DETAILED ACTION
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on May 9th, 2026 has been entered.
This action is in response to the amendments filed on Feb. 24th, 2026. A summary of this action:
Claims 1-7, 9-18, 20-22 have been presented for examination.
Claims 1-7, 9-18, 20-22 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement
Claim(s) 1-7, 11-18, 22 is/are rejected under 35 U.S.C. 103 as being unpatentable over Schwartz et al., US 10,908,588 in view of Cao, “Learning to Rank: From Pairwise Approach to Listwise Approach”, 2007 and in further view of in view of Angelo et al., “A neural network-based build time estimator for layer manufactured objects”, 2011
Claim(s) 9-10, 20-21 is/are rejected under 35 U.S.C. 103 as being unpatentable over Schwartz et al., US 10,908,588 in view of Cao, “Learning to Rank: From Pairwise Approach to Listwise Approach”, 2007 and in further view of in view of Angelo et al., “A neural network-based build time estimator for layer manufactured objects”, 2011 and In further view of Coffman et al., US 2019/0271966
This action is non-final
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 Arguments/Amendments
Regarding the objections
Withdrawn in view of amendments.
Regarding the § 112 Rejection
Withdrawn in view of amendments. New grounds below as necessitated by amendment
Regarding the § 101 Rejection
§ 101 rejection withdrawn in view of recent guidance and new trainings, as the independent claims merely recite a collection of additional elements in view of the updated guidance around example 39 and Ex parte Desjardins, along with the updated guidance around Recentive Analytics.
Regarding the § 102/103 Rejection
Maintained, updated as necessitated by amendment.
See rejection below for how the newly amended subject matter is rejected in view of the relied upon combination of prior art relied upon.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 1-7, 9-18, 20-22 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention. The dependent claims inherit the deficiencies of the claims they depend upon.
See MPEP 2163(II)(A): "For example, in Hyatt v. Dudas, 492 F.3d 1365, 1371, 83 USPQ2d 1373, 1376-1377 (Fed. Cir. 2007), the examiner made a prima facie case by clearly and specifically explaining why applicant’s specification did not support the particular claimed combination of elements, even though applicant’s specification listed each and every element in the claimed combination. The court found the "examiner was explicit that while each element may be individually described in the specification, the deficiency was lack of adequate description of their combination" and, thus, "[t]he burden was then properly shifted to [inventor] to cite to the examiner where adequate written description could be found or to make an amendment to address the deficiency.""
Also, see MPEP 2163(I) for Lockwood v. Amer. Airlines, Inc., 107 F.3d 1565, 1572, 41 USPQ2d 1961, 1966 (Fed. Cir. 1997).
Independent claims (claim 1 as representative) recite:
train the machine learning model as a function of the first training data and at least one manufacturing metric;
generate a ranking for pairs of candidate orientations of the plurality of candidate orientations as a function of the machine learning model and the computer model, wherein ranking each candidate orientation comprises a learning to rank process, wherein an output is a ranking of a plurality of ranked pair combinations;
automatically select an orientation based on at least one of:
the ranking;
correlate the selected orientation to the computer model and the at least one manufacturing metric;
generate a second training data based on the correlation;
and comparatively evaluate, a performance of orientation selection by: generating a machining toolpath for each of a plurality of orientations;
and selecting a preferred orientation as a function of a resulting machining toolpaths in accordance with one of: the at least one manufacturing metric; a per part basis; and one or more of an aggregation method, a probabilistic method and a statistical method.
These above steps are not sufficiently described in this particular ordered combination. See fig. 1-2, see ¶¶ 23-25, see ¶¶ 14-15.
As issues with these claims is that they require a particular ordered combination of features that are not described in sufficient detail for this particular combination. E.g. ¶ 14 for the ranking of pairs for fig. 1 incl.: “In some cases, a plurality of candidate orientations are present wherein all features are accessible by machine tool for formation during manufacturing and machine learning model 112 may be used to select an optimal candidate orientation from this grouping” – then see ¶ 24, in particular: “In some cases, a selected orientation (be it from manual or automatic processes) may be correlated to at least one of computer model 108, candidate orientations 116, and manufacturing metric 128 and used to generate further training data 208.” – but note a lack of description of how that further training data 208 is to be used in fig. 2 ( ¶¶ 21-23, but note in particular in ¶ 24 the modifier newly added of further training data).
