Detailed Action
The following NON-FINAL office action is in response to application 18/646227 filed on
4/25/2024. This communication is the first action on the merits.
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 .
Status of Claims
Claims 1-20 are currently pending and have been rejected as follows.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 4/25/2024 complies with the provisions of 37 CFR 1.97 and is being considered.
Claim Objections
Claims 15-20 are objected to because of the following informalities:
Claim 15 reads “in an engine,” which, presumably should be “is an engine”
Claim 18, line 1 reads “the anomalous data set is contains” which presumably should be “the anomalous data set contains”
Appropriate correction is required.
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.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 16 is rejected as failing to define the invention in the manner required by 35 U.S.C. 112(b) or pre-AIA 35 U.S.C. 112, second paragraph.
The use of the parentheses leaves it unclear as to whether or not the limitations inside of them needs to be present to be read on.
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.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception without significantly more. A subject matter eligibility analysis is set forth below. See MPEP 2106.
Specifically, representative Claim 1 recites:
A machine anomaly detection method comprising:
collecting sparse, event-driven, time series data from one or more physical sensors;
inputting the sparse, event-driven, time series data into a heterogeneous ensemble of at least two disparate and independent anomaly detection algorithms;
receiving an output from each of the disparate and independent anomaly detection algorithms, wherein each output comprises a score and an uncertainty associated with detection of an anomaly; and
combining, with a processor, the outputs into a unified output that comprises an overall score and an overall uncertainty associated with the anomaly.
The claim limitations in the abstract idea have been underlined above; the remaining limitations are “additional elements.” Similar limitations comprise the abstract idea of Claim 14.
Step 1:
Under Step 1 of the analysis, Claims 1 and 14 belong to a statutory category, namely they are both method claims.
Step 2A – Prong I:
Under Step 2A, prong 1: This part of the eligibility analysis evaluates whether the claim recites a judicial exception. As explained in MPEP 2106.04, subsection II, a claim “recites” a judicial exception when the judicial exception is “set forth” or “described” in the claim.
In the instant case, Claims 1 and 14 are found to recite at least one judicial exception (i.e. abstract idea), that being a Mental Process and Mathematical Concept. Inputting the time series data into an ensemble of at least two detection algorithms and receiving an output from each of the anomaly detection algorithms amounts to the use of these algorithms. The use of the algorithms and combining the outputs into a unified output to determine a score or uncertainty in assessing the health of a machine or engine can be accomplished mathematically. Furthermore, the overall score and overall uncertainty are calculated, as demonstrated in Eqs. [1] and [2] and Paragraph [0039] of the Specification.
Step 2A – Prong II:
Step 2A, prong 2 of the eligibility analysis evaluates whether the claim as a whole integrates the recited judicial exception(s) into a practical application of the exception. This evaluation is performed by (a) identifying whether there are any additional elements recited in the claim beyond the judicial exception, and (b) evaluating those additional elements individually and in combination to determine whether the claim as a whole integrates the exception into a practical application.
Claims 1 and 14 do not amount to the recitation of a particular practical application as no improvement to the underlying machine/engine is realized through performance of the abstract idea.
Thus, under Step 2A, prong 2 of the analysis, even when viewed in combination, these additional elements do not integrate the recited judicial exception into a practical application and the claim is directed to the judicial exception.
Step 2B:
Under Step 2B, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements, as described above with respect to Step 2A Prong 2, merely amount to a general purpose computer system that attempts to apply the abstract idea in a technological environment, limiting the abstract idea to a particular field of use, and/or merely performs insignificant extra-solution activit(ies). In addition to the abstract ideas recited in claim 1, the claimed method recites the additional elements indicated above.
However, collecting time series data from one or more physical sensors is described at such a high level of generality as to amount to mere data gathering, input, and output steps and as such, are considered to be insignificant extra-solution activity(ies). See MPEP 2106.05(g) “Insignificant Extra-Solution Activity.”
Such insignificant extra-solution activity, e.g. data gathering and output, when re-evaluated under Step 2B is further found to be well-understood, routine, and conventional as evidenced by MPEP 2106.05(d)(II) (describing conventional activities that include transmitting and receiving data over a network, electronic recordkeeping, storing and retrieving information from memory, and electronically scanning or extracting data from a physical document).
Claim 14 additionally recites the following additional elements: “a machine” and “so as to provide a prognosis of potential issues with the machine so that appropriate maintenance can be performed to avoid catastrophic failures of the machine.” The machine is recited so generically as to amount to no more than attempt to generally link the judicial exception to a broad field that could include machines of any type. See MPEP 2106.05(h).
Therefore, similarly the combination and arrangement of the above identified additional elements when analyzed under Step 2B also fails to necessitate a conclusion that Claim 1 amounts to significantly more than the abstract idea.
With regards to the dependent claims, claims 2-20, merely further expand upon the algorithm/abstract idea and do not set forth further additional elements that integrate the recited abstract idea into a practical application or amount to significantly more. Therefore, these claims are found ineligible for the reasons described for parent Claim 1. Specifically:
Claim 2 recites adjusting a threshold and displaying results. Adjusting the a threshold merely places a limit on the mathematical calculations being performed and is within the abstract idea. Displaying the results is considered to be insignificant extra-solution activity.
Claim 3 further describes the abstract idea itself.
Claim 4 recites a processor, which is a general use computer component, and aggregating the output probabilities and uncertainties from the independent anomaly detection algorithms. This aggregation step can be accomplished via Mathematical Calculation and thus is within the abstract idea of Claim 1.
Claim 5 recites the algorithms accepting a time series data stream. This amounts to mere data gathering and thus is insignificant extra-solution activity that is further found to be well-understood, routine, and conventional in the art.
Claim 6 recites an extension of the abstract idea itself through inclusion of an additional mathematical and/or mental activity algorithm.
