NO347279B1 - Drill bit positioning system - Google Patents

Description

Drill bit positioning system

BACKGROUND

Field of the invention

[0001] The present invention concerns a method for determining a position of a

drill bit during drilling, wherein the method comprises the steps of:

a) recording a pattern of observations from a seismic array on a land surface or a seafloor; b) estimating a drill-bit position;

c) computing a predicted pattern from the estimated drill bit-position

d) comparing the observed and predicted patterns;

e) repeating steps b) - d) until the observed and predicted patterns match with sufficient precision; and

f) repeating steps a) - e) while drilling.

Prior and related art

[0002] In the present application, a formation comprises several layers of diferent materials, e.g. solid rocks, sands and salt deposits. The layers are folded and displaced along faults and fractures, and in some regions form structures of commercial interest hereinafter reservoirs. For example, traps or pockets in the formation may contain hydrocarbons or hot water for the energy industry, and aquifers may be suitable for depositing flue gas and/or CO2. Regardless of application, the reservoirs are accessed through drilled wells.

[0003] Before drilling, a geophysical model of the formation is built using information from various sources. An important source is seismic data acquisition and subsequent inversion. However, inversion does not produce a unique solution - several possible formations can explain the observed data. Hence, an analyst must select the most likely interpretation based on the geology of the area, experience etc. Still, there will be uncertainties in the model that can only be resolved during drilling.

[0004] Drilling a borehole in these applications involves a rotating drill bit on the end of a hollow drill string. The drill string may comprise threaded joints about 9 m (30 feet) each or 'coiled tubing', i.e. a continuous tube with smaller diameter. A drilling mud injected through the drill string cools the drill bit and removes cuttings through an annulus formed between the drill string and the formation. Decades ago, a rotating jointed string with a fixed drill bit was the only option. This method is still widely used for vertical and near vertical wells.

[0005] Today, horizontal wells are typically drilled with the use of a downhole assembly (DHA), which typically comprises a mud motor driven by the drilling mud to rotate the drill bit. The DHA also typically comprise sensors and a controller to adjust the drill hit's angle of attack relative to the DHA housing. This permits geosteering, also called geonavigation. At least 25 companies provide geosteering software and services.

[0006] Geosteering and related applications such as logging-while-drilling (LWD) may comprise resistivity and/or gamma-ray measurements, analysis of cuttings at the surface etc. These measurements are not part of the invention and need no further explanation herein.

[0007] Transmitting data to the surface and/or control signals to the DHA is a challenge. Drill strings with embedded copper or fibre lines are expensive and impractical, especially at threaded connections. Electrical signals conveyed directly through the drill string is severely distorted by capacitances and reflections at every threaded joint, so this alternative is also impractical. Accordingly, mud-pulsing is a widely used alternative to transmit information. However, the transmission rate of mud-pulsing is limited to about 8 b/s or 1 B/s, so much processing is performed in the DHA. This requires accurate data about the formation, which may not be available in a complex geological structure with multiple faults and fractures. Thus, an improved transmission rate would improve flexibility and accuracy in geosteering.

[0008] Eidsvik & Hokstad (2006) [1] and associated patent application W02006/106337 ('337) propose using a seafloor array of seismic receivers for locating a drill bit, in particular by a hyperbolic and a Bayesian approximation. The latter is linearized in a Markov-model and implemented with a Kalman filter (KF). The article [1] also mentions a non-linear filter (Extended KF - EKF), and '337 contains approximate matrices. Embodiments with the drill bit or DHA as source/receiver for complementary receivers and sources on a surface or seafloor are disclosed.

[0009] It is generally acknowledged that EKF can be dificult to tune, mainly because EKF depend on partial derivatives (in a Jacobian), which are approximated by finite differences. Hence, a derivative free unscented KF (UKF), is usually preferred over EKF for estimating the state of a dynamic or non-linear system. In particular, a square-root variety, SR-UKF, is preferred because it guarantees positive definite matrices and eliminates numerical instabilities in very small matrix elements.

[0010] From a signal processing perspective, detecting a signal in a noisy environment is crucial to an improved transmission rate through a drill string and to improve the method of Eidsvik & Hokstad. A few decades ago, advanced sampling was limited to Gibbs sampling and/or the Metropolis-Hastings algorithm. Similarly, noise reduction was limited to filtering, possibly in a transform domain such as FK (2D Fourier transform from time-space TX), tau-p or a wavelet domain. NO201 70488 entitled "Method for denoising seismic data from a seafloor array" and assigned to the present assignee provides an example of filtering that exploits the tau-p transform's ability to distinguish acoustic sources.

