DETAILED ACTION
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 .
Status of Claims
The following is a non-final, first office action in response to the communication filed on 01/27/2023. Claims 1—20 are currently pending.
Information Disclosure Statement
Information Disclosure Statement received 01/10/2024 has been reviewed and considered.
Claim Objections
Claim 8 objected to because of the following informalities: Claim 1 depends from claim 8 where claim 1 recites “accessing a stream of input data from logging tools in a first well-bore, wherein the stream of input data comprises measurements of bore sizes inside the first well-bore.” Claim 8 subsequently recites “wherein the stream of input data comprises logging data encoding a resistivity, a density, a neutron recording, and a gamma ray recording.”
In view of the fact that claim 1 already recites at least one specifically named model feature as part of the input data, it is believed that claim 8 would be improved if amended to recite the following: “wherein the stream of input data further comprises logging data encoding a resistivity, a density, a neutron recording, and a gamma ray recording.”
Appropriate correction is requested.
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.
Claims 7 and 16 are 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 7 recites the limitation “adjusting at least one parameter of the machine learning model.” Claim 7 depends from claim 1 which recites “training a machine learning model using the training set of input data, wherein the machine learning model is configured to predict a bore size parameter based on input features of the training set of input data.” Accordingly, the data used to train the machine learning model are called input data which may be split out into independent variables (e.g., not specified in claim 1) and dependent variables (e.g., parameter which is predicted; bore size parameter). Given that the test and training datasets and the associated components thereof are explicitly named in claim 1, it is unclear what the “one parameter” of claim 7 could be referring to.
Phrased another way, it is unclear what the “at least one parameter,” is with regards to the machine learning model and/or algorithm given that the “at least one parameter” is separately named from the input data (e.g., input features and predicted bore size parameter) and therefore understood to be distinct from the input data. Accordingly, claim 7 is indefinite because it is unclear how the parameter is used with respect to the model and/or how the parameter relates to the model and therefore unclear how the parameter could be used to adjust the model. Claim 16 recites substantially similar limitations and is indefinite for the same reasons as set forth with respect to claim 7
Upon review of the Specification the following statements were found regarding model parameters:
“[a]n iterative approach can iteratively modify data preprocessing and model parameters until the result achieves the desired properties.” (Instant Specification, para. [0028]); and
“[a]fter reaching an acceptable RMSE error in the training data set, the implementations may be deployed in multiple wells across the same field to benchmark against the measured logs. When the RMSE error becomes unsatisfactory, the implementations may further conduct the training of the machine learning model. For example, model parameters may be modified with updated and additional data to supplement the earlier training data.” (Instant Specification, para. [0037])
In view of the foregoing, and for the purposes of examination, the term model parameters of claims 7 and 16 are understood to be the individual values that make up the input data such that when more data is gathered, more model parameters are added to the input data which may be provided to the algorithm to generate the model.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1—20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 of the USPTO’s eligibility analysis entails considering whether the claimed subject matter falls within the four statutory categories of patentable subject matter identified by 35 U.S.C. 101: Process, machine, manufacture, or composition of matter.
Claims 1, 10, and 19 are directed to a method (process), a system (machine or manufacture), and a system (machine or manufacture), respectively. As such, the claims are directed to statutory categories of invention.
If the claim recites a statutory category of invention, the claim requires further analysis in Step 2A. Step 2A of the 2019 Revised Patent SUBJECT Matter Eligibility Guidance is a two-prong inquiry. In Prong One, examiners evaluate whether the claim recites a judicial exception
Claim 1 recites the following limitations which are directed to an abstract idea:
“accessing a stream of input data from logging tools in a first well-bore” (e.g., a mental process);
“splitting the stream of input data into a training set of input data and a testing set of input data” (e.g., a mental process and/or mathematical concept);
“evaluating the machine learning model using the testing set of input data” (e.g., a mental process and/or mathematical concept); and
“applying the machine learning model to a newly received stream of input data from a second well-bore such that the bore size parameter of the second well-bore is determined independent of measurements of bore sizes inside the second well-bore” (e.g., a conditional limitation which is not required to be performed in order to perform claim 1; however, even if the limitation were positively recited, merely utilizing a machine learning model is directed to a mathematical concept).
Claim 10 recites the following limitations which are directed to an abstract idea:
“accessing a stream of input data from logging tools in a first well-bore” (e.g., a mental process);
“splitting the stream of input data into a training set of input data and a testing set of input data” (e.g., a mental process and/or mathematical concept);
“evaluating the machine learning model using the testing set of input data” (e.g., a mental process and/or mathematical concept); and
“applying the machine learning model to a newly received stream of input data from a second well-bore such that the bore size parameter of the second well-bore is determined independent of measurements of bore sizes inside the second well-bore” (e.g., a conditional limitation which is not required to be performed in order to perform claim 10; however, even if the limitation were positively recited, merely utilizing a machine learning model is directed to a mathematical concept).
Claim 19 recites the following limitations which are directed to an abstract idea:
“accessing a stream of input data from logging tools in a first well-bore” (e.g., a mental process);
“splitting the stream of input data into a training set of input data and a testing set of input data” (e.g., a mental process and/or mathematical concept);
“evaluating the machine learning model using the testing set of input data” (e.g., a mental process and/or mathematical concept); and
“applying the machine learning model to a newly received stream of input data from a second well-bore such that the bore size parameter of the second well-bore is determined independent of measurements of bore sizes inside the second well-bore” (e.g., a conditional limitation which is not required to be performed in order to perform claim 19; however, even if the limitation were positively recited, merely utilizing a machine learning model is directed to a mathematical concept).
Under the broadest reasonable interpretation, the limitations identified above constitute abstract ideas because they recite mental processes, mathematical concepts, or a combination thereof. For example, actions including splitting a collection of data into two categories (e.g., test train splitting) and making an evaluation (e.g., evaluating a model or model performance) are directed to actions which are performable by a human mind with or without the benefit of pen and paper. Additionally and/or alternatively, depending on the specific manner in which the actions are performed, the actions may be directed to a mathematical concept.
Due to the manner in which it is claimed, the limitation “accessing a stream of input data from logging tools in a first well-bore” constitutes an abstract idea insofar as it could be performed as a mental process. For example, the limitation does not recite any restrictions around how the data is accessed. Accordingly, under the broadest reasonable interpretation, “accessing a stream of input data from logging tools in a first well-bore” could merely include observing data provided on a display or a sheet of paper (e.g., accessing data can therefore be performed by observing data with human eyes).