Claim 7 recites removing a given anomaly detection algorithm when the algorithm’s output value falls below the confidence value range. Determining whether a value falls within a range and using that to make a decision about whether to include an algorithm amounts to a mental process.
Claim 8 recites removing a given anomaly detection algorithm if patterns are found in the time series data that are known to result in false anomaly detection. Recognizing a pattern and using that to make a decision about whether to include an algorithm amounts to a mental process.
Claim 9 recites a physics-based algorithm, a machine learning algorithm, and a TDA algorithm, which merely places limitations on what mathematical steps or calculations are being performed.
Claim 10 recites that the anomaly is a precursor of a physical component failure. The physical component failure is recited at such a high level of generality, with no indication of what the component may be, that it amounts to no more than an attempt to generally link the abstract idea of claim one to the technological environment of a failing physical component.
Claim 11 recites considering a platform to be monitored when selecting the disparate and independent anomaly detection algorithms. The considering and selecting are both Mental Processes and thus abstract ideas.
Claim 12 recites replacing a component on the platform based on the overall score and the overall uncertainty before the component fails completely. Replacing a component amounts to insignificant post-solution activity that is further found to be well-understood, routine, and conventional in the art [See Segal et. al. (US 20190180527 A1), Paragraph [0038]; See also Tamaki et. al. (US 20110276828 A1), Paragraph [0099]]
Claim 13 recites a processor, which amounts to no more than a general use computer, and combining the outputs through a conformal prediction process, which is a framework consisting of mathematical calculations and is thus an abstract idea.
Claim 15 recites that the machine is an engine, which merely generally links the judicial exception to the technological environment of an engine and amounts to a mere field-of-use limitation which does not serve to amount to significantly more than the recitation of the abstract idea itself.
Claim 16 recites specific functions of the kinematics-based algorithm, including comparing, identifying, and establishing a threshold. The comparing and identifying steps are considered to be mental processes and mathematical calculations, while establishing a threshold value is a mathematical step.
Claim 17 recites adjusting the threshold, which is part of the mathematical calculation of Claim 16 and thus within the judicial exception.
Claims 18 and 19 merely specify a time period for data gathering. Such routine data gathering is insignificant extra-solution activity that, when further evaluated under Step 2B, is found to be well-understood, routine, and conventional in the art.
Claim 20 recites a symbolic aggregation approximation method as one of the ensemble algorithms, which merely places limitations on what mathematical steps or calculations are being performed.
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.
Claim(s) 1-5 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et. al. (US 20170314961 A1) in view of Palani et. al. (US 20210344695 A1).
Regarding Claim 1, Chen discloses a machine anomaly detection method comprising: collecting time series data from one or more physical sensors [Paragraph [0064] – “In one embodiment, data is collected by employing, for example, physical sensors and/or monitoring servers in block 602. The sensors may collect a plurality information (e.g., temperature, pressure, etc.), and time series data may be generated and input in block 604 into the modeling and integration system 606.”].
Chen does not disclose that the time series data is sparse and event-driven.
Palani, however, discloses that the time series data can be provided in batches [Paragraph [0025] – “The time series data 104 can be provided to the anomaly detection system 102 in real-time (e.g., continuously), approximately real-time, and/or in batches (e.g., intermittently), according to various embodiments.”].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to apply the method of Chen to the batched time series data of Palani, as using ensemble data from multiple sources is a known method of modeling data with intermittent or low frequency sampling rates.
The combination of Chen and Palani discloses inputting the sparse, event-driven, time series data [Chen, Paragraph [0064] – “In one embodiment, data is collected by employing, for example, physical sensors and/or monitoring servers in block 602. The sensors may collect a plurality information (e.g., temperature, pressure, etc.), and time series data may be generated and input in block 604 into the modeling and integration system 606.” – using batched data of Palani] into a heterogeneous ensemble of at least two disparate and independent anomaly detection algorithms [Chen, Paragraph [0023] – “In one embodiment, the unique features of data characteristics in physical systems may provide a guideline for developing the analytic engine. Considering that there may be a high diversity of data behaviors, the engine may include an ensemble of analysis models, each of which may explore a specific property from the data (e.g., constant model, periodic model, cumulative sum (CUSUM) model, AutoRegressive model with eXternal input (ARX), etc.). The properties considered in those models may include different compositions of attributes, including those from a single attribute, every pair of attributes, a group of attributes, or the whole data set. Each model in the engine may discover the group of time series that follows the property it is associated with. As the models in the engine may cover all the measurement data, each time series may ultimately find at least one model that can capture the behavior of its evolutions.”; Paragraph [0063] – “A model integrator may be employed to combine information from all the models and to generate an overall report of system operation.”]; and
receiving an output from each of the disparate and independent anomaly detection algorithms, wherein each output comprises a score associated with detection of an anomaly [Chen, Paragraph [0055] – “In one embodiment, different models may have different representations, but they may all include a common metric, which may be represented as the fitness score F. The fitness score may reflect the goodness of fit for a given time series.”; Paragraph [0065] – “The model status checking/anomaly detection module 616 may check the values of associated time series based on their profiles, and may report the status to the model integrator module 618. The model integrator module 618 may combine reports from some or all of the models and may generate and output a global report of the system status in block 620.”].
The combination does not disclose that the output comprises an uncertainty.
Palani, however, discloses that the output comprises an uncertainty [Paragraph [0026] – “The ensemble of deep learning models 110 can comprise two or more models such as, for example, a Long Short-Term Memory (LSTM) model with autoencoders 112, a LSTM model wIth uncertainty estimation 114, and/or a LSTM model with dropouts 116, according to various embodiments.”; Paragraph [0028] – “The LSTM model with uncertainty estimation 114 is an LSTM model that can include an estimation of the uncertainty associated with its predictions.”].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to calculate an uncertainty value, as disclosed by Palani, in the output of the combination of Chen and Palani in order to further validate the anomaly determination.