Then, see ¶ 25 – this describes the selecting of the preferred orientation in fig. 2, but makes no mention of how this is linked with the selected orientation resultant from ¶ 14 from the pairwise learning, or even any linkage whatsoever. ¶ 24 has a self-contained: “For example, in some cases a technician may review ranked list and manually select an orientation for manufacture from the ranked list 124. In some cases, another computer algorithm and/or machine learning process may be used to automatically select an orientation from ranked list 124. In some cases, a selected orientation (be it from manual or automatic processes) may be correlated to at least one of computer model 108, candidate orientations 116, and manufacturing metric 128 and used to generate further training data 208.” – distinct and separate from ¶ 25.
In summary, the present claim amendments require a particular combination of features associated with fig. 1 and 2 separately, and the instant disclosure does not sufficiently describe this particular ordered combination of features now recited in the present claims.
To clarify, this combination requires a particular linkage between numerous particular features associated with fig. 1 and fig. 2 respectively in a particular order, but the instant disclosure does not sufficiently describe this particular combination, but rather only what is in ¶ 24: “In some cases, a selected orientation (be it from manual or automatic processes) may be correlated to at least one of computer model 108, candidate orientations 116, and manufacturing metric 128 and used to generate further training data 208.” – with the additional modifier “further” added to “training data”, not present in other descriptions of # 208, wherein this is from the selected orientation from the result of fig. 1 (incl. ¶ 14).
Further see ¶¶ 7-8: “In an exemplary case, 30 candidate orientations may be considered, a tool path may be generated for each of the 30 candidate orientations and a best toolpath solution could be selected, for example based upon a manufacturing metric, such as manufacturing time. However, generating tool paths is computationally expensive and in many cases, generating multiple tool paths is prohibitively expensive… Aspects of the present disclosure can be used to select a manufacturing orientation from a ranked list of candidate orientations, without needing to generate a tool path for each orientation. This is so, at least in part, because generating of multiple orientation variable toolpaths may be performed only during a training phase to limit computational expenses”
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) 1-7, 11-18, 22 is/are rejected under 35 U.S.C. 103 as being unpatentable over Schwartz et al., US 10,908,588 in view of Cao, “Learning to Rank: From Pairwise Approach to Listwise Approach”, 2007 and in further view of in view of Angelo et al., “A neural network-based build time estimator for layer manufactured objects”, 2011
Regarding Claim 1
Schwartz teaches:
A computer-implemented apparatus of manufacturing orientation selection using machine learning, the apparatus comprising: a processor; and a memory communicatively connected to the processor, wherein the memory contains instructions configuring the processor to: (Schwartz, abstract, then see figure 7 and its accompanying description starting in col. 14)
receive a computer model representative of a part for manufacture, wherein the computer model comprises a virtual representation of the part for manufacture; input the computer model to a machine learning model; create a first training data which comprises determining a plurality of candidate orientations as a function of the machine learning model and the computer model and comparing the plurality of candidate orientations; train the machine learning model as a function of the first training data (Schwartz, fig. 7 # 710-730 – see the accompanying descriptions in col. 14-15, then in col. 15 starting around line 20: “However, the factory server 120 may also identify other candidate orientations for the part. In one embodiment, the factory server 120 provides the feature vector that was generated 720 from the model file to a first machine learning model, and the machine learning model outputs one or more candidate orientations. The first machine learning model may be trained with feedback received from a factory operator, as described further with reference to step 750. In various embodiments, the first machine learning model may be a statistical decision tree, a 30 neural network, or a combination thereof.”
To clarify, col. 17: “When the operator provides input (e.g., approval, rejection, or selection of a candidate orientation), the factory server 120 may use the operator input for training the first 30 machine learning model (which may be used to identify one or more candidate orientations), the second machine learning model (which may be used to score the candidate orientations), or some other algorithm that is used to determine weights for scoring the candidate orientations. Training data may also be collected from previous prints based on the orientations before and after operator input”
… and at least one manufacturing metric;… (Schwartz, fig. 7 as discussed above and its accompanying description; then see col. 9 ¶ 3: “The orientation selector 510 may select an orientation according to a machine learning model that uses features including one or more of projected area, part dimensions, 35 another quantification of the part's geometry, number of part instances per build plate, manufacturing time, and tool properties, as described further with respect to FIG. 7
To clarify, col. 16, last paragraph: “The factory server 120 scores 740 the candidate orientations based on the geometric attributes for each candidate orientation.…For example, the score for a candidate orientation may represent the "cost" or difficulty of printing the part using the candidate orientation. Alternatively, the score for a candidate orientation may increase with decreasing difficulty of printing the part using the candidate orientation so that candidate orientations that allow for "easier" printing have higher scores. As other examples, the score for a candidate orientation may represent the quality, strength, and/or potential failure rate of the part if the part is printed in the candidate orientation.”
