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 .
Style
In this action unitalicized bold is used for claim language, while italicized bold is used for emphasis.
Applicant Reply
“The claims may be amended by canceling particular claims, by presenting new claims, or by rewriting particular claims as indicated in 37 CFR 1.121(c). The requirements of 37 CFR 1.111(b) must be complied with by pointing out the specific distinctions believed to render the claims patentable over the references in presenting arguments in support of new claims and amendments. . . . The prompt development of a clear issue requires that the replies of the applicant meet the objections to and rejections of the claims. Applicant should also specifically point out the support for any amendments made to the disclosure. See MPEP § 2163.06. . . . An amendment which does not comply with the provisions of 37 CFR 1.121(b), (c), (d), and (h) may be held not fully responsive. See MPEP § 714.” MPEP § 714.02. Generic statements or listing of numerous paragraphs do not “specifically point out the support for” claim amendments. “With respect to newly added or amended claims, applicant should show support in the original disclosure for the new or amended claims. See, e.g., Hyatt v. Dudas, 492 F.3d 1365, 1370, n.4, 83 USPQ2d 1373, 1376, n.4 (Fed. Cir. 2007) (citing MPEP § 2163.04 which provides that a ‘simple statement such as ‘applicant has not pointed out where the new (or amended) claim is supported, nor does there appear to be a written description of the claim limitation ‘___’ in the application as filed’ may be sufficient where the claim is a new or amended claim, the support for the limitation is not apparent, and applicant has not pointed out where the limitation is supported.’)” MPEP § 2163(II)(A).
Claim Rejections under both 35 USC §§ 101 and 112 (Incredible Utility)
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.
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.
Claim 2 rejected under 35 U.S.C. 101 because the claimed invention is not supported by either a credible asserted utility or a well-established utility.
Claim 2 recites the method of claim 1“wherein, prior to training the artificial intelligence using the training data sets, the method further comprising using the first artificial intelligence to” carry out various operations. But claim 1 recites “training an artificial intelligence using the training data sets to establish a first artificial intelligence[.]” This is paradoxical. Claim 2 requires using “the first artificial intelligence” “prior to training the artificial intelligence.” But, claim 1 requires the first artificial intelligence is first established by training the artificial intelligence. This combination requires the “first artificial intelligence” to be used (claim 2) prior to the steps resulting in its own creation (claim 1). This is an incredible (i.e. not credible) utility because it is logically impossible to satisfy both conditions.
Claim 2 is also rejected under 35 U.S.C. 112(a) or pre-AIA 35 U.S.C. 112, first paragraph. Specifically, because the claimed invention is not supported by either a credible asserted utility or a well-established utility for the reasons set forth above, one skilled in the art clearly would not know how to use the claimed invention.
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 3-8 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.
Claim 3 recites “running the second trained artificial intelligence using the at least one first calibration data set; recording second assessment data sets of structure-borne noise data from the second trained artificial intelligence based on the at least one first calibration data set; and running the second trained artificial intelligence to assess whether the second assessment data sets are associated with a defective component from the production process.” This appears to recite receiving information (“recording second assessment data sets”) from “the second trained artificial intelligence” and then separately running the same “second trained artificial intelligence” to access whether the data that the model just produced are associated with defective components. That is, data is received from a model, then exactly the same data is fed back into the model, at which point the model ultimately determines that the data it produced the first time around was associated with some defect. No support is found in the original disclosure for this unconventional combination of operations.
Claim 7 recites “training the second artificial intelligence with the second calibration data to establish a third trained artificial intelligence[.]” No explicit support is found in the Specification for a “third trained artificial intelligence.” If the Specification provides implicit support, Applicant may provide a quote of the exact supporting language including reasoning demonstrating implicit support for an unmentioned third model.
All dependent claims are rejected as containing the limitations of the claims from which they depend.
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 1-10 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 pre-AIA the applicant regards as the invention.