The combination discloses combining, with a processor [Chen, Paragraph [0086] – “Embodiments may include a computer program product accessible from a computer-usable or computer-readable medium providing program code for use by or in connection with a computer or any instruction execution system.”; Paragraph [0087] – “A data processing system suitable for storing and/or executing program code may include at least one processor coupled directly or indirectly to memory elements through a system bus” – processor executes code for model ingetrator module 618], the outputs into a unified output that comprises an overall score and an overall uncertainty associated with the anomaly [Chen, Paragraph [0067] – “In one embodiment, at each time t, the model integrator module 618 may receive status reports from all the models 610, 612, 614, each of which may relate to an alert from measurement data. The number of those alerts may reflect the health of the system. In addition, since the alerts may be derived by inputting related time series into the model, the goodness of that fit may reflect the overall reliability of the alert. Therefore, the summation of fitness values from the received status reports may be employed to describe the system status, which may be denoted as the ‘anomaly score’ of the system, and may be detected in block 616. The anomaly score may be based on the sum of alerts at time t with each alert weighted by its associated fitness value.”; Palani, Paragraph [0028] – “The LSTM model with uncertainty estimation 114 is an LSTM model that can include an estimation of the uncertainty associated with its predictions.”].
Regarding Claim 2, the combination of Chen and Palani discloses the anomaly detection method of claim 1, further comprising: adjusting a confidence threshold [Chen, Paragraphs [0070]-[0071] – “The model may select those time series whose fitness F is larger than γ, where γ may be a predefined parameter based on expectations of constant signals, and a common γ can range from 0.3 to 0.8. For the selected time series, the threshold Δ used for the online monitoring may be further computed as follows… which may be defined as the maximum deviation of x.sub.t from mean μ divided by the fitness F. In one embodiment, larger deviations (e.g., |x.sub.t−μ|) may lead to a large bound for error checking. The fitness F may be placed in the denominator of… because those time series with lower fitness may have larger uncertainties and hence may employ wider band in checking their behaviors.” – see also Eqs. [17] and [18] for calculations of fitness threshold, fitness threshold is confidence threshold, which is adjusted when these calculations are performed]; and
displaying to a user only anomaly detection results that meet the threshold [Chen, Paragraph [0027] – “In one embodiment according to the present principles, the system 102 may include one or more displays 108 for viewing. The displays 108 may permit a user to interact with the system 102 and its components and functions.” – see Fig. [1], user interface 110 and display 108 receive model output; Paragraph [0055] – “The fitness score may be employed to remove irrelevant time series from each model. For example, a threshold may be defined and time series whose fitness scores are below that threshold may be pruned out in block 408. As a result, each model may cover a group of time series that follow the data property of that model.”; Paragraph [0056] – “For each selected time series, a profile {θ, F, Δ} may be built, so that its future observations may be checked based on that profile, and the model profiles may be output in block 412.” – output profiles are displayed].
Regarding Claim 3, the combination of Chen and Palani discloses the anomaly detection method of claim 2, wherein each of the disparate and independent anomaly detection algorithms is configured to evaluate different aspects of the sparse, event-driven, time series data to detect the anomaly [Chen, Paragraph [0061] – “In one embodiment, the single attribute analysis 508 may build a plurality of models to describe properties from individual time series, such as the periodic model for signals with periodicity, the constant model for nearly constant signals, the cumulative sum (CUSUM) model for time series with weak dynamics, the autoregressive (AR) model to measure linear signal dynamics, etc.” – as per clustered time series data of Palani].
Regarding Claim 4, the combination of Chen and Palani discloses the anomaly detection method of claim 3, wherein the combining step is performed by using a processor to create an ensemble which aggregates the output probabilities and uncertainties from the disparate and independent anomaly detection algorithms into the unified output [Chen, Paragraph [0067] – “In one embodiment, at each time t, the model integrator module 618 may receive status reports from all the models 610, 612, 614, each of which may relate to an alert from measurement data. The number of those alerts may reflect the health of the system. In addition, since the alerts may be derived by inputting related time series into the model, the goodness of that fit may reflect the overall reliability of the alert. Therefore, the summation of fitness values from the received status reports may be employed to describe the system status, which may be denoted as the ‘anomaly score’ of the system, and may be detected in block 616. The anomaly score may be based on the sum of alerts at time t with each alert weighted by its associated fitness value.”; Paragraph [0084] – “As discussed above, the data analytic engine according to the present principles may be employed for complex physical system self-management. Based on the strong regularity and high diversity data characteristics observed in physical systems, the analytic engine may profile the system monitoring data with an ensemble of models, each of which may have discovered a specific data property. The extracted data profiles may be employed to facilitate a plurality of management tasks, such as system status monitoring and online anomaly detection.” – with uncertainties of Palani].
Regarding Claim 5, the combination of Chen and Palani discloses the anomaly detection method of claim 4, wherein each of the disparate and independent anomaly detection algorithms must accept a time series data stream as an input [Chen, Paragraph [0054] – “Referring now to FIG. 4, a block/flow diagram of a method for generating modeling profiles 400 is illustratively depicted in accordance with the present principles. Time series data may be generated and/or input in block 402, and may be represented by x.sub.t. The model parameters (e.g., θ) may be learned (e.g., estimated) in block 406 based on the input data. In one embodiment, different models may have different types of parameters and may follow different learning processes.”].
