Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendments
Applicant’s amendments filed on 03/12/2026 have been entered. Claims 1-20 are pending in this application of which claims 1, 13, and 19 are independent.
Response to Arguments
Applicant’s arguments, on Page 8 and 9, regarding 35 USC 101 are considered and found persuasive in view of the amendments. Therefore, the 35 USC 101 rejections on claims 19 and 20 are withdrawn.
Applicant’s arguments, on Page 10, 11, 12, and 13, regarding 35 USC 102 and 35 USC 103 are considered and is not found persuasive for the following reasons.
Applicant states in their remarks “the machine learning model of Moseley appears to only receive the velocity model as input” (Remarks Pg. 12), but this overlooks Moseley’s full wavefield embodiment in Section III, which is relied upon. Moseley states “The input to the network is the current and previous wavefield frames concatenated together and the output is a prediction of the wavefield at the next time step.” (Pg. 6), and further that “We condition the network on the input 2D velocity model by concatenating the velocity model to the input of each convolutional layer.”. (Pg. 6). The previous wavefield frames are approximated wavefields generated by propagating through the velocity model and accordingly read on the claimed “guide image based upon at least one attribute of the velocity model,” which is consistent with Applicant’s claim. The 8 Hz Ricker source disclosed in Pg. 5 is the source wavelet data that drives the ground truth. This use of the Ricker source wavelet data encompasses source wavelet data used to train the machine learning system. Therefore, the argument is not persuasive, and the 35 USC 102 and 103 rejections are maintained.
It is further noted that in the remarks applicant argues the claim language “A device, comprising: an input that when in operation receives a velocity model corresponding to at least one attribute of seismic data, source wavelet data corresponding to the seismic data, and a guide image based upon at least one attribute of the velocity model; and a machine learning system that is trained using the velocity model, the source wavelet data, and the guide image to generate a wavefield solution corresponding to the velocity model.”. (Page. 9). However, the entered amended claim 19 is not the same, stating “A device, comprising: an input that when in operation receives a velocity model corresponding to at least one attribute of seismic data, source wavelet data corresponding to the seismic data, and a guide image based upon at least one attribute of the velocity model; and a machine learning system that the velocity model, the source wavelet data, and the guide image to generate a wavefield solution corresponding to the velocity model.”.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION. —The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 19 is rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 19 is indefinite because the claim entered omits the phrase “is trained using” that the applicant argues on Page 9 of the remarks. This omission leaves the claim limitation grammatically incomplete and of indeterminate scope. Correction is required.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 1-9, 13-18, and 19-20 is/are rejected under 35 U.S.C. 102 as being anticipated over Moseley et al. “Fast approximate simulation of seismic waves with deep learning” (2018) [herein “Moseley”].
Regarding Claim 1, Moseley teaches
A method, comprising:receiving a velocity model corresponding to at least one attribute of seismic data;
“We define the input velocity model to be a 1D profile of a horizontally layered Earth velocity model, with a depth of 3.2 km and a sample rate of 12.5 m.” (Pg. 5 Section II B).
“Our proof of concept deep neural network is trained using 50,000 synthetic examples of seismic waves propagating through different 2D horizontally layered velocity models.” (Abstract).
Moseley explicitly states a velocity model as an input with related seismic data.
receiving source wavelet data corresponding to the seismic data;
“We use an 8 Hz Ricker source emitted close to the surface and record the pressure response at 11 receiver locations placed symmetrically around the source, horizontally offset every 200 m (Fig. 1, top left).” (Pg. 5 Section II C).
A source wavelet i.e. ricker wavelet corresponding to seismic data is selected.
generating a guide image based upon at least one attribute of the velocity model;
“Instead of predicting individual receiver responses we design a deep convolutional network which, given the two previous time steps of the pressure wavefield, predicts the next time step of the wavefield over all points in space. This allows the network to be used iteratively to predict the evolution of the full wavefield over multiple time steps, in a fashion similar to FD modelling.” (Pg. 6 Overview Para. 2).
