Zeroing Weight on Bit (WOB) Using Machine Learning

Abstract

Technology is described for determining a point for zeroing a weight on bit (WOB) measurement for a drilling rig. The method can include receiving a plurality of drilling features from sensors associated with the drilling rig. Another operation may be determining a point in a drilling procedure where the weight on bit (WOB) for an off bottom rotating weight, is to be zeroed, by detecting a pattern for drilling features while processing the drilling features. The WOB for the drilling rig at the point may then be zeroed. A WOB may be computed, when a drill of the drilling rig is in contact with a downhole object, based in part on weight added after the WOB has been zeroed for the off bottom rotating weight.

Description

BACKGROUND

[0001] Hydrocarbon producing wellbores extend into the earth's surface and intersect subterranean formations where hydrocarbons are trapped. Casing and tubing that may be added to the wellbores may provide conduits for the hydrocarbons to be brought to the surface. Drill bits may be used to form the wellbores, and the drill bits may be mounted on the ends of drillstrings, as part of a drilling rig. Motorized drive systems on the surface rotate the drillstrings and bits that can drill the rock. Cutting elements on the drill bit may rotate and scrape the bottom of the wellbore and grind away material to deepen the wellbore. Drilling fluid is pumped down the drillstring and directed from the drill bit into the wellbore. The drilling fluid may flow back up the wellbore in the area between the drillstring and walls of the wellbore. A similar process may be used for drilling wells for water too.

[0002] The amount of weight or force applied to the drill bit during drilling, may be referred to as weight on bit (“WOB”), and the amount of WOB affects drilling performance and tool life. Applying an insufficient WOB often reduces the rock penetration rate and increases bit vibration. In contrast, applying excessive WOB can cause mechanical bit failure, by inducing stick slip and increasing shock and vibration experienced by the downhole tools. In this case, energy applied to the system is translated to the drillstring and not to increase Rate of Penetration (ROP). The force exerted by the drilling rig for holding the drillstring is commonly referred to as the hook load. In other words, hook load is the weight of the drillstring in air, the drill collars and any ancillary equipment, reduced by any force that tends to reduce that weight. Some forces that might reduce the weight include friction along the wellbore wall and, importantly, buoyant forces on the drillstring caused by its immersion in drilling fluid. A sensor exists on the deadline that measures how heavy the entire drillstring is from the traveling block. Hook load can change every single stand because more pipe is being added to the drillstring, and depending on the well type or drilling progress, inclination may be added that can increase friction in the system. When the drill bit is on bottom, then extra weight is being applied to cut the rock.

[0003] Traditionally, WOB measurements are based on a difference in hook load between a bit's off bottom rotating weight and on bottom rotating weight. More specifically, when a portion of the hanging drillstring weight is supported by the bit resting on the bottom of the wellbore, hook load is reduced by that amount. This difference between current hook load and a “zero” value is taken as a reference for the amount of weight put on the bit. A zero value is typically obtained by measuring the hook load while suspending the drillstring in the wellbore, and without the drillstring being supported on the bottom. Because the drillstring weight changes as drill pipe segments are added to the drillstring, correctly applying a designated WOB means that the zero weight can be constantly monitored. Terms “zero” or “zeroing” may also be considered equivalent to taring, resetting of the weight on the bit (WOB) or similar terms, as may be used interchangeably in the drilling industry.

[0004] Weight on bit (WOB), as defined in the oil industry, is the amount of downward force exerted on the drill bit and may be measured in thousands of pounds. Weight on bit may be provided by drill collars or stands of pipe, which are thick-walled tubular pieces machined from solid bars of steel, usually plain carbon steel. Gravity and the downward force applied by a topdrive can act on the large mass of the collars to provide the downward force needed for the drill bit to efficiently break the rock. To accurately control the amount of force applied to the bit, the driller may monitor the surface weight measured while the bit is just off the bottom of the wellbore. Next, the drillstring and the drill bit may be slowly and carefully lowered until they touch bottom. After that point, as the driller continues to lower the top of the drillstring, more and more weight is applied to the bit, and correspondingly less weight is measured as hook load. If the surface measurement shows 20,000 pounds less weight than with the bit off bottom, then there should be 20,000 pounds force on the bit (in a vertical hole). Such WOB measurements may be affected by changing forces due to changes in inclination or azimuth while drilling. Some downhole Measurement While Drilling (MWD) sensors can measure downhole Weight on Bit more accurately and transmit the data to the surface.BRIEF DESCRIPTION OF THE DRAWINGS

[0005] FIG. 1 is a schematic diagram illustrating an example of a drilling system.

[0006] FIG. 2 is a block diagram illustrating aspects of a method for determining a point for zeroing a weight on bit (WOB) measurement for a drilling rig.

[0007] FIG. 3 is a flowchart illustrating a method for determining a point for zeroing a weight on bit (WOB) measurement of a drilling rig.DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] Reference will now be made to the examples illustrated in the drawings, and specific language will be used herein to describe the same. It will nevertheless be understood that no limitation of the scope of the technology is thereby intended. Alterations and further modifications of the features illustrated herein, and additional applications of the examples as illustrated herein, which would occur to one skilled in the relevant art and having possession of this disclosure, are to be considered within the scope of the description.

[0009] Drillers who are drilling hydrocarbon wells or other wells desire to have the highest possible on-bottom performance and rotary Rate of Penetration (ROP) when drilling the wells. ROP may be dictated by how much energy is put into the drilling system. ROP may be analyzed by looking at a current ROP, weight on bit (WOB), differential pressure(s), rotary speeds, and torque. As part of this process, drillers may desire to understand current drilling parameters and how that affects ROP. To optimize ROP, drillers may need to modify available setpoints (i.e., a preset value that is the desired value of a drilling parameter) to adjust for drilling conditions.

[0010] Most drilling rigs have an auto driller or a computerized controller for assisting with managing and controlling the drilling process. For example, the auto driller may use electronic controls to manage weight on bit (WOB), differential pressure (DP), ROP, etc., and ROP, WOB, or DP can be utilized as the primary parameters to manage drawworks speed output. If a WOB is desired to be consistent, then the auto driller can utilize the specified WOB Setpoint to set the drawworks speed to be a certain value N to yield the specified WOB (e.g., 30). The drawworks speed may be based on a setting of N or another value.

[0011] WOB may be a subjective calculation but it may be calculated as hook load on bottom vs. hook load off bottom. WOB may represent how much of the weight is actually being applied to the bit vs. how much of the weight is just the drillstring weight within the wellbore. The objective is to measure the energy being put in the system in order to break rock.

[0012] Weight on Bit (WOB) may be a calculated metric defined as the force being applied to the bit to break rock (WOB=Total Weight on Bottom−Weight of the drillstring−friction factors). When drilling, it is valuable to have an accurate and consistent trend in WOB as the system drills to ensure the appropriate amount of force is being used to break the rock.

[0013] Weight on bit (WOB) is an important metric derived from sensor data, specifically hook load, and can be used in optimizing drilling performance. However, the accuracy of WOB measurements relies on timely zeroing, and the measurements are often impacted by inconsistent zeroing methods. For example, such zeroing may introduce + / −10,000 lbf (pounds of force) of error in the computed WOB. Factors like incorrect drilling parameter optimization and unreliable data collection may compound this issue. Accurate WOB information enhances drilling efficiency and is a top-requested feature by operators of a drilling rig.

[0014] In the past, drillers may have been manually ‘zeroing’ the WOB. The zeroing process is not standardized when performed manually and can vary by operator, company, driller, etc. In other words, the human driller decides at what point to zero the WOB. Utilizing a Machine Learning (ML) model or Artificial Intelligence (AI) model to identify and automatically know when to ‘zero’ WOB may provide advantages over manual zeroing. Specifically, the zeroing of the WOB when off bottom and rotating using a machine learning model or pattern matching detected point may result in zeroing at a reliable point in the drill hole or a reliable point in time when appropriate downhole conditions exist. An example of appropriate conditions to detect a zero point within a given window may include one or more of the following conditions: the drill pipe is out of slips, a stable rotary speed, a stable hookload, a stable pump pressure, the off bottom pressure is within a tolerance of the previous stands off bottom pressure, the bit depth is off bottom, no oscillating system is on, if a dynamic zeroing point is being detected then there is a consistent running speed, and / or if a static zeroing point is being detected then there is 0 running speed. Other parameters that represent appropriate conditions to detect a zero point may also be used.

[0015] To improve the accuracy of the WOB, machine learning or AI technology can be used for automated zeroing of the WOB for an off bottom rotating weight (i.e., without user intervention). This technology may use an electronic drilling recorder (EDR), downhole measurements while drilling (MWD) data, surface sensor measurements, along with current WOB values and the error in the Weight on Bit as input data to train a ML model or AI model. The model may be trained to reduce the error which will eventually result in correcting the WOB and automating the zeroing process.

[0016] This technology can use predictive WOB tracking using available drilling features (i.e., parameters) with machine learning techniques to automate the zeroing of weight on bit. The technology and process may use an ML model to automatically zero WOB when appropriate. This removes the dependency on the human driller or manual configuration of automation recipes for zeroing the WOB. Using machine learning, WOB can be a more accurate measurement that the human driller can utilize to optimize ROP.

[0017] Machine learning models and related techniques can be used to predict the deviation of WOB from values predicted by a machine learning algorithm in real time. When the deviation is more than an allowed threshold, the system can use a corrective procedure and automatically zero the WOB.

[0018] Currently, in the absence of automation for WOB (weight on bit) zeroing, the WOB is being zeroed inconsistently or incorrectly and the human error can introduce + / −10,000 lbf errors in the WOB. Down the line calculations which are important for drilling performance like MSE, ROP optimization techniques rely heavily on the WOB (weight on bit) accuracy, and when the WOB is not accurate, this adversely affects the overall drilling performance.

[0019] An example of a drilling system 110 is shown in a side view in FIG. 1. The drilling system 110 may be used for forming a wellbore 112 through a rock formation 114. The drilling system 110 may include a drillstring 116 disposed within wellbore 112, and the drillstring may be made up of segments of drill pipe 118. In one example, the segments of drill pipe 118 may be coupled to one another (e.g., using threading or another method). A drill bit 120 is shown mounted on a lower end of drillstring 116, and the drill bit may include a drill bit body 122 that mounts on the lowest drill pipe 118 of the drillstring 116. Inserts or cutters 124 may be on a surface of drill bit body 122 opposite from where the drill bit body 122 attaches to the drillstring 116. When the drillstring 116 and bit 120 are rotated, the cutters 124 may grind or crush the rock making up the formation 114 thereby forming borehole or wellbore 112.