Regarding Claim 14, Chen discloses an anomaly detection method comprising: collecting sparse, event-driven, time series data from one or more physical sensors connected to a machine [Paragraph [0064] – “In one embodiment, data is collected by employing, for example, physical sensors and/or monitoring servers in block 602. The sensors may collect a plurality information (e.g., temperature, pressure, etc.), and time series data may be generated and input in block 604 into the modeling and integration system 606.”; Paragraph [0091] – “The anomalies detected may be detected using data mined from monitoring the facilities 702, 708, 716 (e.g., from sensors, reports, etc. deployed throughout the facilities/vehicles 702, 708, 716) using a data miner/system monitor 705, and the models generated by the methods 200, 300, 400, and 500 of FIGS. 2, 3, 4, and 5, respectively in accordance with various embodiments. The sensors 704, 710, 718 may include any of a plurality of sensors (e.g., temperature, pressure, etc.) that are capable of being deployed in a particular type of facility/vehicle/etc.” – a vehicle is a machine].
Chen does not disclose that the time series data is sparse and event-driven.
Palani, however, discloses that the time series data can be provided in batches [Paragraph [0025] – “The time series data 104 can be provided to the anomaly detection system 102 in real-time (e.g., continuously), approximately real-time, and/or in batches (e.g., intermittently), according to various embodiments.”].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to apply the method of Chen to the batched time series data of Palani, as using ensemble data from multiple sources is a known method of modeling data with intermittent or low frequency sampling rates.
The combination discloses inputting the sparse, event-driven, time series data [Chen, Paragraph [0064] – “In one embodiment, data is collected by employing, for example, physical sensors and/or monitoring servers in block 602. The sensors may collect a plurality information (e.g., temperature, pressure, etc.), and time series data may be generated and input in block 604 into the modeling and integration system 606.” – using batched data of Palani] into a heterogeneous ensemble of at least two disparate and independent anomaly detection algorithms [Chen, Paragraph [0023] – “In one embodiment, the unique features of data characteristics in physical systems may provide a guideline for developing the analytic engine. Considering that there may be a high diversity of data behaviors, the engine may include an ensemble of analysis models, each of which may explore a specific property from the data (e.g., constant model, periodic model, cumulative sum (CUSUM) model, AutoRegressive model with eXternal input (ARX), etc.). The properties considered in those models may include different compositions of attributes, including those from a single attribute, every pair of attributes, a group of attributes, or the whole data set. Each model in the engine may discover the group of time series that follows the property it is associated with. As the models in the engine may cover all the measurement data, each time series may ultimately find at least one model that can capture the behavior of its evolutions.”; Paragraph [0063] – “A model integrator may be employed to combine information from all the models and to generate an overall report of system operation.”];
receiving an output from each of the disparate and independent anomaly detection algorithms, wherein each output comprises a score [Chen, Paragraph [0055] – “In one embodiment, different models may have different representations, but they may all include a common metric, which may be represented as the fitness score F. The fitness score may reflect the goodness of fit for a given time series.”; Paragraph [0065] – “The model status checking/anomaly detection module 616 may check the values of associated time series based on their profiles, and may report the status to the model integrator module 618. The model integrator module 618 may combine reports from some or all of the models and may generate and output a global report of the system status in block 620.”] associated with detection of an anomaly.
The combination does not disclose that the output comprises an uncertainty.
Palani, however, discloses that the output comprises an uncertainty [Paragraph [0026] – “The ensemble of deep learning models 110 can comprise two or more models such as, for example, a Long Short-Term Memory (LSTM) model with autoencoders 112, a LSTM model wIth uncertainty estimation 114, and/or a LSTM model with dropouts 116, according to various embodiments.”; Paragraph [0028] – “The LSTM model with uncertainty estimation 114 is an LSTM model that can include an estimation of the uncertainty associated with its predictions.”].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to calculate an uncertainty value, as disclosed by Palani, in the output of the combination of Chen and Palani in order to further validate the anomaly determination.
The combination discloses combining, with a processor [Chen, Paragraph [0086] – “Embodiments may include a computer program product accessible from a computer-usable or computer-readable medium providing program code for use by or in connection with a computer or any instruction execution system.”; Paragraph [0087] – “A data processing system suitable for storing and/or executing program code may include at least one processor coupled directly or indirectly to memory elements through a system bus” – processor executes code for model ingetrator module 618], the outputs into a unified output that comprises an overall score and an overall uncertainty associated with the anomaly [Chen, Paragraph [0067] – “In one embodiment, at each time t, the model integrator module 618 may receive status reports from all the models 610, 612, 614, each of which may relate to an alert from measurement data. The number of those alerts may reflect the health of the system. In addition, since the alerts may be derived by inputting related time series into the model, the goodness of that fit may reflect the overall reliability of the alert. Therefore, the summation of fitness values from the received status reports may be employed to describe the system status, which may be denoted as the ‘anomaly score’ of the system, and may be detected in block 616. The anomaly score may be based on the sum of alerts at time t with each alert weighted by its associated fitness value.”; Palani, Paragraph [0028] – “The LSTM model with uncertainty estimation 114 is an LSTM model that can include an estimation of the uncertainty associated with its predictions.”] so as to provide a prognosis of potential issues with the machine so that appropriate maintenance can be performed to avoid catastrophic failures of the machine [Chen, Paragraph [0067] – “A high anomaly score may mean that the system significantly deviates from its normal situations. Once the anomaly score exceeds a predefined threshold, the model integrator module 618 may generate and output an alarm in block 620 so that system operators may be informed regarding any possible problems with the operation of the system.”; Paragraph [0090] – “The remote controller/modeler 701 may include a data analyzer/anomaly detector 703 for determining when an anomaly occurs, and the cause of such an anomaly in the facilities 702, 708, 716. An alert generator/corrective action determiner 707 may be employed to alert facility workers of a detected anomaly using any of a plurality of communication interfaces (e.g., cell phone, email, sound alarm in facility, etc.), and a corrective action to resolve the anomaly condition and/or to prevent future anomalies may be determined in block 707, and performed using the remote controller/modeler 701.”].