The limitation directed to “applying the machine learning model,” is recited in a conditional manner and is therefore not required in order to perform the claims. However, if the limitation were amended to be positively recited, the limitation would be directed to an abstract idea. For example, while training a machine learning model by providing input data to a machine learning algorithm constitutes an additional element, merely utilizing the generated model to generate a model output (e.g., a prediction) is no different from utilizing any computer-based algorithm and would therefore constitute an abstract idea. Accordingly, claims 1, 10 and 19 are identified as reciting abstract ideas (e.g., judicial exceptions).
If the claim recites a judicial exception (i.e., an abstract idea enumerated in Section I of the 2019 Revised Patent Subject Matter Eligibility Guidance, a law of nature, or a natural phenomenon), the claim requires further analysis in Prong Two. In Prong Two, examiners evaluate whether the claim recites additional elements that integrate the exception into a practical application of that exception.
Claim 1 recites the additional element of:
“wherein the stream of input data comprises measurements of bore sizes inside the first well-bore” (e.g., extra-solution activity);
“training a machine learning model using the training set of input data” (e.g., generic machine learning training equivalent to reciting “apply it”).
Claim 10 recites the additional element of:
“one or more hardware processors” (e.g., mere recitation of generic computer components is equivalent to reciting “apply it”);
“wherein the stream of input data comprises measurements of bore sizes inside the first well-bore” (e.g., extra-solution activity); and
“training a machine learning model using the training set of input data” (e.g., generic machine learning training equivalent to reciting “apply it”).
Claim 19 recites the additional element of:
“a non-transitory computer-readable medium” (e.g., mere recitation of generic computer components is equivalent to reciting “apply it”);
“wherein the stream of input data comprises measurements of bore sizes inside the first well-bore” (e.g., extra-solution activity); and
“training a machine learning model using the training set of input data” (e.g., generic machine learning training equivalent to reciting “apply it”).
The above identified limitations of claims 1, 10, and 19 constitute additional elements. However, for the reasons identified above, and discussed further below, the additional elements do not impose any meaningful limits on practicing the abstract idea. Accordingly, the above identified additional elements do not integrate the identified judicial exceptions into a practical application.
If the additional elements do not integrate the exception into a practical application, then the claim is directed to the recited judicial exception, and requires further analysis under Step 2B to determine whether they provide an inventive concept (i.e., whether the additional elements amount to significantly more than the exception itself).
Claims 1, 10, and 19 recite limitations directed to the specific data used in performing the abstract idea. For example, the claims recite “wherein the stream of input data comprises measurements of bore sizes inside the first well-bore.” While such limitations recite additional elements, the limitations constitute court-identified insignificant extra-solution activity. For example, respect to limitations directed to the data used in performing an abstract idea, the MPEP states “[b]elow are examples of activities that the courts have found to be insignificant extra-solution activity:… Selecting a particular data source or type of data to be manipulated… iii. Selecting information, based on types of information and availability of information in a power-grid environment, for collection, analysis and display, Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1354-55, 119 USPQ2d 1739, 1742 (Fed. Cir. 2016)” (MPEP 2106.05(g)). As such claims 1, 10, and 19 do not recite additional elements which integrate the identified abstract ideas into a practical application.
Claims 1, 10, and 19 recite the limitation “training a machine learning model using the training set of input data,” which is directed to training a machine learning model and therefore recites an additional element. However, the limitation is recited at a high level of generality where the generic recitation of a machine learning training step constitutes well-understood, routine, and conventional activity. For example, Published US Patent Application to Soto et al. (US 20180187498 A1), which is directed to detecting events during a drilling operation teaches “[d]ata analytics module 255 is designed to operate with an artificial intelligence software program and/or machine learning software program. Known techniques from data analysis are expected to be applicable here, including machine learning, cognitive systems, pattern recognition, cluster recognition (SVM clustering), genetic algorithms, heuristics, and big data analysis.” (Soto, para. [0058]). Accordingly, data analysis techniques including machine learning (e.g., which requires the generation of a machine learning model) are considered well-understood, routine, and conventional in the technical field of wellbore construction. As discussed in MPEP 2106.05(d), well-understood, routine, and conventional activity cannot provide for a practical application of the identified judicial exceptions. For example, the MPEP states “[i]f, however, the additional element (or combination of elements) is no more than well-understood, routine, conventional activities previously known to the industry, which is recited at a high level of generality, then this consideration does not favor eligibility.” (MPEP 2106.05(d)). Accordingly, claims 1, 10, and 19 do not recite any additional elements which integrate the identified abstract ideas into a practical application.
As further identified above, claims 10 and 19 recite limitation directed to generic computer components. For example, claim 10 recites the limitations “one or more processors,” and “computer-readable medium storing instructions” and claim 19 recites “one or more non-transitory machine-readable storage devices.” While the recitation of computer components constitutes additional elements, the MPEP identifies such limitation as being equivalent to mere directive to apply the judicial exception (e.g., “apply it”). For example, the MPEP states: “ [w]hen determining whether a claim simply recites a judicial exception with the words ‘apply it’ (or an equivalent), such as mere instructions to implement an abstract idea on a computer, examiners may consider the following:… (2) Whether the claim invokes computers or other machinery merely as a tool to perform an existing process. Use of a computer or other machinery in its ordinary capacity for economic or other tasks (e.g., to receive, store, or transmit data) or simply adding a general purpose computer or computer components after the fact to an abstract idea (e.g., a fundamental economic practice or mathematical equation) does not integrate a judicial exception into a practical application or provide significantly more. See Affinity Labs v. DirecTV, 838 F.3d 1253, 1262, 120 USPQ2d 1201, 1207 (Fed. Cir. 2016) (cellular telephone); TLI Communications LLC v. AV Auto, LLC, 823 F.3d 607, 613, 118 USPQ2d 1744, 1748 (Fed. Cir. 2016) (computer server and telephone unit).” (MPEP 2106.05(f)).
Thus, even when viewed as an ordered combination, nothing in the claims add significantly more (i.e., an inventive concept) to the abstract idea.
Claims 2 and 11 are directed to the specific machine learning algorithms utilized to train the machine learning model where the recited algorithms include a Random Forest algorithm and an XGBoost (extreme gradient boosting) algorithm. Both of these models are well-known in the field of statistical learning and/or machine learning as taught by Montoro et al. (US 20170193335 A1) which states “[w]e considered the performance of four well-known machine learning algorithms compared to the best performing WiseNet model: randomForests, generalized linear models (GLM), generalized boosted machines (GBM) and extreme gradient boosting (xgboost).” (Montoro, para. [0072]). Accordingly, while the limitation functions to further limit the training step in a manner which constitutes an additional element, the limitation is directed to well-understood, routine, and conventional activity which cannot provide for a practical application of the judicial exception. For example, the MPEP states “[i]f, however, the additional element (or combination of elements) is no more than well-understood, routine, conventional activities previously known to the industry, which is recited at a high level of generality, then this consideration does not favor eligibility.” (MPEP 2106.05(d)). Accordingly, claims 2 and 11 do not recite any additional elements which integrate the identified abstract ideas into a practical application.
