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 .
Response to Amendment
The amendment filed July 16, 2026 has been entered. Claims 1-9 remain pending in the instant application. Applicant’s amendments to the Abstract, Drawings, and Claims have overcome each and every objection and 112(b) rejection previously set forth in the Non-Final Office Action mailed May 7, 2026.
Response to Arguments
Applicant's arguments, filed July 16, 2026, regarding rejections under 35 U.S.C. 103 have been fully considered but they are not persuasive.
Regarding Claim 1, Applicant argues that 1) Belbute-Peres does not disclose feeding a computational fluid dynamics (CFD) solver simulation to a graph convolutional network (GCN) at each iteration of an iteratively simulated combustion; and 2) Belbute-Peres does not teach feeding predictions of the GCN to a CFD solver. In addition, regarding point 2), Applicant argues that Belbute-Peres does not teach or suggest that a coarse mesh 201 undergoes any modification during training.
Regarding argument 1), the Examiner respectfully disagrees. As illustrated in figure 2 of Belbute-Peres, a CFD solver output is input to a GCN (e.g., solver 203 outputs 205 eventually to GCN 217). This constitutes at least one iteration in which a CFD simulation is input to a GCN. The Examiner notes that the instant claims do not specify a number of iterations that must be performed. Even so, as noted in the instant remarks, paragraph [0035] of Belbute-Peres discloses that additional convolutional layers may be placed before the CFD solver. Paragraph [0014] of Belbute-Peres further discloses that an iterative function may be substituted for a stack of layers of the neural network, wherein paragraph [0024] further discloses that the CFD solver is an additional layer in the network. As Belbute-Peres discloses convolutional layers outputting into a CFD solver which itself outputs into further convolutional layers, one of ordinary skill in the art could have merely replaced the GCN-CFD-GCN pipeline with an iterative looping of GCN to CFD as in the claimed invention. The claimed invention is considered to be merely rearranging and/or duplicating the layers of the neural network disclosed in Belbute-Peres.
Regarding argument 2), the Examiner respectfully disagrees. Paragraph [0035] of Belbute-Peres discloses that additional convolutional layers may be placed before the CFD solver, which discloses an output of a GCN input to a CFD solver. Furthermore, regarding Applicant’s argument that the coarse mesh 201 undergoes modification during training, the Examiner notes that this limitation is not claimed. The training (S2) in the instant claims does not recite modifying a coarse mesh during the training.
An updated rejection under 35 U.S.C. 103, necessitated by Applicant’s amendment, is provided below.
Claim Objections
Claim 5 is objected to because of the following informalities:
Claim 5 recites “wherein said solver (30) performs a Large-Eddy simulations.” The claim should be corrected to either “wherein said solver (30) performs [[a]] Large-Eddy simulations,” or “wherein said solver (30) performs a Large-Eddy simulation.”
Appropriate correction is required.
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 nonobviousness.
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) 1-9 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lapeyre et al. (Lapeyre, Corentin J., Antony Misdariis, Nicolas Cazard, Denis Veynante, and Thierry Poinsot. "Training convolutional neural networks to estimate turbulent sub-grid scale reaction rates." Combustion and Flame 203 (2019): 255-264.), hereinafter Lapeyre, in view of Belbute-Peres et al. (U.S. Pub. No. 2021/0382963 A1), hereinafter Belbute-Peres.
Regarding Claim 1, Lapeyre teaches A method for simulating combustion of a fluid in a combustion chamber, for design of said combustion chamber (“This work presents a new approach for premixed turbulent combustion modeling.”) (e.g., page 1, abstract).
comprising: discretizing (S1) a space of said combustion chamber into a given mesh (“Two meshes are used in this study: a DNS mesh used to perform the reactive simulations, which contains 512 x 256x 256 cells [and] an 'LES' mesh, which represents the same domain but 8 times coarser in every direction, i.e. 64 x 32 x 32 cells.”) (e.g., page 3, column 2, paragraphs 7-9).
vertices of which have, as a value, progress variables predicted by a computational fluid dynamics simulation (20) (“We will limit this study to perfectly premixed combustion where a progress variable c for adiabatic flows is defined as: [equation (2)],” wherein the progress variables may be used in the nodes of the mesh.) (e.g., page 2, column 1, paragraph 5).
