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 .
Information Disclosure Statement
The information disclosure statement (IDS) was submitted on 01/08/2024. The
submission is in compliance with the provisions of 37 CFR 1.97. Accordingly, the
information disclosure statement is being considered by the examiner.
Drawings
[AltContent: textbox (FIG. 6 (partial))]The drawings as filed on 02/27/2024 are objected to as failing to comply with 37 CFR 1.84(p)(5) because they include the following reference character(s) not mentioned in the description:
[AltContent: textbox (FIG. 9 (partial))]FIG. 6 depicts line labeled with “DB”. The term “DB” is not found in the written description (figure included, right, for reference). For examination, this label is interpreted to be an abbreviation for “decision boundary” based on guidance found in specification in at least [0097] .
FIG. 9 depicts arrow labeled with “PCA”. The term “PCA” is not found in the written description (figure included, right, for reference). For examination, this label is interpreted to be an abbreviation for “principal component analysis” based on guidance found in specification in at least [0006].
Corrected drawing sheets in compliance with 37 CFR 1.121(d), or amendment to the specification to add the reference character(s) in the description in compliance with 37 CFR 1.121(b) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either “Replacement Sheet” or “New Sheet” pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
Specification
The disclosure is objected to because of the following informalities:
Examiner notes the following passage in specification:
[0006]: “…errors between original operation data (the normal operation data) and reconstructed operation data (an reconstruction data) calculated after application of a principal component analysis.” It is not clear what is meant by “an reconstruction data”.
For examination, Examiner interprets this to be an insertion error of the word “an”. Appropriate correction is required.
Claim Objections
Claims 1-3, and 14 are objected to because of the following informalities:
Claim 1 recites: “…installed in an oil well and operates; an abnormality detection module…” Examiner suggests the presence of the semicolon makes the claim unclear. For purposes of examination, Examiner interprets this language to read “…installed in an oil well and operates an abnormality detection module…”, i.e., the semicolon is ignored. Appropriate correction is required.
Claim 3 recites: “operation data (the normal operation data) and reconstructed operation data (an reconstruction data)”, where Examiner reads as an insertion error of the word “an”; and “the second abnormality occur when an error level”, where Examiner notes the word “occur” should be the word “occurs”. Appropriate correction is required.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Explanatory Remarks:
Independent CLAIM 1 recites the following limitations invoking interpretation under 112(f):
“a data collection module which collects real-time data ….and operates” and “abnormality detection module which detects three abnormalities” which has corresponding structure of computer software implemented on a processor (see [0048] in the instant application)
Dependent Claims 2 and 3 recite: “abnormality detection module determines” which has corresponding structure of computer software implemented on a processor (see [0048] in the instant application)
Dependent Claim 6 recites: “reporting module providing a notification that an abnormality is detected”” which has corresponding structure of computer software implemented on a processor (see [0048] in the instant application)
Dependent Claim 8 recites: “a training data generation module” which has corresponding structure of computer software implemented on a processor (see [0048] in the instant application)
Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, they are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
Claim Rejections - 35 USC § 112
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.
Claims 1-8, and 12-14 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 applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Specifically, Claim 1 (line 11) recites: “first abnormality appearing before failure occurrence by inputting the real-time data into a first model”. It is not clear as to where or how “first abnormality” ‘appears’. Moreover, the language “failure occurrence” does not appear in the claim prior to this instance. The sentence leaves unclear whether the first abnormality “appears” in real-time data as input into a first model, along with “failure occurrence”, or whether the predictive model also identifies “failure occurrence” based on input of real time data. Examiner understands limitation to define “first abnormality” designated by its “appearance” relative to (“before”) “failure occurrence”. However, the intention and/or meaning of the relationship between the first model, input of real-time data, first abnormality appearance, and the failure occurrence is not clear; additional clarity in use of the word “before” is needed. The language renders the claim indefinite. In order to proceed with examination, Examiner consults written description, finding claim limitation recited in [0012], with no additional clarity. Further [0062] recites that model M1 may predict that “failure with an existing occurrence record will occur in the near future”, where Model M1 has been previously trained with historical data [0057], and where which has been classified by M1 using linear discriminant analysis model [0093], and where training data is a standardized time series [0094]; once M1 is trained, real time data is input to M1, and classified as normal/abnormal [0100]; where “abnormality detection notification” occurs when M1 outputs real data as the abnormal class” and “When the reporting module 25 receives an abnormal class from the abnormality detection module 24, the reporting module 25 may determine that the first abnormality occurs and may provide abnormality detection notification.” And “Through the abnormality detection notification, the administrator 4 may recognize that a failure having an occurrence history will occur” [0101]. Examiner notes that this language does not clarify questions raised above in regard to claim limitations. For purposes of examination and until Applicant either overcomes or cures the deficiency above, the Examiner will interpret claim limitations to mean that a “first abnormality” will be designated by evaluation of real-time data input into M1, and which results in a classification of “abnormal”, where M1 has been previously trained using historical data.
Based on reading of Claim 1, Examiner finds limitations of Claim 2 with dependency to Claim 1 recites inconsistent or unclear language. Specifically, Claim 2 recites “the first abnormality occurs when the real-time data are classified as the abnormal class according to linear discriminant criteria generated by the first model”, in keeping with reading of specification discussed above, but where “first abnormality” is defined in Claim 1 by “appearance” related to “failure occurrence”. It is not clear whether both conditions are required to designate a “first abnormality”, i.e., “appearing before failure occurrence” and “when the real-time data are classified as the abnormal class”. It is not clear whether the limitation of Claim 2 is a separate embodiment or an additional limitation in defining how “first abnormality” is defined or determined. This language renders the claim indefinite. For purposes of examination and until Applicant either overcomes or cures the deficiency above, Examiner will interpret claim limitations to mean that a “first abnormality” will be designated by evaluation of real-time data input into M1, and which results in a classification of “abnormal”, where M1 has been previously trained using historical data, consistent with Claim 1 interpretation.
Claim 7 recites “using normal operation data of another pump…and is regenerated by updating the normal operation data with its own operation data in a normal state”. It is not clear what “its own operation data” is referring to, whether it is the “another pump” or the “a pump before it normally operates”. It is not clear which data is being updated, how updates are performed, and which data is being regenerated. This language renders the claim indefinite. For purposes of examination and until Applicant either overcomes or cures the deficiency above, Examiner will interpret claim limitations to mean that two sets of data are being mathematically combined using broadest reasonable interpretation.
Claim 8 recites (numbering and bold emphasis added): (1) “in such a manner that time series data are standardized in the operation data” and (2) “the training data generation module generates the history data to generate training data of the first model,” The meaning of “standardized” in (1) is not found in claim limitation, nor in Claim 1 to which Claim 8 depends. The language in bold in (2) is not clear, since it has been established prior to this claim that “history data” is acquired rather than generated. This language renders the claim indefinite. For purposes of examination and until Applicant either overcomes or cures the deficiency above, specification was reviewed, finding mention of “standardized” in [0011], [0016], [0072], and [0094], where no further detail or clarity is found as to the intended meaning of “standardized”, leaving ambiguity in how, and/or to what, standard or metric the time series data is being compared to/with for the standardization process. Examiner interprets and evaluates claim using plain meaning to mean any process which systematically structures data to facilitate direct comparison with other data and “training data generation module uses history data to generate…” in order to proceed with comparison with prior art.
Claim 12 recites: (bold emphasis added) “based on the distribution of the feature vectors and the center point of the initial operation data”. Examiner notes Claim 12 carries dependence to Claim 9. The limitation “center point” does not appear in Claim 9 or previous to this line in Claim 12. The limitation “a center point” appears in Claim 1, but Claim 12 does not have dependency to Claim 1. There is insufficient antecedent basis for this limitation in the claim. For purposes of examination and until Applicant either overcomes or cures the deficiency above, Examiner interprets and evaluates claim to have determined a center point as part of the method of Claim 9, similar to limitations recited in Claim 4.
Claim 13 recites: (bold emphasis added): “to generate training data of the first model, generating history data in such a manner that time series data are standardized in the operation data…” The meaning of “standardized” is not found in claim limitation, nor in Claim 9 to which Claim 13 depends. For the same reasons and rationale applied to Claim 8 discussed above, this language renders the claim indefinite. For purposes of examination and until Applicant either overcomes or cures the deficiency above, Examiner interprets and evaluates claim as described above in Claim 8 discussion.