Claim 1 recites “training an artificial intelligence using the training data sets to establish a first artificial intelligence; . . . running the first trained artificial intelligence to determine whether assessing the first assessment data sets using the trained artificial intelligence are associated with a defective component from the production process . . . training the artificial intelligence using the at least one first calibration data set to establish a second trained artificial intelligence, the second trained artificial intelligence being adapted for changes in at least one of a condition and component.” The claims use the convention of renaming a model after training. This is clear and even helpful. But this convention implies that different models are meant when a different name is used. The claims recite training of “an artificial intelligence . . . to establish a first artificial intelligence” and then refer to “the first trained artificial intelligence[.]” The dependent claims use both “first artificial intelligence” and “first trained artificial intelligence.” Compare e.g. claims 2 and 4. For the foregoing reasons, it is not clear from the claims as a whole, whether “the first artificial intelligence” refers to the same model or a different model than “the first trained artificial intelligence.” Note that references to “the second trained artificial intelligence” remain constant throughout the claims, tending to indicate that the same term is used where only a single model is claimed.
Claim 3 recites “recording second assessment data sets of structure-borne noise data from the second trained artificial intelligence based on the at least one first calibration data set[.]” It is not clear what operation is meant by “recording second assessment data sets of structure borne noise data from the second trained artificial intelligence[.]” This could be understood as recording the output of “the second trained artificial intelligence” in response to the calibration data set being input to the model. But that seems inconsistent with language limiting the “second assessment data sets” to being “of structure-borne noise.” Since it is not clear whether the second assessment data sets are model outputs or measured data, the claim scope is indefinite. Further, claim 1 recites very similar language: “running the first trained artificial intelligence to determine whether the first assessment data sets are associated with a defective component from the production process[.]” The limitation of claim 1 tends to support the interpretation where the first assessment data sets are merely input to the first trained artificial intelligence while the plain meaning of the limitation in claim 3 is an argument for interpreting “assessment data sets” as model outputs. This similarity between the limitations of claims 1 and 3 makes both limitations indefinite because construing such similar limitations to have different meanings is relatively less reasonable than construing both limitations to have analogous meanings.
Claim 3 recites “running the second trained artificial intelligence using the at least one first calibration data set[.]” The “first calibration set” was used to train and thereby “establish” “the second trained artificial intelligence.” See Claim 1. Therefore, “running” the “artificial intelligence” (model) appears to refer to using the model to predict its own training data. This is a rather useless task in the world of machine learning because testing a model on its own training data effectively asks the model for answers that the model already has (instead of determining whether the model has generalized the solution or using the model to make online predictions during use of the trained model.) Further, nothing in the Specification indicates any reason for performing this extremely unconventional operation. The combination of ambiguity in the language, use of the generic term “running,” and the futility of testing a model on its own training data with no description of any advantage of this unconventional operation in the Specification tend to support the conclusion that this is not meant, despite the plain meaning of the language. Alternatively, this could refer to validation or testing of the model with a subset of the first calibration data, but that is not what the claim says. Yet another possibility is that this limitation refers to further training of the second trained artificial intelligence, (beyond that required in claim 1.) Since there are multiple possible interpretations, none of which are clearly more likely than the other(s), the claim is indefinite.
Claim 6 recites “wherein the first trained artificial intelligence is trained using the second calibration data sets.” The convention in this claim set is to rename a model after training on a given training data set. See e.g. claim 1 ll. 6-7, 16-17 and claim 7 ll. 4-5. The language of claim 6 does not following this pattern, thereby differentiating itself from the language of the other claims. This indicates that the “second calibration data sets” are used in the creation of the “first trained artificial intelligence.” Note that this interpretation would requires a change in the structure of all instances of “first trained artificial intelligence” recited in claims from which claim 6 depends. Alternatively, this claim could be read as modifying the “first trained artificial intelligence” recited in claim 1 to create another, unnamed, “artificial intelligence.”
Claim 7 recites “training the second artificial intelligence[.]” It is not clear whether or not “the second artificial intelligence” refers back to “the second trained artificial intelligence” of claims 1 and 3.