When taken in view of col. 16, lines 44-46: “The factory server 120 may also generate a geometric attribute corresponding to an estimated printing time [example of manufacturing time] for a part in a given candidate orientation”
automatically select an orientation based on [a] ranking;… (Schwartz, claim 1 and col 17 ¶¶ 1-2 – see the last paragraph in col. 16 to clarify, i.e. there is a ranking of the orientations based on scores associated with each orientation, then in claim 1 see: “selecting, from the subset of the plurality of candidate orientations, a candidate orientation that maximizes a projected area of the part on the build plate and minimizes a number of layers of the part for printing orthogonal to the build plate;” – see col. 9 ¶¶ 2-3 to further clarify)
generate a second training data... (Schwartz, col. 17, ¶¶ 1-2, note the role of the operator to provide “feedback” in the training (example of re-training); note ¶ 24 of the instant disclosure describes a similar embodiment, wherein POSITA would have found it obvious to have automated the role of the operator of Schwartz because this would merely have been broadly automating a manual activity (MPEP § 2144.04(III) for In re Venner) wherein POSITA would have been motivated to do so because computers perform activities much faster than people).
to clarify, col. 15: “The first machine learning model may be trained with feedback received from a factory operator, as described further with reference to step 750.” And col. 17 ¶ 3: “When the operator provides input (e.g., approval, rejection, or selection of a candidate orientation), the factory server 120 may use the operator input for training [re-training, as the model was previously trained so as to provide the initial orientation for feedback] the first 30 machine learning model (which may be used to identify one or more candidate orientations), the second machine learning model (which may be used to score the candidate orientations), or some other algorithm that is used to determine weights for scoring the candidate orientations. Training data may also be collected from previous prints based on
the orientations before and after operator input”
While Schwartz does not explicitly teach the particular manner of ranking of pairs as claimed, Schwartz in view of Cao teaches:
generate a ranking for pairs of candidate orientations of the plurality of candidate orientations as a function of the machine learning model and the computer model, wherein ranking each candidate orientation comprises a learning to rank process, wherein an output is a ranking of a plurality of ranked pair combinations; automatically select an orientation based on at least one of: the ranking; (Schwartz, col. 17, ¶¶ 1-2, and col. 16 last paragraph, and claim 1 along with fig. 7 as discussed above for its ranking based on scores, i.e. its doing the automatic selection by a ranking
in view of Cao, § 1 ¶¶ 12 for “the pairwise approach” to ranking based on scores
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings from Schwartz on “A factory server receives part requests from customer devices and controls one or more manufacturing tools, such as 3D printers, to fabricate the requested parts. The factory server implements several features to streamline the process of fabricating parts using the manufacturing tools…” (Schwartz, abstract) with the teachings from Cao on a pairwise ranking approach (Cao, § 1). The motivation to combine would have been that “There are advantages with taking the pairwise approach. First, existing methodologies on classification can be directly applied. Second, the training instances of document pairs can be easily obtained in certain scenarios (Joachims, 2002).” (Cao, § 1, ¶ 4)
In addition, the KSR rationale of combining prior art elements according to known methods to yield predictable results is applicable – Schwartz discloses a ranking process based on scores in a machine learning environment (col. 15, ¶ 2; col. 16, last paragraph), wherein Rao § 2.1 discusses: “Learning to rank is a new and popular topic in machine learning. There is one major approach to learning to rank, referred to as the pairwise approach in this paper. For other approaches, see (Shashua & Levin, 2002; Crammer & Singer, 2001; Lebanon & Lafferty, 2002), for example…Learning to rank, particularly the pairwise approach, has been successively applied to information retrieval…” – i.e. it’s a successful, popular approach to ranking problems.
While Schwartz does not explicitly teach the following, Schwartz in view of Cao and Angelo teaches: correlate the selected orientation to the computer model and the at least one manufacturing metric…based on the correlation (Schwartz, as discussed above including col. 14-15 and col 17, lines 25-40 including: “Training data may also be collected from previous prints based on the orientations before and after operator input”
As taken in view of Angelo, abstract, Then see § 5: “The neural network in BTES must be trained with real known data related to the build times of a given set of objects that are manufactured with assigned technologies. The training set of samples should be representative of the correlation between each factor and the corresponding build time components”
then see § 3 including the descriptions of equations 6-7 as discussed on pages 218-219: “…The time for deposition tool repositioning (Trep-mat) [example of a manufacturing metric] is a function of the number of repositioning movements (nr-mat) involved in hatching the internal part of the layers. This number depends on the prototype’s orientation around the model building direction (z) with respect to the hatching vector (τ), which defines the direction of the tool path line segments (Fig. 1).”, include seeing the additional clarification on page 219 including: “Table 2 lists the eight build-time driving factors here calculated. The same table reports whether the driving factors affect build time for the different commercial technologies.”