Claim 14 recites: (bold emphasis added) “wherein in the step of training of the second model generating the second model even before a pump normally operates by using normal operation data of another pump, which operates in a similar environment, and regenerating the second model by updating the normal operation data with its own operation data in a normal state.” This language renders the claim indefinite. In order to examine the limitation, Examiner interprets that claim as missing the word “comprises”, based on comparison with similar language found in Claim 3 regarding the second model. For purposes of examination and until Applicant either overcomes or cures the deficiency above, Examiner interprets and evaluates claim to read as follows (bold emphasis added): “wherein in the step of training of the second model comprises generating the second model even before a pump normally operates”.
Regarding Claims 4-7, the claims are also rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite since they are dependents of indefinite independent Claim 1, and their limitations do not overcome the indefiniteness issues identified in the parent claim.
Claim Rejections - 35 USC § 101
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.
As best understood by the examiner, Claims 1-14 are rejected under 35 U.S.C. 101. The claimed invention is directed to the abstract concept of performing mental steps without significantly more.
Claim 1 is held to be patent ineligible, as explained below.
Claim 1 limitations recite abstract ideas (bold emphasis added):
“A pump failure prediction apparatus, comprising:
at least one processor; a storage, which is communicably connected with the processor and stores a program code which operates in the processor;
and a communicator, which is communicably connected with the processor,
wherein the program code comprises:
a data collection module which collects real-time data related to a state of a pump which is installed in an oil well and operates;
an abnormality detection module which detects three abnormalities, the three abnormalities comprising a first abnormality appearing before failure occurrence by inputting the real-time data into a first model, which has performed supervised learning of history data, a second abnormality outside of a normal operation range of the pump by inputting the real-time data to a second model, which has performed unsupervised learning of normal operation data, and a third abnormality outside of an initial normal operation range of the pump by inputting the real-time data to a third model, which has performed unsupervised learning of initial operation data.”
STEP 1: Determination of statutory category
Claim 1 falls within the four statutory categories of patentable subject matter
identified by 35 U.S.C. 101, namely Machine/Manufacture (Apparatus).
STEP 2A-PRONG ONE: Determination regarding whether claim recites a judicial
exception.
Applying broadest reasonable interpretation, Claim limitations noted above with bold emphasis, recite a judicial exception. These limitations include: data collection module…; abnormality detection module…detects three abnormalities; first model, which has performed supervised learning; second model, which has performed unsupervised learning; and third model, which has performed unsupervised learning”. Such limitations constitute a judicial exception of Abstract Idea because under broadest reasonable interpretation and using 2024 Revised Patent Subject Matter Eligibility Guidance, the limitations fall into the grouping of subject matter that covers performing mathematics (MPEP 2106.04(a)(2), I.A,C, III.B,C) Examiner notes execution of the claimed limitations involve performing mathematics using at least some generic computer components.
Specifically, Claim 1 recites the element(s) of using generic computational components and generic artificial intelligence (AI)/machine learning (ML) technology, i.e. “model” which has performed “supervised learning” and/or “unsupervised learning”, to perform input of data, data evaluations or calculations, directed to predicting a fault or determining an abnormality. Claim 1 limitations do not recite details regarding how the AI/ML algorithm or model functions or is trained. Instead, the claims are found to utilize the AI/ML algorithm as a tool that provides nothing more than mere instructions to implement the abstract idea on a general purpose computer. (MPEP 2106.05(f)). Claim 1 recites a judicial exception of Abstract Idea.
STEP 2A-PRONG TWO: Evaluation of additional elements to determine whether
the claim integrates the judicial exception into a practical application of that exception.
Claim 1 does not recite significantly more than the judicial exception to integrate
the recited abstract idea into a practical application because there is no improvement to
another technology or technical field; improvements to the functioning of the computer
itself; a particular machine; or effecting a transformation or reduction of a particular article to a different state or thing. Examiner notes additional elements found in Claim 1 limitations, including: “apparatus comprises at least one processor; a storage, a communicator, a data collection module which collects real-time data, and inputting the real-time data”. Examiner notes these additional elements recite necessary data gathering required to provide data for carrying out the judicial exception as defined in analysis above. As recited in MPEP section 2106.05(g), necessary data gathering (i.e.
receiving data) is considered extra solution activity in light of Mayo, 566 U.S. at 79, 101
USPQ2d at 1968; OIP Techs., Inc. v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1092-93 (Fed. Cir. 2015). Further, Claim 1 limitations recite “data related to a state of a pump which is installed in an oil well and operates”. Such language is considered as generally linking the use of a judicial exception to a particular technological environment or field of use, but does not integrate a judicial exception into a practical application. (MPEP § 2106.05(h)). Further, the additional elements do not integrate the judicial exception into a practical application because there is no improvement to another technology or technical field; improvements to the functioning of the computer itself; a particular machine; effecting a transformation or reduction of a particular article to a different state or thing.
STEP 2B: Consideration of whether the claim amounts to significantly more than
the abstract idea.
Additional elements, as discussed above, do not amount significantly more
than the judicial exception because, as noted above, limitations reciting necessary data gather, even when linked to a particular data source or a type of data, are considered to be insignificant extra solution activity. And, as above, generic computer elements, such as the recited “processor”, “storage”, and “communicator”, along with “abnormality detection module” and “data collection module” are comprised of generic computer elements and not considered significantly more than the abstract idea. As recited in the MPEP, 2106.05(b), merely adding a generic computer, generic computer components, or a programmed computer to perform generic computer functions does not automatically overcome an eligibility rejection. (see Alice Corp. Pty. Ltd. v. CLS Bank Int'l, 134 S. Ct. 2347, 2359-60, 110 USPQ2d 1976, 1984 (2014). See also OIP Techs. v. Amazon.com, 788 F.3d 1359, 1364, 115 USPQ2d 1090, 1093-94.)
As noted, identified additional elements in the Claim 1 are recited in generality and represent insignificant field of use limitations that is not meaningful to indicate a practical application. And, as above, other identified additional elements area considered as necessary data gathering required to perform the abstract idea (i.e. “collects real-time data”) is considered extra solution activity in light of Mayo, 566 U.S. at 79, 101 USPQ2d at 1968; OIP Techs., Inc. v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1092-93 (Fed. Cir. 2015). (MPEP section 2106.05(g))
Thus, Claim 1 is directed to a judicial exception and is held to be patent ineligible.
In consideration of independent Claim 9, following the same steps described above with similar reasoning and rationale reveals:
STEP 1: Claim 9 is in a eligible statutory category: method (process).
STEP 2A-PRONG ONE: Claim 9 recites a judicial exception, reciting limitations parallel to those discussed above in Claim 1, falling into the grouping of subject matter that covers performing mathematics. (MPEP 2106.04(a)(2), I.A,C, III.B,C) Specifically, Claim 9 recites “failure prediction method”, “generating training data by preprocessing operation data”, “training a first model, a second model, and a third model by using the training data”, “detecting”, “first model, which has performed supervised learning”, “second model, which has performed unsupervised learning”, “third model, which has performed unsupervised learning”. As above, these limitations are fall into the grouping of subject matter that covers performing mathematics (MPEP 2106.04(a)(2), I.A,C, III.B,C) Similarly to Claim 1, Claim 9 recites the element(s) of using generic artificial intelligence (AI)/machine learning (ML) technology to perform data evaluations or calculations, namely predicting a fault or determination of an abnormality, and do not recite details regarding how the AI/ML algorithm or model functions or is trained. Instead, the claims are found to utilize the AI/ML algorithm as a tool that provides nothing more than mere instructions to implement the abstract idea on a general purpose computer. (MPEP 2106.05(f)).
STEP 2A-PRONG TWO: Claim 9 recites additional elements similar to those recited in Claim 1. Additional elements, including, “collecting real-time data”, “collected by recording the real-time data over time”, “inputting the real-time data”. As above, such limitations, using reasoning and rationale as presented above in Claim 1 discussion, recite necessary data gathering required to provide data for carrying out the judicial exception. Further, identified additional elements in Claim 9 do not integrate the judicial exception into a practical application because there is no improvement to another technology or technical field; improvements to the functioning of the computer itself; a particular machine; or effecting a transformation or reduction of a particular article to a different state or thing.
