SIEMENS AKTIENGESELLSCHAFT

Patent Trials and Appeals BoardAug 25, 2021

2020002678 (P.T.A.B. Aug. 25, 2021)

UNITED STATES PATENT AND TRADEMARK OFFICE UNITED STATES DEPARTMENT OF COMMERCE United States Patent and Trademark Office Address: COMMISSIONER FOR PATENTS P.O. Box 1450 Alexandria, Virginia 22313-1450 www.uspto.gov APPLICATION NO. FILING DATE FIRST NAMED INVENTOR ATTORNEY DOCKET NO. CONFIRMATION NO. 14/427,418 03/11/2015 Hans-Gerd Brummel P/5070-214 (V33226) 6722 119119 7590 08/25/2021 Ostrolenk Faber LLP 845 THIRD AVENUE 8TH FLOOR New York, NY 10022 EXAMINER RIFKIN, BEN M ART UNIT PAPER NUMBER 2198 NOTIFICATION DATE DELIVERY MODE 08/25/2021 ELECTRONIC Please find below and/or attached an Office communication concerning this application or proceeding. The time period for reply, if any, is set in the attached communication. Notice of the Office communication was sent electronically on above-indicated "Notification Date" to the following e-mail address(es): pat@ostrolenk.com PTOL-90A (Rev. 04/07) UNITED STATES PATENT AND TRADEMARK OFFICE ____________ BEFORE THE PATENT TRIAL AND APPEAL BOARD ____________ Ex parte HANS-GERD BRUMMEL, KAI HEESCHE, UWE PFEIFER, and VOLKMAR STERZING ____________ Appeal 2020-002678 Application 14/427,418 Technology Center 2100 ____________ Before KARL D. EASTHOM, KARA L. SZPONDOWSKI, and STEVEN M. AMUNDSON, Administrative Patent Judges. EASTHOM, Administrative Patent Judge. DECISION ON APPEAL I. STATEMENT OF THE CASE Appellant1 appeals under 35 U.S.C. § 134(a) from the Examiner’s decision finally rejecting claims 1–16 and 18–24, which constitute all of the claims pending in this application. We have jurisdiction under 35 U.S.C. § 6(b). 1 “Appellant” refers to “applicant” as defined in 37 C.F.R. § 1.42. Appellant identifies the real party in interest as Siemens Aktiengesellschaft. Appeal Br. 1. Appeal 2020-002678 Application 14/427,418 2 We AFFIRM. II. DISCLOSED AND CLAIMED SUBJECT MATTER The Specification discloses a method for “computer-assisted monitoring of the operation of a technical system, particularly of an electrical energy-generating installation.” Spec. 1. The Specification explains that “state monitoring of a technical system, such as e.g. an electrical energy-generating installation, and particularly a gas turbine” requires observing “a multiplicity of process variables . . . in order to obtain information from the current values of these variables to indicate whether the technical system is in the intended state.” Id. The Specification also explains that “analytical process models function very well for monitoring specific technical systems,” such as “to detect malfunctions in a technical system which may, in the long term, result in damage to the system.” Id. at 1–2. However, these analytical process models depend “on the mirroring of the configuration of the real technical system,” which limits the precision for installations that “are often configured in a customer-specific and site-specific manner” or where “configuration of the technical system is modified.” Id. at 2. Therefore, the invention endeavors to provide “a simple and efficient method for the computer-assisted monitoring of a technical system.” Id. This solution characterizes the technical system “at corresponding operating times by a state vector comprising a number of input variables and at least one output variable which is to be monitored . . . which influence the operation of the technical system.” Id. at 3. Independent claim 1 follows: 1. A method for computer-assisted monitoring of an operation of a technical system, wherein the technical system is Appeal 2020-002678 Application 14/427,418 3 characterized at corresponding operating times by a state vector comprising a set of input variables and at least one predicted output variable which is to be monitored, the method comprising: a) training a data-driven model using training data from known state vectors, and predicting, with the data-driven model, the at least one predicted output variable for respective operating times on the basis of the set of input variables occurring in the operation of the technical system; b) applying at least one density estimator trained by known input variables of the training data for respective operating times to the set of input variables at the corresponding operating time, so as to calculate a confidence measure, wherein the confidence