Claims 3 and 12 recite the limitation “selecting the input features for the machine learning model,” which is directed to an abstract idea comprising a mental process. For example, under the broadest reasonable interpretation, selecting data to be used in a machine learning model constitutes an action which is well-within the capabilities of a human mind. Accordingly the limitations of claims 3 and 12 recite an abstract idea and do not provide for a practical application of the judicial exception because the limitations themselves are judicial exceptions.
Claims 4, 13, and 20 recite limitations directed which further define the method of evaluating the performance of the machine learning model. As noted above, under the broadest reasonable interpretation evaluating a model constitutes an abstract idea comprising a mental process. The additional limitations which further define the method in which the model is evaluated are directed to “computing a root mean square error (RMSE),” is itself an abstract idea insofar as the limitation is directed to a mathematical concept. Accordingly the limitations of claims 4, 13, and 20 recite an abstract idea and do not provide for a practical application of the judicial exception because the limitations themselves are judicial exceptions.
Claims 5 and 14 recite the limitation “in response to evaluating the machine learning model as unsatisfactory, refining the machine learning model,” which constitutes an additional element directed to model training/model retraining. However, the limitation is recited at a high level of generality where the generic recitation of a model retraining or refining step constitutes well-understood, routine, and conventional activity. For example, published US Patent Application to Despinois et al. (US 20240133293 A1) teaches “[a]s known from the field of machine-learning, each training may comprise iteratively processing a respective dataset, for example mini-batch-by-mini-batch and modifying model parameters of the predictive model along the iterative processing.” (Despinois, para. [0062]). Accordingly, while the limitation functions to further limit the training step in a manner which constitutes an additional element, the limitation is directed to well-understood, routine, and conventional activity which cannot provide for a practical application of the judicial exception. For example, the MPEP states “[i]f, however, the additional element (or combination of elements) is no more than well-understood, routine, conventional activities previously known to the industry, which is recited at a high level of generality, then this consideration does not favor eligibility.” (MPEP 2106.05(d)). Accordingly, claims 5 and 14 do not recite any additional elements which integrate the identified abstract ideas into a practical application.
Claims 6, 8, 9, 15, 17, and 18 recite the limitations directed to defining the type of data which is used to perform to an abstract idea/an analysis (e.g., splitting the data into a test/train split). The limitations are directed to an additional element; however, this type of additional element is classified as court-identified insignificant extra-solution activity. With regards to selecting data for performing an abstract idea, the MPEP states: “[b]elow are examples of activities that the courts have found to be insignificant extra-solution activity:… Selecting a particular data source or type of data to be manipulated… iii. Selecting information, based on types of information and availability of information in a power-grid environment, for collection, analysis and display, Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1354-55, 119 USPQ2d 1739, 1742 (Fed. Cir. 2016)” (MPEP 2106.05(g)). As such claims 6, 8, 9, 15, 17, and 18 do not recite additional elements which integrate the identified abstract ideas into a practical application.
Claims 7 and 16 recite the limitation “adjusting at least one parameter of the machine learning model.” As discussed in the rejection under 35 U.S.C. 112(b), it is unclear what the “at least one parameter,” is with regards to the machine learning model and/or algorithm given that it is separately named from the input data (e.g., input features and predicted bore size parameter) and therefore understood to be distinct from the input data. However, in view of the applied interpretation set forth in the rejection, these limitations are directed to updating the training inputs as new data becomes available. Generically retraining models using updated data without reciting any additional changes to the training constitutes well-understood, routine, and conventional activity for the same reason training the original model constitutes well-understood, routine, and conventional activity. As stated above, Published US Patent Application to Soto et al. (US 20180187498 A1) teaches that machine learning models are well-known methods for data analysis. Accordingly, claims 7 and 16 do not recite any additional elements which integrate the identified abstract ideas into a practical application.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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.
Claim(s) 1, 9, 10, 18, and 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Published US Patent Application to Hori et al., hereinafter “Hori” (US 20220120928 A1) in view of Published US Patent Application to Germain et al., hereinafter “Germain” (US 20140351183 A1).
Regarding claim 1, Hori discloses [a] computer-implemented method (para. [0372], “a method or methods may be executed by a computing system. FIG. 21 shows an example of a system 2100 that can include one or more computing systems 2101-1, 2101-2, 2101-3 and 2101-4, which may be operatively coupled via one or more networks 2109, which may include wired and/or wireless networks.”), comprising:
accessing a stream of input data from logging tools in a first well-bore, wherein the stream of input data comprises measurements of bore sizes inside the first well-bore (para. [0131], “[a]s an example, equipment may be utilized to acquire caliper measurements (e.g., borehole caliper measurements). Knowledge of a borehole diameter while it is being drilled may be utilized to perform one or more remedial actions, optionally in real time. Such an approach can help to diminish non-productive time (NPT), for example, as to tripping a drillstring and conducting open-hole logging activities.”);
splitting the stream of input data into a training set of input data and a testing set of input data (para. [0320], “[a]s an example, a method can include training a machine learning model to generate a trained machine learning model. For example, an interpreter can label data where the labeled data can be utilized for training and, for example, for testing.”);
training a machine learning model using the training set of input data, wherein the machine learning model is configured to predict a bore size parameter based on input features of the training set of input data (para. [0320], “[a]s an example, a method can include training a machine learning model to generate a trained machine learning model… For example, consider a trained neural network model that can receive signals in the time domain and/or the frequency domain and output, probabilistically, one or more characteristics of material and/or borehole geometry.”);
in response to evaluating the machine learning model as satisfactory (this limitation is conditional and does not actually carry any weight in the claim as currently drafted), applying the machine learning model to a newly received stream of input data from a second well-bore such that the bore size parameter (borehole geometry) of the second well-bore is determined independent of measurements of bore sizes inside the second well-bore (para. [0320], “[f]or example, consider a trained neural network model that can receive signals in the time domain and/or the frequency domain and output, probabilistically, one or more characteristics of material and/or borehole geometry.”; para. [0323], “output of a trained machine learning model can be a log, which may be, for example, a real-time log that is generated as a downhole tool used in a borehole to make measurements. For example, a log can be a feature attribute log where the feature attributes may relate to material and/or borehole geometry.” Notably, the limitation directed to “apply the machine learning model…” is predicated on a conditional limitation “in response to evaluating…” which is not positive recited in a manner such that the limitation is required to be performed by the claim. Accordingly, the claims can be rejected without consideration of this limitation; however, for the sake of compact prosecution the limitations have been considered and rejected.).