STEP 2B: Claim 9 does not recite additional elements that amount to significantly more than the abstract idea. Examiner notes Claim 9 includes, in addition to additional elements similar to those found in Claim 1, an additional element of “providing a notification that abnormality is detected”. Such limitation does not integrate the judicial exception into a practical application. Using guidance found in MPEP section 2106.05(g), displaying analysis/results is considered extra solution activity in light of Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1354-55, 119 USPQ2d 1739, 1742 (Fed. Cir. 2016).
Thus, Claim 9 is directed to a judicial exception and is held to be patent ineligible.
Further eligibility consideration includes evaluation of Claims 2-8, with direct or indirect dependency to Claim 1 and Claims 10-14 with dependency to Claim 9. Claims 2-8 and 10-14 recite limitations with further limit performing the mathematical process judicial exception, or additional elements which do not integrate the judicial exception into a practical idea. Limitations found in dependent claims that further limit performing judicial exception include, at least, “model utilizes linear discriminant analysis to classify”, “data are classified as the abnormal class according to linear discriminant criteria generated by the first model”, “model is generated through obtaining an average of a plurality of normal section errors”, “principal component analysis is applied to the original operation data”, “based on the feature vectors, the second model reconstructs reconstruction data”, “obtaining a normal section error”, “obtain the plurality of normal section errors”, “calculates a Mahalanobis distance”, among other limitations which are directed to limiting mathematical processes or calculations to determine quantitative or qualitative results. Further, additional elements recited in dependent claims are not considered to be significantly more than the abstract idea. Such additional elements include, at least, “history data comprise two types of operation data”, “operation data obtained while the pump operates”, “initial operation data are operation data obtained for a predetermined period from time”, among other limitations which are considered as necessary data gathering or insignificant field of use limitations. Examiner notes Claim 6 recites further limitation “reporting module providing a notification that an abnormality”. As discussed above, such language does not integrate the judicial exception into a practical application. (MPEP section 2106.05(g), displaying analysis/results is considered extra solution activity.)
Dependent Claims 2-8 and 10-14 recite additional elements which either further limit the abstract idea without integrating the abstract concept into a practical application or that cannot be considered significantly more than the abstract idea.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. § 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1, 2, 6, and 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over GUPTA (US 20250067164 A1) in view of FERNANDES (Fernandes, et al., “Machine learning techniques applied to mechanical fault diagnosis and fault prognosis in the context of real industrial manufacturing use-cases: a systematic literature review”, Applied Intelligence (2022) 52:14246–14280)
With respect to Claims 1 and 9, GUPTA teaches:
A pump failure prediction apparatus, comprising: (GUPTA is in same technical field, Abstract: “data from pump equipment at the wellsite…using the computational device (i.e. “prediction apparatus”), processing the time series data as input to a trained machine learning model to detect a performance issue (i.e., “failure prediction”) of the pump equipment”; and FIG. 5 with [0083]: “detection techniques 506 (e.g., recognition, detection, prediction, etc.)”)
at least one processor; a storage, which is communicably connected with the processor and stores a program code which operates in the processor; (As above, Abstract; and [0003]: “using the computational device, processing the time series data as input to a trained machine learning model…wellsite system can include a processor; memory accessible to the processor; and processor-executable instructions (i.e., “program code”) stored in the memory to instruct the system”)
and a communicator, which is communicably connected with the processor, (FIG. 22 with [0245]: “computing system 2200 and an example of a networked system… one or more input and/or output devices 2206 and a bus 2208…communication bus (i.e., “communicator”) …user may view output from and interact with a process via an I/O device”
wherein the program code comprises: a data collection module which collects real-time data related to a state of a pump which is installed in an oil well and operates; (Abstract: “receiving by a computational device at a wellsite (i.e., “oil well”), real-time, time series data from pump equipment at the wellsite”; [0039]: “real-time data exchanges (e.g., from one or more types of data acquisition equipment in the field that can acquire data”; )
an abnormality detection module ([0003]: “processing the time series data as input to a trained machine learning model to detect a performance issue (i.e., “abnormality”) of the pump equipment”; and as above, [0083]: “detection techniques 506 (e.g., recognition, detection, prediction, etc.)”; and FIG. 20 “CRM 2021” with [0219]: “the blocks (i.e., “module”)”)
generating training data by preprocessing operation data collected by recording the real-time data over time; (As above, Abstract: “receiving…real-time, time series data from pump equipment at the wellsite…processing the time series data as input to a trained machine learning model”[0110]: “data are required, which can include actual data and/or synthetic data…training a ML model, data may be split into one or more groups, which can include training data and testing data…portion of a dataset can be utilized for training”; and [0184] As to a modeling workflow, a dataset can be preprocessed to remove certain types of inconsistencies in data”; and Claim 16)
providing a notification that abnormality is detected in the pump, based on an output of the detecting of the abnormalities. ([0109]: “”system can output one or more alarms, (i.e., “providing a notification”) which may be directed to humans and/or machines such that one or more control decisions can be taken”; and FIG. 7 with [0131]: “pump suite 620 can be a suite of specialized components for real-time ESP alarms and/or control”; Examiner notes this limitation is specific to Claim 9.)
GUPTA does not explicitly teach:
module which detects three abnormalities, the three abnormalities comprising
training a first model, a second model, and a third model by using the training data;
the three abnormalities comprising: a first abnormality appearing before failure occurrence by inputting the real-time data into a first model, which has performed supervised learning of history data
a second abnormality outside of a normal operation range by inputting the real-time data to a second model, which has performed unsupervised learning of normal operation data, which has performed unsupervised learning of normal operation data,
a third abnormality outside of an initial normal operation range by inputting the real-time data to a third model, which has performed unsupervised learning of initial operation data.
FERNANDES teaches:
module which detects three abnormalities, (FERNANDES is in related technical field, Pg. 14246, “Keywords: Machine learning, Fault detection, Fault prognosis, Predictive maintenance, Manufacturing industry, Industrial case-study”; and Pg 14271, Col1, ¶-1: “methodology was applied in two industrial use-cases: a pump equipment and a robot arm…data was collected from three different pumps and a small amount of failure data was obtained for purposes of model validation”; Pg 14269, Col2, ¶-2: “Three different models were built…evaluated using a dataset containing both normal and fault data to assess their ability to detect data points that diverged from ‘normality’… unsupervised models were used to cluster the fault data…able to create a dataset that contained normal data, as well as instances of three different types of failures (i.e., “three abnormalities”)”
the three abnormalities comprising: a first abnormality appearing before failure occurrence by inputting the real-time data into a first model, ( Pg 14271, Col1, ¶-1: “results demonstrated the value of the health index rose before the occurrence of a failure (i.e., “before failure occurrence”); and Pg. 14273, Col2, ¶-3: “model consisted in a discrete Bayes filter that incorporated expert knowledge, configuration parameters, and real time sensor data”; Examiner notes interpretation of limitations as discussed above regarding rejection of Claim 1 under 35 U.S.C. 112(b))
which has performed supervised learning of history data (Pg 14256, TABLE 11, “Machine learning algorithms and methods employed in the selected primary studies”; and PG. 14260, Col1, SS3.5.3 “Hybrid Models… supervised learning algorithms are more appropriate for classification tasks”; Examiner notes reference is consistent with first model description; and Pg14269, Col2, ¶2: “predictive maintenance approach that combined semi-supervised, unsupervised and supervised learning techniques to detect and classify mechanical faults”; and Pg14273 Col1 PP3: “approach was tested using
seven years of historical data, successfully demonstrating its ability to detect fan malfunctions.”; and Pg14247, Col1, ¶3-4, including “data-driven methods employ historical data”)
a second abnormality outside of a normal operation range (Pg14266, Col1, ¶1: “Kullback-Leibler divergence to construct a health indicator (HI) of multi-sensor systems to represent a system’s deviation (i.e., “abnormality”) from its normal state (i.e., “outside of a normal operation range)”; Examiner interprets “second” to mean a distinct abnormality detection, rather than an ordered number consistent with specification.)