Claims 6 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et. al. in view of Palani et. al., in further view of Bhattarchayya et. al. (US 20200285997 A1).
Regarding Claim 6, the combination of Chen and Palani discloses the anomaly detection method of claim 5.
The combination does not disclose further comprising adding a specific anomaly detection algorithm to the heterogeneous ensemble when the overall score or overall uncertainty associated with the anomaly exceeds a confidence value range.
However, Bhattacharyya discloses further comprising adding a specific anomaly detection algorithm to the heterogeneous ensemble when the overall score or overall uncertainty associated with the anomaly exceeds a confidence value range [Paragraph [0504] – “Drift can be detected for a sensor when models no longer fit the most recent data well and the frequency of type I errors the system detects exceeds an acceptable, pre-specified threshold. Type I errors can be determined by identifying when a model predicts an anomaly and no true anomaly is detected in a defined time window around the predicted anomaly.”; Paragraph [0506] – “In these embodiments, when drift is detected, the system can trigger generation of new models (e.g., of same or different model types) on the most recent data for the sensor. The system can compare the performance of different models or model types on identical test data sampled from the most recent sensor data and put a selected model (e.g., a most effective model) into deployment or production.”].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to add an algorithm, as disclosed by Bhattacharyya, to the ensemble of the combination of Chen and Palani to reduce uncertainty in the ensemble output.
Regarding Claim 15, the combination of Chen and Palani discloses the method of claim 14.
The combination does not disclose that the machine is an engine.
However, Bhattacharyya discloses that the machine is an engine [Paragraph [0346] – “In some embodiments, the machine being analyzed is a diesel engine within a marine vessel, and the analysis system's goal is to identify diesel engine operational anomalies and/or diesel engine sensor anomalies at near real-time latency, using an edge device installed at or near the engine. Of course, other types of vehicles, engines, or machines may similarly be subject to the monitoring and analysis.”].
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Chen et. al. in view of Palani et. al., in view of Bhattarchayya et. al., in further view of Paul et. al. (US 20200116522 A11).
Regarding Claim 7, the combination of Chen, Palani, and Bhattacharyya discloses the anomaly detection method of claim 6.
The combination does not disclose the method further comprising removing a given anomaly detection algorithm from the heterogeneous ensemble.
Paul, however, discloses the method further comprising removing a given anomaly detection algorithm from the heterogeneous ensemble [Paragraphs [0068]-[0069] – “At this time, the model-group learner/updater 3 initializes the candidate models and relearns all candidate models using new training data (step S9). Subsequently, based on the decision accuracies of a plurality of candidate models, the model selector 4 selects one or more candidate models from among the plurality of candidate models to create an applied model and holds the applied model in the applied model holder 11 (step S10).” – candidate models that are not selected are removed from the group].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to remove an underperforming model from the ensemble, as disclosed by Paul, in the method of Chen, Palani, and Bhattacharryya, in order to improve the ensemble model performance and reduce computational load.
The combination of Chen, Palani, Bhattacharryya, and Paul discloses the method further comprising removing a given anomaly detection algorithm [Removing as per Paul] from the heterogeneous ensemble if the given anomaly detection algorithm’s output has a value below the confidence value range [Chen, Paragraph [0055] – “In one embodiment, different models may have different representations, but they may all include a common metric, which may be represented as the fitness score F. The fitness score may reflect the goodness of fit for a given time series. For example, a time series with weak periodicity behaviors may receive a low fitness score in the periodic model. The computation of fitness may vary with each model. In the periodic model, it may be based on the magnitude of dominant components in x.sub.t's frequency distribution, whereas the fitness of pairwise correlation may be based on the average estimation error of the ARX model. The fitness score may be employed to remove irrelevant time series from each model. For example, a threshold may be defined and time series whose fitness scores are below that threshold may be pruned out in block 408.”].
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Chen et. al. in view of Palani et. al., in view of Paul et. al., in further view of Filimonov et. al. (US 20150269050 A1).
Regarding Claim 8, the combination of Chen and Palani discloses the anomaly detection method of claim 5.
The combination does not disclose the method further comprising removing a given anomaly detection algorithm from the heterogeneous ensemble.
Paul, however, discloses the method further comprising removing a given anomaly detection algorithm from the heterogeneous ensemble [Paragraphs [0068]-[0069] – “At this time, the model-group learner/updater 3 initializes the candidate models and relearns all candidate models using new training data (step S9). Subsequently, based on the decision accuracies of a plurality of candidate models, the model selector 4 selects one or more candidate models from among the plurality of candidate models to create an applied model and holds the applied model in the applied model holder 11 (step S10).” – candidate models that are not selected are removed from the group].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to remove an underperforming model from the ensemble, as disclosed by Paul, in the method of Chen and Palani in order to improve the ensemble model performance and reduce computational load.
The combination of Chen, Palani, and Paul discloses the method further comprising removing a given anomaly detection algorithm [Removing as per Paul] from the heterogeneous ensemble [Chen, Paragraph [0055] – “In one embodiment, different models may have different representations, but they may all include a common metric, which may be represented as the fitness score F. The fitness score may reflect the goodness of fit for a given time series. For example, a time series with weak periodicity behaviors may receive a low fitness score in the periodic model. The computation of fitness may vary with each model. In the periodic model, it may be based on the magnitude of dominant components in x.sub.t's frequency distribution, whereas the fitness of pairwise correlation may be based on the average estimation error of the ARX model. The fitness score may be employed to remove irrelevant time series from each model. For example, a threshold may be defined and time series whose fitness scores are below that threshold may be pruned out in block 408.”].