comparatively evaluate, a performance of orientation selection by: generating a machining tool path for each of a plurality of orientations; and selecting a preferred orientation as a function of a resulting machining tool paths in accordance with one of: the at least one manufacturing metric; a per part basis; and one or more of an aggregation method, a probabilistic method and a statistical method (Schwartz, as discussed above including col. 14-15 and col 17, lines 25-40 including: “Training data may also be collected from previous prints based on the orientations before and after operator input” – in particular, note the plural “orientations”, i.e. the previous prints have a plurality of their own orientations
also, note Schwartz col 16. Last two paragraphs: “The factory server 120 may also generate a geometric attribute corresponding to an estimated printing time for a part in a given candidate orientation… The factory server 120 scores 740 the candidate orientations based on the geometric attributes for each candidate orientation… The scoring may be based on weights determined by a second machine learning model ( e.g., statistical, decision tree, neural network, or a combination thereof). For example, the score for a candidate orientation may represent the "cost" or difficulty of printing the part using the candidate orientation.” to col. 17, first few paragraphs: “...In this embodiment, the factory server 120 may optionally present the selected candidate orientation to an operator using the factory client 125, which prompts the operator to approve or reject the presented candidate orientation. If the operator rejects the presented candidate orientation, the operator may input a corrected orientation (e.g., by selecting one of the non- 10 selected candidate orientations). If the operator approves the presented candidate orientation, the presented candidate orientation is retained as the selected candidate orientation…In another embodiment, the factory server 120 first selects a plurality of candidate orientations according to the scores. 15 For example, the factory server 120 ranks the candidate orientations according to the scores and selects one or more candidate orientations based on the ranking (e.g., the three lowest-difficulty orientations). Similar to the previous embodiment, the factory server 120 may optionally present 20 the part according to each of the selected candidate orientations and prompt the operator to choose between of one of the presented candidate orientations. Alternatively, the operator may reject all of the presented candidate orientations, in which case the operator may input a different 25 orientation ( e.g., one of the non-selected candidate orientations).”
As taken in view of Angelo, abstract, then see § 5: “The neural network in BTES must be trained with real known data related to the build times of a given set of objects that are manufactured with assigned technologies. The training set of samples should be representative of the correlation between each factor and the corresponding build time components” – to clarify, see fig. 2 which shows that the “Real Build Time” input into the “Training Samples” is based on an obtained “Geometric Model (STL)”
Then see § 3 including the descriptions of equations 6-7 as discussed on pages 218-219: “…The time for deposition tool repositioning (Trep-mat) [example of a manufacturing metric] is a function of the number of repositioning movements (nr-mat) involved in hatching the internal part of the layers. This number depends on the prototype’s orientation around the model building direction (z) with respect to the hatching vector (τ), which defines the direction of the tool path line segments (Fig. 1).”, include seeing the additional clarification on page 219 including: “Table 2 lists the eight build-time driving factors here calculated. The same table reports whether the driving factors affect build time for the different commercial technologies.” - i.e. a “tool path” was generated for each orientation for this calculation
In particular, then see Angelo fig. 2 for “The process for build time estimation” - note the “calculate build time driving factors” in the iterative loop, i.e. train the neural network, execute it on the basis of the calculated build times, from that estimate a build time, then do an “Analysis of the error” with the “Real build time” as compared to the est. build time, and if the “error > tv” then “Improve the set of training samples”, re-train the NN, re-execute, etc. (the loop) – then see § 5: “…The training set of samples should be representative of the correlation between each factor and the corresponding build time components. In other words, during the training phase, sufficient knowledge should be transferred into the neural network for the build time estimation to be able to be applied also to those cases for which the ANN has not been directly trained…Build-time driving factors are evaluated by computing the selected geometric models, defined in STL standard, with the original software we have developed…orthogonality of the training samples is verified by means of the following correlation function: [see eq. 9, noting in particular: “Xh is the array containing the values for the h-th build time driving factor of the training samples)]…” –and see in § 5 the last few paragraphs incl: “The build time estimation of the six test cases has been performed by training the ANN with different sets of training samples, which have been obtained by adding to the initial set of examples (16 objects) four other groups of new objects, each group consisting of four geometric models. For each stage of the training process (16, 20, 24, 28 and 32 training examples) the build times of the six objects shown in Fig. 4 have been estimated and then compared with the real build times; the results are shown in Fig. 5.”
