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).
Allowable Subject Matter under §§ 102 and 103
Claims 5, 13, and 20 are objected to as being dependent upon a rejected base claim, but would be allowable in view of the prior art, if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
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.
Claims 1, 3, 6-9, 11, 14-15, 17 and 21 are rejected under 35 U.S.C. 103 as being unpatentable over Wang 2019/0378022) and Akouemo (Data Improving in Time Series Using ARX and ANN Models, 2017).
1. A method for producing anomaly-free training data to facilitate machine learning (ML)-based prognostic surveillance operations, comprising: (This language is recited as an intended use. See MPEP §§ 2111.02, 2111.04, and 2103. Further, the producing of anomaly free training data for the purpose of facilitating machine learning based prognostic surveillance operations is obvious in view of the art cited below. receiving a dataset comprising time-series signals obtained from a monitored system during operation of the monitored system; (Wang teaches: “First, the system obtains time-series sensor data.” Wang Abstract.) dividing the dataset into a plurality of subsets; (Wang teaches: “In some embodiments, while dividing the time-series sensor data into the training set and the estimation set, the system first partitions the time-series sensor data into a set of windows. Next, the system selects a subset of alternating windows in the set of windows to be the training set, and selects a remaining subset of unselected alternating windows in the set of windows to be the estimation set.” Wang ¶12. Note that the claimed dividing the dataset into a plurality of subsets reads on both dividing the data into windows and separately reads on dividing data into a training and estimation sets.) identifying subsets that contain anomalies from the target subsets by: (The “target subsets” read on the testing subsets of Wang (and Akouemo first cited below.) The claimed identifying of the target (testing) subsets is addressed below.) training one or more inferential models using combinations of second subsets of the plurality of subsets, wherein one or more of the second subsets used to train the one or more inferential models comprise anomalies, (The “second subsets” read on the training subsets of Wang (and Akouemo first cited below.) The “combinations of second subsets” reads on individual windows of time series training data of Wang. See Wang ¶12 cited above.
The previously cited art does not teach identifying subsets of data that contain anomalies by training a model using combinations of the subsets of data themselves to train a model and using the trained model to detect anomalies in target subsets of a dataset.
Akouemo teaches “The contributions of the proposed algorithms are their ability to extract time series features using an ARX or ANN model, to use the residuals from applying the model to identify anomalous data points, and then to impute replacement values for the identified anomalies.” Akouemo P. 3352. “Typically, additive outliers need to be deleted or replaced because they induce biased variances and estimates [4]. . . . In this work, we focus on additive anomalies. Probabilistic techniques have been used for outlier detection in combination with a rejection threshold, hence yielding many false positives for large data sets [6]–[8]. However, in real data sets, the underlying distribution of the data is not known, and there is not an optimal rule for choosing or calculating a rejection threshold.” Akouemo P. 3352. “Two approaches to identifying and imputing anomalies in energy time series are examined in this paper: ARX and ANN data cleaning models. Because of the potential presence of anomalies, the learned parameters of the models may be biased [9]–[12]. To identify the anomalies, the algorithms calculate the residuals. Hypothesis testing is used to detect anomalies in the residuals and to avoid false positives by taking into account the number of samples in the residual distribution. We assume the residuals are normal. While this assumption is violated in practice, it is useful to develop the theoretical aspect of the technique. Hypothesis testing identifies anomalies in the tails of the distribution. Anomalies in this case are data points considered inconsistent with the distribution of the residual data set. . . . Data cleaning consists of detecting and imputing anomalous data. Therefore, the anomalies identified are imputed using calculated replacement values [31]–[33]. The replacement values also are calculated using ARX or ANN models. . . . As the anomalies are imputed, the estimation of model parameters improves. Therefore, the data cleaning process is implemented iteratively.” Akouemo P. 3353. “The neural network performs a one-step-ahead prediction to keep from learning the anomalies. The ANN is re-trained after an anomaly is imputed.” Akouemo P. 3355. See also Akouemo Algorithm 3 on page 3355. Replacing the anomalous values, as taught in Akouemo, includes removal of the replaced anomalous values. The portions of Akouemo cited above teaches that models are trained using anomalous data which are removed allowing the training dataset to improve, thereby improving the model trained on the improved dataset in an iterative process. This improvement of the model reads on training of the one or more inferential models.