Claim 7 recites “[t]he method of claim 6, further comprising using a second type of component for recording third assessment data sets of structure-borne noise data, and assessing the third assessment data sets with the second calibration data sets after the training of the artificial intelligence.” The meaning of “assessing the third assessment data sets with the second calibration data sets” is also unclear. This could be understood as using the third assessment data sets to assess the model after the second calibration set is used. Alternatively, this may refer to some comparison between the datasets themselves.
All dependent claims are rejected as containing the limitations of the claims from which they depend.
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 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 1, 3-4, 6, 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over Lee (Deep convolutional neural network with new training method and transfer learning for structural fault classification of vehicle instrument panel structure, 2020) and Sun (Deep Transfer Learning Based on Sparse Autoencoder for Remaining Useful Life Prediction of Tool in Manufacturing, 2019.)
1. A computer-implemented method for assessing structure-borne noise, the method comprising the following steps: (Lee teaches “Structural defect have been detected by attaching sensors to all possible defect locations. A new method is proposed to enable the identification of structural defect locations with minimal data collection points using a deep convolutional neural network.” Lee Abstract.) - recording training data sets of structure-borne noise data, the training data sets each being recorded under one or more conditions while a force acts on a component, and the structure-borne noise data depending on the force; - training an artificial intelligence using the training data sets to establish a first artificial intelligence; (“In this study, vibration data were generated by applying white Gaussian random force to a finite element (FE) model of a vehicle’s instrument panel structure, and used for the training and test data of SFC-DCNN. Vibration data were taken from a maximum of 10 measurement points in 50 seconds and augmented by sampling the data in the time domain to have a sufficient number of data.” Lee P. 4489-4490.) - recording first assessment data sets of structure-borne noise data during a production process; (“ . . . vibration data were generated by applying white Gaussian random force to a finite element (FE) model of a vehicle’s instrument panel structure, and used for the training and test data of SFC-DCNN.” Lee PP. 4489-4490.
Lee does not expressly teach recording data sets “during a production process.”
Sun teaches “Data-driven analysis method has proven to be an efficient and effective way for condition monitoring and prediction [1], [2]. In this kind of method, multivariable sensors, like vibration sensor, acoustic emission sensor, and torque sensor, are added to machine tool for sensing operation states of machine. . . . Due to nature of convenience for implementation and no interruption to manufacturing process, data-driven methods have always been a research hotspot for tool condition monitoring and prediction.” Sun p. 2416.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Sun with respect to this limitation because it is useful to detect issues effecting manufacturing in real time (i.e. so you can avoid making more of them.) - running the first trained artificial intelligence the first assessment data sets are associated with defective components from the production process; (“SFC-DCNN was trained and tested[.]” Lee P. 4490.) and - recording at least one first calibration data set of structure-borne noise data, the at least one first calibration data set being recorded under one or more conditions different from the one or more conditions used during the recording of the training sets; - training the artificial intelligence using the at least one first calibration data set to establish a second trained artificial intelligence, the second trained artificial intelligence being adapted for changes in a least one of a condition and component. (“Transfer learning for the initial value was used to increase the accuracy of classifying a more complicated structural defect problem by using the extracted features from a less complicated structural defect problem.” Lee P. 4489. “The first training method is training the whole SFC-DCNN architecture based on the He initial value. The second training method is transfer learning by using the optimized parameters from Sec. 3 as the initial value of in this study. It is common to pretrain a network on a very large dataset, and then to use the optimized parameters of the network as either an initialization or a fixed feature extractor for a case that does not have sufficient training data. Transfer learning that was used in this study is pretraining the network on a relatively easy task of classifying boundary conditions and then the optimized parameters used as an initial value for a more complicated task. The major benefit of this process is that structural characteristics can be extracted well with an easily distinguishable task.” Lee P. 4496. The teaching of using one model as an “initial value” for “transfer learning” “for a case that does not have sufficient training data” would be understood by one of ordinary skill as a teaching of further training of a pretrained model on data in a different domain (i.e. a calibration data set.”
But Lee does not expressly teach “recording” of the calibration data. Note also that the support for this limitation in paragraph 18 of the Specification describes operations that read on transfer learning.