The combination does not disclose patterns in the time series data known to result in false anomaly detections
Filimonov, however, discloses patterns in the time series data known to result in false anomaly detections [Paragraph [0015] – “The unsupervised anomaly detector in accordance with aspects of the subject matter described herein can determine how the application should behave as the application or other type of component runs. Deducing the normal and abnormal behavior of a component quickly means there is typically not enough time to wait for a very large statistical sample to make predictions regarding the characteristics (normal versus anomalous) of a piece of data within a time series. The anomaly detector described herein can calibrate and/or train classifiers based on a specifiable number of data points from as few as ten data points. The anomaly detector described herein can adapt to accommodate a time series that changes dynamically. False positives (normal values mislabeled as anomalous) are minimized and almost no false negatives (anomalous values mislabeled as normal) are produced.”].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to remove an anomaly detection algorithm, as disclosed by Chen, Palani, and Paul, when there are patterns known to produce false anomaly detections in the time series data, as disclosed by Filimonov, in order to avoid mislabeling false positives as anomalies.
The combination of Chen, Palani, Paul, and Filimonov discloses that the time series data are sparse and event driven [Palani, Paragraph [0025] – “The time series data 104 can be provided to the anomaly detection system 102 in real-time (e.g., continuously), approximately real-time, and/or in batches (e.g., intermittently), according to various embodiments.”].
Claims 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et. al. in view of Palani et. al., in further view of Segal et. al. (US 20190180527 A1).
Regarding Claim 9, the combination of Chen and Palani discloses the anomaly detection method of claim 5.
The combination does not disclose that the heterogeneous ensemble includes a physics-based algorithm, a machine learning algorithm, and a TDA algorithm.
Segal, however, discloses that the heterogeneous ensemble includes a physics-based algorithm [Paragraph [0070] – “In order to assign each timestamped vector, the computer system 110 may train one or more machine learning models on a training set of timestamped kinematic sensor vectors from multiple flight datasets. For example, the trained machine learning models may generate kinematic clusters that correspond to different flight phases of FIG. 2,” – kinematics vectors are physics-based], a machine learning algorithm [Paragraph [0070] – “Because cluster analysis is performed independently in the smaller overlapping regions, more complex machine learning methods can be used. In various embodiments, the Mapper algorithm, or another clustering algorithm, may use k-means clustering, hierarchical clustering, or another clustering approach to generate the clusters, and then assign timestamped vectors to the generated clusters.”], and a TDA algorithm [Paragraph [0070] – “In various embodiments, the MAPPER algorithm from the field of Topological Data Analysis may be used to generate the clusters, and then assign timestamped vectors to the generated clusters in 520. The MAPPER algorithm partitions a dataset into overlapping regions, applies a cluster algorithm such as k-means clustering to each region, and combines the clusters to generate a global clustering across the regions. This approach enables better handling of heterogeneous datasets than traditional approaches.”].
It would’ve been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to implement any of the algorithms of Segal listed above on the heterogeneous ensemble of the combination of Chen and Palani in order to more effectively model the machine behavior and identify anomalous behavior.
Regarding Claim 10, the combination of Chen, Palani, and Segal discloses the anomaly detection method of claim 9, wherein the anomaly is a precursor of a physical component failure [Palani, Paragraph [0033] – “The ensemble of deep learning models 110 can identify anomalies 120 from the aggregated time series data 108. Anomalies 120 can indicate hardware failures or degradations, network configuration issues, cyberthreats, software issues (e.g., malfunctions, bugs, etc.), operating system (OS) issues, and so on. ”].
Claims 11-12 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et. al. in view of Palani et. al., in further view of Segal et. al., in further view of Tamaki et. al. (US 20110276828 A1).
Regarding Claim 11, the combination of Chen, Palani, and Segal discloses the anomaly detection method of claim 10.
The combination does not disclose the method further comprising considering a platform to be monitored when selecting the disparate and independent anomaly detection algorithms that make up the heterogeneous ensemble.
Tamaki, however, discloses the method further comprising considering a platform to be monitored when selecting the disparate and independent anomaly detection algorithms that make up the heterogeneous ensemble [Paragraph [0028] – “Firstly, in monitoring the target apparatus having the data items (apparatus-state-measurement data items) from the plurality of sensors being coordinately shifted, a preferred aim is to provide a model and a technique with using the model in which all data items are monitored, and, even if the single data item causes the anomaly span or the plurality of data items cause the anomaly span, the anomaly (anomaly sign) can be detected with high accuracy.”].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to consider the target apparatus being monitored, as disclosed by Tamaki, when selecting one of the algorithms of Chen, Palani, and Segal in order to generate an ensemble of models that more accurately predicts the anomalies.
Regarding Claim 12, the combination of Chen, Palani, Segal, and Tamaki discloses the anomaly detection method of claim 11 further comprising replacing a component on the platform based on the overall score and the overall uncertainty associated with the detected anomaly before the component fails completely [Tamaki, Paragraph [0098]-[0099] – “The maintenance plan module 6 performs a process of the maintenance plan by using the result (the anomalous-part lifetime information D3) from the part lifetime prediction module 5, and outputs maintenance plan information D4. The maintenance operation instruction module 7 performs a process of instructing the maintenance operation based on the direct detection information D1 from the monitoring execution module 3, the maintenance plan information D4 from the maintenance plan module 6, and others. The instruction for the maintenance operation includes outputting the information for instructing the maintenance operation to a used maintenance apparatus (54 in FIG. 2) by a maintenance operator (U1 in FIG. 2). The maintenance operator (U1) follows this instruction to perform the maintenance operation for the apparatus 1, such as replacing a part of a predetermined module.”].
Claim 13 is rejected under 35 U.S.C. 103 as being unpatentable over Chen et. al. in view of Palani et. al., in further view of Segal et. al., in further view of Perneti et. al. (US 20220091915 A1).