by inputting the real-time data to a second model, which has performed unsupervised learning of normal operation data, (Pg 14269, Col2, ¶3: “Three different models were built using normal data (i.e., “real-time” data)exclusively but were evaluated using a dataset containing both normal and fault data to assess their ability to detect data points that diverged from ‘normality’”; and Table 11: “Machine learning algorithms and methods employed in the selected primary studies”, teaching multiple options for unsupervised learning models)
a third abnormality outside of an initial normal operation range by inputting the real-time data to a third model, (Pg 14247,Col1P4: “model-based methods estimate the parameters of interest based on a mathematical model of the system under normal operating conditions”; and ”As above, Pg 14269, Col2, ¶-2: “Three different models … three different types of failures”; Pg14263, Col1, SS3.6, “methods for detecting faults directly from real-time data” and “processed at each step, either as a time interval or as the number of datapoints,”; Applying BRI with guidance from specification, Examiner interprets limitation to mean a third abnormality is designated by an indication of an operational variable with a value outside of an initial normal operating range based on analysis using third model evaluation of real-time data; Further, Examiner interprets “initial normal operation data” to mean “operation data obtained for a predetermined period from time at which the pump is assumed to operate normally in a stable condition” (written description, [0007]), analogous to reference use of normal operation data for a specific time period. Examiner notes FERNANDES points to at least two references [55]-DAS and [69]-NASKOS, included below as important but not directly cited.)
which has performed unsupervised learning of initial operation data. (As above, Pg 14256, TABLE 11, “Machine learning algorithms and methods employed in the selected primary studies”; and PG 14275 Col1P2, “quality that is often necessary is the ability to learn from unlabeled data, i.e., unsupervised learning”: Interpretation of “initial operation data” as above.)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to modify the invention disclosed by GUPTA to include an abnormality detection module which detects three abnormalities, training a first model, a second model, and a third model by using the training data; where the three abnormalities comprise: a first abnormality appearing before failure occurrence by inputting the real-time data into a first model, which has performed supervised learning of history data; a second abnormality outside of a normal operation range by inputting the real-time data to a second model, which has performed unsupervised learning of normal operation data, which has performed unsupervised learning of normal operation data; and a third abnormality outside of an initial normal operation range by inputting the real-time data to a third model, which has performed unsupervised learning of initial operation data, as taught by FERNANDES because the robust details in the FERNANDES reference provide improved and validated methods for implementation of a wide variety of machine learning strategies that would broaden and improve the invention of GUPTA. GUPTA discloses an invention directly related to pump failure diagnosis, while FERNANDES presents a wide variety of technical applications, with details of multiple machine-learning-based approaches for solving the problem of predictive management and failure detection of equipment and/or machines. It would be an obvious and logical combination to include the methods as describe in the FERNANDES reference, a comprehensive literature review, including multiple references therein, to improve the invention of GUPTA with validated machine-learning methods that would allow for more accurate and reliable determination of abnormality/failure either predictively or as identified in real-time by model analysis using real-time and/or historical data. One of ordinary skill would have logical reason and strong motivation to consult FERNANDES as a robust resource to improve the invention disclosed by GUPTA to arrive at key features as recited in the claimed invention.
With respect to Claims 2 and 10, GUPTA in view of FERNANDES, teaches limitations of claims 1 and 9.
GUPTA further teaches:
history data comprise operation data([0125]: “system may be utilized for real-time monitoring and/or control, it may also be used for purposes of assessing historic data…ML model that is used to monitor and/or control the one or more of the pumps may be fed data stored (i.e., “operation data”) as to one or more of the other pumps”)
the first model utilizes linear discriminant analysis to classify the operation data comprised in the history data ([0210]: “system, a method, etc., may utilize one or more machine learning features, which can be implemented using one or more machine learning models…flexible discriminant analysis, linear discriminant analysis, etc.)”)
operation data labeled ([0110]: “data are required, which can include actual data (i.e., “operation data”)…data may be labeled…data may be split into one or more groups…can include training data and testing data”;)
the abnormality detection module determines that the first abnormality occurs according to linear discriminant criteria generated by the first model. (As above, [0210]; and FIG. 5, “detection”; and [0116]: “resulting trained ML model performs suitable for detection of one or more behaviors that can be associated with pump operational issues” )
GUPTA, as modified by FERNANDES and taught above, does not teach:
the history data comprise two types of operation data, the operation data labeled as normal and operation data labeled as abnormal
the first model classifies the operation data comprised in the history data as a normal class or an abnormal class
the abnormality detection module determines that the first abnormality occurs when the real-time data are classified as the abnormal class according criteria generated by the first model.
FERNANDES further teaches:
the history data comprise two types of operation data, the operation data labeled as normal and operation data labeled as abnormal (FERNANDES Pg14247, Col1,P4: “data-driven methods employ historical data and artificial intelligence algorithms”; and Pg14267, Col2, P1: “model was trained using historical data,”; using three months of historical data obtained from a rolling mill machine (i.e. “operation data”)”; and Pg14268, Col2, P2: “condition monitoring data obtained from the machine was segmented into time-series sequences according to three possible equipment health
states, namely “good” (i.e., “normal”),“bad” (i.e., “abnormal”) and “intermediate” operating conditions.” )
the first model classify the operation data comprised in the history data as a normal class or an abnormal class (As above, Pg14267, Col2, P1; and, as above (Claim 1), PG. 14260, Col1, SS3.5.3 “Hybrid Models… supervised learning algorithms are more appropriate for classification tasks”; Examiner notes reference is consistent with first model description, as above; and Pg14269, Col2, ¶2: “predictive maintenance approach that combined semi-supervised, unsupervised and supervised learning techniques to detect and classify mechanical faults”; and Pg14273 Col1 PP3: “approach was tested using seven years of historical data, successfully demonstrating its ability to detect fan malfunctions.”; and Pg14247, Col1, ¶3-4, including “data-driven methods employ historical data”); and Pg14266 Col2Para1: “Data was collected from the machine’s central control system, as well as from external sensors…used to infer when failure events occurred and label the data accordingly”)
the abnormality detection module determines that the first abnormality occurs when the real-time data are classified as the abnormal class according to linear discriminant criteria generated by the first model.(As above, Pg14268, Col2, P2: “condition monitoring data obtained from the machine was segmented into time-series sequences according to three possible equipment health states, namely “good” (i.e., “normal”),“bad” (i.e., “abnormal”) and “intermediate” operating conditions.”; and Table 11, teaching multiple options for supervised learning models, where one of ordinary skill would understand supervised learning models require labeled training data, Pg14260, Col1, SS3.5.3: “demonstrated that supervised learning algorithms are more appropriate for classification tasks but require labeled data”, pointing to reference [45], included below as important and germane to instant application. )
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to modify the invention disclosed by GUPTA to include history data comprised of two types of operation data, the operation data labeled as normal and operation data labeled as abnormal; a first model that classifies the operation data comprised in the history data as a normal class or an abnormal class; and the abnormality detection module determines that the first abnormality occurs when the real-time data are classified as the abnormal class. according criteria generated by the first model, as taught by FERNANDES because the model depends on supervised learning and this would be seen as an effective and efficient improvement to the machine learning detection method/system taught by GUPTA. By using the teaching of FERNANDES to include explicitly labeled “normal” and “abnormal” data for model training, false positive results are avoided through the definition of clear boundaries.
With respect to Claim 6, GUPTA, in view of FERNANDES, teaches limitations of claim 1.
GUPTA further teaches:
wherein the program code further comprises a reporting module providing a notification that an abnormality is detected in the pump based on an output of the abnormality detection module. (As above, Abstract: “trained machine learning model (i.e. “program code”) to detect a performance issue of the pump equipment; and issuing a signal responsive to detection of the performance issue”; FIG 1 with [0040]: “outputs (i.e., “notification”) from the workspace framework 110 (i.e. “reporting module”) can be utilized for directing, controlling, etc.,”; [0108]: “trained machine learning (ML) model can output a result…output a likelihood of occurrence of a particular issue or issues (i.e., “abnormality”)” [0109]: “system can provide for output generation, which may be directed to a controller, controllers, a dashboard, a network interface, etc…output one or more alarms”).
Claims 3, 7-8, 11, and 13-14 are rejected under 35 U.S.C. 103 as being unpatentable over GUPTA in view of FERNANDES, as applied to Claim 1 above, and further in view of GIVNAN (Givnan, et al., “Anomaly Detection Using Autoencoder Reconstruction upon Industrial Motors”, Sensors 2022, 22, 3166) and ABBASZADEH (US 20200244677 A1).