Sun teaches “An SAE [Sparse Auto Encoder] network is first trained by run-to-failure data with RUL information of a cutting tool in an off-line process. The trained network is then transferred to a new tool under operation for on-line RUL [Remaining Useful Life] prediction. The prediction result with high accuracy shows advantage of the DTL method for RUL prediction.” Sun 2416. “In this kind of method, multivariable sensors, like vibration sensor, acoustic emission sensor, and torque sensor, are added to machine tool for sensing operation states of machine.” Sun 2416. Further, the algorithm on page 2421 of Sun teaches training and SAE before steps described as “1) Copy the trained SAE to a new deep SAE network. 2) Input data of the new tool to the new SAE network, performing feature transfer learning and weight update to update the new SAE.” Sun p. 2421. “Monitoring data of tool2 are input to this network for feature extraction. Features learned for tool2 by the DTL network after weight transfer, features transfer learning, and weight update is shown in Fig. 7.” Sun p. 2422. In other words, the data from the new tool is input to a model trained on another tool, thereby forming a new model trained on data from both tools in a form of transfer learning.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Sun because collecting new data of a under a different condition (from the original training data) and using the new to update the model is a data efficient way of training models when there is minimal data available for a specific use case.
3. The method of claim 1, wherein after training the artificial intelligence, the method comprising: running the second trained artificial intelligence using the at least one first calibration data set; recording second assessment data sets of structure-borne noise data from the second trained artificial intelligence based on the at least one first calibration set; and running the second trained artificial intelligence to assess whether the second assessment data sets are associated with a defective component from the production process. (There is no clear way of interpreting this claim language. “First, where the degree of uncertainty is not great, and where the claim is subject to more than one interpretation and at least one interpretation would render the claim unpatentable over the prior art, an appropriate course of action would be for the examiner to enter two rejections: (A) a rejection based on indefiniteness under 35 U.S.C. 112(b) or pre-AIA 35 U.S.C. 112, second paragraph; and (B) a rejection over the prior art based on the interpretation of the claims which renders the prior art applicable. See, e.g., Ex parte Ionescu, 222 USPQ 537 (Bd. App. 1984). When making a rejection over prior art in these circumstances, it is important for the examiner to point out how the claim is being interpreted. Second, where there is a great deal of confusion and uncertainty as to the proper interpretation of the limitations of a claim, it would not be proper to reject such a claim on the basis of prior art. As stated in In re Steele, 305 F.2d 859, 134 USPQ 292 (CCPA 1962), a rejection under 35 U.S.C. 103 should not be based on considerable speculation about the meaning of terms employed in a claim or assumptions that must be made as to the scope of the claims.” MPEP § 2173.06(II). See rejections under §112b, above. In an attempt at compact prosecution, art is cited for the best guess at the scope of this claim. The claim recites Given the use of “running” of the “second trained artificial intelligence using” the same data used to train the model. This reads on both training of the model using the same dataset (see rejection of claim 1) and on testing the model (see rejection of claim 1.) The claimed “recording of the . . . structure borne noise data . . . from the second trained artificial intelligence” and separately running the same model to determine defects based on the same data recorded from itself, appears on its face to be nonsensical. As a best guess, this may be meant to recite evaluating data by the second trained artificial intelligence, which was trained based on the first calibration data set. That aspect is taught in the art cited in the rejection of claim 1.)
4. The method of claim 3, further comprising using the first trained artificial intelligence for assessing the first and/or second assessment data sets as being associated with defect-free components or as being associated with defective components. (See Lee P. 4494 Tables 5 and 6.)