[0020] Above the wellbore 112, a derrick 126 may be mounted on a surface 128 or the ground. Equipment as part of the drilling system 110 may be provided for manipulating the drillstring 116, which includes a drawworks 130. The drawworks 130 may selectively pull or release a cable 132 shown engaging sheaves 134 that are rotatably mounted on an upper end of the derrick 126. Additional cables may run through the sheaves 134, and support a traveling block 136, that in conjunction with a hook 138 and swivel 140, couple with the drillstring 116 for raising and lowering the drillstring 116.

[0021] In some examples, a kelly 142 can axially couple to a lower end of swivel 140 and is rotatable with respect to swivel 140. A lower end of the kelly 142 projects through a rotary table 144, which engages outer surfaces of the kelly 142 and rotates to exert a rotational force onto drillstring 116. The rotary table 144 is formed on a rig floor 146 that attaches to derrick 126 and is set above surface 128. Below the rig floor 146 and at surface 128 is a wellhead housing 148 that is mounted in the opening of wellbore 112. On top of the wellhead housing 148 is a blowout preventer (“BOP”) 152.

[0022] Further shown on surface 128 are stands of pipe 154 that are supported by a rack 156 illustrated on one of the side beams of derrick 126. On the rig floor 146, a driller's console 158 with gauges or electronic readouts may be provided to report downhole conditions, and controls for operating the drilling assembly, such as the drawworks 130. A controller 160 is logically illustrated as being in communication with the console 158. Communications between the controller 160 and console 158 can be wireless, fiber optic, or wired. Embodiments may exist wherein the controller 160 is included within the console 158.

[0023] The weight on bit (“WOB”) exerted by drillstring 116 on the bottom of wellbore 112 can be controlled by an operator on the rig floor 146 and in conjunction with the console 158. Operators can adjust the drawworks 130 so that an upward force on drillstring 116 can be exerted on traveling block 136, hook 138, swivel 140, and kelly 142. These functions may be from software commands from the controller 160. In one example, WOB is estimated based on a hook load, which is the axial force exerted on hook 138, or other components that provide an axial supporting force for drillstring 116. Sensors (not shown) can provide a signal that when viewed at a console 158 represents the axial load by which drillstring 116 is supported by the remaining portions of the drilling system 110, i.e. the hook load.

[0024] It is noted that while the above described drilling system 110 employing a kelly-type drive system may be used with the technology described herein, skilled artisans will recognize after reading the entire disclosure that other drilling systems, particularly other drive systems or mechanisms, can be used with the technology described herein. For instance, in another example, the drilling system 110 can instead comprise a top drive type of drive mechanism. In still other examples, a rack and pinion type of drive mechanism, or other known drilling components operable with the drilling system 110 can be used.

[0025] FIG. 2 is a block diagram illustrating aspects of a method for determining a point for zeroing a weight on bit (WOB) measurement for a drilling rig. A plurality of drilling features may be received from sensors 210 associated with the drilling rig. The drilling features may be received from sensors that are above ground such as a hook load sensor, etc. In addition, the drilling features may be features from a downhole sensor(s) that collects downhole and / or WOB data. These features may be used as input for a detection process to determine a WOB.

[0026] A WOB at a point during drilling where the bit is not contacting rock may be determined based on an off-bottom rotating hook load. The WOB may be the free hanging weight of the bottom hole assembly (BHA) while rotary conditions exist. The rotary conditions may include rotation of the drill bit, pressure, and / or hook load. Hook load may be a measurement from a surface sensor representing the hook load at the time the sensor is read. The WOB value while off bottom is expected to be zero if the zeroing point is appropriate.

[0027] A point in a drilling procedure where the weight on bit (WOB) for an off bottom rotating weight is to be zeroed 212 may be determined by detecting a pattern for drilling features while processing the drilling features using a machine learning model. The WOB may be zeroed for the drilling rig at the point. The zeroing can reset the WOB at the detected point. Thus, the hook load zeroing point can be used to calculate the WOB for an on bottom rotating weight. The hook load may be used to calculate the WOB. One sensor that may be used in the drilling process is a hook load sensor and WOB can be measured based in part from the hook load. Other sensors or sensor types may be used in calculating the WOB.

[0028] Detecting a pattern for drilling features may be performed using one of a variety of detection methods or machine learning models 214. For example, types of models that may be used may include at least one of: a deep neural network, transformers, a convolutional neural network (CNN), a recurrent neural network (RNN), a genetic model, a UBoost model, regression, reinforcement learning, supervised machine learning, unsupervised machine learning, classification models, predictive modeling, pattern recognition, expert rules, condition based rules, or a heuristic to detect when the WOB is to be zeroed. For example, a genetic model may be used to try to determine the patterns in the features that provide a better zero point for the WOB and then a zeroing may be triggered when the desired pattern is identified. In another example, UBoost may use a series of machine learning classifiers to identifying a useful zeroing point. In addition, other models may be used that can process temporal, event based, sensor obtained, high frequency data from multiple data sources using any combination of ML, heuristic, and / or physics based techniques to identify the accurate WOB zeroing point.

[0029] One useful example machine learning model for zeroing is an RNN. The RNN may receive a time series of the drilling features as input features to the RNN. RNNs are able to process a sequence of data over time by storing previous data points in memory, and thereby, the data that is time sampled during drilling can be processed by the RNN. For example, a variable window length for drilling features input to the RNN may span a range of minutes (e.g., between 1 minute and 10 minutes) which may be useful where the off bottom duration is quite variable. Multiple samples for a feature from a sensor maybe included a time window. An RNN can provide time series predictions with a moving window as described.

[0030] A machine learning model that is used may be trained using a training data set of the drilling features that are known to represent a desired weight on bit (WOB) for an off bottom rotating weight. This training data set may be a ground truth to train a machine learning model or AI model and the training data set may train the machine learning with patterns representing where the WOB for an off bottom rotating position is to be auto zeroed out. For example, when appropriate downhole conditions exist, there may be a specific rate of change (e.g., wobble) for the surface weight on bit which suggests the pattern for zeroing out.

[0031] The models or machine learning model can use a feedback loop and scan the sensor channels for the drilling rig. Feedback may be used to train the models. For example, the machine learning model may be trained or learn based on the down hole data that is collected from the drilling rig. Down hole data can represent a ground truth about the downhole environment. In addition to what the model learns during the training process, the model may also be customized to the well, downhole environment and / or drilling rig which the model is being used on. Each of the rigs and wells may be unique depending the downhole environment, rig differences and related parameters.

[0032] The deep neural network can also be trained on data patterns that are obtained from the surface sensors. The deep neural network or machine learning model can also be trained using any parameters measurable or computable during the drilling process. These other parameters may include additional sensor readings, in addition to hook load, that contribute to actual WOB such as: pump pressures, Surface RPMs (revolutions per minute), downhole RPMs, and other parameters. Multiple internal features for the drilling rig may be used within each multidimensional input to a deep neural network (e.g., RNN).

[0033] In one embodiment, the machine learning models may be trained on data from multiple wells, wells from different regions, drilling on different rock formations and sections. The machine learning model may be created with as generic of a model as possible so that it may be used with many types of drilling rigs in many different location types. Downhole data may be obtained from drilling companies across a variety of wells. This data from many drillers can be used to check how accurate the model is, to train the model(s) and to continuously improve the model(s).

[0034] The downhole data for a specific drilled hole can be used as feedback into the model where the model is not performing as well as desired. The downhole data can be used to help re-label the zero points and where the model should have zeroed. Then if the surface data is used in detecting the zero point, the downhole data may have been used to train the model regarding the actual zero point selected.

[0035] As discussed, there are a number of different features (i.e., parameters) that may be used in training the machine learning models. Examples of these features may include: downhole RPM (rotations per minute), rate of penetration (ROP), downhole pressures, downhole WOB, downhole differential pressures, surface weight on bit, surface differential pressures, depth of the bit, a depth of the well, a block position, geological formation information or service data such as inclinations and azimuth. Calculated features may also be used with the machine learning model. The calculated features may be based on the time based features. This technology may use calculated features and the time based calculated features. For example, a time based feature may represent the differential pressure after 1 minute or 5 minutes. The RNN may create thousands of its own internal features, as is known in RNN networks.

[0036] This technology can automate the zeroing process and avoid a human driller having to manually determine the zeroing point. Automatically detecting a zeroing point is useful because a human can become distracted or not be able to detect a good zeroing point for the drilling rig. Thus, having a recognition process or machine learning model that is part of the zeroing process may provide reliable detection of a zeroing point. The system can be trained to anticipate a point when zeroing of the WOB, when off bottom, is desired to occur.

[0037] This technology also avoids the situation where drillers can hide weight by zeroing late (e.g., when 10,000 lbs. is already applied to the bit). In this situation, the driller may or may not know that when the instruments are reading 20,000 lbf it is really 30,000 lbf of WOB. Drillers and operators desire to optimize ROP for any given section of the well. However, when ROP optimization uses machine learning, AI or physics models, if an accurate or consistent measure of WOB is not available, then wells cannot be compared and the results are not consistent. It can be valuable in drilling optimization to say that that well 1 and well 2 have consistent zeroing practices for WOB with similar BHA (bottom hole assembly), and drilling programs otherwise comparing drilling parameters for the two wells to plan future wells may be a wasted effort. Uniformity in computing the WOB helps improve such desired drilling optimizations.

[0038] In one embodiment, the downhole states that are identified as good states for zeroing out can be used and the machine learning or AI is trained using those states as a ground truth for when a zeroing is desired to occur. Thus, the prediction is for when the WOB should be zeroed. The ground truth for the WOB may be identified and the machine learning model and / or process can use that ground truth to zero the WOB that is off bottom and rotating.

[0039] As discussed, the machine learning process may receive a moving window of data inputs, and the pattern of data inputs can be identified by the machine learning to zero the WOB when off bottom 220, as in FIG. 2. The pattern of what is identified may include the hook load or related features and this same pattern can be identified across all the stands (e.g., two or three joints of drill pipe or collars connected and stood in the derrick vertically, usually while connecting the pipe). The zeroing may occur consistently in every stand if appropriate conditions exist with the machine learning by identifying the pattern in the time window, and when the pattern is identified then the machine learning application can signal to zero the weight (or zero hook load and related inputs). Consistent zeroing provides more accurate WOB measurements and provides better overall drilling results.

[0040] After an accurate WOB zero point is detected and a drill of the drilling rig is then moved into contact with a downhole object, the WOB can be determined based in part on weight added after the WOB was zeroed for the off bottom rotating weight. The zeroed weight may be used to calculate the weight when the drill bit hits on the bottom of the hole being drilled 230. This zeroed weight may represent a hook load and related factors. Accordingly, Weight on Bit (WOB) can be computed using the hook load weight when the drill bit is not on bottom. The difference between the hook load weights is that there is a hook load weight before on bottom and when the drill bit is on bottom. When the WOB is computed, the user and any applications or process automation utilized may be notified 240 of this completed computation. This may be displaying the WOB of the drill when the drill is drilling through rock on the bottom of the well hole or wellbore.