While Hori discloses splitting the data into training and testing groups, Hori does not explicitly disclose using the non-training data for model evaluation. However, Germain, which is in the same field of endeavor as the instant application insofar as it is directed to machine learning models used in drilling operations, teaches the deficient limitation. For example, Germain teaches:
“FIG. 7 shows, in block diagram form, various example steps in performing the model building 304 workflow… With respect to one more candidate models to be created, the example workflow involves a training 700 workflow, a testing 702 workflow, and a validation and scoring workflow 704. Each will be discussed in turn.” (Germain, para. [0122]);
“[t]he candidate models, even trained candidate models, are not necessarily always ‘good’ models. In fact, multiple candidate models of the same underlying type (e.g., neural networks with varying numbers input, output, and/or “hidden” nodes) may differently predict the very same operational outcome when trained on the same training subset. Thus, in accordance with various example embodiments, the next illustrative workflow is testing 702 the one more candidate models.” (Germain, para. [0125]);
“in accordance with yet still further embodiments, the predictive outputs of the candidate models (i.e., the predicted operational outcomes) may also be tested against the third subset 626 (hereafter the validation subset). Various error metrics are generated for each candidate model based on the validation subset, such as root mean square error (RMSE), mean absolute percentage error (MAPE), and other custom metrics.” (para. [0129]); and
“[o]ne or more candidate models may be trained and tested, and the ROP predictions of the candidate models may be compared against the ‘actual’ ROP of the validation subset. The various error metrics may be determined for the predicted versus actual ROP… In some cases, one or models may be discarded based on the error metrics. In other cases, however, the candidate models may be ranked for future use.” (para. [0132]).
It would have been obvious to one of ordinary skill in the art before the effective filing of the claimed invention to add the testing and validation (e.g., “evaluation”) step of Germain to the machine learning model training workflow as set forth in Hori. The evaluation step could be performed by known methods, including considerations paid to error metrics such as the Root Mean Squared Error (“RMSE”) and/or Mean Absolute Error (“MAE”). The evaluation would be performed in the same manner in the workflow of Hori as the workflow of Germain to achieve the predictable result of evaluating a machine learning model prior to deployment. Furthermore, one would be motivated to perform model testing and validation (e.g., “evaluation”) prior to model deployment, where, as acknowledged by Germain, “trained models are not necessarily always good models.”
Regarding claim 9, Hori modified by Germain teaches wherein the bore size parameter comprises at least one of: a maximum size, or a minimum size (Hori, para. [0131], “[a]s an example, if a diameter of a borehole is below gauge or nominal size, such fact may indicate that the bit is worn and to be replaced so as to help reduce demand for later well reaming activities, etc. Borehole diameter data may be utilized to help reduce risk of sticking (e.g., stuck pipe).” The nominal size, which correlates to the bit diameter, constitutes a minimum acceptable size for a wellbore).
Regarding claim 10, Hori discloses [a] computer system comprising one or more hardware processors configured to perform operations of (“[a]s an example, training may be performed using GPU-based workstations, clusters and/or clouds (e.g., consider NVIDIA GPU Cloud and Amazon EC2 GPU instances, etc.).” (para. [0326]):
accessing a stream of input data from logging tools in a first well-bore, wherein the stream of input data comprises measurements of bore sizes inside the first well-bore (para. [0131], “[a]s an example, equipment may be utilized to acquire caliper measurements (e.g., borehole caliper measurements). Knowledge of a borehole diameter while it is being drilled may be utilized to perform one or more remedial actions, optionally in real time. Such an approach can help to diminish non-productive time (NPT), for example, as to tripping a drillstring and conducting open-hole logging activities.”);
splitting the stream of input data into a training set of input data and a testing set of input data (para. [0320], “[a]s an example, a method can include training a machine learning model to generate a trained machine learning model. For example, an interpreter can label data where the labeled data can be utilized for training and, for example, for testing.”);
training a machine learning model using the training set of input data, wherein the machine learning model is configured to predict a bore size parameter based on input features of the training set of input data (para. [0320], “[a]s an example, a method can include training a machine learning model to generate a trained machine learning model… For example, consider a trained neural network model that can receive signals in the time domain and/or the frequency domain and output, probabilistically, one or more characteristics of material and/or borehole geometry.”);
in response to evaluating the machine learning model as satisfactory, applying the machine learning model to a newly received stream of input data from a second well-bore such that the bore size parameter (borehole geometry) of the second well-bore is determined independent of measurements of bore sizes inside the second well-bore (para. [0320], “[f]or example, consider a trained neural network model that can receive signals in the time domain and/or the frequency domain and output, probabilistically, one or more characteristics of material and/or borehole geometry.”; para. [0323], “output of a trained machine learning model can be a log, which may be, for example, a real-time log that is generated as a downhole tool used in a borehole to make measurements. For example, a log can be a feature attribute log where the feature attributes may relate to material and/or borehole geometry.” Notably, the limitation directed to “apply the machine learning model…” is predicated on a conditional limitation “in response to evaluating…” which is not positive recited in a manner such that the limitation is required to be performed by the claim. Accordingly, the claims can be rejected without consideration of this limitation; however, for the sake of compact prosecution the limitations have been considered and rejected.).
While Hori discloses splitting the data into training and testing groups, Hori does not explicitly disclose using the non-training data for model evaluation. However, Germain, which is in the same field of endeavor as the instant application insofar as it is directed to machine learning models used in drilling operations, teaches the deficient limitation. For example, Germain teaches:
“FIG. 7 shows, in block diagram form, various example steps in performing the model building 304 workflow… With respect to one more candidate models to be created, the example workflow involves a training 700 workflow, a testing 702 workflow, and a validation and scoring workflow 704. Each will be discussed in turn.” (Germain, para. [0122]);
“[t]he candidate models, even trained candidate models, are not necessarily always ‘good’ models. In fact, multiple candidate models of the same underlying type (e.g., neural networks with varying numbers input, output, and/or “hidden” nodes) may differently predict the very same operational outcome when trained on the same training subset. Thus, in accordance with various example embodiments, the next illustrative workflow is testing 702 the one more candidate models.” (Germain, para. [0125]);
“in accordance with yet still further embodiments, the predictive outputs of the candidate models (i.e., the predicted operational outcomes) may also be tested against the third subset 626 (hereafter the validation subset). Various error metrics are generated for each candidate model based on the validation subset, such as root mean square error (RMSE), mean absolute percentage error (MAPE), and other custom metrics.” (para. [0129]); and
“[o]ne or more candidate models may be trained and tested, and the ROP predictions of the candidate models may be compared against the ‘actual’ ROP of the validation subset. The various error metrics may be determined for the predicted versus actual ROP… In some cases, one or models may be discarded based on the error metrics. In other cases, however, the candidate models may be ranked for future use.” (para. [0132]).