Regarding Claim 13, the combination of Chen and Palani discloses the anomaly detection method of claim 1, wherein the step of combining, with a processor, the outputs into a unified output that comprises an overall score and an overall uncertainty [Chen, Paragraph [0086] – “Embodiments may include a computer program product accessible from a computer-usable or computer-readable medium providing program code for use by or in connection with a computer or any instruction execution system.”; Paragraph [0087] – “A data processing system suitable for storing and/or executing program code may include at least one processor coupled directly or indirectly to memory elements through a system bus” – processor executes code for model ingetrator module 618; Chen, Paragraph [0067] – “In one embodiment, at each time t, the model integrator module 618 may receive status reports from all the models 610, 612, 614, each of which may relate to an alert from measurement data. The number of those alerts may reflect the health of the system. In addition, since the alerts may be derived by inputting related time series into the model, the goodness of that fit may reflect the overall reliability of the alert. Therefore, the summation of fitness values from the received status reports may be employed to describe the system status, which may be denoted as the ‘anomaly score’ of the system, and may be detected in block 616. The anomaly score may be based on the sum of alerts at time t with each alert weighted by its associated fitness value.”; Palani, Paragraph [0028] – “The LSTM model with uncertainty estimation 114 is an LSTM model that can include an estimation of the uncertainty associated with its predictions.”].
The combination does not disclose that the values are associated with the anomaly is performed through a conformal prediction process.
Perenti, however, discloses that the values are associated with the anomaly is performed through a conformal prediction process [Paragraph [0035]-[0036] – “The device failure prediction service 102 in the FIG. 1 embodiment is assumed to be implemented using at least one processing device. Each such processing device generally comprises at least one processor and an associated memory, and implements one or more functional modules for controlling certain features of the device failure prediction service 102. In the FIG. 1 embodiment, the device failure prediction service 102 comprises a feature ranking and selection module 112, a device classification and failure prediction module 114, and a conformal prediction analysis module 116…The conformal prediction analysis module 116 is configured to implement a conformal prediction framework that generates measures of the confidence and credibility of the classifications. Such information is utilized to generate alerts or notifications sent to the client devices 104, and/or to initiate remedial action to address the failure or predicted failure of devices.”].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to use the conformal prediction process of Perneti to determine the unified output and uncertainty score of the combination of Chen and Palani in order to better validate the combined output.
Claims 16-17 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et. al. in view of Palani et. al., in view of Bhattarchayya et. al., in further view of Segal et. al.
Regarding Claim 16, Chen, Palani, and Bhattacharyya disclose the method of claim 15.
The combination does not disclose that the ensemble of disparate and independent anomaly detection algorithms includes a kinematics-based algorithm that comprises the following steps: comparing an anomalous data set (consisting of characteristic values from a group of similar machines that experienced a known anomalous event) and a non-anomalous data set (consisting of characteristic values from a non-anomaly group of similar machines) by plotting the anomalous and non-anomalous data sets on a histogram; identifying systematic differences in distributions between the anomalous and non-anomalous data sets; and establishing a threshold value of one or more characteristic values that correlates to an anomalous event.
Segal, however, discloses that the ensemble of disparate and independent anomaly detection algorithms includes a kinematics-based algorithm [Paragraph [0029] – “The disclosed technology includes machine learning based systems and methods for diagnostics, prognostics and health management for aircraft using kinematic clusters to maintain aircraft, ground vehicles, surface ships, and underwater vessels.”] that comprises the following steps: comparing an anomalous data set (consisting of characteristic values from a group of similar machines that experienced a known anomalous event) and a non-anomalous data set (consisting of characteristic values from a non-anomaly group of similar machines) by plotting the anomalous and non-anomalous data sets on a histogram [Paragraph [0065] – “FIG. 4E brings together FIGS. 4B, 4C, and 4D onto a single graph, and illustrates that only the lower tail 422 of PDF 420 for the first kinematic cluster and the upper tail 444 of PDF 440 for the third kinematic cluster would be recognized as anomalous or out-of-bounds by PDF 410. The remaining tails of PDF 420, PDF 430, and PDF 440 would not be recognized as anomalous our out-of-bounds by PDF 410. These figures illustrate the importance of considering flight phases and their corresponding kinematic clusters when determining if a behavioral sensor reading is anomalous or out-of-bounds.” – See Figs. [4a-e], taking the PDFs to be histograms with infinitesimally small bin widths; PDFs are of an aircraft during different flight phases, making the distributions those of similar machines];
identifying systematic differences in distributions between the anomalous and non-anomalous data sets [Paragraph [0060] – “Additional kinematic sensor measures, or combinations of these measures, may be used by machine learning methods to generate kinematic clusters, corresponding to flight phases, and assign timestamped vectors to the kinematic clusters once they have been assigned. This makes is possible to estimate a probability distribution function (PDF) for a behavioral sensor for each kinematic cluster, or a combination of behavioral sensors for each kinematic cluster, instead of just estimating the PDF across all clusters. Anomalous or out-of-bounds behavioral sensor reading thresholds can be customized for each kinematic cluster or corresponding flight phase, as illustrated in FIGS. 4A-4E.”]; and
establishing a threshold value of one or more characteristic values that correlates to an anomalous event [Paragraph [0060] – “Anomalous or out-of-bounds behavioral sensor reading thresholds can be customized for each kinematic cluster or corresponding flight phase, as illustrated in FIGS. 4A-4E.”].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to compare the anomalous and non-anomalous distributions of Segal using the ensemble methods of the combination of Chen, Palani, and Bhattacharyya in order to improve the identification of anomalous and non-anomalous behavior.
Regarding Claim 17, the combination of Chen, Palani, Bhattacharyya, and Segal discloses the method of claim 16.
The combination does not disclose the method further comprising adjusting the threshold value based on a type of machine being monitored.