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Akouemo because using data that has been filtered to remove the bad (i.e. erroneous) data tends to result in a model that is more accurate. See e.g. Akouemo P. 3352 (“Training accurate models in the energy domain requires anomaly-free training signals, where anomalies refer to data points that are considerably dissimilar to the remaining points in the data set [2].”))) and wherein one or more of the second subsets used to train the one or more inferential models do not contain anomalies, and (Wang teaches “Next, MSET model 108 is “trained” to learn patterns of correlation among all of the time-series signals 104. This training process involves a one-time, computationally intensive computation, which is performed offline with accumulated data that contains no anomalies.” Wang ¶30. “[0035] A problem with the training process is that whatever mechanisms are causing missing values in the time-series data, it is just as likely that there will be missing values in the training dataset as in the analysis dataset. This is problematic, because it is not possible to train an MSET model (or another ML model) with a dataset possessing missing values, if the trained model is to be effectively applied to the analysis dataset to perform optimal MVI. [0036] To overcome this problem, we introduce a new two-phase procedure. We start by detecting missing values in the training dataset and temporarily replacing the missing values with corresponding values computed through conventional interpolation. We then train an MSET model using the training dataset with the temporary interpolations. Next, we apply the trained MSET model to the analysis dataset to detect and replace missing values in the analysis dataset with MSET estimates. In the second phase of the procedure, we use the analysis dataset (with optimal MVI values inserted) to train another model, which is used to replace the temporary interpolations in the training dataset with MSET estimates.” Wang ¶¶35-36. This teaches training a model on data that does not contain anomalies to replace “missing values.” The claimed “anomalies” reads on missing values.) and using the one or more trained inferential models to detect anomalies in the target subsets of the dataset; and removing one or more subsets of the target subsets having the detected anomalies from the dataset. (“The outlier detection and imputation results for the ANN data cleaning algorithm are presented in Fig. 4 and Table II. . . . The ratio for randomly dividing the data set is selected to be 70% for training, 15% for validation, and 15% for testing. Fig. 4 and Table II present the anomalies found along with their original and imputed values.” Akouemo P. 3356. Note that figure 4 shows removal of anomalies and replacing them with other data. Note also that Akouemo teaches “Training accurate models in the energy domain requires anomaly-free training signals, where anomalies refer to data points that are considerably dissimilar to the remaining points in the data set [2].” Akouemo P. 3353.)
3. The method of claim 1, wherein training the one or more inferential models using combinations of the subsets comprises training an inferential model for every possible combination of the subsets. (See rejection of claim 1. Note that the claim language only requires two subsets. Also, while the intended scope of the claim may be training of the models using every possible combination of the subsets as training and testing data, the language “for every possible combination of the subsets” is written as an intended use. See MPEP § 2013 and 2111.04.)
6. The method of claim 1, wherein the method further comprises: during a training mode, using the anomaly-free training data to train an inferential model; (Wang teaches: “Next, MSET model 108 is “trained” to learn patterns of correlation among all of the time-series signals 104. This training process involves a one-time, computationally intensive computation, which is performed offline with accumulated data that contains no anomalies.” Wang ¶ 30.) and during a surveillance mode, using a trained inferential model of the one or more trained inferential models to generate estimated values for the time-series signals received from the monitored system based on cross-correlations between the time-series signals, (“The pattern-recognition system is then placed into a “real-time surveillance mode,” wherein the trained MSET model 108 predicts what each signal should be, based on other correlated variables; these are the “estimated signal values” 110 illustrated in FIG. 1.” Wang ¶ 30.) performing pairwise differencing operations between actual values and the estimated values for the time-series signals set to produce residuals, and analyzing the residuals to detect incipient anomalies in the monitored system. (“Next, the system uses a difference module 112 to perform a pairwise differencing operation between the actual signal values and the estimated signal values to produce residuals 114. The system then performs a “detection operation” on the residuals 114 by using SPRT module 116 to detect anomalies and possibly to generate an alarm 118.” Wang ¶ 30.)
7. The method of claim 6, wherein analyzing the residuals comprises: performing a sequential probability ratio test (SPRT) on the residuals to produce SPRT alarms; and detecting the incipient anomalies based on the SPRT alarms. (“The system then performs a “detection operation” on the residuals 114 by using SPRT module 116 to detect anomalies and possibly to generate an alarm 118. In this way, prognostic-surveillance system 100 can proactively alert system operators of incipient anomalies, such as impending failures, hopefully with enough lead time so that such problems can be avoided or proactively fixed.” Wang ¶ 30.)
8. The method of claim 1, wherein an inferential model of the one or more inferential models comprises a multivariate state estimation technique (MSET) model. (See rejection of claim 6.)
11. The non-transitory computer-readable storage medium of claim 9, wherein training the one or more inferential models using combinations of the subsets comprises training an inferential model for every possible combination of the subsets, (See rejection of claim 3.) and wherein the inferential model of the one or more inferential models comprises a multivariate state estimation technique (MSET) model. (See rejection of claim 6.)