measure is higher the greater a similarity of the set of input variables at the corresponding operating time to the known input variables from the training data; c) for respective cycles for a plurality of consecutive operating times, calculating a deviation weighted by the confidence measure, averaged over the set of state vectors in the respective cycle, between the at least one predicted output variable and the at least one actual output variable occurring in the operation of the technical system, wherein state vectors whose set of input variables have low confidence measures are weighted less in the average weighted deviation than state vectors whose set of input variables have high confidence measures; d) determining a malfunction of the technical system when the average weighted deviation for a number of consecutive cycles exceeds a predefined threshold value; and controlling the operation of the technical system based on the determined malfunction. Appeal 2020-002678 Application 14/427,418 4 III. REFERENCES The prior art relied upon by the Examiner as evidence in rejecting the claims on appeal is: Name Reference Date Svensson US 3,875,403 Apr. 1, 1975 Cardner US 5,402,333 Mar. 28, 1995 Strackeljan et al. (“Strackeljan”) US 2002/0013664 A1 Jan. 31, 2002 Pokorny et al. (“Pokorny”) US 2003/0150908 A1 Aug. 14, 2003 Feeney et al. (“Feeney”) US 2005/0261820 A1 Nov. 24, 2005 Schafer et al. (“Schafer”) US 2009/0271344 A1 Oct. 29, 2009 Renals “Learning and Generalization Neural Networks” Feb. 27, 1998 Lang et al. (“Lang”) “Neural Clouds for Monitoring of Complex Systems” June 24, 2008 Nozari et al. (“Nozari”) “Model-based robust fault detection and isolation of an industrial gas turbine prototype using soft computing techniques” Mar. 24, 2012 CareerRide “Difference between a Vector and an Array” Feb. 28, 2010 Appeal 2020-002678 Application 14/427,418 5 IV. REJECTIONS Claim 18 stands rejected under 35 U.S.C. § 112(b) for failing to particularly point out and distinctly claim the subject matter regarded as the invention. Final Act. 2. Claims 1–6, 9, 10, 15, 16, 19, and 20 stand rejected under 35 U.S.C. § 103 as unpatentable over Nozari, CareerRide, Renals, Cardner, and Svensson. Final Act. 3.2 Claims 7, 23, and 24 stand rejected under 35 U.S.C. § 103 as unpatentable over Nozari, CareerRide, Renals, Cardner, Svensson, and Lang. Final Act. 15. Claim 8 stands rejected under 35 U.S.C. § 103 as unpatentable over Nozari, CareerRide, Renals, Cardner, Svensson, and Strackeljan. Final Act. 18. Claims 11–13 and 22 stand rejected under 35 U.S.C. § 103 as unpatentable over Nozari, CareerRide, Renals, Cardner, Svensson, and Feeney. Final Act. 19. Claim 14 stands rejected under 35 U.S.C. § 103 as unpatentable over Nozari, CareerRide, Renals, Cardner, Svensson, and Pokorny. Final Act. 22. Claim 21 stands rejected under 35 U.S.C. § 103 as unpatentable over Nozari, CareerRide, Renals, Cardner, Svensson, and Schafer. Final Act. 23. 2 The Final Action lists claims “1–6, 9–10, 15–20” as rejected for obviousness based on Nozari, CareerRide, Renals, Cardner, and Svensson. Id. at 3. This listing is an obvious error, because claim 17 is not pending and the Final Action does not address claims 17 and 18 in the body of the obviousness rejection. See id. at 3–15. Appeal 2020-002678 Application 14/427,418 6 V. OPINION Section 112(b) Rejection The Examiner finds that the “status of” claim 18 “is unclear due to the relationship with other claims.” Final Act. 2. Appellant argues that claim 18 was amended and therefore “this rejection is believed to be withdrawn by the Examiner.” Appeal Br. 5. However, the rejection has not been withdrawn and remains standing, and Appellant has not addressed it as it stands. Therefore, we summarily affirm the rejection of claim 18 under 35 U.S.C. § 112(b) as indefinite.3 Section 103 Rejection The Examiner rejects claims 1–16 and 19–24 as obvious over the combined teachings as set forth above. See Final Act. 3–24. Appellant treats the claims as a group and relies on arguments it presents for claim 1. See Appeal Br. 9–11; Reply Br. 4. Therefore, claim 1 represents the claims on appeal. Claim 1 recites “training a data-driven model using training data from known state vectors, and predicting, with the data-driven model, the at least one predicted output variable for respective operating times on the basis of 3 This rejection applies