[0041] The driller is responsible for zeroing the weight on bit (WOB). There are various procedures for doing this depending on where a driller is in the well, operator preferences, and / or tools in the hole. Even 4 different drillers on the same hole may have different preferred procedures for addressing this issue. However, the ability to zero out the WOB is only as good as the process that the driller is using, the point at which they obtain the zero and their knowledge of the drilling system. This technology can take the variability out of the zeroing process and provide a system that can learn when and where zeroing should take place. Using this technology, the drilling system can be notified to zero out at a specific time as indicated by output from the machine learning model.

[0042] When manual zeroing is used, the human drillers sometimes forget to set a zero point or set an incorrect zero point. If the driller only zeros once during the entire drilling process, then the WOB may be incorrect because the WOB will keep increasing when weight is added, then the WOB will not be correct because an incorrect WOB is being used. Drillers need to zero consistently when each of the stands or pipes are added. Each time a pipe is added, this means there is more weight on the drill bit.

[0043] It can be hard to zero manually every stand and get an accurate WOB point. It is also time consuming and sometimes drillers are tempted to skip this step in order to improve their drilling speed. During drilling, the present automation technology can automatically zero, and the WOB can be zeroed. This technology can also take a real time measurement from a hook load to identify a correct hook load zeroing point based on a defined procedure. Hook load can change a lot during a time period but this technology can sample at, for example, 50 Hz data frequency, 1 Hz frequency or 0.1 Hz frequency. Regardless of data sampling frequency, the system can perform the needed zeroing.

[0044] In one example, a driller may monitor the weight and then select a zero point. If the driller zeros at the incorrect time, when the hook load is too light or heavy, this may introduce error. In contrast, using machine learning, an accurate point can be selected for zeroing the WOB and the zeroing can occur consistently across many different wells.

[0045] FIG. 3 illustrates a method for determining a point for zeroing a weight on bit (WOB) measurement of a drilling rig. The method may include the operation of receiving a plurality of drilling features from sensors associated with the drilling rig, as in block 310. In one example, the drilling features may be data that includes at least one of: a hook load weight, data from an electronic drilling recorder (EDR), surface sensor measurements or downhole measurement while drilling (MWD). In another example, a plurality of drilling features that are received may include downhole features from sensors associated with the drilling rig, a well plan, BHA (bottom hole assembly), drilling mud information, and / or geological formation information. The use of downhole features may depend on whether downhole sensors are available in real time (during the drilling process).

[0046] The point in a drilling procedure where the weight on bit (WOB), having an off bottom rotating weight, is to be zeroed, may be determined using a machine learning model (e.g., deep neural network) to process the drilling features, as in block 320. In the example of a deep neural network, a deep neural network may be a recurrent neural network (RNN), transformer, convolutional neural network (CNN), etc. to detect when the WOB is to be zeroed. In addition, a time series of drilling features are submitted to the deep neural network. A time window for drilling features input to the RNN is between 1 minute and 10 minutes, as an example. The deep neural network may be trained using a training data set for the plurality of drilling features that are known to have a desired WOB zero point for an off bottom rotating weight.

[0047] In one configuration, the machine learning model is a first machine learning model (e.g., a deep neural network) used to process surface features obtained from surface sensors to identify the point to zero the WOB and an additional machine learning model (e.g., an additional deep neural network) receives an output of the first deep neural network to identify whether the point to zero the WOB is within a valid window for the predictions. This relationship between the first machine learning model and the second machine learning model may be reversed where the window of valid predictions is identified first and a prediction for a point during drilling is made within the window. Alternatively, there may be two or more machine learning models or neural networks in a relationship for sending and receiving of information between the machine learning models (e.g., neural networks). Further, the outputs from at least one machine learning model(s) or deep neural network(s) may be fed to at least one receiving machine learning model or deep neural network(s) in multiple or various combinations.

[0048] The WOB for the drilling rig may be zero at the point that is identified by the deep neural network, as in block 330. Certain procedural compliance criteria may need to be met before the zero occurs. This means that for the WOB to be zeroed, the machine learning has first predicted an accurate point for zeroing. However, procedural compliance is also important. The zeroing may have other constraints that are to be met, such as only zeroing: when the drill bit is off bottom, when rotary is on, certain pressures are met, and / or additional conditions that are desired. In one embodiment, when both the prediction by the machine learning occurs and the procedural requirements are met, then the zeroing may occur.

[0049] A WOB representing the on bottom rotating weight minus the off bottom rotating weight that has been zeroed can be computed, when a drill of the drilling rig is in contact with a downhole object, as in block 340. In the computation, the WOB may be calculated using a total weight on bottom and subtracting the off bottom rotating weight. The WOB in either case may include at least one of: a hook load weight, differential pressure or friction factors. In other words, the calculated measurement being zeroed out is WOB and this represents how much weight is applied when the drill bit is on bottom and drilling ahead.

[0050] WOB is the weight that has been transferred to the bit on bottom. WOB is not always straight subtraction. There are other forces to incorporate into WOB calculations such as flows, pressure, etc. and WOB can be less than expected. This technology can determine at what point the WOB can be zeroed so the exact hook load measure while off bottom vs. when the bit is on bottom can be calculated. Hook load can be a continuous measurement but a point for the hook load that can be used for the WOB can be determined.

[0051] In another example, the off bottom rotating weight may be identified as zero WOB. When a stable off bottom rotating weight is identified then that is where the machine learning may identify the zero point. This means whatever weight was added from the last stand to the drilling rig is zeroed out. Any additional weight applied from the zero point becomes the WOB or the weight on the bottom when drilling rock.

[0052] A user of the drilling rig and any applications or process automation utilized may also be notified of the point where the WOB was zeroed and the user may also be notified when the WOB on bottom is finally calculated.

[0053] An integrated rate of penetration (ROP) Optimizer may also be provided when using the technology. This may be a feature built into a Automated Rig Control Drilling System that will significantly improve performance for all rigs utilizing the feature. Having this feature may improve on bottom performance drilling rigs and drive incremental value when used in conjunction with ROP Optimizers.

[0054] This technology allows the operations of FIG. 3 to be performed using automation and without manual human input. As a result, the less reliable behaviors of humans may be avoided and the WOB (off bottom) is zeroed at a machine detected point. In addition, the drilling operator may save time during drilling by using machine learning to automate the zeroing process.

[0055] A number of example workflows can be provided for this technology (e.g., four workflow cases) that may be used by a driller. There are a few inputs to manage through an application so that the automated rig control system can appropriately interact with the machine learning model and only alert the driller when appropriate. Building confidence in the technology is useful for long term success with the drillers. So, building a workflow that is not intrusive and clearly communicates a quality of the zeros and when / if the driller needs to take action is valuable. It is also useful for a driller to receive minimal informative alerts to avoid alert fatigue.

[0056] The driller may have a setup screen available in an application that highlights automated zero settings and shows the zeroing trends. For each workflow or configuration that is advisory, the system will only alert the driller if the zero taken (by automation or manually) is outside of a user defined tolerance.

[0057] An automated zero, as discussed below, may be taken by the automated drilling control system by following a sequential (i.e., deterministic) process by taking and zeroing the WOB at a programed point. In contrast, a modeled zero, as discussed below, may be the zeroing point that the machine learning model identifies.

[0058] In a first workflow, there may be no automation and no modeled zero. This is a base case the represents zeroing that is done by a driller alone. In this case, a driller is drilling the well manually and taking the zeros per best drilling practices.

[0059] In a second workflow, there may be manual zeroing that takes place and a machine learning model advisory that may occur or be provided to the driller. At any time, a driller can navigate to the setup screen to adjust tolerance and see the zeroing trend. In this machine learning model, the driller may be utilizing automated rig controls to drill the well, but the automation is in a paused state while the driller is taking a zero, and the machine learning model is running to independently identify the appropriate zero point. In this situation, the modeled zero is purely advisory and does not override the manual zero. While the automation is paused, the driller is controlling the rig and managing the procedure to zero WOB. At this point, the zero is taken and the modeled zero is also identified. If the manual zero is within a user defined tolerance, no alert is presented to the user. On the other hand, if the manual zero is out of tolerance, an alert may be presented to the driller. If the driller or user dismisses the alert, then no action will be taken. If driller or user requests more details, then the application will navigate the driller to the setup screen to see how far out of tolerance the driller is. In either scenario, the driller is responsible for taking a new zero, if desired.

[0060] A third workflow may provide an automated zero and a machine learning model advisory. At any time during drilling, the driller can navigate to the setup screen to adjust the zeroing tolerance and see the zeroing trend. The driller may be utilizing the automated rig controls to drill the well, and the automation is following a standard procedure to capture a zero, and the machine learning model is running to independently identify the appropriate zero point. The modeled zero is purely advisory and does not override the automated zero. At one point, the zero is taken and the modeled zero is identified. If the automated zero is within a user defined tolerance, no alert is presented to the driller user. If the automated zero is out of tolerance, an alert is presented to the driller or user. If the alert is dismissed then no action is taken. If user clicks details, the application will navigate the driller to the setup screen to see how far out of tolerance from the model zero the automated zero was. For either scenario, driller is responsible for taking a new zero.

[0061] The fourth workflow can provide an automated zeroing and model zero ‘in control’ when outside of the user defined threshold. At any time, a driller can navigate to the setup screen to adjust tolerance and see the zeroing trend. The driller may be utilizing the automated rig controls to drill the well, and the automation is following a standard deterministic procedure to capture a zero. As a result, the machine learning model is running to independently identify the appropriate zero point. At one point, a zero is taken and modeled zero is identified. If the automated zero is within a user defined tolerance, no alert is presented to the user. If the automated zero is out of tolerance, an alert is presented to the driller (this may go away over time). If the alert is dismissed, then no action is taken. If the driller or user clicks on the details, the application will navigate the driller to the setup screen to see how far out of tolerance the automated zero was. In either alert scenario, the modeled zero will overwrite the automated zero and be utilized by the autodriller to provide instructions for or control ROP.

[0062] Many types of feedback can be used to provide a feedback loop and training or re-training of the machine learning model(s). For example, driller feedback from interactions with the UI / UX (user interface / user experience) and driller actions taken after out of tolerance zeros are identified can be used for on-going learning and retraining of the machine learning model as needed. As discussed earlier, downhole data for a specific drilled hole can also be used as feedback into the machine learning model where the machine learning model is not performing as desired. The downhole data can be used to help re-label the zero points and where the machine learning model should have zeroed.

[0063] It is also helpful to understand the context of obtaining automated zeros. Automated zeros are dependent on a good procedure. Automated zeros will start to fall out of tolerance when hole conditions or well trajectories change and the driller does not update the automated procedure to capture the zero.