[0011] Notable contributors to advances in signal and image processing since the 1980s include Ingrid Daubechies, David Donoho, Emmanuel Candes and their co-workers. For example, the JPEG2000 standard uses Daubechies wavelets to transform an image and store the most significant wavelet coeficients. Even with a compression rate 1:43, an inverse transform (synthesis) recover the original image with barely visible degradation. Donoho's work includes development of efective methods in multiscale geometric analysis. In the early 2000s, Donoho and Candés published a series of papers on certain wavelets called curvelets and associated curvelet transforms useful for seismic applications and image denoising.

[0012] The current field of compressed sensing (CS aka sparse sampling) may be traced back to a seminal paper by Candés et al. (2004) [2], which disclosed statistical conditions tore covery of a signal from highly incomplete frequency data, i.e. a sparse representation without an explicit synthesis formula such as an inverse wavelet transform. A comprehensive list of Candés' papers is available at https://statweb.stanford.edu/~candes/publications.html. A google search reveals contributions from Daubechies, Donoho and others to CS.

[0013] In general, advances in electronics, computer technology, information technology, statistics etc. provide an increasingly large toolbox for building applications. For example, computer networks and statistical inference with a dash of Bayesian statistics enable generally known applications such as Google and Facebook to provide a user with customized advertisements and/or search results in near real time based on the user's search history. In addition to these ' big data' applications, machine-learning algorithms using support vectors, forests and neural networks use statistical techniques. Essentially, statistical techniques may extract a small amount of information from a large ’noisy' dataset based on similarities in the required information. In contrast, an explicit search in a structured database or unstructured data requires explicit a priori information, i.e. known or assumed similarities in the required information.

[0014] In the present context, lifting algorithms enable wavelet constructions without explicit expressions for the wavelet, and sparse sampling may replace a need for a priori noise reduction in a full dataset Specifically, increased randomness and sparsity (up to a limit) increases the probability of recovery. See, for example, Dragotti (tutorial 2015) [3] for examples and further references.

[0015] Riel et al. (2014) [4] describes efficient transient detection using linear sparse estimation techniques, in particular to detect onset and duration of slips of a few mm in geodetic time series over several years in arrays of GPS-stations. The onsets occur randomly, and are not modelled by a Gaussian distribution. Rather, Bayesian sampling provides uncertainties for transient amplitudes and durations. The research article also describes spatial weighting filters for spatially coherent signals.

[0016] US Patent. No. 4460059 describes a method for determining the position of a drill bit in the earth without interrupting the drilling operations. In operation, rotation of the drill bit against the formation being drilled generates coherent acoustical signals which are recorded at the surface of the earth by a plurality of spaced detectors. The signals recorded at each of the different detectors are time shifted relative to each other. These time shifts correspond to possible locations of the drill bit within the earth and are controlled to some degree by the length of drill pipe in the borehole. After the acoustic signals are shifted in time their coherency is determined. This procedure is repeated for a number of assumed locations of the drill bit. The drill bit position is determined to be at the location having the highest coherency value. In this manner the surface detectors are focused on precise positions within the earth avoiding interference from acoustical signals generated at the surface or from the drill pipe.

[0017] The article “Advanced Data Communications for Downhole Data Logging and Control Applications in the Oil Industry” by C. T. Spracklen and T. Aslam describes a high-speed downhole communications system that utilizes the (metallic) wall of a gas or oil pipeline or a drill ‘string’ as the communications ‘channel’ to control or monitor equipment or sensors used in the oil industry. The system uses Orthogonal Frequency Division Multiplexing (OFDM) coupled with Wideband Frequency Division Multiple Access (W-CDMA).

[0018] US Patent No. 5148048 describes a method of acoustically transmitting data signals where the optimum transmission frequencies for transmitting and receiving acoustic data signals are determined by use of at least two spaced acoustic transmitter/receiver pairs located at or near opposed ends of the drillstring. One acoustic transmitter will transmit at different frequencies while transmitted signal characteristics are monitored by the acoustic receiver at the other end of the drillstring. As a result, the optimum frequencies are determined for that particular drillstring geometry.

[0019] A main objective of the present invention is to provide a method for improved drill bit positioning while retaining benefits from prior art.