The previous wavefield frame helps predict the next which is akin to a “guide image” to predict the wavefield solution.
transmitting the velocity model, the source wavelet data, and the guide image to a machine learning system; and
“FIG. 7. Full wavefield simulation over time using our deep convolutional network. We recursively predict the evolution of an initial wavefield (left-most frame) in our validation set using our deep convolutional network (middle), compared to the ground truth FD modelling (top). We show the prediction at t = 0.00, 0.08, 0.16 and 0.24 s (left to right) and its difference to ground truth (bottom). The input velocity model is also shown (bottom right).” (Pg. 7 Fig. 7).
“We use an 8 Hz Ricker source emitted close to the surface and record the pressure response at 11 receiver locations placed symmetrically around the source, horizontally offset every 200 m (Fig. 1, top left).” (Pg. 5 Section II C).
training the machine learning system into a trained machine learning system using the velocity model, the source wavelet data, and the guide image.
“We condition the network on the input 2D velocity model by concatenating the velocity model to the input of each convolutional layer.” (Pg. 6 Section III B).
“For each training example the network is used to recursively predict 11 time steps ahead, using the output prediction at each time step as the current wavefield input for the next time step. Our L2 loss function is then given by” (Pg. 7 III C).
“We use an 8 Hz Ricker source emitted close to the surface and record the pressure response at 11 receiver locations placed symmetrically around the source, horizontally offset every 200 m (Fig. 1, top left).” (Pg. 5 Section II C).
Moseley discloses a deep learning method that receives an input velocity model, utilizes a Ricker source i.e. wavelet to generate ground truth simulations for training, and employs the previous frames a guide image for wavefield evolution.
Regarding Claim 2, Moseley teaches
The method of claim 1, comprising generating, at the trained machine learning system, a wavefield solution corresponding to the velocity model.
“Training data is generated using the same workflow as Section II C. For these simulations we also randomly vary the location of the source as well as the velocity model. We generate 5000 simulations and from each simulation extract 8 training examples. Each training example contains the previous wavefield, the current wavefield and 11 future wavefields, over different starting time steps.” (Section III C).“Our training loss converges and we assess the performance of our trained network using a validation set of 200 unseen examples. The full wavefield prediction over multiple time steps for 1 randomly selected example in the validation set is shown in Fig. 7. For the example shown the trained convolutional network is able to approximate the update equation given by Eq. 4. The predicted wavefield expands outward and reflections occur at velocity boundaries. The speed and shape of the wavefront also changes when entering different velocity layers, as expected.” (Section III D).
“Full wavefield simulation over time using our deep convolutional network. We recursively predict the evolution of an initial wavefield (left-most frame) in our validation set using our deep convolutional network (middle), compared to the ground truth FD modelling (top). We show the prediction at t = 0.00, 0.08, 0.16 and 0.24 s (left to right) and its difference to ground truth (bottom). The input velocity model is also shown (bottom right).”. (Fig. 7).
Moseley teaches generating a wavefield solution corresponding the velocity model by prediction.
Regarding Claim 3, Moseley Teaches
The method of claim 2, comprising applying the wavefield solution in a migration operation to characterize a reservoir in a subsurface region of Earth.
“Seismic simulations are invaluable in many areas of geophysics. In earthquake monitoring, they are a key tool for quantifying the ground motion of potential earthquakes [1]. In oil and gas prospecting, they are used to understand the seismic response of hydrocarbon reservoirs [2, 3]. In geophysical surveying, they show how the subsurface is illuminated by different survey designs [4]. In global geophysics, seismic simulations are invaluable for obtaining snapshots of the Earth’s interior dynamics [5] and for deciphering source or path effects from individual seismograms [6].” (Section I Para. 1).
Regarding Claim 4, Moseley teaches
The method of claim 2, comprising receiving a second velocity model and transmitting the second velocity model to the trained machine learning system.
“During training both the test loss and the training loss converge to similar values, suggesting the network is able to generalise over different input velocity models.” (Pg. 5 Section II E).