[0064] Some of the functional units described in this specification have been labeled as modules, in order to more particularly emphasize their implementation independence. For example, a module may be implemented as a hardware circuit comprising custom VLSI circuits or gate arrays, off-the-shelf semiconductors such as logic chips, transistors, or other discrete components. A module may also be implemented in programmable hardware devices such as field programmable gate arrays, programmable array logic, programmable logic devices or the like.

[0065] Modules may also be implemented in software for execution by various types of processors. An identified module of executable code may, for instance, comprise one or more blocks of computer instructions, which may be organized as an object, procedure, or function. Nevertheless, the executables of an identified module need not be physically located together, but may comprise disparate instructions stored in different locations which comprise the module and achieve the stated purpose for the module when joined logically together.

[0066] Indeed, a module of executable code may be a single instruction, or many instructions, and may even be distributed over several different code segments, among different programs, and across several memory devices. Similarly, operational data may be identified and illustrated herein within modules, and may be embodied in any suitable form and organized within any suitable type of data structure. The operational data may be collected as a single data set, or may be distributed over different locations including over different storage devices. The modules may be passive or active, including agents operable to perform desired functions.

[0067] The technology described here can also be stored on a computer readable storage medium that includes volatile and non-volatile, removable and non-removable media implemented with any technology for the storage of information such as computer readable instructions, data structures, program modules, or other data. Computer readable storage media include, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other computer storage medium which can be used to store the desired information and described technology.

[0068] The devices described herein may also contain communication connections or networking apparatus and networking connections that allow the devices to communicate with other devices. Communication connections are an example of communication media. Communication media typically embodies computer readable instructions, data structures, program modules and other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media. A “modulated data signal” means a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, communication media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, radio frequency, infrared, and other wireless media. The term computer readable media as used herein includes communication media.

[0069] Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more examples. In the preceding description, numerous specific details were provided, such as examples of various configurations to provide a thorough understanding of examples of the described technology. One skilled in the relevant art will recognize, however, that the technology can be practiced without one or more of the specific details, or with other methods, components, devices, etc. In other instances, well-known structures or operations are not shown or described in detail to avoid obscuring aspects of the technology.

[0070] The following examples are further illustrative of several embodiments of the present technology:

[0071] A method for determining a point for zeroing a weight on bit (WOB) measurement for a drilling rig, comprising:receiving a plurality of drilling features from sensors associated with the drilling rig;determining a point in a drilling procedure where the weight on bit (WOB) for an off bottom rotating weight, is to be zeroed, by detecting a pattern for drilling features while processing the drilling features;zeroing the WOB for the drilling rig at the point; andcomputing a WOB, when a drill of the drilling rig is in contact with a downhole object, based in part on weight added after the WOB has been zeroed for the off bottom rotating weight.

[0072] The method as in example 1, further comprising notifying a user of the drilling rig, a process or applications that the WOB that has been computed.

[0073] The method as in any preceding example, wherein detecting a pattern for drilling features is performed using a machine learning model that is at least one of: a recurrent neural network (RNN), a genetic model, gradient boost, CNN, a UBoost model, regression, reinforcement learning, supervised machine learning, unsupervised machine learning, predictive modeling, pattern recognition, expert rules, classification models, condition based rules, or a heuristic to detect when the WOB is to be zeroed.

[0074] The method as in any preceding example, wherein a time series of the drilling features are input features to the RNN.

[0075] The method as in any preceding example, wherein a time window for drilling features input to the RNN is a defined number of seconds or minutes.

[0076] The method as in any preceding example, wherein the machine learning model is trained using a training data set of the drilling features that are known to have a desired weight on bit (WOB) for an off bottom rotating weight.

[0077] The method as in any preceding example, wherein the machine learning model includes at least one machine learning model is used to process surface features obtained from surface sensors to identify the point when the WOB is to be zeroed and at least one additional machine learning model to receive an output of the at least one machine learning model to identify whether the point to zero the WOB is within a valid window for predictions..

[0078] The method as in any preceding example, further comprising:receiving a manual zeroing;determining whether the manual zeroing is out of tolerance; andsending a machine learning model advisory to a driller, if the manual zeroing is out of tolerance.

[0079] The method as in any preceding example, further comprising:receiving an automated zeroing;determining whether the automated zeroing is out of tolerance; andsending a machine learning model advisory to a driller, if the automated zero is out of tolerance.

[0080] The method as in any preceding example, further comprising:receiving an automated zeroing;determining whether the automated zeroing is out of tolerance; andreceiving zeroing instructions from the machine learning model, if the automated zeroing is out of tolerance.

[0081] A method for determining a point for zeroing a weight on bit (WOB) measurement of a drilling rig, comprising:receiving a plurality of drilling features from sensors associated with the drilling rig;determining the point in a drilling procedure where the weight on bit (WOB), having an off bottom rotating weight, is to be zeroed, using a deep neural network to process the drilling features;zeroing the WOB for the drilling rig at the point; andcomputing a WOB representing the on bottom rotating weight minus the off bottom rotating weight that has been zeroed, when a drill of the drilling rig is in contact with a downhole object.

[0082] The method as in any preceding example, further comprising using the deep neural network that is a recurrent neural network (RNN) to detect when the WOB is to be zeroed.

[0083] The method as in any preceding example, wherein a time series of drilling features are submitted to the RNN.

[0084] The method as in any preceding example, wherein a time window for drilling features input to the RNN is a defined number of seconds or minutes.

[0085] The method as in any preceding example, wherein the deep neural network is trained using a training data set of the plurality of drilling features that are known to have a desired WOB zero point, at an off bottom rotating weight.

[0086] The method as in any preceding example, wherein the WOB is calculated using a total weight on bottom and subtracting the off bottom rotating weight that includes at least one of: a hook load weight, differential pressure or friction factors.

[0087] The method as in any preceding example, wherein the deep neural network includes at least one deep neural network used to process surface features obtained from surface sensors to identify the point when the WOB is to be zeroed and at least one additional deep neural network to receive an output of the at least one additional deep neural network to identify whether the point to zero the WOB is within a valid window for predictions.

[0088] The method as in any preceding example, further comprising receiving drilling features that are at least one of: a hook load weight, data from an electronic drilling recorder (EDR), a well plan, BHA (bottom hole assembly), drilling mud information, and / or geological formation information or downhole measurement while drilling (MWD).

[0089] The method as in any preceding example, further comprising notifying a user of the drilling rig, a process or applications of the point where the WOB was zeroed and notifying the user of the WOB.

[0090] The method as in any preceding example, further comprising performing the steps of claim 8 using process automation and without manual human input.

[0091] The method as in any preceding example, further comprising receiving a plurality of drilling features, including downhole features, from sensors associated with the drilling rig.

[0092] The method as in any preceding example, determining that procedural compliance criteria are met before the zeroing occurs.

[0093] A system for determining a point for zeroing a weight on bit (WOB) measurement for a drilling rig, the system comprising:

[0094] at least one processor;

[0095] at least one memory device including a data store to store a plurality of data and instructions that, when executed, cause the system to:receive a plurality of drilling features from sensors associated with the drilling rig;determine the point in a drilling procedure where the weight on bit (WOB), including an off bottom rotating weight, is to be zeroed using a machine learning model to process the drilling features;zero the WOB for the drilling rig at the point; andcompute a WOB based in part on the on bottom rotating weight minus the off bottom rotating weight that has been zeroed, when a drill of the drilling rig is in contact with a downhole object.

[0096] The system as in any preceding example, wherein a time series of drilling features are input features to the machine learning model that is a recurring neural network (RNN).

[0097] The system as in any preceding example, wherein a time window for drilling features input to the RNN is a defined number of seconds or minutes.

[0098] The system as in any preceding example, wherein a first machine learning model is used to process surface features obtained from surface sensors to identify a WOB point for zeroing and a second machine learning model receives an output of the first machine learning model to identify whether the point to zero the WOB is within a valid window for predictions.

[0099] Although the subject matter has been described in language specific to structural features and / or operations, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features and operations described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims. Numerous modifications and alternative arrangements can be devised without departing from the spirit and scope of the described technology.

Examples

Embodiment Construction

[0008]Reference will now be made to the examples illustrated in the drawings, and specific language will be used herein to describe the same. It will nevertheless be understood that no limitation of the scope of the technology is thereby intended. Alterations and further modifications of the features illustrated herein, and additional applications of the examples as illustrated herein, which would occur to one skilled in the relevant art and having possession of this disclosure, are to be considered within the scope of the description.

[0009]Drillers who are drilling hydrocarbon wells or other wells desire to have the highest possible on-bottom performance and rotary Rate of Penetration (ROP) when drilling the wells. ROP may be dictated by how much energy is put into the drilling system. ROP may be analyzed by looking at a current ROP, weight on bit (WOB), differential pressure(s), rotary speeds, and torque. As part of this process, drillers may desire to understand current drilling ...

BACKGROUND

[0001] Hydrocarbon producing wellbores extend into the earth's surface and intersect subterranean formations where hydrocarbons are trapped. Casing and tubing that may be added to the wellbores may provide conduits for the hydrocarbons to be brought to the surface. Drill bits may be used to form the wellbores, and the drill bits may be mounted on the ends of drillstrings, as part of a drilling rig. Motorized drive systems on the surface rotate the drillstrings and bits that can drill the rock. Cutting elements on the drill bit may rotate and scrape the bottom of the wellbore and grind away material to deepen the wellbore. Drilling fluid is pumped down the drillstring and directed from the drill bit into the wellbore. The drilling fluid may flow back up the wellbore in the area between the drillstring and walls of the wellbore. A similar process may be used for drilling wells for water too.

[0002] The amount of weight or force applied to the drill bit during drilling, may be referred to as weight on bit (“WOB”), and the amount of WOB affects drilling performance and tool life. Applying an insufficient WOB often reduces the rock penetration rate and increases bit vibration. In contrast, applying excessive WOB can cause mechanical bit failure, by inducing stick slip and increasing shock and vibration experienced by the downhole tools. In this case, energy applied to the system is translated to the drillstring and not to increase Rate of Penetration (ROP). The force exerted by the drilling rig for holding the drillstring is commonly referred to as the hook load. In other words, hook load is the weight of the drillstring in air, the drill collars and any ancillary equipment, reduced by any force that tends to reduce that weight. Some forces that might reduce the weight include friction along the wellbore wall and, importantly, buoyant forces on the drillstring caused by its immersion in drilling fluid. A sensor exists on the deadline that measures how heavy the entire drillstring is from the traveling block. Hook load can change every single stand because more pipe is being added to the drillstring, and depending on the well type or drilling progress, inclination may be added that can increase friction in the system. When the drill bit is on bottom, then extra weight is being applied to cut the rock.