With respect to Claims 3 and 11, GUPTA, in view of FERNANDES, teaches limitations of claims 1 and 9.
GUPTA further teaches:
wherein the normal operation data are operation data obtained while the pump operates in a normal state, (As above, Abstract: “real-time, time series data from pump equipment”; and [0040]: “received via one or more interfaces in one or more forms (e.g., acquired data as to operational conditions”; and [0110]: “data can be labeled and referred to as labeled data”; and FIG. 7 with [0140]: “under substantially normal operation”; Examiner interprets “normal state” as discussed above.)
model is generated after application of a principal component analysis, (As above, and [0191]: “dimensionality reduction technique (e.g., principal component analysis (PCA)”; and [0228]: “decisions trees are built using principal component analysis.”)
FERNANDES teaches as above, use of three models (Col2, ¶-2: “Three different models were built”; Examiner interprets numbering on models as discussed above to indicate distinction between models rather than order.)
GUPTA as modified by FERNANDES and taught above, does not teach:
model is generated through obtaining an average of a plurality of normal section errors between original operation data (the normal operation data) and reconstructed operation data (an reconstruction data) calculated;
based on the feature vectors, the second model reconstructs reconstruction data, and a process of obtaining a normal section error between the original operation data and the reconstructed operation data;
is repeated at every time step of the operation data in order to obtain the plurality of normal section errors,
based on the real-time feature vectors, real-time reconstruction data are obtained;
a real-time error between the real-time data and the real-time reconstruction data is obtained;
the error level is obtained by dividing the real-time error by an average of the plurality of normal section errors generated by the second model;
and the error level is recorded over time.
GIVNAN teaches:
model is generated through obtaining an average of a plurality of normal section errors between original operation data (the normal operation data) and reconstructed operation data (an reconstruction data) calculated; (GIVNAN is in same technical field, Pg2/21, Para3: “detection of anomalies in real-time”; Page4/21Para3: “evaluation metrics used in the experiment to calculate the reconstruction error for anomaly detection”; and Pg4/21 Section 3.1: “data provided include a large proportion of known healthy data (i.e. “normal operation data”); and Pg 6/21 Section 3.3 “Machine Learning and Modelling: signal is fed into an anomaly detection model and abnormal behaviours are identified by calculating the reconstruction error
based on the feature vectors, the model reconstructs reconstruction data, and a process of obtaining a normal section error between the original operation data and the reconstructed operation data (Pg 6/21 Section 3.3 “Machine Learning and Modelling: signal is fed into an anomaly detection model and abnormal behaviours are identified by calculating the reconstruction error.”; and Pg 7/21 Para3: “backpropagation is used to minimise the overall cost function throughout the network, here this is the overall reconstruction error using mean squared error”; and FIG. 5 element B, with Pg8/21: “device is embedded with a pre-trained stacked autoencoder with a set of rules to investigate anomalies based on the reconstruction of sensor data” Pg9/21Para2: “final set of features (70) obtained from the SAE is used to reconstruct the output”; and Pg 9/21 Para4: “model will be unable to recognise unusual data points, as such the reconstruction error will be higher. This is used to detect anomalies in machine operation and is defined” follow by specific comparative equation used.”)
is repeated at every time step of the operation data in order to obtain the plurality of normal section errors, (Abstract: “approach is evaluated on windowed real machine sensor data to observe normal and abnormal behaviour.”; and Pg4/21, Para4: “dataset contains, time-series (i.e. “time step”) sensor data”; and see Figure 2, Pg5/21)
based on the real-time feature vectors, real-time reconstruction data are obtained; (Pg3/21 Para 2, 3: “learnt features are used to reconstruct the input vector” and “the model takes an initial input, such as a data window”; and Pg10/s1 Para 2: “Anomalies are identified by the feature vectors of the input (i.e., “real-time” feature vectors) and output of the final model”; and Pg11/21, Para4: “Data reconstruction for each window is used”)
a real-time error between the real-time data and the real-time reconstruction data is obtained; (Pg6/21 Para5: “signal (i.e., “real-time data”) is fed into an anomaly detection model and abnormal behaviours are identified by calculating the reconstruction error (i.e., “error between real-time data and real-time reconstruction data”))
the error level is obtained by dividing the real-time error by an average of the plurality of normal section errors generated by the second model; (Pg 10/21, Para5: “of squared differences between the given input values, and the predicted output values, and minimises the error across the whole training dataset.”; Examiner notes that GIVNAN teaches an averaging, the error analysis using fractional uncertainty as defined here would be well known to one of ordinary skill. For example, see Taylor, John. An Introduction to Error Analysis, 2nd. ed. University Science Books: Sausalito, 1997.)
the error level is recorded over time. (Pg2/21, Para5: “Modelling healthy data and detecting changes over time is performed at the edge while only anomalies are returned thus saving bandwidth”)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to further modify the invention disclosed by GUPTA, as modified by FERNANDES to include the method of generation of a model through obtaining an average of a plurality of normal section errors between original operation data (the normal operation data) and reconstructed operation data; and where, based on the feature vectors, the model reconstructs reconstruction data, and a process of obtaining a normal section error between the original operation data and the reconstructed operation data; and where the comparative analysis is repeated at every time step of the operation data in order to obtain the plurality of normal section errors, and based on the real-time feature vectors, real-time reconstruction data are obtained; with real-time error between the real-time data and the real-time reconstruction data is obtained; and the error level is recorded over time, as taught by GIVNAN because these detailed error analysis and reconstruction techniques result, as reported by GIVNAN, in improved efficiency and reliability in detection of anomaly/abnormality from real-time data. One of ordinary skill would be motivated to consult the GIVNAN reference in solving the problem of developing a new and improved method/system for detection of faults/anomaly/abnormality based on the similarity of the problem to be solved and the success demonstrated by the techniques detailed by GIVNAN. The combination would provide an obvious and advantageous improvement to the invention of GUPTA as modified by FERNANDES particularly since the method would not require additional or different data acquisition and little to no additional computational effort and allow for real-time automated monitoring of system status with regard to performance issues. One of ordinary skill would understand that these techniques allow for an efficient way to use reconstruction error to evaluate and quickly identify abnormal data, as taught by GIVNAN (Pg11).
GUPTA, as modified by FERNANDES and GIVNAN and taught above, does not teach:
principal component analysis is applied to the original operation data to obtain feature vectors
the abnormality detection module determines that the second abnormality occur when an error level of real-time data increases over time when real-time feature vectors are obtained by applying the principal component analysis to the real-time data;
ABBASZADEH teaches:
the principal component analysis is applied to the original operation data to obtain feature vectors (ABBASZADEH is in same technical field, [0003] “time series data as input to a trained machine learning model to detect a performance issue of the pump equipment; and issue a signal responsive to detection of the performance issue”; FIG.. 12 with [0054]: “At S1220, the system may compute features and form feature vectors… system might use weights from a principal component analysis as features.”; [0134]: “one or more ML modeling approaches can be directly applied to flag each timestamp” (i.e. “applied to original operation data”))
the abnormality detection module determines that the second abnormality occur when an error level of real-time data increases over time when real-time feature vectors are obtained by applying the principal component analysis to the real-time data; (FIG.. 12 with [0054]: “At S1220, the system may compute features and form feature vectors… system might use weights from a principal component analysis as features”; Abstract: “time series data as input to a trained machine learning model to detect a performance issue”)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to further modify the invention disclosed by GUPTA, as modified by FERNANDES and GIVNAN and taught above, to include application of principal component analysis to the original operation data to obtain feature vectors, and an abnormality detection module determines that the second abnormality occur when an error level of real-time data increases over time when real-time feature vectors are obtained by applying the principal component analysis (PCA) to the real-time data, as taught by ABBASZDEH the method of PCA is a trusted standard tool for implementation of data reduction that allows for more efficient learning and predictive modeling in machine-learning environments. As detailed by ABBASZDEH, the technique reduces data to its most essential elements, and would be seen by one of ordinary skill as an advantageous combination with the invention of GUPTA as modified by FERNANDES in efficiency with an expectation of improving overall accuracy in detection of abnormal condition.
With respect to Claims 7 and 14, GUPTA, in view of FERNANDES and further in view of GAVNIN and ABBASZDEH, teaches limitations of claims 3 and 9.