“Similar to Section II, we only predict the 2D acoustic pressure response (Eq. 1) and keep the density and the size of the Earth model fixed. We train the network to predict the wavefield evolution over time for different 2D horizontally layered velocity models and different starting wavefields as input.” (Pg. 6 Section III A).
“We expect the network to generalise well over unseen velocity models.” (Pg. 4 Section II A).
Moseley teaches being able to generally use different velocity models i.e. “second model” to produce corresponding wavefield results.
Regarding Claim 5, Moseley teaches
The method of claim 4, comprising generating, at the trained machine learning system, a second wavefield solution corresponding to the second velocity model.
“We expect the network to generalise well over unseen velocity models.” (Pg. 4 Section II A).
Moseley teaches being able to generally use different velocity models i.e. “second model” to produce corresponding wavefield results.
Regarding Claim 6, Moseley teaches
The method of claim 5, comprising applying the second wavefield solution in a migration operation to characterize a reservoir in a subsurface region of Earth.
“Seismic simulations are invaluable in many areas of geophysics. In earthquake monitoring, they are a key tool for quantifying the ground motion of potential earthquakes [1]. In oil and gas prospecting, they are used to understand the seismic response of hydrocarbon reservoirs [2, 3]. In geophysical surveying, they show how the subsurface is illuminated by different survey designs [4]. In global geophysics, seismic simulations are invaluable for obtaining snapshots of the Earth’s interior dynamics [5] and for deciphering source or path effects from individual seismograms [6].” (Pg.1 Section I).
“In Full Waveform Inversion (FWI), a strategy quickly becoming widespread in the field of seismic imaging, forward simulations are used thousands of times to iteratively estimate a medium’s elastic properties [8].” (Pg.1 Section I).
Moseley teaches iteratively applying generated wavefield solutions to characterize a reservoir subsurface.
Regarding Claim 7, Moseley teaches
The method of claim 5, wherein generating the second wavefield solution comprises utilizing the guide image at the trained machine learning system.
“The input to the network is the current and previous wavefield frames concatenated together and the output is a prediction of the wavefield at the next time step.” (Pg. 6 Section B).
“We train the network to predict the wavefield evolution over time for different 2D horizontally layered velocity models and different starting wavefields as input.” (Pg. 6 Section A).
Moseley teaches utilizing a “guide image” (frames) for the second wavefield solution as the design allows for different wavefields and different velocity models.
Regarding Claim 8, Moseley teaches
The method of claim 5, wherein generating the second wavefield solution comprises utilizing second source wavelet data corresponding to the seismic data at the trained machine learning system.
“For these simulations we also randomly vary the location of the source as well as the velocity model. We generate 5000 simulations and from each simulation extract 8 training examples. Each training example contains the previous wavefield, the current wavefield and 11 future wavefields, over different starting time steps.” (Pg. 7 Section III C).
“Our training loss converges and we assess the performance of our trained network using a validation set of 200 unseen examples.” (Pg. 7 Section III D).
Moseley shows that its system allows for a second source wavelet as it can use different source locations for source data for each velocity model that is evaluated.
Regarding Claim 9, Moseley teaches
The method of claim 1, wherein the attribute of the velocity model comprises an approximated wavefield of the velocity model.
“The input to the network is the current and previous wavefield frames concatenated together and the output is a prediction of the wavefield at the next time step.” (Pg. 6 Section III B).
“The predicted wavefield expands outward and reflections occur at velocity boundaries. The speed and shape of the wavefront also changes when entering different velocity layers, as expected.” (Pg. 7 Section III D).
Moseley shows a “guide image” (previous wavefield) is an approximate wavefield whose properties are of the velocity model. As the approximation is based on the velocity model.
Claim 13 recites substantially the same limitations as a combination of claims 1-2 except these claims are directed to a “A tangible and non-transitory machine readable medium, comprising instructions to cause a machine learning system to:” Therefore, these claims are rejected under the same rationale as addressed above.
Claims 14-18 recite substantially the same limitations as claims 1-9 except these claims are directed to a “The tangible and non-transitory machine readable medium of claim [X], comprising instructions to cause the machine learning system to”. Therefore, these claims are rejected under the same rationale as addressed above.