to claim 18 as filed on April 2, 2019 and as rejected by the Examiner in the Final Action mailed June 5, 2019. Subsequent to the Final Action and before the Notice of Appeal filed September 5, 2019, Appellant proposed an amendment of claim 18 in a Supplemental Amendment filed August 21, 2019. Per the Manual of Patent Examining Procedure, “amendments filed after the final action are not entered unless approved by the examiner.” MPEP § 714.12; see id. § 706.07(f), § 714.13, § 1206. Here, the Examiner did not approve or enter the after-final amendment of claim 18 so we do not consider it. See Ans. 3 (maintaining the rejections in the Final Action of June 5, 2019). Appeal 2020-002678 Application 14/427,418 7 the set of input variables occurring in the operation of the technical system,” “applying at least one density estimator trained by known input variables of the training data for respective operating times to the set of input variables at the corresponding operating time, so as to calculate a confidence measure” that “is higher the greater a similarity of the set of input variables at the corresponding operating time to the known input variables from the training data,” and “calculating a deviation weighted by the confidence measure, averaged over the set of state vectors in the respective cycle, between the at least one predicted output variable and the at least one actual output variable occurring in the operation of the technical system.” The Examiner finds that Nozari teaches “using data to train the networks” and a “data-driven model.” Final Act. 4 (citing Nozari 32); see Ans. 4–5 (citing Nozari 33–34). The Examiner also finds that Nozari teaches “build[ing] an error model for the neural network” with a “density estimator” that “determines [availability] based on the input data availability to determine the certainty of the network outputs,” and also teaches “giving a confidence with upper and lower thresholds to match the training data.” Final Act. 4 (citing Nozari 32–33); see Ans. 6–7. The Examiner relies on Renals to teach “using training to perform generalization so inputs similar to those seen in the training set will be correctly responded to.” Final Act. 5 (citing Renals 15); see Ans. 6. According to the Examiner, the “combined references” of Nozari and Renals teach the claimed calculating a confidence measure by looking at the trained input data. Ans. 6–7. The Examiner relies on Cardner “to show that weighting deviations/errors based upon confidence was obvious to one of ordinary skill in the art.” Id. at 8; see also Final Act. 5–6 (citing Cardner, col. 8, ll. 21–35). Appeal 2020-002678 Application 14/427,418 8 Nozari discloses “robust fault detection” by identifying “an input- output model” after “selecting of the proper inputs and output.” Nozari 33. Inputs to the neuro fuzzy network include process inputs and also residuals “generated by comparing the corresponding predicted and system outputs.” Id. at 32–33. “The output of these error models are added with corresponding nominal model output in order to generate centre of uncertainty region” with “upper and lower bands [that] are built by some statistical extension to the generated uncertainty centres.” Id. In an example, Nozari discloses manipulating the sum of the output of the system model and the related error model “considering standard deviation of the output of [the] error model on the inputs . . . and a value assigned to a given confidence level which leads to creating adaptive upper and lower thresholds.” Id. The system uses the bands and system outputs to decide “on the occurrence of fault.” Id. Renals, as relied upon by the Examiner, discloses using “the training process (e.g., backprop) . . . to estimate the parameters of the function (i.e. the weights of the network) so that it replicates the data as well as possible – and generalizes to new data well.” Renals 15 (emphasis added). Cardner, as relied upon by the Examiner, discloses that a “correction factor for each flow signal” produces “an error signal” and is then “multiplied by two weights; a weight reflecting the degree of confidence in the measurement, and a weight that reflects the degree of correction required for the flow.” Cardner, col. 8, ll. 28–35. Appellant argues that “Nozari does not calculate a confidence measure C(t) based on similarity of input values to known (training set) input