with local combustion quantities at said vertices (“Σ = | ∇c | is the generalized flame surface density [...] Figure 2 shows a typical instantaneous snapshot of the configuration in the (x −y ) plane: Σ+ varies between ≈1 near the inlet, where turbulence injection has not yet wrinkled the flame, and a maximum of ≈3 in some local pockets.”) (e.g., page 2, column 2, paragraph 2; page 4, column 1, last paragraph).
each vertex having a corresponding value of said progress variable, obtained by said solver, in order to obtain predicted values of said local combustion quantity at said vertices (Figure 2 shows “Fully resolved progress variable c (top). From this data, the input of the neural network c (middle) and target output to be learned Σ (bottom) are produced.”) (e.g., page 4, column 2, figure 2).
However, Lapeyre does not appear to specifically teach training (S2) a multilayer neural network (10) by means of a learning set associating a graph corresponding to said given mesh […] and according to a loss function (50) configured to minimize an error between output of said neural network and said local combustion quantities; and iteratively simulating (S3) said combustion which involves, at each iteration: providing as input to a solver (30) the values predicted by said multilayer neural network (10) of a local combustion quantity at the vertices of said given mesh, in order to obtain a value of a progress variable at each vertex of said given mesh; and providing to said multilayer neural network a graph corresponding to the vertices of said given mesh.
On the other hand, Belbute-Peres, which relates similarly to computational fluid dynamics (CFD), does teach training (S2) a multilayer neural network (10) by means of a learning set associating a graph corresponding to said given mesh (“The processor subsystem 160 may be further configured to iteratively train the neural network using the training data 192 [...] The network operates over two different graphs, a "fine" mesh 211 over which to compute the CFD simulation, and a "coarse" mesh 201 (initially a simple coarsened version of the fine mesh, but eventually tuned by the model) that acts as input to the CFD solver 203.”) (e.g., paragraphs [0014] and [0015]).
and according to a loss function (50) configured to minimize an error between output of said neural network and said local combustion quantities (“The adjoint method in SU2 uses reverse-mode differentiation, so gradients can be efficiently computed with respect to a scalar-valued loss such as the overall predictive error of the CFD-GCN.”)
and iteratively simulating (S3) said combustion (“The system 100 may further comprise a processor subsystem 160 which may be configured to, during operation of the system 100, provide an iterative function as a substitute for a stack of layers of the neural network to be trained.” The CFD solver and GCN layers may be looped to provide the output of the GCN layers back as input to a CFD layer.) (e.g., paragraph [0014]).
which involves, at each iteration: providing as input to a solver (30) the values predicted by said multilayer neural network (10) of a local combustion quantity at the vertices of said given mesh, in order to obtain a value of a progress variable at each vertex of said given mesh (“The network may operate by first running a CFD simulation on the coarse mesh input 201, while simultaneously processing the graph defined by the fine mesh 211 with one or more GCNs 213. The system may then upsample 207 the results of the simulation, and concatenates these with an intermediate output 215 from a GCN at step 209 [...] Intuitively, the CFD solver 203 (e.g., SU2 solver) should be thought of as an additional layer in our network.”) (e.g., paragraphs [0016] and [0024]).
and providing to said multilayer neural network a graph corresponding to the vertices of said given mesh (“Finally, the system may apply additional GCN layers 217, 213 to these joint features, ultimately predicting the desired output values 225.”) (e.g., paragraph [0017]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the Applicant's claimed invention to combine Lapeyre with Belbute-Peres. The claimed invention is considered to be substituting one known element for another to obtain predictable results, see MPEP § 2143(I)(B). Lapeyre teaches a method for simulating combustion using direct numerical simulation on a mesh and a convolutional neural network. However, Lapeyre does not appear to specifically teach using a graph neural network and iteratively simulating the combustion. On the other hand, Belbute-Peres does teach a method for fluid dynamics simulations that iteratively combines a graph convolutional network (GCN) and a differential solver. As both Lapeyre and Belbute-Peres relate to using convolutional neural networks to supplement fluid mechanic modeling, one of ordinary skill in the art could have substituted the graph convolutional neural network of Belbute-Peres for the convolutional neural network of Lapeyre and used the progress variables and combustion quantities as node features for the graphs of Belbute-Peres. Furthermore, Lapeyre discloses that “subgrid-scale (SGS) closure models for Reynolds averaged Navier-Stokes (RANS) and Large Eddy Simulation (LES) could be addressed using several machine learning algorithms, including ANNs” (e.g., Lapeyre; page 1, column 2, paragraph 3). Thus, one of ordinary skill in the art would have recognized the substitution of the convolutional neural network in Lapeyre with the graph convolutional network in Belbute-Peres as producing predictable results. Therefore, it would have been obvious to a person of ordinary skill in the art to combine Lapeyre with Belbute-Peres in order to use a machine learning model better suited to processing a simulation graph/mesh as a direct input.