To clarify, see § 6: “This paper presents a new approach to estimate the build time of layer manufactured objects. The driving factors, which typically affect build time in the main layer manufacturing technologies, are identified. In order to automatically evaluate these factors, starting from the STL standard file of the object to be manufactured, the methods to analyse the significant geometric features are proposed. Therefore, and by means of a specifically designed artificial neural network, we are able to obtain an approximation of the build time, which is a very complex and non-linear function of the previously defined build time driving factors.” And abstract: “A correct prediction of build time is essential to calculate the accurate cost of a layer manufactured object.”
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings from Schwartz on “A factory server receives part requests from customer devices and controls one or more manufacturing tools, such as 3D printers, to fabricate the requested parts…” (Schwartz, abstract)with the teachings from Angelo on “In order to overcome these limitations, this paper proposes a parametric approach which uses a more complete set of build-time driving factors. Furthermore, considering the complexity of the parametric build time function, an artificial neural network is used so as to improve the method flexibility.” (Angelo, abstract) The motivation to combine would have been that “The analysis of the test cases shows that the proposed approach provides a quite accurate estimation of build time even in critical cases and when supports are required.” (Angelo, abstract), also see Angelo’s conclusions in § 6 including: “Particularly, the new method has been applied to calculate the build time of some test cases that are characterised by some critical aspects which add to the difficulty in predicting build time. The results obtained show a good performance, especially when they are compared with those of the methods dedicated to specific technologies presented in literature”
Regarding Claim 2
Schwartz teaches:
The apparatus of claim 1, further comprising:
wherein the automatically selecting comprises selecting a candidate orientation from the plurality of candidate orientations.. (Schwartz, as discussed above including col. 17, ¶ 1: “The factory server 120 selects 750 one of the candidate orientations based on the scores. In one embodiment, the factory server 120 selects a single "best" candidate orientation having the lowest "cost" score. In this embodiment, the factory server 120 may optionally present the selected candidate orientation to an operator using the factory client 125, which prompts the operator to approve or reject the presented candidate orientation.”)
Regarding Claim 3
Schwartz teaches:
The apparatus of claim 1, wherein the generate a ranking of a plurality of candidate operation orientations is according to manufacturing time. (Schwartz, as discussed above including col. 16, last paragraph: “The factory server 120 scores 740 the candidate orientations based on the geometric attributes for each candidate orientation.…For example, the score for a candidate orientation may represent the "cost" or difficulty of printing the part using the candidate orientation. Alternatively, the score for a candidate orientation may increase with decreasing difficulty of printing the part using the candidate orientation so that candidate orientations that allow for "easier" printing have higher scores. As other examples, the score for a candidate orientation may represent the quality, strength, and/or potential failure rate of the part if the part is printed in the candidate orientation.”
When taken in view of col. 16, lines 44-46: “The factory server 120 may also generate a geometric attribute corresponding to an estimated printing time for a part in a given candidate orientation”
Regarding Claim 4
Schwartz teaches:
The apparatus of claim 1, further comprising:
wherein the generate a ranking of a plurality of candidate operation orientations is according to completeness of manufacture.. (Schwartz, as discussed above including col. 16, last paragraph: “The factory server 120 scores 740 the candidate orientations based on the geometric attributes for each candidate orientation.…For example, the score for a candidate orientation may represent the "cost" or difficulty of printing the part using the candidate orientation. Alternatively, the score for a candidate orientation may increase with decreasing difficulty of printing the part using the candidate orientation so that candidate orientations that allow for "easier" printing have higher scores. As other examples, the score for a candidate orientation may represent the quality, strength, and/or potential failure rate of the part if the part is printed in the candidate orientation.”; as clarified by col. 21, lines 45-60: “It is advantageous to define 920 the printable area in this manner because, for example, failure rates may be higher for part instances that are fabricated closer to an edge of the build plate (e.g., due to warping or dislodging of the part during fabrication because the build plate is not perfectly level). Thus, by defining 920 a printable area that stops short of the edges and defining a non-printable area along the edges, the factory server 120 generates a layout that does not place any part instances within a threshold distance of an edge of the part, which reduces the overall failure rate of the part instances in the layout.”