[0003] Traditionally, WOB measurements are based on a difference in hook load between a bit's off bottom rotating weight and on bottom rotating weight. More specifically, when a portion of the hanging drillstring weight is supported by the bit resting on the bottom of the wellbore, hook load is reduced by that amount. This difference between current hook load and a “zero” value is taken as a reference for the amount of weight put on the bit. A zero value is typically obtained by measuring the hook load while suspending the drillstring in the wellbore, and without the drillstring being supported on the bottom. Because the drillstring weight changes as drill pipe segments are added to the drillstring, correctly applying a designated WOB means that the zero weight can be constantly monitored. Terms “zero” or “zeroing” may also be considered equivalent to taring, resetting of the weight on the bit (WOB) or similar terms, as may be used interchangeably in the drilling industry.

[0004] Weight on bit (WOB), as defined in the oil industry, is the amount of downward force exerted on the drill bit and may be measured in thousands of pounds. Weight on bit may be provided by drill collars or stands of pipe, which are thick-walled tubular pieces machined from solid bars of steel, usually plain carbon steel. Gravity and the downward force applied by a topdrive can act on the large mass of the collars to provide the downward force needed for the drill bit to efficiently break the rock. To accurately control the amount of force applied to the bit, the driller may monitor the surface weight measured while the bit is just off the bottom of the wellbore. Next, the drillstring and the drill bit may be slowly and carefully lowered until they touch bottom. After that point, as the driller continues to lower the top of the drillstring, more and more weight is applied to the bit, and correspondingly less weight is measured as hook load. If the surface measurement shows 20,000 pounds less weight than with the bit off bottom, then there should be 20,000 pounds force on the bit (in a vertical hole). Such WOB measurements may be affected by changing forces due to changes in inclination or azimuth while drilling. Some downhole Measurement While Drilling (MWD) sensors can measure downhole Weight on Bit more accurately and transmit the data to the surface.BRIEF DESCRIPTION OF THE DRAWINGS

[0005] FIG. 1 is a schematic diagram illustrating an example of a drilling system.

[0006] FIG. 2 is a block diagram illustrating aspects of a method for determining a point for zeroing a weight on bit (WOB) measurement for a drilling rig.

[0007] FIG. 3 is a flowchart illustrating a method for determining a point for zeroing a weight on bit (WOB) measurement of a drilling rig.DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] Reference will now be made to the examples illustrated in the drawings, and specific language will be used herein to describe the same. It will nevertheless be understood that no limitation of the scope of the technology is thereby intended. Alterations and further modifications of the features illustrated herein, and additional applications of the examples as illustrated herein, which would occur to one skilled in the relevant art and having possession of this disclosure, are to be considered within the scope of the description.

[0009] Drillers who are drilling hydrocarbon wells or other wells desire to have the highest possible on-bottom performance and rotary Rate of Penetration (ROP) when drilling the wells. ROP may be dictated by how much energy is put into the drilling system. ROP may be analyzed by looking at a current ROP, weight on bit (WOB), differential pressure(s), rotary speeds, and torque. As part of this process, drillers may desire to understand current drilling parameters and how that affects ROP. To optimize ROP, drillers may need to modify available setpoints (i.e., a preset value that is the desired value of a drilling parameter) to adjust for drilling conditions.

[0010] Most drilling rigs have an auto driller or a computerized controller for assisting with managing and controlling the drilling process. For example, the auto driller may use electronic controls to manage weight on bit (WOB), differential pressure (DP), ROP, etc., and ROP, WOB, or DP can be utilized as the primary parameters to manage drawworks speed output. If a WOB is desired to be consistent, then the auto driller can utilize the specified WOB Setpoint to set the drawworks speed to be a certain value N to yield the specified WOB (e.g., 30). The drawworks speed may be based on a setting of N or another value.

[0011] WOB may be a subjective calculation but it may be calculated as hook load on bottom vs. hook load off bottom. WOB may represent how much of the weight is actually being applied to the bit vs. how much of the weight is just the drillstring weight within the wellbore. The objective is to measure the energy being put in the system in order to break rock.

[0012] Weight on Bit (WOB) may be a calculated metric defined as the force being applied to the bit to break rock (WOB=Total Weight on Bottom−Weight of the drillstring−friction factors). When drilling, it is valuable to have an accurate and consistent trend in WOB as the system drills to ensure the appropriate amount of force is being used to break the rock.

[0013] Weight on bit (WOB) is an important metric derived from sensor data, specifically hook load, and can be used in optimizing drilling performance. However, the accuracy of WOB measurements relies on timely zeroing, and the measurements are often impacted by inconsistent zeroing methods. For example, such zeroing may introduce + / −10,000 lbf (pounds of force) of error in the computed WOB. Factors like incorrect drilling parameter optimization and unreliable data collection may compound this issue. Accurate WOB information enhances drilling efficiency and is a top-requested feature by operators of a drilling rig.

[0014] In the past, drillers may have been manually ‘zeroing’ the WOB. The zeroing process is not standardized when performed manually and can vary by operator, company, driller, etc. In other words, the human driller decides at what point to zero the WOB. Utilizing a Machine Learning (ML) model or Artificial Intelligence (AI) model to identify and automatically know when to ‘zero’ WOB may provide advantages over manual zeroing. Specifically, the zeroing of the WOB when off bottom and rotating using a machine learning model or pattern matching detected point may result in zeroing at a reliable point in the drill hole or a reliable point in time when appropriate downhole conditions exist. An example of appropriate conditions to detect a zero point within a given window may include one or more of the following conditions: the drill pipe is out of slips, a stable rotary speed, a stable hookload, a stable pump pressure, the off bottom pressure is within a tolerance of the previous stands off bottom pressure, the bit depth is off bottom, no oscillating system is on, if a dynamic zeroing point is being detected then there is a consistent running speed, and / or if a static zeroing point is being detected then there is 0 running speed. Other parameters that represent appropriate conditions to detect a zero point may also be used.

[0015] To improve the accuracy of the WOB, machine learning or AI technology can be used for automated zeroing of the WOB for an off bottom rotating weight (i.e., without user intervention). This technology may use an electronic drilling recorder (EDR), downhole measurements while drilling (MWD) data, surface sensor measurements, along with current WOB values and the error in the Weight on Bit as input data to train a ML model or AI model. The model may be trained to reduce the error which will eventually result in correcting the WOB and automating the zeroing process.

[0016] This technology can use predictive WOB tracking using available drilling features (i.e., parameters) with machine learning techniques to automate the zeroing of weight on bit. The technology and process may use an ML model to automatically zero WOB when appropriate. This removes the dependency on the human driller or manual configuration of automation recipes for zeroing the WOB. Using machine learning, WOB can be a more accurate measurement that the human driller can utilize to optimize ROP.

[0017] Machine learning models and related techniques can be used to predict the deviation of WOB from values predicted by a machine learning algorithm in real time. When the deviation is more than an allowed threshold, the system can use a corrective procedure and automatically zero the WOB.

[0018] Currently, in the absence of automation for WOB (weight on bit) zeroing, the WOB is being zeroed inconsistently or incorrectly and the human error can introduce + / −10,000 lbf errors in the WOB. Down the line calculations which are important for drilling performance like MSE, ROP optimization techniques rely heavily on the WOB (weight on bit) accuracy, and when the WOB is not accurate, this adversely affects the overall drilling performance.

[0019] An example of a drilling system 110 is shown in a side view in FIG. 1. The drilling system 110 may be used for forming a wellbore 112 through a rock formation 114. The drilling system 110 may include a drillstring 116 disposed within wellbore 112, and the drillstring may be made up of segments of drill pipe 118. In one example, the segments of drill pipe 118 may be coupled to one another (e.g., using threading or another method). A drill bit 120 is shown mounted on a lower end of drillstring 116, and the drill bit may include a drill bit body 122 that mounts on the lowest drill pipe 118 of the drillstring 116. Inserts or cutters 124 may be on a surface of drill bit body 122 opposite from where the drill bit body 122 attaches to the drillstring 116. When the drillstring 116 and bit 120 are rotated, the cutters 124 may grind or crush the rock making up the formation 114 thereby forming borehole or wellbore 112.

[0020] Above the wellbore 112, a derrick 126 may be mounted on a surface 128 or the ground. Equipment as part of the drilling system 110 may be provided for manipulating the drillstring 116, which includes a drawworks 130. The drawworks 130 may selectively pull or release a cable 132 shown engaging sheaves 134 that are rotatably mounted on an upper end of the derrick 126. Additional cables may run through the sheaves 134, and support a traveling block 136, that in conjunction with a hook 138 and swivel 140, couple with the drillstring 116 for raising and lowering the drillstring 116.

[0021] In some examples, a kelly 142 can axially couple to a lower end of swivel 140 and is rotatable with respect to swivel 140. A lower end of the kelly 142 projects through a rotary table 144, which engages outer surfaces of the kelly 142 and rotates to exert a rotational force onto drillstring 116. The rotary table 144 is formed on a rig floor 146 that attaches to derrick 126 and is set above surface 128. Below the rig floor 146 and at surface 128 is a wellhead housing 148 that is mounted in the opening of wellbore 112. On top of the wellhead housing 148 is a blowout preventer (“BOP”) 152.

[0022] Further shown on surface 128 are stands of pipe 154 that are supported by a rack 156 illustrated on one of the side beams of derrick 126. On the rig floor 146, a driller's console 158 with gauges or electronic readouts may be provided to report downhole conditions, and controls for operating the drilling assembly, such as the drawworks 130. A controller 160 is logically illustrated as being in communication with the console 158. Communications between the controller 160 and console 158 can be wireless, fiber optic, or wired. Embodiments may exist wherein the controller 160 is included within the console 158.

[0023] The weight on bit (“WOB”) exerted by drillstring 116 on the bottom of wellbore 112 can be controlled by an operator on the rig floor 146 and in conjunction with the console 158. Operators can adjust the drawworks 130 so that an upward force on drillstring 116 can be exerted on traveling block 136, hook 138, swivel 140, and kelly 142. These functions may be from software commands from the controller 160. In one example, WOB is estimated based on a hook load, which is the axial force exerted on hook 138, or other components that provide an axial supporting force for drillstring 116. Sensors (not shown) can provide a signal that when viewed at a console 158 represents the axial load by which drillstring 116 is supported by the remaining portions of the drilling system 110, i.e. the hook load.

[0024] It is noted that while the above described drilling system 110 employing a kelly-type drive system may be used with the technology described herein, skilled artisans will recognize after reading the entire disclosure that other drilling systems, particularly other drive systems or mechanisms, can be used with the technology described herein. For instance, in another example, the drilling system 110 can instead comprise a top drive type of drive mechanism. In still other examples, a rack and pinion type of drive mechanism, or other known drilling components operable with the drilling system 110 can be used.

[0025] FIG. 2 is a block diagram illustrating aspects of a method for determining a point for zeroing a weight on bit (WOB) measurement for a drilling rig. A plurality of drilling features may be received from sensors 210 associated with the drilling rig. The drilling features may be received from sensors that are above ground such as a hook load sensor, etc. In addition, the drilling features may be features from a downhole sensor(s) that collects downhole and / or WOB data. These features may be used as input for a detection process to determine a WOB.