Regarding Claim 2, Lapeyre in view of Belbute-Peres teaches The method according to claim 1. Lapeyre further teaches wherein said local combustion quantity is approximated by a flame surface density (“The output flame surface density value Σ however spans from 0 far from the flame (both in fresh and burnt gases) to a maximum value that depends on the amount of SGS wrinkling of the flame.”) (e.g., page 4, column 1, paragraph 3).
Regarding Claim 3, Lapeyre in view of Belbute-Peres teaches The method according to claim 1. Lapeyre further teaches further comprising discretizing said space into a second mesh, substantially finer than said given mesh (“Two meshes are used in this study: a DNS mesh used to perform the reactive simulations, which contains 512 x 256x 256 cells [and] an 'LES' mesh, which represents the same domain but 8 times coarser in every direction, i.e. 64 x 32 x 32 cells.”) (e.g., page 3, column 2, paragraphs 7-9).
and wherein said training the multilayer neural network comprises a direct numerical simulation (20) of said combustion, in order to obtain said progress variables and said local combustion quantities at the vertices of said second mesh (“Table 1 describes the two DNS simulations performed in this study and used to train the CNN [...] Fig. 2. (x −y ) slice view of the last field from DNS2 (“snapshot 0”in Fig. 5 ).”) (e.g., page 3, column 2, paragraph 5).
then a filtering (40) in order to obtain said progress variables and said local combustion quantities at the vertices of said given mesh (“The DNS field of c is filtered to produce c and Σ+ , then sampled on the 8 times coarser LES mesh,” wherein Σ+ describes local combustion quantities which may be used in the mesh of Belbute-Peres.) (e.g. page 4, column 1, last paragraph).
Regarding Claim 4, Lapeyre in view of Belbute-Peres teaches The method according to claim 1. Belbute-Peres further teaches wherein nodes of said graph match the vertices of said given mesh (“The network operates over two different graphs, a "fine" mesh 211 over which to compute the CFD simulation, and a "coarse" mesh 201 (initially a simple coarsened version of the fine mesh, but eventually tuned by the model) that acts as input to the CFD solver 203.” The graph is a mesh and vice versa, thus the graph processed by the GCN is interpreted as having nodes that match the vertices of the given mesh.) (e.g., paragraph [0015]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the Applicant's claimed invention to combine Lapeyre with Belbute-Peres. The claimed invention is considered to be using a known technique to improve similar devices (methods, or products) in the same way, see MPEP § 2143(I)(C). Lapeyre teaches a method for simulating combustion using direct numerical simulation on a mesh and a convolutional neural network. However, Lapeyre does not appear to specifically teach using a graph neural network, wherein nodes of the graph are vertices of a simulation mesh, and a state of a node in the graph is computed based on neighbor nodes. On the other hand, Belbute-Peres does teach a method for fluid dynamics simulations that iteratively combines a graph convolutional network and a differential solver (GCN), wherein the GCN operates on a graph/mesh, and a status of nodes of the graph is computed based on a variable k nearest neighbor nodes. As both Lapeyre and Belbute-Peres relate to using convolutional neural networks to supplement fluid mechanic modeling, one of ordinary skill in the art could have improved the method of Lapeyre by substituting the GCN of Belbute-Peres for the convolutional neural network of Lapeyre, and computing the states of nodes using the GCN by taking into account a variable number of neighboring nodes. Furthermore, Lapeyre discloses that “subgrid-scale (SGS) closure models for Reynolds averaged Navier-Stokes (RANS) and Large Eddy Simulation (LES) could be addressed using several machine learning algorithms, including ANNs” (e.g., Lapeyre; page 1, column 2, paragraph 3). Thus, one of ordinary skill in the art could have applied the known GCN and k nearest neighbor machine learning algorithms to the method of Lapeyre, and the results would have been predictable to one of ordinary skill in the art. Therefore, it would have been obvious to a person of ordinary skill in the art to combine Lapeyre with Belbute-Peres in order to model interactions between nodes of a simulation mesh.