It would have been obvious to one of ordinary skill in the art before the effective filing of the claimed invention to add the testing and validation (e.g., “evaluation”) step of Germain to the machine learning model training workflow as set forth in Hori. The evaluation step could be performed by known methods, including considerations paid to error metrics such as the Root Mean Squared Error (“RMSE”) and/or Mean Absolute Error (“MAE”). The evaluation would be performed in the same manner in the workflow of Hori as the workflow of Germain to achieve the predictable result of evaluating a machine learning model prior to deployment. Furthermore, one would be motivated to perform model testing and validation (e.g., “evaluation”) prior to model deployment, where, as acknowledged by Germain, “trained models are not necessarily always good models.”
Regarding claim 18, Hori modified by Germain teaches wherein the bore size parameter comprises at least one of: a maximum size, or a minimum size (Hori, para. [0131], “[a]s an example, if a diameter of a borehole is below gauge or nominal size, such fact may indicate that the bit is worn and to be replaced so as to help reduce demand for later well reaming activities, etc. Borehole diameter data may be utilized to help reduce risk of sticking (e.g., stuck pipe).” The nominal size, which correlates to the bit diameter, constitutes a minimum acceptable size for a wellbore).
Regarding claim 19, Hori discloses [a] non-transitory computer-readable medium comprising software instructions which, when executed by one or more computer processors (para. [0372], “a method or methods may be executed by a computing system. FIG. 21 shows an example of a system 2100 that can include one or more computing systems 2101-1, 2101-2, 2101-3 and 2101-4, which may be operatively coupled via one or more networks 2109, which may include wired and/or wireless networks.”; para. [0373], “the computer system 2101-1 can include one or more modules 2102, which may be or include processor-executable instructions, for example, executable to perform various tasks (e.g., receiving information, requesting information, processing information, simulation, outputting information, etc.).”; ), cause the one or more computer processors to perform operations of:
accessing a stream of input data from logging tools in a first well-bore, wherein the stream of input data comprises measurements of bore sizes inside the first well-bore (para. [0131], “[a]s an example, equipment may be utilized to acquire caliper measurements (e.g., borehole caliper measurements). Knowledge of a borehole diameter while it is being drilled may be utilized to perform one or more remedial actions, optionally in real time. Such an approach can help to diminish non-productive time (NPT), for example, as to tripping a drillstring and conducting open-hole logging activities.”);
splitting the stream of input data into a training set of input data and a testing set of input data (para. [0320], “[a]s an example, a method can include training a machine learning model to generate a trained machine learning model. For example, an interpreter can label data where the labeled data can be utilized for training and, for example, for testing.”);
training a machine learning model using the training set of input data, wherein the machine learning model is configured to predict a bore size parameter based on input features of the training set of input data (para. [0320], “[a]s an example, a method can include training a machine learning model to generate a trained machine learning model… For example, consider a trained neural network model that can receive signals in the time domain and/or the frequency domain and output, probabilistically, one or more characteristics of material and/or borehole geometry.”);
in response to evaluating the machine learning model as satisfactory, applying the machine learning model to a newly received stream of input data from a second well-bore such that the bore size parameter (borehole geometry) of the second well-bore is determined independent of measurements of bore sizes inside the second well-bore (para. [0320], “[f]or example, consider a trained neural network model that can receive signals in the time domain and/or the frequency domain and output, probabilistically, one or more characteristics of material and/or borehole geometry.”; para. [0323], “output of a trained machine learning model can be a log, which may be, for example, a real-time log that is generated as a downhole tool used in a borehole to make measurements. For example, a log can be a feature attribute log where the feature attributes may relate to material and/or borehole geometry.” Notably, the limitation directed to “apply the machine learning model…” is predicated on a conditional limitation “in response to evaluating…” which is not positive recited in a manner such that the limitation is required to be performed by the claim. Accordingly, the claims can be rejected without consideration of this limitation; however, for the sake of compact prosecution the limitations have been considered and rejected.).
While Hori discloses splitting the data into training and testing groups, Hori does not explicitly disclose using the non-training data for model evaluation. However, Germain, which is in the same field of endeavor as the instant application insofar as it is directed to machine learning models used in drilling operations, teaches the deficient limitation. For example, Germain teaches:
“FIG. 7 shows, in block diagram form, various example steps in performing the model building 304 workflow… With respect to one more candidate models to be created, the example workflow involves a training 700 workflow, a testing 702 workflow, and a validation and scoring workflow 704. Each will be discussed in turn.” (Germain, para. [0122]);
“[t]he candidate models, even trained candidate models, are not necessarily always ‘good’ models. In fact, multiple candidate models of the same underlying type (e.g., neural networks with varying numbers input, output, and/or “hidden” nodes) may differently predict the very same operational outcome when trained on the same training subset. Thus, in accordance with various example embodiments, the next illustrative workflow is testing 702 the one more candidate models.” (Germain, para. [0125]);
“in accordance with yet still further embodiments, the predictive outputs of the candidate models (i.e., the predicted operational outcomes) may also be tested against the third subset 626 (hereafter the validation subset). Various error metrics are generated for each candidate model based on the validation subset, such as root mean square error (RMSE), mean absolute percentage error (MAPE), and other custom metrics.” (para. [0129]); and
“[o]ne or more candidate models may be trained and tested, and the ROP predictions of the candidate models may be compared against the ‘actual’ ROP of the validation subset. The various error metrics may be determined for the predicted versus actual ROP… In some cases, one or models may be discarded based on the error metrics. In other cases, however, the candidate models may be ranked for future use.” (para. [0132]).
It would have been obvious to one of ordinary skill in the art before the effective filing of the claimed invention to add the testing and validation (e.g., “evaluation”) step of Germain to the machine learning model training workflow as set forth in Hori. The evaluation step could be performed by known methods, including considerations paid to error metrics such as the Root Mean Squared Error (“RMSE”) and/or Mean Absolute Error (“MAE”). The evaluation would be performed in the same manner in the workflow of Hori as the workflow of Germain to achieve the predictable result of evaluating a machine learning model prior to deployment. Furthermore, one would be motivated to perform model testing and validation (e.g., “evaluation”) prior to model deployment, where, as acknowledged by Germain, “trained models are not necessarily always good models.”
Claim(s) 2, 3, 11, and 12 is/are rejected under 35 U.S.C. 103 as being unpatentable over Published US Patent Application to Hori et al., hereinafter “Hori” (US 20220120928 A1) in view of Published US Patent Application to Germain et al., hereinafter “Germain” (US 20140351183 A1) as applied to claims 1 and 10 above, and further in view of Published US Patent Application to Montoro et al., hereinafter “Montoro” (US 20170193335 A1).