[0026] A WOB at a point during drilling where the bit is not contacting rock may be determined based on an off-bottom rotating hook load. The WOB may be the free hanging weight of the bottom hole assembly (BHA) while rotary conditions exist. The rotary conditions may include rotation of the drill bit, pressure, and / or hook load. Hook load may be a measurement from a surface sensor representing the hook load at the time the sensor is read. The WOB value while off bottom is expected to be zero if the zeroing point is appropriate.

[0027] A point in a drilling procedure where the weight on bit (WOB) for an off bottom rotating weight is to be zeroed 212 may be determined by detecting a pattern for drilling features while processing the drilling features using a machine learning model. The WOB may be zeroed for the drilling rig at the point. The zeroing can reset the WOB at the detected point. Thus, the hook load zeroing point can be used to calculate the WOB for an on bottom rotating weight. The hook load may be used to calculate the WOB. One sensor that may be used in the drilling process is a hook load sensor and WOB can be measured based in part from the hook load. Other sensors or sensor types may be used in calculating the WOB.

[0028] Detecting a pattern for drilling features may be performed using one of a variety of detection methods or machine learning models 214. For example, types of models that may be used may include at least one of: a deep neural network, transformers, a convolutional neural network (CNN), a recurrent neural network (RNN), a genetic model, a UBoost model, regression, reinforcement learning, supervised machine learning, unsupervised machine learning, classification models, predictive modeling, pattern recognition, expert rules, condition based rules, or a heuristic to detect when the WOB is to be zeroed. For example, a genetic model may be used to try to determine the patterns in the features that provide a better zero point for the WOB and then a zeroing may be triggered when the desired pattern is identified. In another example, UBoost may use a series of machine learning classifiers to identifying a useful zeroing point. In addition, other models may be used that can process temporal, event based, sensor obtained, high frequency data from multiple data sources using any combination of ML, heuristic, and / or physics based techniques to identify the accurate WOB zeroing point.

[0029] One useful example machine learning model for zeroing is an RNN. The RNN may receive a time series of the drilling features as input features to the RNN. RNNs are able to process a sequence of data over time by storing previous data points in memory, and thereby, the data that is time sampled during drilling can be processed by the RNN. For example, a variable window length for drilling features input to the RNN may span a range of minutes (e.g., between 1 minute and 10 minutes) which may be useful where the off bottom duration is quite variable. Multiple samples for a feature from a sensor maybe included a time window. An RNN can provide time series predictions with a moving window as described.

[0030] A machine learning model that is used may be trained using a training data set of the drilling features that are known to represent a desired weight on bit (WOB) for an off bottom rotating weight. This training data set may be a ground truth to train a machine learning model or AI model and the training data set may train the machine learning with patterns representing where the WOB for an off bottom rotating position is to be auto zeroed out. For example, when appropriate downhole conditions exist, there may be a specific rate of change (e.g., wobble) for the surface weight on bit which suggests the pattern for zeroing out.

[0031] The models or machine learning model can use a feedback loop and scan the sensor channels for the drilling rig. Feedback may be used to train the models. For example, the machine learning model may be trained or learn based on the down hole data that is collected from the drilling rig. Down hole data can represent a ground truth about the downhole environment. In addition to what the model learns during the training process, the model may also be customized to the well, downhole environment and / or drilling rig which the model is being used on. Each of the rigs and wells may be unique depending the downhole environment, rig differences and related parameters.

[0032] The deep neural network can also be trained on data patterns that are obtained from the surface sensors. The deep neural network or machine learning model can also be trained using any parameters measurable or computable during the drilling process. These other parameters may include additional sensor readings, in addition to hook load, that contribute to actual WOB such as: pump pressures, Surface RPMs (revolutions per minute), downhole RPMs, and other parameters. Multiple internal features for the drilling rig may be used within each multidimensional input to a deep neural network (e.g., RNN).

[0033] In one embodiment, the machine learning models may be trained on data from multiple wells, wells from different regions, drilling on different rock formations and sections. The machine learning model may be created with as generic of a model as possible so that it may be used with many types of drilling rigs in many different location types. Downhole data may be obtained from drilling companies across a variety of wells. This data from many drillers can be used to check how accurate the model is, to train the model(s) and to continuously improve the model(s).

[0034] The downhole data for a specific drilled hole can be used as feedback into the model where the model is not performing as well as desired. The downhole data can be used to help re-label the zero points and where the model should have zeroed. Then if the surface data is used in detecting the zero point, the downhole data may have been used to train the model regarding the actual zero point selected.

[0035] As discussed, there are a number of different features (i.e., parameters) that may be used in training the machine learning models. Examples of these features may include: downhole RPM (rotations per minute), rate of penetration (ROP), downhole pressures, downhole WOB, downhole differential pressures, surface weight on bit, surface differential pressures, depth of the bit, a depth of the well, a block position, geological formation information or service data such as inclinations and azimuth. Calculated features may also be used with the machine learning model. The calculated features may be based on the time based features. This technology may use calculated features and the time based calculated features. For example, a time based feature may represent the differential pressure after 1 minute or 5 minutes. The RNN may create thousands of its own internal features, as is known in RNN networks.

[0036] This technology can automate the zeroing process and avoid a human driller having to manually determine the zeroing point. Automatically detecting a zeroing point is useful because a human can become distracted or not be able to detect a good zeroing point for the drilling rig. Thus, having a recognition process or machine learning model that is part of the zeroing process may provide reliable detection of a zeroing point. The system can be trained to anticipate a point when zeroing of the WOB, when off bottom, is desired to occur.

[0037] This technology also avoids the situation where drillers can hide weight by zeroing late (e.g., when 10,000 lbs. is already applied to the bit). In this situation, the driller may or may not know that when the instruments are reading 20,000 lbf it is really 30,000 lbf of WOB. Drillers and operators desire to optimize ROP for any given section of the well. However, when ROP optimization uses machine learning, AI or physics models, if an accurate or consistent measure of WOB is not available, then wells cannot be compared and the results are not consistent. It can be valuable in drilling optimization to say that that well 1 and well 2 have consistent zeroing practices for WOB with similar BHA (bottom hole assembly), and drilling programs otherwise comparing drilling parameters for the two wells to plan future wells may be a wasted effort. Uniformity in computing the WOB helps improve such desired drilling optimizations.

[0038] In one embodiment, the downhole states that are identified as good states for zeroing out can be used and the machine learning or AI is trained using those states as a ground truth for when a zeroing is desired to occur. Thus, the prediction is for when the WOB should be zeroed. The ground truth for the WOB may be identified and the machine learning model and / or process can use that ground truth to zero the WOB that is off bottom and rotating.

[0039] As discussed, the machine learning process may receive a moving window of data inputs, and the pattern of data inputs can be identified by the machine learning to zero the WOB when off bottom 220, as in FIG. 2. The pattern of what is identified may include the hook load or related features and this same pattern can be identified across all the stands (e.g., two or three joints of drill pipe or collars connected and stood in the derrick vertically, usually while connecting the pipe). The zeroing may occur consistently in every stand if appropriate conditions exist with the machine learning by identifying the pattern in the time window, and when the pattern is identified then the machine learning application can signal to zero the weight (or zero hook load and related inputs). Consistent zeroing provides more accurate WOB measurements and provides better overall drilling results.

[0040] After an accurate WOB zero point is detected and a drill of the drilling rig is then moved into contact with a downhole object, the WOB can be determined based in part on weight added after the WOB was zeroed for the off bottom rotating weight. The zeroed weight may be used to calculate the weight when the drill bit hits on the bottom of the hole being drilled 230. This zeroed weight may represent a hook load and related factors. Accordingly, Weight on Bit (WOB) can be computed using the hook load weight when the drill bit is not on bottom. The difference between the hook load weights is that there is a hook load weight before on bottom and when the drill bit is on bottom. When the WOB is computed, the user and any applications or process automation utilized may be notified 240 of this completed computation. This may be displaying the WOB of the drill when the drill is drilling through rock on the bottom of the well hole or wellbore.

[0041] The driller is responsible for zeroing the weight on bit (WOB). There are various procedures for doing this depending on where a driller is in the well, operator preferences, and / or tools in the hole. Even 4 different drillers on the same hole may have different preferred procedures for addressing this issue. However, the ability to zero out the WOB is only as good as the process that the driller is using, the point at which they obtain the zero and their knowledge of the drilling system. This technology can take the variability out of the zeroing process and provide a system that can learn when and where zeroing should take place. Using this technology, the drilling system can be notified to zero out at a specific time as indicated by output from the machine learning model.

[0042] When manual zeroing is used, the human drillers sometimes forget to set a zero point or set an incorrect zero point. If the driller only zeros once during the entire drilling process, then the WOB may be incorrect because the WOB will keep increasing when weight is added, then the WOB will not be correct because an incorrect WOB is being used. Drillers need to zero consistently when each of the stands or pipes are added. Each time a pipe is added, this means there is more weight on the drill bit.

[0043] It can be hard to zero manually every stand and get an accurate WOB point. It is also time consuming and sometimes drillers are tempted to skip this step in order to improve their drilling speed. During drilling, the present automation technology can automatically zero, and the WOB can be zeroed. This technology can also take a real time measurement from a hook load to identify a correct hook load zeroing point based on a defined procedure. Hook load can change a lot during a time period but this technology can sample at, for example, 50 Hz data frequency, 1 Hz frequency or 0.1 Hz frequency. Regardless of data sampling frequency, the system can perform the needed zeroing.

[0044] In one example, a driller may monitor the weight and then select a zero point. If the driller zeros at the incorrect time, when the hook load is too light or heavy, this may introduce error. In contrast, using machine learning, an accurate point can be selected for zeroing the WOB and the zeroing can occur consistently across many different wells.

[0045] FIG. 3 illustrates a method for determining a point for zeroing a weight on bit (WOB) measurement of a drilling rig. The method may include the operation of receiving a plurality of drilling features from sensors associated with the drilling rig, as in block 310. In one example, the drilling features may be data that includes at least one of: a hook load weight, data from an electronic drilling recorder (EDR), surface sensor measurements or downhole measurement while drilling (MWD). In another example, a plurality of drilling features that are received may include downhole features from sensors associated with the drilling rig, a well plan, BHA (bottom hole assembly), drilling mud information, and / or geological formation information. The use of downhole features may depend on whether downhole sensors are available in real time (during the drilling process).

[0046] The point in a drilling procedure where the weight on bit (WOB), having an off bottom rotating weight, is to be zeroed, may be determined using a machine learning model (e.g., deep neural network) to process the drilling features, as in block 320. In the example of a deep neural network, a deep neural network may be a recurrent neural network (RNN), transformer, convolutional neural network (CNN), etc. to detect when the WOB is to be zeroed. In addition, a time series of drilling features are submitted to the deep neural network. A time window for drilling features input to the RNN is between 1 minute and 10 minutes, as an example. The deep neural network may be trained using a training data set for the plurality of drilling features that are known to have a desired WOB zero point for an off bottom rotating weight.