Regarding Claim 5, Lapeyre in view of Belbute-Peres teaches The method according to claim 1. Lapeyre further teaches wherein said solver (30) performs a Large-Eddy (“In this study, a convolutional neural network (CNN) inspired from a U-net architecture was used to predict SGS flame surface density for a premixed turbulent flame on an underresolved mesh typical of large eddy simulation. Results show excellent agreement for this case between the prediction by the neural network and the baseline value from the DNS.”) (e.g., page 9, column 1, paragraph 1).
Regarding Claim 6, Lapeyre in view of Belbute-Peres teaches The method according to claim 1. Belbute-Peres further teaches wherein said multilayer neural network (10) takes into account said graph by computing a state of a neuron from states of neighboring neurons of a neighborhood determined by said graph (“the system may find the set {n1 , ... , nk} containing the indices of the k closes nodes to the corresponding node positions in the coarser graph Xn. The system may define each node feature in the fine mesh U(i). Thus, given a coarse mesh with positions Xe and features Uc, and fine mesh positions Xp, this operation produces fine mesh features UF.”) (e.g., paragraph [0027]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the Applicant's claimed invention to combine Lapeyre with Belbute-Peres for the same reasons as in Claim 4, above.
Regarding Claim 7, Lapeyre in view of Belbute-Peres teaches The method according to claim 6. Belbute-Peres further teaches wherein the state of the neuron is computed based on an influence of said neighboring neurons that is variable according to said neighboring neurons (“the system may find the set {n1 , ... , nk} containing the indices of the k closes nodes to the corresponding node positions in the coarser graph Xn. The system may define each node feature in the fine mesh U(i). Thus, given a coarse mesh with positions Xe and features Uc, and fine mesh positions Xp, this operation produces fine mesh features UF.” The number of k closest nodes may be variable.) (e.g., paragraph [0027]).
Regarding Claim 8, Lapeyre in view of Belbute-Peres teaches The method according to claim 1. Belbute-Peres further teaches further including a step (S4) of determining a set of design parameters for said combustion chamber from results of said iterative simulating for a plurality of envisaged sets of design parameters (“the CFD-GCN model may be used in an application that utilizes temperature and angle of attack or other design parameters to determine fluid flow and turbulence through pipes, or an application that utilizes geometric or shape optimization parameters to determine noise or turbulence generated by blades of a fan or other moving parts of an object.”) (e.g., paragraph [0015]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the Applicant's claimed invention to combine Lapeyre with Belbute-Peres. The claimed invention is considered to be merely combining prior art elements according to known methods to yield predictable results, see MPEP § 2143(I)(A). Lapeyre teaches a method for simulating combustion using direct numerical simulation on a mesh and a convolutional neural network. However, Lapeyre does not appear to specifically teach determining a set of design parameters based on the iterative simulation. On the other hand, Belbute-Peres does teach a method for using fluid dynamics simulations for shape optimization. As both Lapeyre and Belbute-Peres relate to using convolutional neural networks to supplement fluid mechanic modeling, one of ordinary skill in the art could have merely added the step of optimizing a shape based on simulations from Belbute-Peres to the method of Lapeyre. In combination, the simulation of Lapeyre and design parameters of Belbute-Peres merely perform the same functions as the do separately, and one of ordinary skill in the art would have recognized the results of the combination as predictably outputting design parameters based on a numerical/GCN simulation. Therefore, it would have been obvious to a person of ordinary skill in the art to combine Lapeyre with Belbute-Peres in order to apply the results of the simulations towards an actual physical design.