21. The method according to claim 1, wherein the identifying the subsets that contain anomalies from target subsets of the dataset, further comprises, prior to the training, receiving one or more of the second subsets comprising anomalies and one or more of the second subsets that do not contain anomalies. (The claimed subsets read on sets of data taught in the prior art. See rejection of claim 1. One of ordinary skill in the art would understand a teaching of using data to train a model as implicitly teaching receipt of that data.) “[I]n considering the disclosure of a reference, it is proper to take into account not only specific teachings of the reference but also the inferences which one skilled in the art would reasonably be expected to draw therefrom.” MPEP § 2144.01.
Claims 9 and 14-15 are rejected for the reasons given in the rejections of claims 1 and 6-7, respectively.
Claim 17 is rejected for the reasons given in the rejection of claim 1.
Claims 2, 10, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Wang, Akouemo, and Applicant Admitted Prior Art.
2. The method of claim 1, wherein removing identified subsets from the dataset comprises: confirming that the identified subsets contain anomalies; and removing the identified subsets that are confirmed to contain anomalies. (The previously cited art does not expressly teach confirming that the identified subsets contain anomalies.
The background section titled “Related Art” indicates this is AAPA. “ML-based prognostic-surveillance techniques operate by learning patterns in undegraded training data, which is obtained when no degradation is present in the monitored assets, and subsequently detecting anomalies in those patterns during normal system operation. Note that this undegraded training data is not necessarily pristine data from a brand new asset; it is data from an asset for which a subject-matter expert (SME) has determined that no degradation modes were active during the time of training data collection.” Spec. ¶ 5. See also Provisional Spec. (63/196328) Page 2, under “Prior Industry Practices.”
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine this AAPA because just asking an expert, when possible, can provide undegraded training data with minimal technical creativity or effort (i.e. asking someone is easy.))
Claims 10 and 18 are rejected for the reasons given in the rejection of claim 2.
Claims 4, 12, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Wang, Akouemo, and Nowak (Statistical Signal Processing; 2011).
4. The method of claim 1, wherein using the one or more trained inferential models to detect anomalies in the target subsets of the dataset comprises: using the one or more trained inferential models to perform prognostic surveillance operations on the target subsets; and identifying target subsets that contain anomalies based on a number of alerts produced during the prognostic-surveillance operations. (Note that “a number of alerts” reads on one alert (i.e. the claim language does not require a threshold number of alerts.) Wang teaches: “The pattern-recognition system is then placed into a “real-time surveillance mode,” wherein the trained MSET model 108 predicts what each signal should be, based on other correlated variables; these are the “estimated signal values” 110 illustrated in FIG. 1. Next, the system uses a difference module 112 to perform a pairwise differencing operation between the actual signal values and the estimated signal values to produce residuals 114. The system then performs a “detection operation” on the residuals 114 by using SPRT module 116 to detect anomalies and possibly to generate an alarm 118.” Wang ¶ 30.
While the claim language in its current for reads on the art cited above, in the interest of compact prosecution, a secondary reference is provided to explain why one of ordinary skill in the art would be motivate to use multiple alerts and SPRT to identify anomalies in data.
Nowak teaches: “The SPRT is based on considering the likelihood ratio as a function of the number of observations. . . . The goal of the SPRT is to decide which hypothesis is correct as soon as possible (i.e., for the smallest value of k). To do this the SPRT requires two thresholds, [Gamma]1 >= [Gamma]0. The SPRT stops’ as soon as [Lambda]k >= [Gamma]1, and we then decide H1 is correct, or when [Lambda]k <= [Gamma]0, and we then decide H0 is correct. The key is to set the thresholds so that we are guaranteed a certain levels of error. . . . We will try to set the thresholds to provide desired probabilities of detection PD and false-alarm PFA.” Nowak Page 1. “The expected stopping time of the SPRT that we determined above is optimal. No other test can achieve the same PD and PFA with a smaller expected number of samples, under either hypothesis[.]” Nowak Page 3.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Nowak because using more observations improves the level of error that can be guaranteed.)
Claims 12 and 19 are rejected for the reasons given in the rejection of claim 4.
Response to Arguments
Applicant's arguments filed 09/11/2025 have been fully considered but they are not persuasive.
Rejections under § 103
The amended claims additionally recite training model using non-anomalous data. See rejection of claim 1, citing Wang ¶30.
The Remarks assert that both cited references fail to teach of using subsets of data that include anomalies to train a model for removal of anomalies. This does not address the position articulated in the rejection of claim 1. Akouemo teaches training a model on portions of a dataset, whereby anomalies are iteratively removed from the dataset. One of ordinary skill would infer that a model being trained on a dataset to iteratively remove anomalies from that dataset, presupposes anomalies within the dataset.
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. 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. KNIGHT
/PAUL M KNIGHT/Examiner, Art Unit 2148