6. The method of claim 3, further comprising using a first type of component for recording the training data sets, the first and second assessment data sets, and the at least one first calibration data set, and, after the assessment of the second assessment data sets, (There is no antecedent basis for “the second assessment data sets” or for “the assessment of the second assessment data sets.” Without any specific distinction between the “first and second assessment data sets” this language reads on subsets of the testing data of Lee. Lee teaches “The response is acceleration from 10 measurement points that are assigned in IP structure shown in Fig. 3. Acceleration data from the instrument panel structure’s FE model were stacked in sequence at each measurement point. Input data were augmented by cutting the total 50 seconds of data into 0.2 second lengths. The time step is 0.001 seconds, and 1000 input data samples were prepared. 80% of the data were used for training, and 20% was used for testing.” Lee P. 4492. “The overall procedure is the same as in the case study in Sec. 3. Classification with SFC-DCNN was conducted with 800 training sets and 200 test sample sets.” Lee P. 4496.) the method further comprises using a second type of component for recording second calibration data sets, wherein, upon recording of the training data sets, the first and second assessment data sets and the at least one first calibration data set are not used for a second type of component, and wherein the first trained artificial intelligence is trained using the second calibration data sets. (Note that Lee teaches multiple experiments for sensing multiple structural flaws test data an data used for transfer learning (calibration data.) “SFC-DCNN was trained and tested with three cases of structural defects and one case of multi-channel acoustic noise data for classifying gear fault from public dataset [4]. The first structural defect case is for classifying the boundary conditions of an instrument panel structure, the second and third case is for classifying the conditions for partial weld elimination and cracks respectively.” Lee P. 4490.)
9. The method of claim 1, wherein the components are part of motor vehicles. (See rejection of claim 1.)
10. A system comprising a digital electronic storage medium and a digital electronic processing unit, wherein instructions are stored in the storage medium, the processing unit is configured to read out and execute the instructions, and the processing unit is configured to perform the method of claim 1 when executing the instructions. (See rejection of claim 1. One of ordinary skill in the art would understand the number of operations required to train the convolutional neural network of figure 1 table 1, as a teaching of a model implemented in software using a processor and memory. See also Lee P. 4496 (“Classification with SFC-DCNN was conducted with 800 training sets and 200 test sample sets. The batch number is 50, the learning rate is 0.001, and 1000 epochs were used in this study.”)
Claims 2 is rejected under 35 U.S.C. 103 as being unpatentable over Lee, Sun and Sohail (A Comprehensive Introduction to Clustering Methods, 2018.)
2. The method of claim 1, wherein, prior to training the artificial intelligence using the training data sets, the method further comprising: using the first artificial intelligence to apply a quality score to each training data set and classify the training data sets into clusters based on the applied quality score of the training data sets, wherein only training data sets with a quality score above a limit are used for training the first artificial intelligence. (The previously cited art does not teach classifying training data sets into clusters.
Sohail teaches “In terms of unsupervised learning methods, some of the most well researched and common methods can be grouped under clustering. The basic idea is simple. If you can figure out how to define distances between data points, then data points that are closer together may exhibit some kind of group characteristic we could exploit for modeling or extract new understanding from. Some examples include patients with similar blood test results that have the same disease, consumers with a similar purchase history that are part of the same socioeconomic class or occupation, and flowers with similar colors and petal lengths that are part of the same species. Obviously, some of these problems can be solved as classification problems, but this is only possible if the labels are available. Clustering, as with other unsupervised methods, operate without a label of interest.” Sohail PP. 1-2. See also Sohail P. 10-11 showing points that are closer being grouped in the same group. That the data exhibits an observable clustering pattern, indicates a level of “quality score” in the training data.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Sohail because clustering is one way of preparing unsupervised data for use in training machine learning models and avoid the task of labeling data.)
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Lee, Sun, and Wang (Detecting anomalies in time series data from a manufacturing system using recurrent neural networks, 2020)
5. The method of claim 4, wherein only structure-borne noise data from defect- free components is used to record the at least one first calibration data set. (The previously cited art does not expressly state that only data from defect-free components is used to record the calibration set.