GUPTA further teaches:
the second model is generated and trained even before a pump normally operates by using normal operation data of another pump, which operates in a similar environment, and is regenerated by updating the normal operation data with its own operation data in a normal state. ([0125]: “consider a field site with multiple pumps for multiple wells where a trial implementation is commenced for one of the pumps for one of the wells…ML model that is used to monitor and/or control the one or more of the pumps may be fed data stored as to one or more of the other pumps.”;
With respect to Claim 8, GUPTA, in view of FERNANDES, teaches limitations of claim 1.
GUPTA further teaches:
wherein the program code further comprises: a training data generation module which generates training data for training by using preprocessed operation data in such a manner that the real-time data are recorded over time to generate operation data, and the operation data are preprocessed (As above, Claim 1 Abstract: “receiving…real-time, time series data from pump equipment at the wellsite…processing the time series data as input to a trained machine learning model”; [0110]: “data are required, which can include actual data and/or synthetic data…training a ML model, data may be split into one or more groups, which can include training data and testing data…portion of a dataset can be utilized for training”; and [0184] As to a modeling workflow, a dataset can be preprocessed to remove certain types of inconsistencies in data”; and Claim 16)
the training data generation module generates the history data to generate training data of the first model, in such a manner that time series data are standardized in the operation data, ([0110]: “include actual data and/or synthetic data…portion of a dataset can be utilized for training to generate a trained ML model that can then be tested using another portion of the dataset”; [0135]: “clustering based upon historical labeled events”; and [0144]: “various workflows, which may be performed continuously…on real-time data and/or in a historical assessment manner on historical data”; and [0188]: “workflow can include feature engineering, which may be performed on signals in a dataset…features can be engineered based on rolling window computations, for example, using different window sizes, where functions utilized can range from relatively simple descriptive statistics like the mean and min-max difference to more complex Theil-Sen slopes, median absolute deviation, etc.”; Examiner asserts this would be understood by one of ordinary skill as teaching options for standardization of data which would produce uniform comparable format among data sets.)
wherein operation data corresponds to a failure state recorded previously are labeled as abnormal, and data of a normal operation section are labeled as normal; ([0123]: “labeled data may be utilized where data can be labeled by domain experts…one or more types of pump issues”; and FIG. 16 with [0180]: “types of data along with labeling…binary labels of 1 and 0 (i.e., “normal” / “abnormal”)”)
the training data generation module further generates the normal operation data to generate training data of the second model, (FIG. 18 with [0200]: “data can be utilized as input and/or can be processed to generate input for a ML model”; Examiner interprets limitation as above to mean operation data is used to generate additional training data, analogous to reference.)
in such a manner that a principal component analysis is applied to the operation data at every time step thereof to extract feature vectors; ([0191]: “dimensionality reduction technique (e.g., principal component analysis (PCA)”; and [0228]: “decisions trees are built using principal component analysis.”)
FERNANDEZ teaches as above, as discussed in claim 1 above:
the first model, the second model, and the third model (Pg 14269, Col2, ¶-2
“Three different models were built”)
FERNANDES further teaches:
the training data generation module further generates operation data, which are collected for a predetermined period from time (As above, claim 1, Pg 14247,Col1P4: “model-based methods estimate the parameters of interest based on a mathematical model of the system under normal operating conditions”; and Pg14263, Col1, SS3.6, “methods for detecting faults directly from real-time data” and “processed at each step, either as a time interval or as the number of datapoints”;)
at which the pump operates in a normal state after the pump is installed, as the initial operation data for generating training data of the third model.( Pg 14247,Col1P4: “model-based methods estimate the parameters of interest based on a mathematical model of the system under normal operating conditions”; and ”As above, Pg 14269, Col2, ¶-2: “Three different models … three different types of failures”; Pg14263, Col1, SS3.6, “methods for detecting faults directly from real-time data” and “processed at each step, either as a time interval or as the number of datapoints,”; where “initial operation data” is interpreted as above, applying BRI with guidance from specification, and where numbering indicates a distinct model rather than order.)
to generate training data of the third model, collecting operation data for a predetermined period from time at which the pump operates in a normal state after the pump is installed are generated as the initial operation data (As above, Pg 14247,Col1P4: “model-based methods estimate the parameters of interest based on a mathematical model of the system under normal operating conditions”; and ”As above, Pg 14269, Col2, ¶-2: “Three different models … three different types of failures”; Pg14263, Col1, SS3.6, “methods for detecting faults directly from real-time data” and “processed at each step, either as a time interval or as the number of datapoints,”; Interpretation as discussed above.)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to further modify the invention disclosed by GUPTA, in view of FERNANDES and further in view of GAVNIN and ABBASZDEH, to include a training data generation module that further generates operation data, which are collected for a predetermined period from time; at which the pump operates in a normal state after the pump is installed, as the initial operation data for generating training data of the third model; to generate training data of the third model, collecting operation data for a predetermined period from time at which the pump operates in a normal state after the pump is installed are generated as the initial operation data, as further taught by FERNANDES because these steps would be a logical way to improve the method/system of GUPTA as modified and taught above by refining available data in such a way to optimize model performance for speed and accuracy. One of ordinary skill would understand the value of improving the invention of GUPTA, as modified, to specifically use data collected for a predetermined time in order to provide sufficient data without accumulating unnecessary information which would demand longer processing time for model building.
GUPTA as modified by FERNANDES and taught above, does not teach:
reconstruction data are obtained from reconstructing the feature vectors;
wherein an error between the reconstruction data and the operation data is calculated;
GIVNAN teaches:
wherein reconstruction data are obtained from reconstructing the feature vectors; (Pg9/21, Para4: “reconstruction of the data using the 70 features”(i.e., “feature vectors”))
wherein an error between the reconstruction data and the operation data is calculated; (Pg. 6 Section 3.3: “signal is fed into an anomaly detection model and abnormal behaviours are identified by calculating the reconstruction error.”; Pg7/21 Para4: “Backpropagation is used to minimise the overall cost function…here this is the overall reconstruction error using mean squared error ...achieved by optimising the network and calculating the gradient of the cost function with respect to the weights and biases, then adjusting the weights and biases at each layer to improve the overall reconstruction error.”; Pg8/21 section 3.4: “Training continues until an optimal reconstruction error is achieved”)
wherein when the error is smaller than or equal to a reference error, the operation data are comprised in the normal operation data; and when the error is greater than the reference error, the operation data are excluded from the normal operation data; and (Pg11/21 Section 4.: “allows a threshold to be determined based on the reconstruction error calculated. This will be the basis of the anomaly detection”)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to further modify the invention disclosed by GUPTA, as modified by FERNANDES and taught above, to combine the techniques of using reconstruction data are obtained from reconstructing the feature vectors wherein an error between the reconstruction data and the operation data is calculated, as taught by GIVNAN because the technique, which would be known typically to involve an autoencoder, would be a way to confirm differences between an original feature vector and a reconstructed feature vector, and would represent an advance method to improve the invention of GUPTA as modified above. This method allows for reduction in poor training that may be caused by reconstruction of anomalous training data, providing a quantitative measure of the discrepancy and ultimately resulting in a more reliable and accurate predictive model.
With respect to Claim 13, GUPTA, in view of FERNANDES, teaches limitations of claim 9.
GUPTA further teaches:
in the step of generating the training data, to generate training data of the first model, generating history data in such a manner that time series data are standardized in the operation data, ([0125]: “system may be utilized for real-time (i.e., “time series”) monitoring and/or control, it may also be used for purposes of assessing historic data…ML model that is used to monitor and/or control the one or more of the pumps may be fed data stored (i.e., “operation data”) as to one or more of the other pumps”)
data labeling ([0110]: “data are required, which can include actual data (i.e., “operation data”)…data may be labeled…data may be split into one or more groups …can include training data and testing data”)
FERNANDES further teaches:
operation data corresponds to a failure state recorded previously are labeled as abnormal, and data of a normal operation section are labeled as normal, (As above, Pg14247, Col1,P4: “data-driven methods employ historical data and artificial intelligence algorithms”; and Pg14267, Col2, P1: “model was trained using historical data,”; using three months of historical data obtained from a rolling mill machine (i.e. “operation data”)”; and Pg14268, Col2, P2: “condition monitoring data obtained from the machine was segmented into time-series sequences according to three possible equipment health states, namely “good” (i.e., “normal”),“bad” (i.e., “abnormal”) and “intermediate” operating conditions.” )
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to further modify the invention disclosed by GUPTA, as modified by FERNANDES and taught above, to include operation data corresponds to a failure state recorded previously are labeled as abnormal, and data of a normal operation section are labeled as normal because this is the standard way to generate a robust set of training data based on known features as input for training a predictive model. One of ordinary skill would see the advantage of combining this with the machine-learning based detection method/system of GUPTA, as modified above, to improve the accuracy of the trained model.