To clarify on the BRI of completeness of manufacture, instant disclosure, ¶ 15: “As used in this disclosure, "completeness of manufacture" is a quantifiable indication, for example proportionality, of similarity between a manufactured part and a complete part. For example, in some cases, complete manufacture of a part is not possible in one step or operation. For example, depending on orientation features of manufactured part may not be able to be formed.” - e.g. the “quality, strength, and/or potential failure rate of the part”
Regarding Claim 5
Schwartz, in view of Angelo teaches:
The apparatus of claim 1, further comprising:
receiving third training data, wherein the third training data correlates the at least one manufacturing metric to a sample part candidate orientation for a plurality of sample parts; inputting the third training data to a machine learning algorithm; and training the machine learning model as a function of the machine learning algorithm and the third training data.. (Schwartz, as discussed above including col. 14-15: “After obtaining 710 the model file, the factory server 120 generates 720 a feature vector based on the model file…The factory server 120 identifies 730 one or more candidate orientations for the part based on the feature vector for the part…In one embodiment, the factory server 120 provides the feature vector that was generated 720 from the model file to a first machine learning model, and the machine learning model outputs one or more candidate orientations. The first machine learning model may be trained with feedback received from a factory operator, as described further with reference to step 750. In various embodiments, the first machine learning model may be a statistical decision tree, a 30 neural network, or a combination thereof” and col 17, lines 25-40: “When the operator provides input (e.g., approval, rejection, or selection of a candidate orientation), the factory server 120 may use the operator input for training the first machine learning model (which may be used to identify one or more candidate orientations), the second machine learning model (which may be used to score the candidate orientations), or some other algorithm that is used to determine weights for scoring the candidate orientations. Training data may also be collected from previous prints based on the orientations before and after operator input.”
As taken in view of Angelo, abstract, Then see § 5: “The neural network in BTES must be trained with real known data related to the build times of a given set of objects that are manufactured with assigned technologies. The training set of samples should be representative of the correlation between each factor and the corresponding build time components”
then see § 3 including the descriptions of equations 6-7 as discussed on pages 218-219: “…The time for deposition tool repositioning (Trep-mat) [example of a manufacturing metric] is a function of the number of repositioning movements (nr-mat) involved in hatching the internal part of the layers. This number depends on the prototype’s orientation around the model building direction (z) with respect to the hatching vector (τ), which defines the direction of the tool path line segments (Fig. 1).”, include seeing the additional clarification on page 219 including: “Table 2 lists the eight build-time driving factors here calculated. The same table reports whether the driving factors affect build time for the different commercial technologies.” - see remaining citations to Angelo above to further clarify, including the iterative loop on the training process and § 5 last few paragraphs in particular: “The build time estimation of the six test cases has been performed by training the ANN with different sets of training samples, which have been obtained by adding to the initial set of examples (16 objects) four other groups of new objects, each group consisting of four geometric models. For each stage of the training process (16, 20, 24, 28 and 32 training examples) the build times of the six objects shown in Fig. 4 have been estimated and then compared with the real build times; the results are shown in Fig. 5.” The rationale to combine is the same as discussed above.
Regarding Claim 6
Schwartz, in view of Angelo as discussed above teaches:
The apparatus of claim 5, further comprising:
inputting a sample computer model representing a sample part of the plurality of sample parts to a computer aided manufacturing (CAM) resource; (Schwartz, as discussed above including col. 14-15 and col 17, lines 25-40 including: “Training data may also be collected from previous prints based on the orientations before and after operator input”
As taken in view of Angelo, abstract, Then see § 5: “The neural network in BTES must be trained with real known data related to the build times of a given set of objects that are manufactured with assigned technologies. The training set of samples should be representative of the correlation between each factor and the corresponding build time components” – to clarify, see fig. 2 which shows that the “Real Build Time” input into the “Training Samples” is based on an obtained “Geometric Model (STL)”)
generating a first toolpath as a function of the sample computer model, the CAM resource, and a first candidate orientation; generating a second toolpath as a function of the sample computer model, the CAM resource, and a second candidate orientation; determining at least a first manufacturing metric of the at least one manufacturing metric as a function of the first toolpath and at least a second manufacturing metric of the at least one manufacturing metric as a function of the second toolpath; and generating fourth training data, wherein the fourth training data correlates the first candidate orientation to the at least a first manufacturing metric and the second candidate orientation to the at least a second manufacturing metric. (Schwartz, as discussed above including col. 14-15 and col 17, lines 25-40 including: “Training data may also be collected from previous prints based on the orientations before and after operator input” – in particular, note the plural “orientations”, i.e. the previous prints have a plurality of their own orientations
As taken in view of Angelo, abstract, then see § 5: “The neural network in BTES must be trained with real known data related to the build times of a given set of objects that are manufactured with assigned technologies. The training set of samples should be representative of the correlation between each factor and the corresponding build time components” – to clarify, see fig. 2 which shows that the “Real Build Time” input into the “Training Samples” is based on an obtained “Geometric Model (STL)”
Then see § 3 including the descriptions of equations 6-7 as discussed on pages 218-219: “…The time for deposition tool repositioning (Trep-mat) [example of a manufacturing metric] is a function of the number of repositioning movements (nr-mat) involved in hatching the internal part of the layers. This number depends on the prototype’s orientation around the model building direction (z) with respect to the hatching vector (τ), which defines the direction of the tool path line segments (Fig. 1).”, include seeing the additional clarification on page 219 including: “Table 2 lists the eight build-time driving factors here calculated. The same table reports whether the driving factors affect build time for the different commercial technologies.” - i.e. a “tool path” was generated for each orientation for this calculation, see remaining citations to Angelo above for more clarification, including the iterative loop on the training process and § 5 last few paragraphs in particular: “The build time estimation of the six test cases has been performed by training the ANN with different sets of training samples, which have been obtained by adding to the initial set of examples (16 objects) four other groups of new objects, each group consisting of four geometric models. For each stage of the training process (16, 20, 24, 28 and 32 training examples) the build times of the six objects shown in Fig. 4 have been estimated and then compared with the real build times; the results are shown in Fig. 5.”