Regarding claim 2, Hori modified by Germain teaches a plurality of algorithms which may be used to train the machine learning model of Hori including decision trees (e.g., see para. [0324] of Hori, “[a]s an example, a machine model may be built using a computational framework with a library, a toolbox, etc., such as, for example, those of the MATLAB framework (MathWorks, Inc., Natick, Mass.). The MATLAB framework includes a toolbox that provides supervised and unsupervised machine learning algorithms, including support vector machines (SVMs), boosted and bagged decision trees…”) however, Hori modified by Germain does not explicitly teach “wherein the machine learning model comprises at least one of: a Random Forest (RF) model, or a XGBoost (eXtreme Gradient Boosting) model.”
Montoro, which is in the same field of endeavor as the instant application insofar as it is directed to training machine learning models teaches the deficient limitation. For example, Montoro teaches “[w]e considered the performance of four well-known machine learning algorithms compared to the best performing WiseNet model: randomForests, generalized linear models (GLM), generalized boosted machines (GBM) and extreme gradient boosting (xgboost). FIG. 13 illustrates the performance comparison for all models ranked by AUC in decreasing order from top to bottom.” (Montoro, para. [0072]). Notably, Montoro, which was published in 2017, teaches that random forests and XGBoost are identified as “well-known” machine learning models.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have replaced the machine learning algorithm of Hori with another well-known algorithm used for regression such as a random forest or an XGBoost. As stated by Montoro, both algorithms are well-known in the art of machine learning such that the algorithm functions are also well-known. The substitution would create the predictable result of a machine learning algorithm capable of generating a machine learning model.
Regarding claim 3, Hori modified by Germain and Montoro teach selecting the input features for the machine learning model (Hori, para. [0320], “For example, consider a trained neural network model that can receive signals in the time domain and/or the frequency domain and output, probabilistically, one or more characteristics of material and/or borehole geometry.” Examiner notes that if a machine learning model used for regression is trained on a dataset, it is implicitly understood that the data which is used to train the model has to be selected. Moreover, the model of Hori takes at least time domain and/or frequency domain inputs such that it is implicitly understood that, at least, these features are also what the model was trained on.).
Regarding claim 11, Hori modified by Germain teaches a multitude of algorithms which may be used to train the machine learning model of Hori including decision trees (e.g., see para. [0324] of Hori, “[a]s an example, a machine model may be built using a computational framework with a library, a toolbox, etc., such as, for example, those of the MATLAB framework (MathWorks, Inc., Natick, Mass.). The MATLAB framework includes a toolbox that provides supervised and unsupervised machine learning algorithms, including support vector machines (SVMs), boosted and bagged decision trees…”) however, Hori modified by Germain does not explicitly teach “wherein the machine learning model comprises at least one of: a Random Forest (RF) model, or a XGBoost (eXtreme Gradient Boosting) model.”
Montoro, which is in the same field of endeavor as the instant application insofar as it is directed to training machine learning models teaches the deficient limitation. For example, Montoro teaches “[w]e considered the performance of four well-known machine learning algorithms compared to the best performing WiseNet model: randomForests, generalized linear models (GLM), generalized boosted machines (GBM) and extreme gradient boosting (xgboost). FIG. 13 illustrates the performance comparison for all models ranked by AUC in decreasing order from top to bottom.” (Montoro, para. [0072]). Notably, Montoro, which was published in 2017, teaches that random forests and XGBoost are identified as “well-known” machine learning models.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have replaced the machine learning algorithm of Hori with another well-known algorithm used for regression such as a random forest or an XGBoost. As stated by Montoro, both algorithms are well-known in the art of machine learning such that the algorithm functions are also well-known. The substitution would create the predictable result of a machine learning algorithm capable of generating a machine learning model.
Regarding claim 12, Hori modified by Germain and Montoro teach selecting the input features for the machine learning model (Hori, para. [0320], “For example, consider a trained neural network model that can receive signals in the time domain and/or the frequency domain and output, probabilistically, one or more characteristics of material and/or borehole geometry.” Examiner notes that if a machine learning model used for regression is trained on a dataset, it is implicitly understood that the data which is used to train the model has to be selected. Moreover, the model of Hori takes at least time domain and/or frequency domain inputs such that it is implicitly understood that, at least, these features are also what the model was trained on.).
Claim(s) 4, 5, 6, and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Published US Patent Application to Hori et al., hereinafter “Hori” (US 20220120928 A1) in view of Published US Patent Application to Germain et al., hereinafter “Germain” (US 20140351183 A1) as applied to claims 1 and 19 above, and further in view of Published US Patent Application to Madasu et al., hereinafter “Madasu” (US 20210148213 A1).
Regarding claim 4 Hori modified by Germain teaches wherein evaluating the machine learning model comprises: computing a Root Mean Square Error (RMSE) between the predicted bore size parameter and an actual measurement (Germain, para. [0129], “in accordance with yet still further embodiments, the predictive outputs of the candidate models (i.e., the predicted operational outcomes) may also be tested against the third subset 626 (hereafter the validation subset). Various error metrics are generated for each candidate model based on the validation subset, such as root mean square error (RMSE), mean absolute percentage error (MAPE), and other custom metrics.”; Germain, para. [0132], “[o]ne or more candidate models may be trained and tested, and the ROP predictions of the candidate models may be compared against the ‘actual’ ROP of the validation subset. The various error metrics may be determined for the predicted versus actual ROP… In some cases, one or models may be discarded based on the error metrics. In other cases, however, the candidate models may be ranked for future use.”).
However, Hori modified by Germain may not explicitly teach comparing the RMSE with a pre-determined threshold. Madasu, which is in the same field of endeavor as the instant application insofar as it is directed to machine learning models used in drilling operations, teaches the deficient limitations. For example, Madasu teaches “[i]t may also be assumed for purposes of this example that no retraining of the SWNN was required, e.g., because the predictions produced by the SWNN met the retraining criterion or error tolerance threshold. For example, the retraining criterion or error threshold may be a specified root mean square error value (e.g., 0.2), and the difference between an actual value of the operating variable and the response value predicted using the SWNN may be less than this root mean square error value.” (Madasu, para. [0077]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have added a threshold value associated with the error metric in order to evaluate whether the error metric indicates the model is usable for its intended purpose. The use of a threshold as applied to an error metric threshold associated with the RMSE could be combined into Hori as modified by Germain by known methods where the error metric threshold functions the same in Madasu as it would in combination with Hori modified by Germain. The combination would generate the predictable result of a method for determining whether or not a model was usable and/or of the model requires retraining.