GUPTA, as modified by FERNANDES and taught above, does not teach:
to generate training data of the second model, generating the normal operation data in such a manner that a principal component analysis is applied to the operation data every specific time thereof to extract feature vectors;
reconstruction data are obtained from the feature vectors;
an error between the reconstruction data and the operation data is calculated;
when the error is smaller than or equal to a reference error, the operation data are comprised in the normal operation data; and when the error is greater than the reference error, the operation data are excluded from the normal operation data,
GIVNAN teaches:
reconstruction data are obtained from the feature vectors; (As above, Claim 8, Pg2/21, Para3: “paper primarily focuses on the detection of anomalies in real-time using an edge device, which reduces the overall transmission of data sent for analysis”; Pg9/21, Para4: “reconstruction of the data using the 70 features”)
an error between the reconstruction data and the operation data is calculated; (Pg2/21, Para3: “detection of anomalies in real-time”; Page4/21Para3: “evaluation metrics used in the experiment to calculate the reconstruction error for anomaly detection”; and Pg4/21 Section 3.1: “data provided include a large proportion of known healthy data (i.e. “normal operation data”); and Pg 6/21 Section 3.3 “Machine Learning and Modelling: signal is fed into an anomaly detection model and abnormal behaviours are identified by calculating the reconstruction error”)
when the error is smaller than or equal to a reference error, the operation data are comprised in the normal operation data; and when the error is greater than the reference error, the operation data are excluded from the normal operation data; (Abstract: “approach is evaluated on windowed real machine sensor data to observe normal and abnormal behaviour.”; and Pg11/21 Section 4.: “allows a threshold to be determined based on the reconstruction error calculated. This will be the basis of the anomaly detection”; Examiner asserts reference teaches comparative analysis based on thresholding which would be understood as comparative lesser or greater evaluation relative to a predetermined value deemed as decision boundary.)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to further modify the invention disclosed by GUPTA, as modified by FERNANDES and taught above, to include, the method steps where reconstruction data are obtained from the feature vectors; an error between the reconstruction data and the operation data is calculated; when the error is smaller than or equal to a reference error, the operation data are comprised in the normal operation data; and when the error is greater than the reference error, the operation data are excluded from the normal operation data, as taught by GIVNAN because the detailed technique of GIVNAN would improve the invention of GUPTA modified by FERNANDES by using an additional validation measure that would reduce computing time while resulting in a more accurate and reliable detection model. One of ordinary skill would realize the benefit of including GIVNAN reference to establish a better baseline for what would be considered “normal” performance/behavior and thus allow an improved determination of anomaly/abnormality.
GUPTA, as modified by FERNANDES and GIVNAN and taught above, does not teach:
to generate training data of the second model, generating the normal operation data in such a manner that a principal component analysis is applied to the operation data every specific time thereof to extract feature vectors;
ABBASZADEH teaches:
to generate training data of the second model, generating the normal operation data in such a manner that a principal component analysis is applied to the operation data every specific time thereof to extract feature vectors; ([0110]: “In unsupervised learning, data may be labeled and/or unlabeled…data may be split into one or more groups, which can include training data and testing data…a portion of a dataset can be utilized for training to generate a trained ML model that can then be tested using another portion of the dataset”; As above, [0134]: “one or more ML modeling approaches can be directly applied to flag each timestamp”; Examiner notes interpretation of “normal operation data” as above.)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to further modify the invention disclosed by GUPTA, as modified by FERNANDES and taught above, to generate training data of the second model, generating the normal operation data in such a manner that a principal component analysis is applied to the operation data every specific time thereof to extract feature vectors, as taught by ABBASZADEH, because this technique for generation of training data would be understood as an improvement to the method disclosed by GUPTA, as modified by FERNANDES. Using data known as “normal” allows a model to be better formulated to identify abnormal behaviors, which would result in an improved detection of fault/abnormality.
Claims 4 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over GUPTA in view of FERNANDES, as applied to claims 1 and 9 above, and further in view of ABBASZADEH and BHATTACHARYYA (US 20200285997 A1).
With respect to Claims 4 and 12, GUPTA in view of FERNANDES teaches limitations of claims 1 and 9.
FERNANDES further teaches:
the initial operation data are operation data obtained for a predetermined period from time at which the pump is assumed to operate normally in a stable condition without any issues, (As above, Pg 14247,Col1P4: “model-based methods estimate the parameters of interest based on a mathematical model of the system under normal operating conditions”; and Pg14263, Col1, SS3.6, “methods for detecting faults directly from real-time data” and “processed at each step, either as a time interval or as the number of datapoints”; where “initial operation data” is interpreted as above, Claim 1)
the third model is trained by applying principal component analysis to the data, (As above, Pg 14247,Col1P4: “model-based methods estimate the parameters of interest based on a mathematical model of the system under normal operating conditions”; and Pg 14269, Col2, ¶-2: “Three different models … three different types of failures”; Pg14263, Col1, SS3.6, “methods for detecting faults directly from real-time data”; with interpretation as above.; and Pg14268, Col1, ¶1 “Principal component analysis was used”)
the principal component analysis is applied to the real-time operation data (FERNANDES Pg14268, Col1, ¶1 “Principal component analysis was used to reduce the dimensionality of time series data collected from the conveyor system (i.e. “real-time data”)”)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to further modify the invention disclosed by GUPTA, as previously modified by FERNANDES and taught above, to include the initial operation data are operation data obtained for a predetermined period from time at which the pump is assumed to operate normally in a stable condition without any issues; the third model is trained by applying principal component analysis to the data; and the principal component analysis is applied to the real-time operation data, as further taught by FERNANDES because these steps are known as efficient methods associated with effective development and use of predictive modeling processes and would further improve the method/system of GUPTA to achieve reliable detection of abnormality/fault/anomaly. Incorporating the teaching of FERNANDES to explicitly include initial operation data acquired over a set time period and using well-known principal component analysis methods to identify key features and reduce a potentially large time-dependent data set to the most essential components would result in a more efficient and accurate predictive modeling process while simultaneously reducing cost by saving time.
GUPTA, as modified by FERNANDES and as taught above, does not teach:
model is trained by applying principal component analysis to the data, feature vectors which consist of at least two components are obtained feature vectors which consist of at least two components are obtained;
a center point of the feature vectors is obtained based on distribution of the vectors in multi- dimensional space, and based on the distribution of the feature vectors and the center point of the initial operation data;
third model calculates a Mahalanobis distance from the center point to each feature vectors of the data,
from the range of the Mahalanobis distance, a reference distance is statistically determined,
the third model calculates Mahalanobis distance from the center point to each real-time data and determines whether the state of a pump is abnormal if the Mahalanobis distance is greater than the reference distance.
ABBASZADEH teaches:
model is trained by applying principal component analysis to the data, feature vectors which consist of at least two components are obtained feature vectors which consist of at least two components are obtained,(ABBASZADEH is in same technical field, Abstract: “determinations may be repeated until an abnormality is localized”; FIG.. 12 with [0054]: “At S1220, the system may compute features and form feature vectors… system might use weights from a principal component analysis as features.”; [0055]: “many different types of features (i.e. “components”) may be utilized in accordance with any of the embodiments described herein, including principal components (weights constructed with natural basis sets) and statistical features (e.g., mean, variance, skewness, kurtosis, maximum, minimum values of time series signals, location of maximum and minimum values, independent components, etc.) (i.e. “at least two”); )
a center point of the feature vectors is obtained based on distribution of the vectors in multi-dimensional space, and based on the distribution of the feature vectors and the center point of the initial operation data, ([0038: “system might use the distance of the node data to the cluster centroids (e.g., based on Euclidian or Mahalanobis distance)... monitoring node may have one or more time-series associated with its normal (or both normal and abnormal) behavior acquired as historical field data or generated offline for training.”; [0051]… an appropriate set of multi-dimensional feature vectors, which may be extracted automatically (e.g., via an algorithm) and/or be manually input, might comprise a good predictor of measured data”; Examiner interprets “initial operation data” as above, analogous to referencing centroid determination to “normal” time series historical data as in reference.)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to further modify the invention disclosed by GUPTA, as modified by FERNANDES and taught above, to include training based on principal component analysis where feature vectors of at least two components are generated and determination of a parameter based on distance between centers of key feature distribution in initial operation data, as taught by ABBASZADEH, because as noted above, principal component analysis adds efficient and trusted data reduction capacity, and allowed for accurate identification of feature vectors, while computing the additional metric of distance to distribution centers further empowers the model to produce accurate results without consideration of every acquired data point through comparison with an expected value of a reference vector or reference distance. One of ordinary skill would find logical reason to incorporate the invention disclosed by ABBASZADEH with that of GUPTA as modified by FERNANDES because it would make the modeling and detection process more robust and reliable.