Wang teaches “The model takes time series data as an input and reconstructs the input data. Time series data with an anomaly would causes patterns in the reconstruction errors that are inconsistent with error patterns of anomaly-free data.” Wang Abstract. “In general, anomalous data show patterns that are different from anomaly-free data (i.e., data collected during routine functioning of the system). For example, as shown in Fig. 1, in assembling a crank, the rotational angle of the driving equipment and the corresponding torque applied to rotate the crank can be recorded as a time series.” Wang P. 823 col. 2. “But these methods need a large number of labeled anomalies (the anomalies must be identified a priori) to train the model; such detailed information is usually not available [9]. . . . A second group of studies uses semi-supervised learning or unsupervised learning to solve the problems caused by the lack of labeled anomalies [9]. The basic idea of these methods is to use time series data collected under routine operation to establish a time series model (including estimates of parameters). . . . Since the model was trained using data having no anomalies, the reconstruction errors corresponding to data with anomalies would be large [21].” Wang P. 824 col. 1. “The proposed model was trained using sensor data from routine operation of a diesel engine assembly line (anomaly-free data set).” Wang P. 829 col. 1.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Wang because defect free data is easier to acquire than data associated with a given defect.)
Claims 7-8 are rejected under 35 U.S.C. 103 as being unpatentable over Lee, Sun, and Wang2 (Transfer learning for enhanced machine fault diagnosis in manufacturing)
7. The method of claim 6, further comprising: using the second type of component for recording third assessment data sets of structure-borne noise data; training the second artificial intelligence with the second calibration data to establish a third trained artificial intelligence; (The Specification describes a third data set used to analyze a different component. See Spec. ¶¶18-19 (“For example, when the first type of component is used for a first motor vehicle model for a particular purpose, and the second type of component is used for the same purpose in a second motor vehicle model, the training with the second calibration data sets can be sufficient for assessing the structure-borne noise data of the second type of component. Advantageously, a completely new training of the artificial intelligence or another artificial intelligence can then be omitted. [0019] According to some embodiments, third assessment data sets of structure-borne noise data can be recorded using a second type of component. The third assessment data sets can be assessed after training the artificial intelligence with the second calibration data sets. For example, the assessment can comprise the third assessment data sets being assessed as belonging to defective or defect-free components.”) Without any limitation on the type of data in the calibration data set, this limitation reads on merely further training of the second artificial intelligence with more “calibration data” from another “type” of component. Sun teaches “An SAE [Sparse Auto Encoder] network is first trained by run-to-failure data with RUL information of a cutting tool in an off-line process. The trained network is then transferred to a new tool under operation for on-line RUL [Remaining Useful Life] prediction. The prediction result with high accuracy shows advantage of the DTL method for RUL prediction.” Sun 2416. “In this kind of method, multivariable sensors, like vibration sensor, acoustic emission sensor, and torque sensor, are added to machine tool for sensing operation states of machine.” Sun 2416. Further, the algorithm on page 2421 of Sun teaches training and SAE before steps described as “1) Copy the trained SAE to a new deep SAE network. 2) Input data of the new tool to the new SAE network, performing feature transfer learning and weight update to update the new SAE.” Sun p. 2421. “Monitoring data of tool2 are input to this network for feature extraction. Features learned for tool2 by the DTL network after weight transfer, features transfer learning, and weight update is shown in Fig. 7.” Sun p. 2422. In other words, the data from the new tool is input to a model trained on another tool, thereby forming a new model trained on data from both tools in a form of transfer learning.
The previously cited art does not expressly teach transfer learning between two types of components.