Regarding Claim 19, Moseley teaches
A device, comprising:
an input that when in operation receives a velocity model corresponding to at least one attribute of seismic data,
“We define the input velocity model to be a 1D profile of a horizontally layered Earth velocity model, with a depth of 3.2 km and a sample rate of 12.5 m.” (Pg. 5 Section II B).
“Our proof of concept deep neural network is trained using 50,000 synthetic examples of seismic waves propagating through different 2D horizontally layered velocity models.” (Abstract).
Moseley explicitly states a velocity model as an input with related seismic data.source wavelet data corresponding to the seismic data,
“We use an 8 Hz Ricker source emitted close to the surface and record the pressure response at 11 receiver locations placed symmetrically around the source, horizontally offset every 200 m (Fig. 1, top left).” (Pg. 5 Section II C).
A source wavelet i.e. ricker wavelet corresponding to seismic data is selected.and a guide image based upon at least one attribute of the velocity model; and
“Instead of predicting individual receiver responses we design a deep convolutional network which, given the two previous time steps of the pressure wavefield, predicts the next time step of the wavefield over all points in space. This allows the network to be used iteratively to predict the evolution of the full wavefield over multiple time steps, in a fashion similar to FD modelling.” (Pg. 6 Overview Para. 2).
The previous wavefield frame helps predict the next which is akin to a “guide image” to predict the wavefield solution.
a machine learning system that the velocity model, the source wavelet data, and the guide image to generate a wavefield solution corresponding to the velocity model.
“Training data is generated using the same workflow as Section II C. For these simulations we also randomly vary the location of the source as well as the velocity model. We generate 5000 simulations and from each simulation extract 8 training examples. Each training example contains the previous wavefield, the current wavefield and 11 future wavefields, over different starting time steps.” (Section III C).“Our training loss converges and we assess the performance of our trained network using a validation set of 200 unseen examples. The full wavefield prediction over multiple time steps for 1 randomly selected example in the validation set is shown in Fig. 7. For the example shown the trained convolutional network is able to approximate the update equation given by Eq. 4. The predicted wavefield expands outward and reflections occur at velocity boundaries. The speed and shape of the wavefront also changes when entering different velocity layers, as expected.” (Section III D).
“Full wavefield simulation over time using our deep convolutional network. We recursively predict the evolution of an initial wavefield (left-most frame) in our validation set using our deep convolutional network (middle), compared to the ground truth FD modelling (top). We show the prediction at t = 0.00, 0.08, 0.16 and 0.24 s (left to right) and its difference to ground truth (bottom). The input velocity model is also shown (bottom right).”. (Fig. 7).
“The input to the network is the current and previous wavefield frames concatenated together and the output is a prediction of the wavefield at the next time step.” (Pg. 6).
“We condition the network on the input 2D velocity model by concatenating the velocity model to the input of each convolutional layer.”. (Pg. 6).
“We use an 8 Hz Ricker source emitted close to the surface and record the pressure response at 11 receiver locations placed symmetrically around the source…”. (Pg. 5).
Moseley teaches generating a wavefield solution using a wavelet, guide image, and velocity model.
Regarding Claim 20, Moseley teaches
The device of claim 19, comprising an output that when in operation transmits the wavefield solution for use in a migration operation to characterize a reservoir in a subsurface region of Earth.
“Seismic simulations are invaluable in many areas of geophysics. In earthquake monitoring, they are a key tool for quantifying the ground motion of potential earthquakes [1]. In oil and gas prospecting, they are used to understand the seismic response of hydrocarbon reservoirs [2, 3]. In geophysical surveying, they show how the subsurface is illuminated by different survey designs [4]. In global geophysics, seismic simulations are invaluable for obtaining snapshots of the Earth’s interior dynamics [5] and for deciphering source or path effects from individual seismograms [6].” (Section I Para. 1).
Moseley explains the purpose of wavefield solutions in pursuit of characterizing subsurface regions of the earth.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or non-obviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claim(s) 10-11 is/are rejected under 35 U.S.C. 103 as being unpatentable over Moseley et al.