SUMMARY OF THE INVENTION

[0020] According to the invention, a method is provided, for determining a position of a drill bit during drilling:

a subsurface surrounding the drill bit is divided into a network of cells, and a sound burst from each cell having a unique arrival time pattern at the surface, said measured arrival time is recorded, and the suficient precision ~ which is a maximum coherence between the measured arrival time and a precalculated theoretical value for each cell - is found.

[0021] It should be noted that the drill bit is positioned in real time / near real time. This also gives the application distinctive character and is an important element for the value of the invention.

[0022] Further, the drill bit application can be combined with real-time detection of fracturing of formations (caused by a drilling operation). An example where the combination may be important is in connection with the loss of drilling fluid into a formation. If one measures the drill bit position and at the same time manage to locate where a loose of drilling fluid in the borehole takes place (along with conventional methods for measuring drill bit depth and monitoring return of circulated drilling fluid / drilling fluid pressure) one will be able to use these observations to make faster decisions.

[0023] According to one embodiment, recording a pattern of observations includes denoising collected data.

[0024] According to one embodiment, estimating the drill bit-position includes using physical constraints.

[0025] According to one embodiment, computing a predicted pattern includes selecting a characteristic frequency.

[0026] According to one embodiment, the method comprises a step of estimating the characteristic frequency in a downhole assembly (DHA).

[0027] According to one embodiment, the method further comprises the step of transmitting data acoustically through a drill string.

[0028] According to one embodiment, the method further comprises the step of dividing a frequency band used for transmission into several orthogonal subchannels.

[0029] According to one embodiment, the method further comprises the step of maintaining a symbol in a subchannel during a timeslot suficiently long to permit summation of several samples.

[0030] According to one embodiment, a subsurface surrounding the drill bit is divided into a network of cells, and a sound burst from each cell have a unique arrival time pattern at the surface.

[0031] According to one embodiment, a calculated drill bit position can be/is a fed back to the drilling system for adjustment of drilling parameters in order to correct for any deviations from a planned drilling path or to adjust for any obstacles that are detected.

[0032] The invention is a method for determining a position of a drill bit during drilling, based on recording a pattern of acoustic observations from a seismic sensor network placed on a land surface or a seafloor above the drill bit and estimating the drill-bit position from the noise emitted by the drill bit Predicted patterns of the drill bit is provided and a position is calculated from the estimated drill bit-position, comparing the observed and predicted patterns. This is repeated until the observed and predicted patterns match with suficient precision. Denoising of the data by use of various filtering techniques may be performed prior to calculation of drill bit position, in order to increase the signal-to-noise-ratio of the signal.

[0033] The method used in the invention include the use of a network of sensors on the seafloor directly connected to a data acquisition and analysis system on the drilling rig. The layout of the network is carefully planned to optimize drill bit detection capabilities and drill bit detection position accuracies. A computer program reads data when acquired and scans through for signal of possible significance (noise from the drill bit). Having a detailed sound velocity model for the drilling location, modelling can be performed and a detailed database of travel times from each planned location in the subsurface can be calculated. Dividing the subsurface surrounding the drill bit into a network of cells, a sound burst from each cell will have a unique arrival time pattern at the surface. Using a pattern recognition scheme, a measured pattern can be compared to a planned pattern and a detection can be confirmed according to a pre-set required coherence level. Prior to detection, acoustic noise from other sources like vessel noise, noise form the rig, weather noise or seismic shooting should be removed- The sensor array layout will be optimized for noise removal. Special consideration with respect noise aliasing and sensor spacing, array geometry etc. are important. The calculated drill bit position can thus be fed back to the drilling system for adjustment of drilling parameters in order to correct for any deviations from planned drilling path or to adjust for any obstacles that are detected.

BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The invention will be explained with reference to exemplary embodiments and the accompanying drawings, in which:

Fig. 1 (prior art) illustrates communication over several subchannels using phase shift keying (PSK).

Fig. 2 (prior art) illustrates an observed noisy signal in two adjacent timeslots.

Fig. 3 illustrates mean frequencies estimated at the DHA during drilling.

Fig. 4 illustrates the diferent steps of the invention.

Fig. 5 illustrates the position of an oilrig in relation to a selected area and a cell structure.

DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT

[0035] The drawings are schematic and not necessarily to scale. For ease of understanding, numerous details known to the skilled person are omitted from the drawings and following description.

[0036] Fig. 1 (prior art) illustrates an orthogonal frequency multiplexing (OFDM) scheme, e.g. as used in mobile communication. In the present example, a frequency band is divided into eight subchannels C1-C8, and a harmonic or sinusoid signal may have one of four phases Pl-P4 during a timeslot Si, each lasting long enough to detect the phase P1-P4 reliably, e.g. ~ls. Fig. 1 shows three timeslots S1-S3. Noise in one channel does not necessarily affect the signal in adjacent channels.