GUPTA, as modified by FERNANDES and further modified by ABBASZADEH and as taught above, does not teach:
third model calculates a Mahalanobis distance from the center point to each feature vectors of the data,
from the range of the Mahalanobis distance, a reference distance is statistically determined,
the third model calculates Mahalanobis distance from the center point to each real-time data and determines whether the state of a pump is abnormal if the Mahalanobis distance is greater than the reference distance.
BHATTACHARYYA teaches:
third model calculates a Mahalanobis distance from the center point to each feature vectors of the data, (BHATTACHARYYA is in same technical field, [0002]: “relates to the field of anomaly detection in machines”; and [0010] with table therein, “symmetric divergence reduce to the Mahalanobis distance”, followed by detailed description Mahalanobis distance calculation; Pg 36, “Algorithm 6: create Mahalanobis distances”: ; Examiner asserts one of ordinary skill would be familiar with this limitation as the basic definition and use of Mahalanobis distance in the technical field of machine learning, for example see Wikipedia, and would be known to one of ordinary skill, by definition, to be distance to the mean (centroid) of a distribution of feature vectors, for example see reference ALLAHDADIAN (US 20220156578 A1) included below as important but not cited prior art)
wherein from the range of the Mahalanobis distance, a reference distance is statistically determined ([0422]: “system calculates robust Mahalanobis distances (and/or Bhattacharyya distances) from the z-scores of error data from multiple engine sensors of interests and stores the calculated range”; [0482]: “system calculates the z-scores of error data from the engine sensor data time series then optionally calculates the robust Mahalanobis distance (i.e., “reference distance”)…value is compared against the range of Mahalanobis distances”; and Page 36 Algorithm 6; Examiner asserts one of ordinary skill would understand “robust” Mahalanobis distance to mean a Mahalanobis distance calculated using masking or other statistical methods to remove impact of outliers.)
the third model calculates Mahalanobis distance from the center point to each real-time data and determines whether the state of a pump is abnormal if the Mahalanobis distance is greater than the reference distance. ([0483]: “set of data (i.e., “real-time data”) as an input to a user interface (e.g., analysis gauges) in the form of standardized error values for each sensor and/or the combined Mahalanobis distance (or Bhattacharyya distance) for each sensor...allows users to understand why data were classified as failures or anomalies”; and Page 36 Algorithms 6 and 7, with [0517]: “Algorithm 6 details the calculation of the Mahalanobis distance and/or robust Mahalanobis distance (i.e. “reference distance”), can be used along with Algorithm 7 to classify anomalies …At real time, Algorithm 7 may be used to calculate and match test data with the tags created during training thus providing a means of understanding which anomaly conditions may lead to failure conditions.”)
It would have been obvious to one of ordinary skill in the art before effective filing date of the claimed invention to further modify the invention disclosed by GUPTA, as modified by FERNANDES and ABBASZADEH and taught above, to include, steps involving Mahalanobis distance calculations based on feature vector distribution and determination of a reference distance, where the reference distance is used in comparative analysis with Mahalanobis distances computed with real-time data to detect an abnormality, as taught by BHATTACHARYYA, because using the Mahalanobis distance technique is known as a way to allow for multi-dimensional data transformation into a standard scale, and allows for using reliable consideration of “standard deviation” style analysis related to proximity of a data point from a distribution center (or centroid). One of ordinary skill would see incorporating the Mahalanobis method as taught by BHATTACHARYYA as a way to make the method/system of GUPTA, as modified by FERNANDES and ABBASHADEH even more robust and reliable in prediction/detection of abnormalities by examination of real-time data. principal component analysis adds efficient and trusted data reduction capacity, and allowed for accurate identification of feature vectors, while computing the additional metric of distance to distribution centers further empowers the model to produce accurate results without consideration of every acquired data point through comparison with an expected value of a reference vector or reference distance. One of ordinary skill would find logical reason to incorporate the invention disclosed by ABBASZADEH with that of GUPTA as modified by FERNANDES because it would make the modeling and detection process more robust and reliable.
Allowable Subject Matter
Claim 5 is objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
The following is a statement of reasons indicating allowable subject matter:
Claim 5 carries dependency to Claim 4 and indirectly to Claim 1 adding limits determination of a reference distance, reciting: “ the reference distance is determined as a multiple of an average value of a plurality of Mahalanobis distances obtained from the initial operation data.” Examiner notes, as above in rejection of Claim 4, BHATTACHARYYA teaches a detailed method for using Mahalanobis technique for model development for fault prediction, including averaging of Mahalanobis distances, but does not teach use of “a multiple” of such averages.
Any comments considered necessary by applicant must be submitted no later than the payment of the issue fee and, to avoid processing delays, should preferably accompany the issue fee. Such submissions should be clearly labeled “Comments on Statement of Reasons for Allowance.”
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
ALLAHDADIAN (US 20220156578 A1) – teaches anomaly detection using Mahalanobis methods, autoencoding and reconstructive models , moving average of reconstruction errors, including thresholding for determination of result validity; [0118]: “Mahalanobis distance automatically computes the distance ton mean (centroid)”.
DAS, et al., "A self-evolving mutually-operative recurrent network-based model for online tool condition monitoring in delay scenario", 2020 Proceedings-26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp 27752-783 (Year: 2020) – teaches automated tool health monitoring and prediction methods using a range of ML techniques and model types, including data-driven methods.
GU, et al., "Fault diagnosis method of rolling bearing using principal component analysis and support vector machine", Journal of Mechanical Science and Technology 32 (11) (2018) 5079-5088. (Year: 2018) – teaches faulty feature extraction for industrial setting using principal component analysis and SVM with reduced computational complexity and high accuracy.
JIN, et al., “An intelligent fault diagnosis method of rolling bearings based on Welch power spectrum transformation with radial basis function neural network. JVC/Journal of Vibration and Control, 26(9-10), 629-642. (Year: 2020) – teaches fault detection method with focus on avoiding information loss in feature extraction using data-driven methods for model training.
LEE (KR 102618023 B1) – teaches method for monitoring, predicting and diagnosing failures in system performance using real time data for training and testing of machine learning models
NASKOS, et al., "Detecting anomalous behavior towards predictive maintenance", Advanced Information Systems Engineering Workshops (CAISE 2019), Lecture Notes in Business Information Processing, vol 349, pp 73-82 (Year: 2019) - teaches methods directed toward effective predictive maintenance modeling using runtime analysis and real time reporting and data driven training sets for building predictive model.
Strauß , et al., "Enabling of Predictive Maintenance in the Brownfield through Low-Cost Sensors, an IIoT Architecture and Machine Learning", Proceedings IEEE 2018 IEEE International Conference on Big Data, Springer, pp 1474-1483. (Year: 2018) – teaches enablement of industrial machinery to perform predictive modeling using data-driven methods; implementation with existing equipment; low cost sensor devices.
Mahalanobis distance – Wikipedia or Prasanta Chandra Mahalanobis Wikipedia – provides detailed information on history, use, and mathematical relationships related to Mahalanobis distance for development of improved machine learning models.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to TONI D SAUNCY whose telephone number is (703)756-4589. The examiner can normally be reached Monday - Friday 8:30 a.m. - 5:30 p.m. 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, Catherine Rastovski can be reached at 571-270-0349. 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.
/TONI D SAUNCY/Examiner, Art Unit 2857
/Catherine T. Rastovski/Supervisory Primary Examiner, Art Unit 2857
Prosecution Insights