[0047] In one configuration, the machine learning model is a first machine learning model (e.g., a deep neural network) used to process surface features obtained from surface sensors to identify the point to zero the WOB and an additional machine learning model (e.g., an additional deep neural network) receives an output of the first deep neural network to identify whether the point to zero the WOB is within a valid window for the predictions. This relationship between the first machine learning model and the second machine learning model may be reversed where the window of valid predictions is identified first and a prediction for a point during drilling is made within the window. Alternatively, there may be two or more machine learning models or neural networks in a relationship for sending and receiving of information between the machine learning models (e.g., neural networks). Further, the outputs from at least one machine learning model(s) or deep neural network(s) may be fed to at least one receiving machine learning model or deep neural network(s) in multiple or various combinations.

[0048] The WOB for the drilling rig may be zero at the point that is identified by the deep neural network, as in block 330. Certain procedural compliance criteria may need to be met before the zero occurs. This means that for the WOB to be zeroed, the machine learning has first predicted an accurate point for zeroing. However, procedural compliance is also important. The zeroing may have other constraints that are to be met, such as only zeroing: when the drill bit is off bottom, when rotary is on, certain pressures are met, and / or additional conditions that are desired. In one embodiment, when both the prediction by the machine learning occurs and the procedural requirements are met, then the zeroing may occur.

[0049] A WOB representing the on bottom rotating weight minus the off bottom rotating weight that has been zeroed can be computed, when a drill of the drilling rig is in contact with a downhole object, as in block 340. In the computation, the WOB may be calculated using a total weight on bottom and subtracting the off bottom rotating weight. The WOB in either case may include at least one of: a hook load weight, differential pressure or friction factors. In other words, the calculated measurement being zeroed out is WOB and this represents how much weight is applied when the drill bit is on bottom and drilling ahead.

[0050] WOB is the weight that has been transferred to the bit on bottom. WOB is not always straight subtraction. There are other forces to incorporate into WOB calculations such as flows, pressure, etc. and WOB can be less than expected. This technology can determine at what point the WOB can be zeroed so the exact hook load measure while off bottom vs. when the bit is on bottom can be calculated. Hook load can be a continuous measurement but a point for the hook load that can be used for the WOB can be determined.

[0051] In another example, the off bottom rotating weight may be identified as zero WOB. When a stable off bottom rotating weight is identified then that is where the machine learning may identify the zero point. This means whatever weight was added from the last stand to the drilling rig is zeroed out. Any additional weight applied from the zero point becomes the WOB or the weight on the bottom when drilling rock.

[0052] A user of the drilling rig and any applications or process automation utilized may also be notified of the point where the WOB was zeroed and the user may also be notified when the WOB on bottom is finally calculated.

[0053] An integrated rate of penetration (ROP) Optimizer may also be provided when using the technology. This may be a feature built into a Automated Rig Control Drilling System that will significantly improve performance for all rigs utilizing the feature. Having this feature may improve on bottom performance drilling rigs and drive incremental value when used in conjunction with ROP Optimizers.

[0054] This technology allows the operations of FIG. 3 to be performed using automation and without manual human input. As a result, the less reliable behaviors of humans may be avoided and the WOB (off bottom) is zeroed at a machine detected point. In addition, the drilling operator may save time during drilling by using machine learning to automate the zeroing process.

[0055] A number of example workflows can be provided for this technology (e.g., four workflow cases) that may be used by a driller. There are a few inputs to manage through an application so that the automated rig control system can appropriately interact with the machine learning model and only alert the driller when appropriate. Building confidence in the technology is useful for long term success with the drillers. So, building a workflow that is not intrusive and clearly communicates a quality of the zeros and when / if the driller needs to take action is valuable. It is also useful for a driller to receive minimal informative alerts to avoid alert fatigue.

[0056] The driller may have a setup screen available in an application that highlights automated zero settings and shows the zeroing trends. For each workflow or configuration that is advisory, the system will only alert the driller if the zero taken (by automation or manually) is outside of a user defined tolerance.

[0057] An automated zero, as discussed below, may be taken by the automated drilling control system by following a sequential (i.e., deterministic) process by taking and zeroing the WOB at a programed point. In contrast, a modeled zero, as discussed below, may be the zeroing point that the machine learning model identifies.

[0058] In a first workflow, there may be no automation and no modeled zero. This is a base case the represents zeroing that is done by a driller alone. In this case, a driller is drilling the well manually and taking the zeros per best drilling practices.

[0059] In a second workflow, there may be manual zeroing that takes place and a machine learning model advisory that may occur or be provided to the driller. At any time, a driller can navigate to the setup screen to adjust tolerance and see the zeroing trend. In this machine learning model, the driller may be utilizing automated rig controls to drill the well, but the automation is in a paused state while the driller is taking a zero, and the machine learning model is running to independently identify the appropriate zero point. In this situation, the modeled zero is purely advisory and does not override the manual zero. While the automation is paused, the driller is controlling the rig and managing the procedure to zero WOB. At this point, the zero is taken and the modeled zero is also identified. If the manual zero is within a user defined tolerance, no alert is presented to the user. On the other hand, if the manual zero is out of tolerance, an alert may be presented to the driller. If the driller or user dismisses the alert, then no action will be taken. If driller or user requests more details, then the application will navigate the driller to the setup screen to see how far out of tolerance the driller is. In either scenario, the driller is responsible for taking a new zero, if desired.

[0060] A third workflow may provide an automated zero and a machine learning model advisory. At any time during drilling, the driller can navigate to the setup screen to adjust the zeroing tolerance and see the zeroing trend. The driller may be utilizing the automated rig controls to drill the well, and the automation is following a standard procedure to capture a zero, and the machine learning model is running to independently identify the appropriate zero point. The modeled zero is purely advisory and does not override the automated zero. At one point, the zero is taken and the modeled zero is identified. If the automated zero is within a user defined tolerance, no alert is presented to the driller user. If the automated zero is out of tolerance, an alert is presented to the driller or user. If the alert is dismissed then no action is taken. If user clicks details, the application will navigate the driller to the setup screen to see how far out of tolerance from the model zero the automated zero was. For either scenario, driller is responsible for taking a new zero.

[0061] The fourth workflow can provide an automated zeroing and model zero ‘in control’ when outside of the user defined threshold. At any time, a driller can navigate to the setup screen to adjust tolerance and see the zeroing trend. The driller may be utilizing the automated rig controls to drill the well, and the automation is following a standard deterministic procedure to capture a zero. As a result, the machine learning model is running to independently identify the appropriate zero point. At one point, a zero is taken and modeled zero is identified. If the automated zero is within a user defined tolerance, no alert is presented to the user. If the automated zero is out of tolerance, an alert is presented to the driller (this may go away over time). If the alert is dismissed, then no action is taken. If the driller or user clicks on the details, the application will navigate the driller to the setup screen to see how far out of tolerance the automated zero was. In either alert scenario, the modeled zero will overwrite the automated zero and be utilized by the autodriller to provide instructions for or control ROP.

[0062] Many types of feedback can be used to provide a feedback loop and training or re-training of the machine learning model(s). For example, driller feedback from interactions with the UI / UX (user interface / user experience) and driller actions taken after out of tolerance zeros are identified can be used for on-going learning and retraining of the machine learning model as needed. As discussed earlier, downhole data for a specific drilled hole can also be used as feedback into the machine learning model where the machine learning model is not performing as desired. The downhole data can be used to help re-label the zero points and where the machine learning model should have zeroed.

[0063] It is also helpful to understand the context of obtaining automated zeros. Automated zeros are dependent on a good procedure. Automated zeros will start to fall out of tolerance when hole conditions or well trajectories change and the driller does not update the automated procedure to capture the zero.

[0064] Some of the functional units described in this specification have been labeled as modules, in order to more particularly emphasize their implementation independence. For example, a module may be implemented as a hardware circuit comprising custom VLSI circuits or gate arrays, off-the-shelf semiconductors such as logic chips, transistors, or other discrete components. A module may also be implemented in programmable hardware devices such as field programmable gate arrays, programmable array logic, programmable logic devices or the like.

[0065] Modules may also be implemented in software for execution by various types of processors. An identified module of executable code may, for instance, comprise one or more blocks of computer instructions, which may be organized as an object, procedure, or function. Nevertheless, the executables of an identified module need not be physically located together, but may comprise disparate instructions stored in different locations which comprise the module and achieve the stated purpose for the module when joined logically together.

[0066] Indeed, a module of executable code may be a single instruction, or many instructions, and may even be distributed over several different code segments, among different programs, and across several memory devices. Similarly, operational data may be identified and illustrated herein within modules, and may be embodied in any suitable form and organized within any suitable type of data structure. The operational data may be collected as a single data set, or may be distributed over different locations including over different storage devices. The modules may be passive or active, including agents operable to perform desired functions.

[0067] The technology described here can also be stored on a computer readable storage medium that includes volatile and non-volatile, removable and non-removable media implemented with any technology for the storage of information such as computer readable instructions, data structures, program modules, or other data. Computer readable storage media include, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other computer storage medium which can be used to store the desired information and described technology.

[0068] The devices described herein may also contain communication connections or networking apparatus and networking connections that allow the devices to communicate with other devices. Communication connections are an example of communication media. Communication media typically embodies computer readable instructions, data structures, program modules and other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media. A “modulated data signal” means a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, communication media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, radio frequency, infrared, and other wireless media. The term computer readable media as used herein includes communication media.

[0069] Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more examples. In the preceding description, numerous specific details were provided, such as examples of various configurations to provide a thorough understanding of examples of the described technology. One skilled in the relevant art will recognize, however, that the technology can be practiced without one or more of the specific details, or with other methods, components, devices, etc. In other instances, well-known structures or operations are not shown or described in detail to avoid obscuring aspects of the technology.

[0070] The following examples are further illustrative of several embodiments of the present technology:

[0071] A method for determining a point for zeroing a weight on bit (WOB) measurement for a drilling rig, comprising:receiving a plurality of drilling features from sensors associated with the drilling rig;determining a point in a drilling procedure where the weight on bit (WOB) for an off bottom rotating weight, is to be zeroed, by detecting a pattern for drilling features while processing the drilling features;zeroing the WOB for the drilling rig at the point; andcomputing a WOB, when a drill of the drilling rig is in contact with a downhole object, based in part on weight added after the WOB has been zeroed for the off bottom rotating weight.

[0072] The method as in example 1, further comprising notifying a user of the drilling rig, a process or applications that the WOB that has been computed.