The rationale to combine would have been the same as discussed above.
Regarding Claim 7
Schwartz, in view of Angelo as discussed above teaches:
The apparatus of claim 6, wherein the first toolpath and the second toolpath are machining toolpaths. (Schwartz, as was taken in view of Angelo above, including Angelo § 3 including the descriptions of equations 6-7 as discussed on pages 218-219: “…The time for deposition tool repositioning (Trep-mat) [example of a manufacturing metric] is a function of the number of repositioning movements (nr-mat) involved in hatching the internal part of the layers. This number depends on the prototype’s orientation around the model building direction (z) with respect to the hatching vector (τ), which defines the direction of the tool path line segments (Fig. 1).” – i.e. it is the toolpath of the machine with a “deposition tool”, and each “tool path line segment” is an example of a toolpath as its repositioned between each tool path line segment
The rationale to combine would have been the same as discussed above.
Regarding Claim 11.
Schwartz teaches:
The apparatus of claim 1, wherein receiving the computer model further comprises receiving the computer model from a user device; and the apparatus further comprises: (Schwartz, col. 14, lines 40-45: “The factory server 120 obtains 710 a model file for a part. For example, the customer interacts with the customer interface 205 to upload a model file.”, as taken in view of fig. 1 including # 120 and # 110)
inputting the computer model to a computer aided manufacturing (CAM) resource; (Schwartz, col. 14, lines 40-45: “The factory server 120 obtains 710 a model file for a part. For example, the customer interacts with the customer interface 205 to upload a model file.”, as taken in view of fig. 1 including # 120 and # 110; as taken in view of the abstract: “A factory server receives part requests from customer devices and controls one or more manufacturing tools, such as 3D printers, to fabricate the requested parts.” )
generating a toolpath as a function of the computer model, the CAM resource, and a candidate orientation of the plurality of candidate orientations; and transmitting the toolpath to a tool. (Schwartz, abstract: “…The factory server can also select an orientation in which to fabricate the part and determine print settings to use when fabricating the part…” then see col. 18, ¶ 2: “The factory server 120 instructs a printer 130 to print 770 the part in the selected orientation and using the determined print settings. For example, the factory server 120 generates manufacturing instructions (e.g., G-code) [example of a generated toolpath] and sends the manufacturing instructions to the printer 130 along with the determined print settings” – to clarify, see fig. 9 and its accompanying description, including col. 23 ¶¶ 1-2: “For example, the factory server 120 sends the manufacturing instructions for the layout to a 3D printer 130, and the 3D printer 130 executes the manufacturing instructions to print 970 the layout.” - to clarify, a skilled person would have inferred that the G-code generated for this included a generated toolpath for the 3D printer because the 3D printer, by executing the G-code, printed the layout
Regarding Claim 12
Rejected under similar rationale as claim 1 above.
Regarding Claim 13.
This claim is rejected under a similar rationale as claim 2 above.
Regarding Claim 14.
This claim is rejected under a similar rationale as claim 3 above.
Regarding Claim 15.
This claim is rejected under a similar rationale as claim 4 above.
Regarding Claim 16.
This claim is rejected under a similar rationale as claim 5 above.
Regarding Claim 17.
This claim is rejected under a similar rationale as claim 6 above.
Regarding Claim 18.
This claim is rejected under a similar rationale as claim 7 above.
Regarding Claim 22.
This claim is rejected under a similar rationale as claim 11 above.