[0037] Each phase P1-P4 represent a 2-bit combination, namely 00, 01, 10 and 11. This is an example of phase shift keying (PSK), in particular 4-PSK. Current mobile communication us 16-PSK and 32-PSK where the 16=24 or 32=2s phases represent 4, respectively 5, bits.

[0038] Assuming that each timeslot S1-S3 lasts one second, the scheme in Fig 1 transmits 2 bits times 8 channels = 16 bits per second (b/s). This number includes a payload and parity bits for error correction. Forward error correction codes are well known, and include Hamming [7, 4] (1950), Reed-Solomon (1960), BCH (1959 - 1960), LDPC (1960) and Turbo (early 1990s). The fundamental patent for turbo coding, US 5,446,747, expired in 2013. Error correction and other aspects of coding theory are outside the scope of the present invention.

Here, we observe that the efective transmission rate exceeds 8 b/s whenever the number of parity bits are less than the number of payload bits. Moreover, the number of channels need not be limited to 8, and 8-PSK may replace 4-PSK.

[0039] Applied to the task of transmitting signals between a DHA and equipment on the surface, acoustic signals through the drillstring are a potentially suitable channel as it avoids the impracticalities of separate communication lines and the distortions associated with electrical signals through a jointed string and acoustic signals through a highly non-linear formation. Acoustic signals through the drillstring should be compared to the currently used mud-pulsing.

[040] The scheme in Fig. 1 depends on the ability to separate subchannels C1 -C8. This is conventionally done by quadrature filters, which incidentally led to the development of wavelet transforms and lifting algorithms. To produce the signal, assume one 'hammer' per channel pinging on the drillstring. Disregarding the actual Bessel-signals produced by a hammer, the hammering produces an envelope signal that may be approximated by a sinusoid. To reduce the effect of reflections at the threaded joints, suitable carrier frequencies are in the range 500-600 Hz. These carrier frequencies correspond to wavelengths greater than the length of joints, i.e, > ~9m (30 feet), and depend on the speed of sound -6000 m/s in (stainless) steel. In other words, the wavelength of the envelope signal makes the threaded joints 'invisible'.

[0041] Considering the length of the timeslots S1 -S3 about 1 s, a sampling frequency of 10 Hz permits addition of about 10 samples. The summation enhances a coherent signal and removes incoherent (random) noise, which contributes negatively equally well as positively. The addition of several signals also reduces the need for large amplitudes in the transmitted signal. Thus, depending on the sampling frequency, piezoelectric devices with smaller amplitudes may replace 'hammers', e.g. driven by induction that provide larger amplitudes.

[0042] The scheme in Fig. 1 also depends on the ability to separate phases P1-P4. Fig. 2 illustrates a transient φ(1) shifting from phase P2 to P1 at a known point in time, nT, corresponding to a boundary between adjacent timeslots Si, S(i+1) in Fig. 1. A shaded area illustrates one standard deviation from the transient due to ambient Gaussian (white) noise. Sparse sampling may recover the transient signal φ(t) suficiently well to determine whether 30 the phase in one timeslot is P2 or PI. Noise reduction before signal processing is not strictly required. Similar statistical techniques form the basis of support vectors in machine learning. Applied to Fig.2, such an algorithm would determine whether an observation falls into 'class' P2 or Pi.

[0043] Fig. 3 illustrates mean frequencies f(t) estimated at the DHA during drilling. The observed signal is noisy, and the noise is well approximated by a Gaussian distribution. The dip in frequencies at t1 corresponds to the drill bit entering a fault. The raise at t2 is caused by the drill bit entering a harder rock, which is indicated by higher frequencies. The events at tl and t2 cannot be approximated by a simple global distribution, and they do not appear at fixed intervals such as the phase transitions in Fig. 2. However, tl and t2 can be predicted from a geophysical model of the formation, and they can be used to update the model. For example, a fracture may have a diferent width than expected, and some fractures may be discovered. Such information is crucial to avoid accidents due to leakage from an oil or gas reservoir.

[0044] Although current (signal) processors in the DHA may provide the most significant components in a frequency spectrum, the times tl and t2 etc., there is a limit to the transmission capacity to the surface. Perhaps more important, there is a need for a feedback loop through the formation in order to determine the position of the drill bit and to verify the geophysical model in an iterative process.