Regarding Claim 9, Lapeyre teaches A device for simulating combustion of a fluid in a combustion chamber, for design of said combustion chamber (“This work presents a new approach for premixed turbulent combustion modeling [...] The network converges in ≈150 epochs, for a total training time of 20 min on an Nvidia Tesla V100 GPU.”) (e.g., page 1, abstract; page 7, column 1, paragraph 1).
vertices of which have, as a value, progress variables predicted by a computational fluid dynamics simulation (20) (“We will limit this study to perfectly premixed combustion where a progress variable c for adiabatic flows is defined as: [equation (2)],” wherein the progress variables may be used in the nodes of the mesh.) (e.g., page 2, column 1, paragraph 5).
with local combustion quantities at said vertices (“Σ = | ∇c | is the generalized flame surface density [...] Figure 2 shows a typical instantaneous snapshot of the configuration in the (x −y ) plane: Σ+ varies between ≈1 near the inlet, where turbulence injection has not yet wrinkled the flame, and a maximum of ≈3 in some local pockets.”) (e.g., page 2, column 2, paragraph 2; page 4, column 1, last paragraph).
each vertex having a corresponding value of said progress variable, obtained by said solver, in order to obtain predicted values of said local combustion quantity at said vertices (Figure 2 shows “Fully resolved progress variable c (top). From this data, the input of the neural network c (middle) and target output to be learned Σ (bottom) are produced.”) (e.g., page 4, column 2, figure 2).
However, Lapeyre does not appear to specifically teach a solver (30) suitable for receiving as input the values predicted by a multilayer neural network (10) of a local combustion quantity at the vertices of said given mesh, in order to obtain a value of a progress variable at each vertex of said given mesh, said multilayer neural network (10) being trained by means of a learning set associating a graph corresponding to a given mesh of a space of said combustion chamber […] and according to a loss function (50) configured to minimize an error between output of said neural network and said local combustion quantities, and wherein said multilayer neural network is suitable for receiving as input a graph corresponding to the vertices of said given mesh.
On the other hand, Belbute-Peres, which relates similarly to computational fluid dynamics (CFD), does teach a solver (30) suitable for receiving as input the values predicted by a multilayer neural network (10) of a local combustion quantity at the vertices of said given mesh, in order to obtain a value of a progress variable at each vertex of said given mesh (“The network may operate by first running a CFD simulation on the coarse mesh input 201, while simultaneously processing the graph defined by the fine mesh 211 with one or more GCNs 213. The system may then upsample 207 the results of the simulation, and concatenates these with an intermediate output 215 from a GCN at step 209 [...] Intuitively, the CFD solver 203 (e.g., SU2 solver) should be thought of as an additional layer in our network.”) (e.g., paragraphs [0016] and [0024]).
said multilayer neural network (10) being trained by means of a learning set associating a graph corresponding to a given mesh of a space of said combustion chamber (“The processor subsystem 160 may be further configured to iteratively train the neural network using the training data 192 [...] The network operates over two different graphs, a "fine" mesh 211 over which to compute the CFD simulation, and a "coarse" mesh 201 (initially a simple coarsened version of the fine mesh, but eventually tuned by the model) that acts as input to the CFD solver 203.”) (e.g., paragraphs [0014] and [0015]).
and according to a loss function (50) configured to minimize an error between output of said neural network and said local combustion quantities (“The adjoint method in SU2 uses reverse-mode differentiation, so gradients can be efficiently computed with respect to a scalar-valued loss such as the overall predictive error of the CFD-GCN.”)
and wherein said multilayer neural network is suitable for receiving as input a graph corresponding to the vertices of said given mesh (“Finally, the system may apply additional GCN layers 217, 213 to these joint features, ultimately predicting the desired output values 225.”) (e.g., paragraph [0017]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the Applicant's claimed invention to combine Lapeyre with Belbute-Peres for the same reasons as in Claim 1.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 KYLE HWA-KAI TSENG whose telephone number is (571)272-3731. The examiner can normally be reached M-F 9A-5P PST.
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.
/K.H.T./ Examiner, Art Unit 2189
/REHANA PERVEEN/ Supervisory Patent Examiner, Art Unit 2189