[0073] The method as in any preceding example, wherein detecting a pattern for drilling features is performed using a machine learning model that is at least one of: a recurrent neural network (RNN), a genetic model, gradient boost, CNN, a UBoost model, regression, reinforcement learning, supervised machine learning, unsupervised machine learning, predictive modeling, pattern recognition, expert rules, classification models, condition based rules, or a heuristic to detect when the WOB is to be zeroed.

[0074] The method as in any preceding example, wherein a time series of the drilling features are input features to the RNN.

[0075] The method as in any preceding example, wherein a time window for drilling features input to the RNN is a defined number of seconds or minutes.

[0076] The method as in any preceding example, wherein the machine learning model is trained using a training data set of the drilling features that are known to have a desired weight on bit (WOB) for an off bottom rotating weight.

[0077] The method as in any preceding example, wherein the machine learning model includes at least one machine learning model is used to process surface features obtained from surface sensors to identify the point when the WOB is to be zeroed and at least one additional machine learning model to receive an output of the at least one machine learning model to identify whether the point to zero the WOB is within a valid window for predictions..

[0078] The method as in any preceding example, further comprising:receiving a manual zeroing;determining whether the manual zeroing is out of tolerance; andsending a machine learning model advisory to a driller, if the manual zeroing is out of tolerance.

[0079] The method as in any preceding example, further comprising:receiving an automated zeroing;determining whether the automated zeroing is out of tolerance; andsending a machine learning model advisory to a driller, if the automated zero is out of tolerance.

[0080] The method as in any preceding example, further comprising:receiving an automated zeroing;determining whether the automated zeroing is out of tolerance; andreceiving zeroing instructions from the machine learning model, if the automated zeroing is out of tolerance.

[0081] A method for determining a point for zeroing a weight on bit (WOB) measurement of a drilling rig, comprising:receiving a plurality of drilling features from sensors associated with the drilling rig;determining the point in a drilling procedure where the weight on bit (WOB), having an off bottom rotating weight, is to be zeroed, using a deep neural network to process the drilling features;zeroing the WOB for the drilling rig at the point; andcomputing a WOB representing the on bottom rotating weight minus the off bottom rotating weight that has been zeroed, when a drill of the drilling rig is in contact with a downhole object.

[0082] The method as in any preceding example, further comprising using the deep neural network that is a recurrent neural network (RNN) to detect when the WOB is to be zeroed.

[0083] The method as in any preceding example, wherein a time series of drilling features are submitted to the RNN.

[0084] The method as in any preceding example, wherein a time window for drilling features input to the RNN is a defined number of seconds or minutes.

[0085] The method as in any preceding example, wherein the deep neural network is trained using a training data set of the plurality of drilling features that are known to have a desired WOB zero point, at an off bottom rotating weight.

[0086] The method as in any preceding example, wherein the WOB is calculated using a total weight on bottom and subtracting the off bottom rotating weight that includes at least one of: a hook load weight, differential pressure or friction factors.

[0087] The method as in any preceding example, wherein the deep neural network includes at least one deep neural network used to process surface features obtained from surface sensors to identify the point when the WOB is to be zeroed and at least one additional deep neural network to receive an output of the at least one additional deep neural network to identify whether the point to zero the WOB is within a valid window for predictions.

[0088] The method as in any preceding example, further comprising receiving drilling features that are at least one of: a hook load weight, data from an electronic drilling recorder (EDR), a well plan, BHA (bottom hole assembly), drilling mud information, and / or geological formation information or downhole measurement while drilling (MWD).

[0089] The method as in any preceding example, further comprising notifying a user of the drilling rig, a process or applications of the point where the WOB was zeroed and notifying the user of the WOB.

[0090] The method as in any preceding example, further comprising performing the steps of claim 8 using process automation and without manual human input.

[0091] The method as in any preceding example, further comprising receiving a plurality of drilling features, including downhole features, from sensors associated with the drilling rig.

[0092] The method as in any preceding example, determining that procedural compliance criteria are met before the zeroing occurs.

[0093] A system for determining a point for zeroing a weight on bit (WOB) measurement for a drilling rig, the system comprising:

[0094] at least one processor;

[0095] at least one memory device including a data store to store a plurality of data and instructions that, when executed, cause the system to:receive a plurality of drilling features from sensors associated with the drilling rig;determine the point in a drilling procedure where the weight on bit (WOB), including an off bottom rotating weight, is to be zeroed using a machine learning model to process the drilling features;zero the WOB for the drilling rig at the point; andcompute a WOB based in part on the on bottom rotating weight minus the off bottom rotating weight that has been zeroed, when a drill of the drilling rig is in contact with a downhole object.

[0096] The system as in any preceding example, wherein a time series of drilling features are input features to the machine learning model that is a recurring neural network (RNN).

[0097] The system as in any preceding example, wherein a time window for drilling features input to the RNN is a defined number of seconds or minutes.

[0098] The system as in any preceding example, wherein a first machine learning model is used to process surface features obtained from surface sensors to identify a WOB point for zeroing and a second machine learning model receives an output of the first machine learning model to identify whether the point to zero the WOB is within a valid window for predictions.

[0099] Although the subject matter has been described in language specific to structural features and / or operations, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features and operations described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims. Numerous modifications and alternative arrangements can be devised without departing from the spirit and scope of the described technology.

Examples

Embodiment Construction

[0008]Reference will now be made to the examples illustrated in the drawings, and specific language will be used herein to describe the same. It will nevertheless be understood that no limitation of the scope of the technology is thereby intended. Alterations and further modifications of the features illustrated herein, and additional applications of the examples as illustrated herein, which would occur to one skilled in the relevant art and having possession of this disclosure, are to be considered within the scope of the description.

[0009]Drillers who are drilling hydrocarbon wells or other wells desire to have the highest possible on-bottom performance and rotary Rate of Penetration (ROP) when drilling the wells. ROP may be dictated by how much energy is put into the drilling system. ROP may be analyzed by looking at a current ROP, weight on bit (WOB), differential pressure(s), rotary speeds, and torque. As part of this process, drillers may desire to understand current drilling ...

Claims

1. A method for determining a point for zeroing a weight on bit (WOB) measurement for a drilling rig, comprising:receiving a plurality of drilling features from sensors associated with the drilling rig;determining a point in a drilling procedure where the weight on bit (WOB) for an off bottom rotating weight, is to be zeroed, by detecting a pattern for drilling features while processing the drilling features;zeroing the WOB for the drilling rig at the point; andcomputing a WOB, when a drill of the drilling rig is in contact with a downhole object, based in part on weight added after the WOB has been zeroed for the off bottom rotating weight.

2. The method as in claim 1, further comprising notifying a user of the drilling rig, a process or applications that the WOB that has been computed.

3. The method as in claim 1, wherein detecting a pattern for drilling features is performed using a machine learning model that is at least one of: a recurrent neural network (RNN), a genetic model, gradient boost, CNN, a UBoost model, regression, reinforcement learning, supervised machine learning, unsupervised machine learning, predictive modeling, pattern recognition, expert rules, classification models, condition based rules, or a heuristic to detect when the WOB is to be zeroed.

4. The method as in claim 3, wherein a time series of the drilling features are input features to the RNN.

5. The method as in claim 4, wherein a time window for drilling features input to the RNN is a defined number of seconds or minutes.

6. The method as in claim 3, wherein the machine learning model is trained using a training data set of the drilling features that are known to have a desired weight on bit (WOB) for an off bottom rotating weight.

7. The method as in claim 3, wherein the machine learning model includes at least one machine learning model is used to process surface features obtained from surface sensors to identify the point when the WOB is to be zeroed and at least one additional machine learning model to receive an output of the at least one machine learning model to identify whether the point to zero the WOB is within a valid window for predictions..

8. A method for determining a point for zeroing a weight on bit (WOB) measurement of a drilling rig, comprising:receiving a plurality of drilling features from sensors associated with the drilling rig;determining the point in a drilling procedure where the weight on bit (WOB), having an off bottom rotating weight, is to be zeroed, using a deep neural network to process the drilling features;zeroing the WOB for the drilling rig at the point; andcomputing a WOB representing the on bottom rotating weight minus the off bottom rotating weight that has been zeroed, when a drill of the drilling rig is in contact with a downhole object.

9. The method as in claim 8, further comprising using the deep neural network that is a recurrent neural network (RNN) to detect when the WOB is to be zeroed.

10. The method as in claim 9, wherein a time series of drilling features are submitted to the RNN.

11. The method as in claim 9, wherein a time window for drilling features input to the RNN is a defined number of seconds or minutes.

12. The method as in claim 11, wherein the deep neural network is trained using a training data set of the plurality of drilling features that are known to have a desired WOB zero point, at an off bottom rotating weight.

13. The method as in claim 11, wherein the WOB is calculated using a total weight on bottom and subtracting the off bottom rotating weight that includes at least one of: a hook load weight, differential pressure or friction factors.

14. The method as in claim 11, wherein the deep neural network includes at least one deep neural network used to process surface features obtained from surface sensors to identify the point when the WOB is to be zeroed and at least one additional deep neural network to receive an output of the at least one additional deep neural network to identify whether the point to zero the WOB is within a valid window for predictions.

15. The method as in claim 11, further comprising receiving drilling features that are at least one of: a hook load weight, data from an electronic drilling recorder (EDR), a well plan, BHA (bottom hole assembly), drilling mud information, and / or geological formation information or downhole measurement while drilling (MWD).

16. The method as in claim 11, further comprising notifying a user of the drilling rig, a process or applications of the point where the WOB was zeroed and notifying the user of the WOB.

17. The method as in claim 11, further comprising performing the steps of claim 8 using process automation and without manual human input.

18. The method as in claim 11, further comprising receiving a plurality of drilling features, including downhole features, from sensors associated with the drilling rig.

19. The method as in claim 11, determining that procedural compliance criteria are met before the zeroing occurs.

20. A system for determining a point for zeroing a weight on bit (WOB) measurement for a drilling rig, the system comprising:at least one processor;at least one memory device including a data store to store a plurality of data and instructions that, when executed, cause the system to:receive a plurality of drilling features from sensors associated with the drilling rig;determine the point in a drilling procedure where the weight on bit (WOB), including an off bottom rotating weight, is to be zeroed using a machine learning model to process the drilling features;zero the WOB for the drilling rig at the point; andcompute a WOB based in part on the on bottom rotating weight minus the off bottom rotating weight that has been zeroed, when a drill of the drilling rig is in contact with a downhole object.

21. The system as in claim 20, wherein a time series of drilling features are input features to the machine learning model that is a recurring neural network (RNN).

22. The system as in claim 21, wherein a time window for drilling features input to the RNN is a defined number of seconds or minutes.

23. The system as in claim 20, wherein a first machine learning model is used to process surface features obtained from surface sensors to identify a WOB point for zeroing and a second machine learning model receives an output of the first machine learning model to identify whether the point to zero the WOB is within a valid window for predictions.

Images