Segal, however, discloses adjusting the threshold value based on kinematic cluster and flight phase [Paragraph [0040] – “A vehicle health indicator may be a numerical value representing how close to ‘normal’ a vehicle is operating, relative to pre-defined behavioral sensor value thresholds or to historical behaviors of this or other vehicles.”; Paragraph [0060] – “Anomalous or out-of-bounds behavioral sensor reading thresholds can be customized for each kinematic cluster or corresponding flight phase, as illustrated in FIGS. 4A-4E.”]
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to adjust the thresholds, as disclosed by Segal, in the method of Chen, Palani, Bhattacharyya, and Segal to have individual thresholds that reflect the physical parameters and behaviors of different machines.
Claims 18-19 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et. al. in view of Palani et. al., in view of Bhattarchayya et. al., in view of Segal et. al., in further view of Tholar et. al. (US 20250106231 A1).
Regarding Claim 18, the combination of Chen, Palani, Bhattacharyya, and Segal discloses the method of claim 17.
The combination does not disclose that the anomalous data set is contains data gathered for a time period before the known anomalous event.
Tholar, however, discloses that the anomalous data set is contains data gathered for a time period before the known anomalous event [Paragraph [0081] – “FIG. 9 illustrates an example result of a context anomaly detection procedure according to some embodiments of the invention. In some embodiments, a machine learning anomaly detection algorithm and/or protocol (such as for example the iForest algorithm) may be applied, e.g., to monthly temperature data measured over ˜3 years. While fluctuations in temperature may result in local minima (such as for example around the month of December, or in t.sub.1 910), anomaly detection algorithms and/or protocols according to some embodiments of the invention may not label such local minima as anomalies—which may be, e.g., intuitively explained based on the fact that the depth and width of “wells” included in the graph and including the minima is consistent, and that similar wells are repeating along the graph. In contrast, a sharp peak in t.sub.2 920, around the 3rd month of June included in the graph may be inconsistent and may not repeat itself along the graph and measured data (and it can thus be suspected to reflect, e.g., measurement errors or defected data entries). Anomaly detection algorithms and/or protocols according to some embodiments of the invention may label or classify this peak as an anomaly in the input temperature data.” – see also Fig. 9].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to gather data for a time period prior to the anomalous event, as disclosed by Tholar, when identifying anomalies according to the method of the combination of Chen, Palani, Bhattacharyya, and Segal in order to capture and avoid misidentifying non-anomalous patterns and fluctuations in the data as anomalies.
Regarding Claim 19, the combination of Chen, Palani, Bhattacharyya, Segal, and Tholar discloses the method of claim 18, wherein the time period is three months up to and including a date of the known anomalous event [Paragraph [0081] – “FIG. 9 illustrates an example result of a context anomaly detection procedure according to some embodiments of the invention. In some embodiments, a machine learning anomaly detection algorithm and/or protocol (such as for example the iForest algorithm) may be applied, e.g., to monthly temperature data measured over ˜3 years. While fluctuations in temperature may result in local minima (such as for example around the month of December, or in t.sub.1 910), anomaly detection algorithms and/or protocols according to some embodiments of the invention may not label such local minima as anomalies—which may be, e.g., intuitively explained based on the fact that the depth and width of “wells” included in the graph and including the minima is consistent, and that similar wells are repeating along the graph. In contrast, a sharp peak in t.sub.2 920, around the 3rd month of June included in the graph may be inconsistent and may not repeat itself along the graph and measured data (and it can thus be suspected to reflect, e.g., measurement errors or defected data entries). Anomaly detection algorithms and/or protocols according to some embodiments of the invention may label or classify this peak as an anomaly in the input temperature data.” – see also Fig. 9, with data collection beginning in third March].
Claim 20 is rejected under 35 U.S.C. 103 as being over Chen et. al. in view of Palani et. al., in view of Bhattarchayya et. al., in view of Segal et. al., in further view of Tholar et. al. in further view of Yang et. al. (US 20190044912 A1).
Regarding Claim 20, the combination of Chen, Palani, Bhattacharyya, Segal, and Tholar discloses the method of claim 18.
The combination does not disclose that the ensemble of disparate and independent anomaly detection algorithms includes a symbolic aggregation approximation (SAX) method.
Yang, however, discloses a symbolic aggregation approximation (SAX) method [Paragraph [0023] – “While machine learning algorithms may be suitable for this purpose, they require attack samples during a training phase, which means that they may not be able to detect attacks in real-time. Thus, to provide an improved IDS, the present specification illustrates a symbolic aggregation approximation (SAX)-based method that uses multiple streams of time series to observe inherent patters in the data. This enables the system to detect anomalies in the CAN bus (or other network), and to identify those anomalies as possible intrusions or other errors.“].
It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to use the SAX method of Yang as one of the disparate and independent anomaly detection algorithms of the combination of Chen, Palani, Bhattacharyya, Segal, and Tholar in order to improve detection of patterns in the time series data.
Pertinent Prior Art
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
US-20250187748-A1, METHODS FOR DETERMINING LIKELIHOODS OF IGNITION HAZARDS
US-20220318684-A1, SPARSE ENSEMBLING OF UNSUPERVISED MODELS
US-20210124983-A1, DEVICE AND METHOD FOR ANOMALY DETECTION ON AN INPUT STREAM OF EVENTS
US-20180096261-A1, UNSUPERVISED MACHINE LEARNING ENSEMBLE FOR ANOMALY DETECTION
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JANELLE A HOLMES whose telephone number is (571)272-4336. The examiner can normally be reached Monday - Friday 8:00 am - 5:00 pm.
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, Arleen M Vazquez can be reached at (571) 272-2619. 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.
/J.A.H./Examiner, Art Unit 2857
/ARLEEN M VAZQUEZ/Supervisory Patent Examiner, Art Unit 2857
Prosecution Insights