Wang2 teaches “One limiting factor for successful DL applications is the availability of sufficient amount of data of relevance to the specific application. A solution is presented in this paper for cross domain data learning and effective network training, enabled by the generalization of the DL’s feature learning capability, which is independent of the specific application domains. The developed method is experimentally verified by transferring a DL model trained by non-manufacturing data to manufacturing machine condition monitoring, and transferring model among different working conditions and machines.” Wang2 Abstract. “Transfer Learning (TL) [7] may provide a promising solution to the problems by (1) adapting a network that is well trained using generic data to manufacturing domain problems, and (2) further transferring the adapted network across multiple scenarios (such as different types of machines or fault severity levels) within the manufacturing domain to solve a specific, manufacturing related problem.” Wang2 p. 413. “The paper presents a CNN-based TL technique, using vibration analysis for rolling bearing fault diagnosis as a case study.” Wang2 P. 413. “Specifically, by adapting a pretrained CNN structure using non-manufacturing data (i.e., model transfer) and transferring the adapted network structure to different fault severity levels and bearing types (i.e., feature transfer), a new approach to solving the two existing problems for fault diagnosis of rolling bearings is demonstrated.” Wang2 pp. 413-414. “Subsequent to the model transfer, feature transfer has been performed to generalize the adapted VGG19 to different operating conditions, fault types, severities, and machine types. This is achieved by further modifying the FC layers to extract common latent features that are representative of different data under different scenarios (e.g. from different bearings).” Wang2 P. 415. See also Wang2 Fig. 1.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Wang2 because known work in one field of endeavor may prompt variations of it for use in either the same field or a different one based on design incentives or other market forces if the variations are predictable to one of ordinary skill in the art. The scope and content of the prior art, whether in the same field of endeavor as that of the applicant’s invention or a different field of endeavor, included a similar or analogous device (method, or product) (Wang teaches use of transfer learning to create models for one set of industrial machinery based on previously created models trained on a different type of machine (with another type of bearing.) This is analogous to the claimed use of transfer learning for using a model trained on a different type of component.) There were design incentives or market forces which would have prompted adaptation of the known device (method, or product) (as explained in Wang, there was design incentive: the lack of data specific to a given type of device or machine part. The differences between the claimed invention and the prior art were encompassed in known variations or in a principle known in the prior art (the difference between using this technique on a manufactured part and using this technique on a part of a machine in the factory are encompassed in the concept explained in Wang2. See Wang2 Abstract and Wang2 p. 413. One of ordinary skill in the art, in view of the identified design incentives or other market forces, could have implemented the claimed variation of the prior art, and the claimed variation would have been predictable to one of ordinary skill in the art. See MPEP § 2143(I)(F).) and assessing the third assessment data sets with the second calibration data sets after the training of the artificial intelligence. (As best understood, this claim recites yet another component used for collecting data and training a model in the same way as claim 6. See rejection of claim 6 showing Lee to teach at least three different types of data used to train three different models.)
8. The method of claim 7, wherein the assessment of the first, second, and/or third assessment data sets comprises assessing, a movement of the respective component relative to another component, a strength of the respective component, and/or the presence of an imbalance. (See Lee P. 4494 teaching the sensing of cracks and missing welds.)
Response to Arguments
Applicant's arguments filed 11/13/2025 have been fully considered but they are not persuasive.
Rejections under § 112b:
No specific arguments are submitted.
Note: The amendments led to various rejections under this section. In the interest of compact prosecution, it is submitted that limitations which clearly indicate whether operations are part of training, testing, or online use (i.e. after deployment) together with clearly indicating whether the various data sets are being input to or output from the various AI models may improve clarity. Further, consistent use of names (i.e. avoid using both “trained model 1” and “model 1” when referring to the same model) and consistent use of language (i.e. if training “model 1” results in creation of “model 2” in one location, then training of “model n” should result in “model n+1” (or similar) throughout the claims.)
Rejections under § 103:
Applicant characterizes the combined teachings of Lee and Brownlee, juxtaposes this characterization against the amended claim language and concludes that the claims are non-obvious. See Rem. 6-8. Notably absent from the Remarks, is any articulated distinction between the claimed invention and the combined teachings of the prior art. Without any clearly articulated distinction between the claimed invention and the combined teachings of the prior art, the Remarks are not persuasive.
The omission of any characterization of Applicant’s invention may be strategically sound. However, arguments which articulate the importance of a claimed combination, and explain how that combination is different than the combined teachings of the prior art tend to advance prosecution.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PAUL M KNIGHT whose telephone number is (571) 272-8646. The examiner can normally be reached Monday - Friday 9-5 ET.
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, Michelle Bechtold can be reached on (571) 431-0762. 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.
PAUL M. KNIGHTExaminerArt Unit 2148
/PAUL M KNIGHT/Examiner, Art Unit 2148
Prosecution Insights