0045] This objective can be realized by precomputing how a wave with a characteristic frequency, e.g. fl in Fig. 3, would propagate through the formation and be received at an array on the surface or the seafloor. More specifically, the wave may be an S-wave as estimated at the DHA, and the computing might yield an expected pattern, i.e. a prediction of what the array should measure based on the assumed bit-position.

[0046] Fig. 4 illustrates a method 100 according to the invention in greater detail.

[0047] Step 110, recording a pattern, involves collecting data from a seismic array on a land surface or a seafloor and arrange the data in a suitable observed pattern. This step may include step 111, denoising the collected data as an explicit step before further processing. Step 11 1 is optional because denoising may be implicit in CS-techniques.

[0048] Step 120 involves estimating a drill bit position. The estimation may, for example, involve a Kalman filter, in which the a posteriori estimate is an a priori estimate updated with a weighted sum of an observation and a prediction. Step 121 illustrates that the possible positions are limited by physical constraints such as last position, drilling speed and/or inclination of the DHA.

[0049] Step 130, computing a predicted pattern from a drill bit-position, involves propagating a wave from an assumed drill bit position through the geophysical model using known methods. Variational methods may speed up the computation.

[0050] Preferably, step 130 includes a step 131: Selecting a characteristic frequency. The characteristic frequency should be 'typical' for the rock being drilled such as fl in Fig.3.

[0051] In step 140, comparing observed and predicted patterns, a simple comparison involves subtracting one dataset from the other. Alternatively, any known pattern recognition technique may be employed. Franek et al. (2007) [5] describes a fast algorithm and contains references to other known algorithms.

[0052] Test 150 determines whether the predicted pattern matches the observed pattern with suficient precision. If not, the process loops back 151 to step 120 for a new estimate of the drill bit-position. This process is repeated until the patterns match. Then, the process loops 152 to step 1 10 to collect new observations from the array.

[0053] Fig. 5 illustrates the position of an oilrig 200 in relation to a selected area 202 and for providing a focus area divided into a cell structure around the planned drilling wellpath. A drill bit 204 is connected to the oilrig 200 through a drillstring 201. The drill bit 204 is emitting signals 205 - which is noise arising from the drilling. The signals 205 are picked up by a seafloor sensor network/array 206 placed at the seafloor 203.

[0054] A focus area and a cell structure is selected. Then a curve for theoretical arrival time for events for each cell to each receiver is calculated. A measured arrival time is recorded and the maximum coherence between this value and the precalculated theoretical value is found.

[0055] While the invention has been described by way of examples, its scope is defined by the attached claims.

Claims (9)

Claims

1. A method (100) for determining a position of a drill bit during drilling, comprising the steps of:

a) Recording (110) a pattern of observations from a seismic array on a land surface or a seafloor; b) Estimating (120) a drill-bit position;

c) Computing (130) a predicted pattern from the estimated drill bit-position

d) Comparing (140) the observed and predicted patterns;

e) Repeating (151) steps b) – d) until the observed and predicted patterns match with sufficient precision; and

f) Repeating (152) steps a) – e) while drilling characterized in

g) that a subsurface surrounding the drill bit is divided into a network of cells, and a sound burst from each cell having a unique arrival time pattern at the surface, said measured arrival time is recorded, and the sufficient precision - which is a maximum coherence between the measured arrival time and a precalculated theoretical value for each cell - is found.

2. The method according to claim 1, wherein recording (110) a pattern of observations includes denoising (111) collected data.

3. The method according to claim 1 or 2, wherein estimating (120) the drill bit-position includes using (121) physical constraints.

4. The method according to any preceding claim, wherein computing (130) a predicted pattern includes selecting (131) a characteristic frequency (f1).

5. The method according to claim 4, further comprising a step of estimating the characteristic frequency (f1) in a downhole assembly (DHA).

6. The method according to any preceding claim, further comprising the step of transmitting data acoustically through a drill string.

7. The method according to claim 6, further comprising the step of dividing a frequency band used for transmission into several orthogonal subchannels (C1-C8).

8. The method according to claim 6 or 7, further comprising the step of maintaining a symbol in a subchannel (C1-C8) during a timeslot (S1-S3) sufficiently long to permit summation of several samples.

9. The method according to any preceding claim, wherein a calculated drill bit position can be/is a fed back to the drilling system for adjustment of drilling parameters in order to correct for any deviations from a planned drilling path or to adjust for any obstacles that are detected.