values, as required by claim 1.” Appeal Br. 6. Appellant argues that Appeal 2020-002678 Application 14/427,418 9 “Cardner does not disclose or suggest that the degree of confidence is based on input data obtained now compared with known input variables from the training data” or “that this degree of confidence is used to weight the deviation between actual and predicted outputs.” Id. at 8 (emphasis added); see also Reply Br. 3. However, the Examiner relies on Renals to teach “using training data” and data “trained by known input variables” as claimed. Final Act. 4–5 (citing Nozari 32–34; Renals 15); see Ans. 6–7. Therefore, by failing to address the Examiner’s showing based on the combination of Nozari and Renals, Appellant does not undermine the Examiner’s showing. As cited by the Examiner, Nozari discloses using process inputs and residuals as inputs in “an input-output model” for error modeling that uses “proper inputs and output” to “generate centre of uncertainty region” that “consider[s] standard deviation of the output of [the] error model on the inputs . . . and a value assigned to a given confidence level which leads to creating adaptive upper and lower thresholds.” Nozari 33. Nozari further discloses building a “predictor model” by feeding “the past samples of the process inputs as well as of the process output . . . into the model as inputs.” Id. at 34. As relied on by the Examiner, Renals discloses using “the training process (e.g., backprop) . . . to estimate the parameters of the function (i.e. the weights of the network).” Renals 15. In other words, Nozari teaches inputs and outputs, including past samples, for a data-driven model and calculating a confidence level value; and Renals teaches using known training data (estimated values based on a training process). Appeal 2020-002678 Application 14/427,418 10 Appellant has not explained why the claimed “training a data-driven model using training data from known state vectors” and “predicting, with the data-driven model, the at least one predicted output variable” and using these “trained” and “known input variables . . . to calculate a confidence measure” precludes the combination of Nozari’s “input-output model,” a data-driven model, utilizing “past samples or the process inputs as well as of the process output,” to determine confidence level, with Renals’s using a training process to estimate values. Appellant also asserts that “Nozari does not disclose or suggest using this confidence measure C(t) to weight the average error (obtained as a difference between the actual output and the neural network or data-driven model output), as required by claim 1.” Appeal Br. 6–7. Appellant further argues that Renals teaches “using many patterns of training data to train the neural network so as to produce weights that map to meaningful outputs is better than using fewer patterns of such training data,” but otherwise “is silent as to a weighted deviation calculated in such a way based on such a confidence measure for the predicted output variable and the actual output variable” or “averaged over a set of state vectors” as claimed. Id. at 7–8. However, the Examiner relies on Cardner “to show that weighting deviations/errors based upon confidence was obvious to one of ordinary skill in the art.” Ans. 8; see also Final Act. 5–6 (citing Cardner, col. 8, ll. 21–35). Therefore, by failing to address the Examiner’s showing based on Cardner, Appellant does not undermine the Examiner’s showing. As cited by the Examiner, Cardner discloses multiplying an error signal “by two weights; a weight reflecting the degree of confidence in the measurement, and a weight that reflects the degree of correction required for the flow.” Appeal 2020-002678 Application 14/427,418 11 Cardner, col. 8, ll. 28–35. As also relied on by the Examiner, Nozari discloses “considering standard deviation of the output of [the] error model on the inputs . . . and a value assigned to a given confidence level.” Nozari 33. In other words, Nozari teaches calculations for the outputs including deviation and confidence level; and Cardner teaches calculations being weighted by a confidence level. Appellant does not explain why the claimed “calculating a deviation weighted by the confidence measure . . . between the at least one predicted output variable and the at least one output variable” precludes the combination of Nozari’s