, and implementing a declarative statement comprises a) applying a time window to the received drilling data such that the one or more outputs are only generated in dependence on drilling data received in that time window, wherein the time window is a predetermined and moving time period of ten minutes immediately preceding real-time, b) filtering the drilling data over the time window such that the one or more outputs are dependent only on the filtered drilling data, wherein the filtering is in dependence on a condition that bit depth equals hole depth, and c) applying one or more functions to the filtered drilling data, wherein the one or more functions is an averaging function, and wherein the one or more generated outputs comprises average rate of penetration.

24. A method according to any preceding claim, wherein the method further comprises storing the received drilling data in order to generate a historical database.

25. A method according to claim 24, wherein the method further comprises storing past data from another database in the historical database.

26. The method according to any preceding claim, the method further comprising processing the drilling data by one or more neural networks, wherein the one or more real-time outputs for aiding the drilling operation are dependent on output(s) of the one or more neural networks.

27. The method according to claim 26, wherein the output(s) of the one or more neural networks are dependent on the real-time data stream of drilling data.

28. The method according to claim 26 or 27, wherein the output(s) of the one or more neural networks are dependent on past drilling data.

29. The method according to any of claims 26 to 28, wherein the generated one or more real-time outputs for aiding the drilling operation are prioritised in dependence on the output(s) of the one or more neural networks.

30. The method according to any of claims 26 to 29, wherein the processing of the drilling data using complex event processing is dependent on the output(s) of the one or more neural networks.

31. A method according to any preceding claim, wherein the method further comprises obtaining, from a historical database, a reference set of historical data that is measured from an actual drilling operation, and generating one or more outputs in dependence on a direct comparison of the drilling data and the reference set of historical data.

32. A method according to claim 31 , wherein the method further comprises performing filtering and/or complex event processing on the obtained reference set of historical data, wherein the complex event processing comprises implementing one or more declarative statements according to any preceding claim on the obtained reference set of historical data.

33. A method according to claim 32, wherein the one or more generated outputs are dependent on a direct comparison of the processed drilling data and the processed historical data.

34. A method according to claim 32 or 33, wherein the one or more generated outputs comprise the obtained reference set of historical data and/or a recommendation generated in dependence on the obtained reference set of historical data.

35. A method according to any one of claims 31 to 34, wherein the obtained reference set of historical data is obtained in dependence on one or more contextual factors that include: geological formation of oil well, oil field, geological location, oil well profile, bit size of drill, bottom hole assembly make-up, drilling fluid properties and equipment installed on rig.

36. A method according to claim 35, wherein the method further comprises selecting a subset of the reference set of historical data that correlates to a desired performance improvement.

37. A method according to claim 36, wherein the desired performance improvement is a desired output and preferably the desired output is a desired drilling rate of penetration.

38. A method according to claim 37, wherein the method further comprises comparing the input parameter(s) associated with a current drilling operation and the input parameter(s) associated with desired performance improvement

39. A method according to claim 38, wherein generating one or more outputs is further in dependence on the comparison of input parameters.

40. A method according to any preceding claim, wherein the method further comprises presenting the one or more outputs to an operator of a drilling operation.

41. A method according to any preceding claim, wherein the method further comprises automatically controlling the drilling operation in dependence on the generated one or more real-time outputs.

42. A processor for analysing real-time drilling data, wherein the processor is configured to perform the method of any of claims 1 to 41.

43. A system for controlling a drilling operation, the system comprising:

a cloud based server system configured to receive an input from at least one sensor at a drilling operation, wherein the cloud based server system is remote from the drilling operation, wherein the cloud based server system comprises a processor for analysing real-time drilling data that is configured to perform the method of any of claims 1 to 41.

44. A system according to claim 43, wherein the system further comprises a buffer for storing time-indexed real-time drilling data.

45. A system according to claim 43 or 44, wherein the system further comprises a database for storing historical drilling data.

46. A system according to any of claims 43 to 45, the system further comprising a drilling operation that, in use, generates drilling data that is said received input of cloud based server system.

47. A method for operating a drilling operation, the method comprising:

generating a real-time data stream of drilling data at a drilling operation, wherein the drilling data is time-indexed;

sending the drilling data from the drilling operation to a cloud based server system;

performing the method of any one of claims 1 to 41 at the cloud based server system; and

sending a recommendation, that is one more outputs generated by the method of any one of claims 1 to 41 , to an operator of the drilling operation.