Regarding claim 5, Hori modified by Germain and Madasu teach in response to evaluating the machine learning model as unsatisfactory, refining the machine learning model (Madasu, para. [0077], “[i]t may also be assumed for purposes of this example that no retraining of the SWNN was required, e.g., because the predictions produced by the SWNN met the retraining criterion or error tolerance threshold. For example, the retraining criterion or error threshold may be a specified root mean square error value (e.g., 0.2), and the difference between an actual value of the operating variable and the response value predicted using the SWNN may be less than this root mean square error value.” If the error metric of the model does not succeed in meeting the error metric threshold, then the model is retrained).
Regarding claim 6, Hori modified by Germain and Madasu teach wherein refining the machine learning model comprises at least one of: providing at least one additional input feature to the machine learning model (Madasu, para. [0078], “[a]ccordingly, the accuracy and/or efficiency of the DNN model for predicting the ROP response in this example may be further improved by increasing the number of input variables that are used to appropriately train or retrain the model. For example, the model may be retrained using additional input variables, e.g., reservoir properties or other information relating to the characteristics of the subsurface formation, which may affect ROP during a drilling operation.”), or replacing at least one input feature with a different input feature.
Regarding claim 20 Hori modified by Germain teaches wherein evaluating the machine learning model comprises: computing a Root Mean Square Error (RMSE) between the predicted bore size parameter and an actual measurement (Germain, para. [0129], “in accordance with yet still further embodiments, the predictive outputs of the candidate models (i.e., the predicted operational outcomes) may also be tested against the third subset 626 (hereafter the validation subset). Various error metrics are generated for each candidate model based on the validation subset, such as root mean square error (RMSE), mean absolute percentage error (MAPE), and other custom metrics.”; Germain, para. [0132], “[o]ne or more candidate models may be trained and tested, and the ROP predictions of the candidate models may be compared against the ‘actual’ ROP of the validation subset. The various error metrics may be determined for the predicted versus actual ROP… In some cases, one or models may be discarded based on the error metrics. In other cases, however, the candidate models may be ranked for future use.”).
However, Hori modified by Germain may not explicitly teach comparing the RMSE with a pre-determined threshold. Madasu, which is in the same field of endeavor as the instant application insofar as it is directed to machine learning models used in drilling operations, teaches the deficient limitations. For example, Madasu teaches “[i]t may also be assumed for purposes of this example that no retraining of the SWNN was required, e.g., because the predictions produced by the SWNN met the retraining criterion or error tolerance threshold. For example, the retraining criterion or error threshold may be a specified root mean square error value (e.g., 0.2), and the difference between an actual value of the operating variable and the response value predicted using the SWNN may be less than this root mean square error value.” (Madasu, para. [0077]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have added a threshold value associated with the error metric in order to evaluate whether the error metric indicates the model is usable for its intended purpose. The use of a threshold as applied to an error metric threshold associated with the RMSE could be combined into Hori as modified by Germain by known methods where the error metric threshold functions the same in Madasu as it would in combination with Hori modified by Germain. The combination would generate the predictable result of a method for determining whether or not a model was usable and/or of the model requires retraining.
Claim(s) 7 is/are rejected under 35 U.S.C. 103 as being unpatentable over Published US Patent Application to Hori et al., hereinafter “Hori” (US 20220120928 A1) in view of Published US Patent Application to Germain et al., hereinafter “Germain” (US 20140351183 A1) and Published US Patent Application to Madasu et al., hereinafter “Madasu” (US 20210148213 A1) as applied to claims 6 above, and further in view of Published US Patent Application to Dursun et al., hereinafter “Dursun” (US 20170191359 A1).
In view of the claim interpretation set forth in the rejection under 35 U.S.C. 112(b), claim 7 is understood to recite refining or retraining a machine learning model when new data becomes available. While Hori modified by Germain and Madasu does not explicitly disclose this feature, Dursun, which is in the same field of endeavor as the instant application insofar as it is directed to machine learning models used in drilling operations, teaches the deficient feature. For example, Dursun states “[a]s more data becomes available, prediction model training can be repeated or updated to improve prediction results.” (Dursun, para. [0020]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have added a retraining step (e.g., as taught by Dursun) to the machine learning workflow of Hori modified by Germain and Madasu. Retraining a model when new data is available cand be performed by known methods (e.g., the same method that was performed when the initial model was trained) and is used in the same manner in Dursun as it would be used in combination with Hori as modified by Germain and Madasu. The combination would generate the predictable result of incorporate new data into a machine learning model by retraining by including the new data in the model training process. Furthermore, one would be motivated to make the combination because, as stated by Dursun, including additional data points can improve the prediction results.
Claim(s) 13, 14, and 15 is/are rejected under 35 U.S.C. 103 as being unpatentable over Published US Patent Application to Hori et al., hereinafter “Hori” (US 20220120928 A1) in view of Published US Patent Application to Germain et al., hereinafter “Germain” (US 20140351183 A1) and Published US Patent Application to Montoro et al., hereinafter “Montoro” (US 20170193335 A1) as applied to claim 11 above, and further in view of Published US Patent Application to Madasu et al., hereinafter “Madasu” (US 20210148213 A1).
Regarding claim 13 Hori modified by Germain and Montoro teaches wherein evaluating the machine learning model comprises: computing a Root Mean Square Error (RMSE) between the predicted bore size parameter and an actual measurement (Germain, para. [0129], “in accordance with yet still further embodiments, the predictive outputs of the candidate models (i.e., the predicted operational outcomes) may also be tested against the third subset 626 (hereafter the validation subset). Various error metrics are generated for each candidate model based on the validation subset, such as root mean square error (RMSE), mean absolute percentage error (MAPE), and other custom metrics.”; Germain, para. [0132], “[o]ne or more candidate models may be trained and tested, and the ROP predictions of the candidate models may be compared against the ‘actual’ ROP of the validation subset. The various error metrics may be determined for the predicted versus actual ROP… In some cases, one or models may be discarded based on the error metrics. In other cases, however, the candidate models may be ranked for future use.”).
However, Hori modified by Germain and Montoro may not explicitly teach comparing the RMSE with a pre-determined threshold. Madasu, which is in the same field of endeavor as the instant application insofar as it is directed to machine learning models used in drilling operations, teaches the deficient limitations. For example, Madasu teaches “[i]t may also be assumed for purposes of this example that no retraining of the SWNN was required, e.g., because the predictions produced by the SWNN met the retraining criterion or error tolerance threshold. For example, the retraining criterion or error threshold may be a specified root mean square error value (e.g., 0.2), and the difference between an actual value of the operating variable and the response value predicted using the SWNN may be less than this root mean square error value.” (Madasu, para. [0077]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have added a threshold value associated with the error metric in order to evaluate whether the error metric indicates the model is usable for its intended purpose. The use of a threshold as applied to an error metric threshold associated with the RMSE could be combined into Hori as modified by Germain by known methods where the error metric threshold functions the same in Madasu as it would in combination with Hori modified by Germain. The combination would generate the predictable result of a method for determining whether or not a model was usable and/or of the model requires retraining.