“Fast approximate simulation of seismic waves with deep learning” (2018) [herein “Moseley”], and “WAVEFIELD RECONSTRUCTION INVERSION VIA PHYSICS-INFORMED NEURAL NETWORKS” (2021) by Song et al [herein “Song”].
Regarding Claim 10, Moseley et al do not explicitly teach but Song teaches
The method of claim 9, comprising determining the approximated wavefield of the velocity model based on a straight line travel time of a wave of the velocity model.
“We use the scattered wavefield, δu = u − u0, as an alternative solution to get the wavefield. The Lippmann Schwinger form of the acoustic wave equation is shown as [65]:” (Section 2.2).
(Please refer to Equations 4 - 7).
It would have been obvious to one skilled in the art before the effective filing date of the
claimed invention to incorporate the teachings of Song’s straight line travel time of a wave for wavefield estimation with Moseley’s machine learning system. The motivation for doing so would have been to use “… the underlying physical laws as loss functions to train the neural network (NN), and it has shown its effectiveness in solving the Helmholtz equation and generating Green’s functions, specifically for the scattered wavefield.”. (Abstract).
Regarding Claim 11, Moseley et al do not explicitly teach but Song teaches
The method of claim 9, comprising determining the approximated wavefield of the velocity model based on a travel time of a diagonal or another chosen direction of a wave of the velocity model.
“For 3D isotropic case, the analytical solution for constant velocity and a point source located at xs is given by:”. (Section 2.2). (Please refer to Equations 4 - 7).
It would have been obvious to one skilled in the art before the effective filing date of the claimed invention to incorporate the teachings of Song’s directional travel time of a wave for wavefield estimation with Moseley’s machine learning system. The motivation for doing so would have been to use “… the underlying physical laws as loss functions to train the neural network (NN), and it has shown its effectiveness in solving the Helmholtz equation and generating Green’s functions, specifically for the scattered wavefield.”. (Abstract).
Claim(s) 12 is/are rejected under 35 U.S.C. 103 as being unpatentable over Moseley et al.
“Fast approximate simulation of seismic waves with deep learning” (2018) [herein “Moseley”], and US 20170038490 A1 (2017) by HU et al [herein “HU”].
Regarding Claim 12, Moseley et al do not explicitly teach but Hu teaches
The method of claim 9, comprising determining the approximated wavefield of the velocity model based on a stretched wavefield travel time of a wave of the velocity model.
“This, and other aspects, can include one or more of the following features. The ray-equation method is implemented for the Kirchhoff integral method. In some instances, a multi-parameter Green's function can be computed based on ray-tracing in either depth or converted/stretched time domain. In some implementations, the ray-equation based Kirchhoff integral method for composite velocity model is used to image both free surface multiples and primary reflections of VSP data simultaneously. In some instances, ray parameters can be computed based on gradients of travel time fields computed based on a velocity model for the VSP data geometry... “. (005).
It would have been obvious to one skilled in the art before the effective filing date of the claimed invention to incorporate the teachings of HU’s stretched travel time of a green’s function calculation for wavefield estimation with Moseley’s machine learning system. The motivation for doing so would have been to “… analyze location and geology of reservoirs that contain hydrocarbons and can be used to design a drilling process for placing wellbores in the earth to maximize oil or gas production.”. (0036).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
US-9946974-B2 by BURCH et al teaches a method that includes training, via a computing system, a well performance predictor based on field data corresponding to a hydrocarbon field in which a well is to be drilled.
US-20210181362-A1 by JIANG et al teaches a fault prediction system using a deep learning neural network. The fault prediction system utilizes as input seismic data, and then derives various seismic attributes from the seismic data.
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NARCISO EDUARDO MONTES whose telephone number is (571)272-5773. The examiner can normally be reached Mon-Fri 8-5.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, REHANA PERVEEN can be reached at (571) 272-3676. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/N.E.M./Examiner, Art Unit 2189
/REHANA PERVEEN/Supervisory Patent Examiner, Art Unit 2189
Prosecution Insights