Claim(s) 9-10, 20-21 is/are rejected under 35 U.S.C. 103 as being unpatentable over Schwartz et al., US 10,908,588 in view of Cao, “Learning to Rank: From Pairwise Approach to Listwise Approach”, 2007 and in further view of in view of Angelo et al., “A neural network-based build time estimator for layer manufactured objects”, 2011 and In further view of Coffman et al., US 2019/0271966
Regarding Claim 9
While Schwartz, in view of Cao and Angelo, does not explicitly teach the following feature, Schwartz, in view of Cao, Angelo, and in further view of Coffman teaches:
The apparatus of claim 1, further comprising: receiving an element of part data; and selecting the machine learning model as a function of the element of part data or the computer model. (Schwartz, col. 5, lines 35-55: “In the direct print mode, the customer interface 205 35 accepts request parameters including one or more of customer contact information, a model file, model file scale (e.g., distance units), part quantity, part color, and part material [examples of elements of part data received]…”
To clarify on the BRI of an element of part data, ¶¶ 18-19: “As used in this disclosure, "part data" is information related to a part. Non-limiting examples of part data include quantity, material, requested lead time, surface finish, manufacturing process, and dimensional tolerance…Continuing to refer to FIG. 1, manufacturing request datum may include at least an element of part data…”
Then see Schwartz, as discussed above including col. 14-15 including: “In one embodiment, the factory server 120 provides the feature vector that was generated 720 from the model file to a first machine learning model, and the machine learning model outputs one or more candidate orientations.” And col. 17, lines 25-40: “Training data may also be collected from previous prints based on the orientations before and after operator input” - The see Schwartz, col. 5, lines 35-55: “In the direct print mode, the customer interface 205 35 accepts request parameters including one or more of customer contact information, a model file, model file scale (e.g., distance units), part quantity, part color, and part material [examples of elements of part data received]…” – then see col. 3-4, the paragraph split between the columns including: “Conversion of the model file to manufacturing instructions depends at least in part on manufacturing settings. Manufacturing settings depend on the type of manufacturing tool ( e.g., the type of printer 130) and may be specified at least in part by a customer's part request. For example, the part request specifies one or more colors for the part as well as a material (e.g., plastic, rubber, metal) for the part. The factory server 120 may configure manufacturing settings for manufacturing a part specified by a model file…” – a skilled person would have inferred, or at least found it obvious, that the data collected from previous prints for use in the training data would have included the data that was used to perform the previous printing acts such as “color” and “materials”, and the “model file”, because the previous prints would have been printed based on the “model file” and the other parameters from the user such as “material” and “colors” for the previous prints
As taken in view of Coffman, ¶¶ 97-100: “…The array data structure stores a set of data values corresponding to geometric and/or physical attributes of such a shape… PSMP server 109 generates, at 603, a training set by executing one or more operations over the corpus received at 601….In some implementations, one or more machine learning models are trained at 605. The machine learning models trained at 605 are selected based on different criteria depending on the problem to be solved and/or data available in the training set...”, as see the abstract: “The subject technology is related to methods and apparatus for discretization and manufacturability analysis of computer assisted design models…The computer-based method determines geometric and physical attributes from a discretized version of the digital model, a cloud point version of the digital model, and symbolic functions generated through evolutionary algorithms”
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings from Schwartz on “A factory server receives part requests from customer devices and controls one or more manufacturing tools, such as 3D printers, to fabricate the requested parts…” (Schwartz, abstract) with the teachings from Hoffman on “The subject technology is related to methods and apparatus for discretization and manufacturability analysis of computer assisted design models.”
The motivations to combine would have been that “Advantageously, the subject technology provides manufacture predictions in near real-time. Specifically, in some instances, the time delay introduced by the compute device implementing the computer-based methods and network transmissions described herein is on a millisecond order between 150 ms and 1000 ms. Thus, the responsiveness of the computer-based method is immediate and perceived as a part of a mechanical action ( e.g., a request submitted via a user interface) made by a user. The subject technology provides objective, accurate, and consistent classifications and/or predictions regarding manufacturing process, including estimated times, optimal costs, comparisons of fabrication materials, and other suitable information. The classifications and/or predictions are reliable; that is, assessments for the manufacture of similar products result in similar or equivalent outcomes. The subject technology operates in near real-time and thus, optimizes manufacturing process by decreasing overhead associated with human-based estimations.” (Hoffman, ¶¶ 27-28)
Regarding Claim 10.
Rejected under a similar rationale as claim 9 above.
Regarding Claim 20.
Rejected under a similar rationale as claim 9 above.
Regarding Claim 21.
Rejected under a similar rationale as claim 9-10 above.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DAVID A. HOPKINS whose telephone number is (571)272-0537. The examiner can normally be reached Monday to Friday, 10AM to 7 PM EST.
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, Ryan Pitaro can be reached at (571) 272-4071. 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.
/David A Hopkins/Primary Examiner, Art Unit 2188
Prosecution Insights