Regarding claim 14, Hori modified by Germain, Montoro, and Madasu teach in response to evaluating the machine learning model as unsatisfactory, refining the machine learning model (Madasu, para. [0077], “[i]t may also be assumed for purposes of this example that no retraining of the SWNN was required, e.g., because the predictions produced by the SWNN met the retraining criterion or error tolerance threshold. For example, the retraining criterion or error threshold may be a specified root mean square error value (e.g., 0.2), and the difference between an actual value of the operating variable and the response value predicted using the SWNN may be less than this root mean square error value.” If the error metric of the model does not succeed in meeting the error metric threshold, then the model is retrained).
Regarding claim 15, Hori modified by Germain, Montoro, and Madasu teach wherein refining the machine learning model comprises at least one of: providing at least one additional input feature to the machine learning model (Madasu, para. [0078], “[a]ccordingly, the accuracy and/or efficiency of the DNN model for predicting the ROP response in this example may be further improved by increasing the number of input variables that are used to appropriately train or retrain the model. For example, the model may be retrained using additional input variables, e.g., reservoir properties or other information relating to the characteristics of the subsurface formation, which may affect ROP during a drilling operation.”), or replacing at least one input feature with a different input feature.
Claim(s) 16 is/are rejected under 35 U.S.C. 103 as being unpatentable over Published US Patent Application to Hori et al., hereinafter “Hori” (US 20220120928 A1) in view of Published US Patent Application to Germain et al., hereinafter “Germain” (US 20140351183 A1), Published US Patent Application to Madasu et al., hereinafter “Madasu” (US 20210148213 A1), and Published US Patent Application to Montoro et al., hereinafter “Montoro” (US 20170193335 A1) as applied to claims 15 above, and further in view of Published US Patent Application to Dursun et al., hereinafter “Dursun” (US 20170191359 A1).
In view of the claim interpretation set forth in the rejection under 35 U.S.C. 112(b), claim 16 is understood to recite refining or retraining a machine learning model when new data becomes available. While Hori modified by Germain, Montoro, and Madasu does not explicitly disclose this feature, Dursun, which is in the same field of endeavor as the instant application insofar as it is directed to machine learning models used in drilling operations, teaches the deficient feature. For example, Dursun states “[a]s more data becomes available, prediction model training can be repeated or updated to improve prediction results.” (Dursun, para. [0020]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have added a retraining step (e.g., as taught by Dursun) to the machine learning workflow of Hori modified by Germain and Madasu. Retraining a model when new data is available cand be performed by known methods (e.g., the same method that was performed when the initial model was trained) and is used in the same manner in Dursun as it would be used in combination with Hori as modified by Germain and Madasu. The combination would generate the predictable result of incorporate new data into a machine learning model by retraining by including the new data in the model training process. Furthermore, one would be motivated to make the combination because, as stated by Dursun, including additional data points can improve the prediction results.
Claim(s) 8 and 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Published US Patent Application to Hori et al., hereinafter “Hori” (US 20220120928 A1) in view of Published US Patent Application to Germain et al., hereinafter “Germain” (US 20140351183 A1) as applied to claims 1 and 10 above, and further in view of Published US Patent Application to Madasu et al., hereinafter “Madasu” (US 20210148213 A1).
Regarding claim 8, Hori modified by Madasu teaches “[i]n LWD operations, certain measurements (e.g., nuclear, sonic, resistivity measurements of a formation) can be sensitive to borehole diameter. Knowledge of the borehole diameter under certain circumstances can be helpful for validating or adjusting such measurements.” (Hori, para. [0133]). Accordingly, there is at least a known relationship between the above stated logging parameters and the wellbore diameter. However, Hori modified by Madasu, and specifically Hori, does not disclose utilizing a resistivity, a density, a neutron recording, and a gamma ray recording as training data to train the model of Hori. However, Dursun, which is in the same field of endeavor as the instant application insofar as it is directed to building machine learning models used for drilling operations teaches the deficient limitation. For example Dursun teaches: “[w]hile various examples are provided herein, it should be appreciated that any available input attribute that can be correlated with a target risk attribute can be used to train the prediction model.” (Dursun, para. [0020]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have included the nuclear, sonic, and resistivity measurements of Hori into the training data of Hori used to build the model of Hori as modified by Madasu where, as taught by Hori the features (e.g., nuclear, sonic, and resistivity measurements) are known to have a relationship with the predictor of Hori (e.g., the borehole geometry) and as taught by Dursun, any features which can be correlated with the target value (e.g., predictor) could be used to train the prediction model. The combination would generate the predicted result of a model which includes features which correlate to the target variable (e.g., objective, predictor, solution, dependent variable…).
Regarding claim 17, Hori modified by Madasu teaches “[i]n LWD operations, certain measurements (e.g., nuclear, sonic, resistivity measurements of a formation) can be sensitive to borehole diameter. Knowledge of the borehole diameter under certain circumstances can be helpful for validating or adjusting such measurements.” (Hori, para. [0133]). Accordingly, there is at least a known relationship between the above stated logging parameters and the wellbore diameter. However, Hori modified by Madasu, and specifically Hori, does not disclose utilizing a resistivity, a density, a neutron recording, and a gamma ray recording as training data to train the model of Hori. However, Dursun, which is in the same field of endeavor as the instant application insofar as it is directed to building machine learning models used for drilling operations teaches the deficient limitation. For example Dursun teaches: “[w]hile various examples are provided herein, it should be appreciated that any available input attribute that can be correlated with a target risk attribute can be used to train the prediction model.” (Dursun, para. [0020]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have included the nuclear, sonic, and resistivity measurements of Hori into the training data of Hori used to build the model of Hori as modified by Madasu where, as taught by Hori the features (e.g., nuclear, sonic, and resistivity measurements) are known to have a relationship with the predictor of Hori (e.g., the borehole geometry) and as taught by Dursun, any features which can be correlated with the target value (e.g., predictor) could be used to train the prediction model. The combination would generate the predicted result of a model which includes features which correlate to the target variable (e.g., objective, predictor, solution, dependent variable…).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to URSULA NORRIS whose telephone number is (703)756-4731. The examiner can normally be reached Monday to Friday, 7 AM to 4 PM.
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, TARA SCHIMPF can be reached at 571-270-7741. 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.
/U.L.N./Examiner, Art Unit 3676
/TARA SCHIMPF/Supervisory Patent Examiner, Art Unit 3676
Prosecution Insights