input-output data-driven model, utilizing past samples and determining confidence level, with Cardner’s weighing a signal by a confidence measurement. In summary, Appellant does not persuasively explain how the Examiner erred in finding that the combination of Nozari, Renals, and Cardner, with CareerRide and Svensson, teaches or suggests “training a data-driven model using training data from known state vectors, and predicting, with the data-driven model, the at least one predicted output variable for respective operating times on the basis of the set of input variables occurring in the operation of the technical system,” “applying at least one density estimator trained by known input variables of the training data for respective operating times to the set of input variables at the corresponding operating time, so as to calculate a confidence measure” that “is higher the greater a similarity of the set of input variables at the corresponding operating time to the known input variables from the training data,” and “calculating a deviation weighted by the confidence measure, averaged over the set of state vectors in the respective cycle, between the at Appeal 2020-002678 Application 14/427,418 12 least one predicted output variable and the at least one actual output variable occurring in the operation of the technical system,” as recited in claim 1. In the Reply Brief, Appellant asserts that Nozari does not teach “calculating a confidence measure that is higher the greater a similarity of the set of input variables to the known input variables from the training data.” Reply Br. 2. Appellant also argues that “a person of ordinary skill in the art would have been hard pressed to take this teaching of Nozari” concerned with “classifying a residual vector pattern to a pre-specified class of faulty conditions” with “any other teaching from Renals and Cardner to arrive at the” disputed limitations of claim 1. Id. Appellant also argues that “Nozari teaches away from the method recited in claim 1 by describing a complex method of creating upper and lower thresholds for a center of uncertainty region.” Id. Appellant advances these arguments for the first time in the Reply Brief. Such arguments “will not be considered by the Board” unless an appellant shows good cause. See 37 C.F.R. § 41.41(b)(2); see also Ex parte Borden, 93 USPQ2d 1473, 1475 (BPAI 2010) (informative) (discussing procedural difficulties with belated arguments). Here, Appellant has not shown good cause for the belated arguments in the Reply Brief. Hence, we decline to consider them. “Considering an argument advanced for the first time in a reply brief . . . is not only unfair to an appellee, but also entails the risk of an improvident or ill-advised opinion on the legal issues tendered.” McBride v. Merrell Dow & Pharm., Inc., 800 F.2d 1208, 1211 (D.C. Cir. 1986). Appeal 2020-002678 Application 14/427,418 13 Based on the foregoing discussion, Appellant does not show error in the Examiner’s findings and determination of the obviousness of claim 1.4 As noted above, Appellant does not challenge claims 2–16 and 19–24 independently from claim 1. Accordingly, claims 2–16 and 19–24 fall with claim 1. VI. CONCLUSION We affirm the Examiner’s decision rejecting claim 18 under § 112. We affirm the Examiner’s decision rejecting claims 1–16 and 19–24 under § 103. VII. DECISION SUMMARY In summary: Claim(s) Rejected 35 U.S.C. § Reference(s)/Basis Affirmed Reversed 18 112 Indefiniteness 18 1–6, 9, 10, 15, 16, 19, 20 103 Nozari, CareerRide, Renals, Cardner, Svensson 1–6, 9, 10, 15, 16, 19, 20 7, 23, 24 103 Nozari, CareerRide, Renals, Cardner, Svensson, Lang 7, 23, 24 8 103 Nozari, CareerRide, Renals, Cardner, Svensson, Strackeljan 8 11–13, 22 103 Nozari, CareerRide, Renals, Cardner, Svensson, Feeney 11–13, 22 4 In the event of further prosecution, the Examiner may consider whether claim 1 lacks clarity or otherwise presents ambiguity. The term “the set of state vectors” in step c lacks antecedent basis. Appeal 2020-002678 Application 14/427,418 14 14 103 Nozari, CareerRide, Renals, Cardner, Svensson, Pokorny 14 21 103 Nozari, CareerRide, Renals, Cardner, Svensson, Schafer 21 Overall Outcome 1–16, 18– 24 TIME PERIOD FOR RESPONSE No time period for taking any subsequent action in connection with this appeal may be extended under 37 C.F.R. § 1.136(a). See 37 C.F.R. § 1.136(a)(1)(iv) (2019). AFFIRMED