DETAILED ACTION
Claims 1-20 (filed 12/28/2023) have been considered in this action. Claims 1-20 are newly filed.
Specification
The disclosure is objected to because of the following informalities:
The title of the invention is not descriptive. A new title is required that is clearly indicative of the invention to which the claims are directed.
The examiner recommends a title that is reflective of the core features of the claimed invention, such as: SYSTEM AND METHOD FOR CONTORLLING AN INDUSTRIAL ASSET OF AN ASSET FAMILY USING A MULTI-VARIATE ANOMALY SCORE DETERMINED FROM ANALYSIS OF A POWER SPECTRAL DENSITY
Appropriate correction is required.
Allowable Subject Matter
Claims 7-14 are 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 for the indication of allowable subject matter: Based upon a thorough searching of the prior art, no individual reference or obvious combination of references has been found to teach the totality of the plurality of features required by claim 7. All other claims found allowable are dependent upon claim 7.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-3, 6, and 15-19 are rejected under 35 U.S.C. 102(a)(2) as anticipated by or, in the alternative, under 35 U.S.C. 103 as obvious over Beckerman et al. (US 20210033066, hereinafter Beckerman).
In regards to Claim 1, Beckerman teaches “A method for controlling an industrial asset of an asset family, wherein the asset family comprises a plurality of industrial assets, the method comprising” ([0009] In practice, a data analytics platform may be configured to monitor and/or analyze the operation of wind turbines in a real-world environment (e.g., a set of wind turbines at a given wind site). [0028] If the model outputs an indication that there does appear to be a rotor imbalance issue at the given wind turbine, the data analytics platform may generate an output (e.g., an alert or other notification) and provide the output to an individual responsible for overseeing the operation of the given wind turbine, such as by presenting a representation of an alert at a work station, mobile device, or the like. In some cases, the data analytics platform may be configured to perform various other functions as a result of executing the model for detecting rotor imbalance as well) “determining, via a controller, a plurality of frequency-parameter pairings corresponding to at least one power spectral density of the industrial asset, each frequency-parameter pairing comprising an energy-level distribution for a parameter of the industrial asset across a plurality of frequency intervals of a portion of the at least one power spectral density;” ([0167] At block 506, asset data platform 102 may be configured to, for each wind turbine, transform each of a plurality of time segments of the historical vibration data into a frequency-domain representation. In example implementations, this function may involve asset data platform 102, on a wind-turbine-by-wind-turbine basis, (i) breaking the given wind turbine's set of historical time-series vibration data into a plurality of time segments and (ii) converting each respective time segment into the frequency domain; wherein the frequency domain representation of vibration magnitude/power is a power spectral density; [0169] To illustrate, FIG. 7 provides a conceptual illustration of two bursts of the set of historical time-series vibration data from FIG. 6 transformed into a frequency-domain representation. In particular, data plots 700 and 710 depict a Fourier representation of the x-dimension vibration data (data plot 700) and y-dimension vibration data (data plot 710) of burst B.sub.1 from FIG. 6. Similarly, data plots 720 and 730 depict a Fourier representation of the x-dimension vibration data (data plot 720) and y-dimension vibration data (data plot 730) of burst B.sub.2 from FIG. 6. In practice, asset data platform 102 computes a similar Fourier transform for each of the N bursts of vibration data for the given wind turbine. Each data plot 700, 710, 720, and 730 includes a respective horizontal axis 701, 711, 721, and 731 corresponding to frequency and a respective vertical axis 702, 712, 722, and 732 corresponding to vibration magnitude (e.g., Fourier coefficient value). As shown in FIG. 7, data plots 700, 710, 720, and 730 are divided into multiple frequency-ranges of interest (e.g., a 1P, 3P, and 6P range), each of which corresponds to a different harmonic mode (e.g., integer multiples of the given wind turbine's rotor frequency); wherein each corresponding harmonic frequency to the magnitude in a particular vibration direction is a frequency-parameter pair) “determining, via the controller, a deviation score for each of the plurality of frequency-parameter pairings, wherein each of the deviation scores is indicative of a magnitude difference between the energy-level distribution of each frequency- parameter pairing and a corresponding energy-level distribution of a nominal frequency-parameter pairing of the asset family;” ([0020] As yet another possibility, the model for detecting rotor imbalance may take the form of a rules-based model that is configured to (i) receive one or more sets of “harmonic mode” values that have been derived based on multi-dimensional vibration data captured at a given wind turbine and perhaps also historical data related to the operation of the given wind turbine (e.g., rotor frequency, windspeed, ambient temperature, etc.) and (ii) output an indication of whether a rotor imbalance issue has been detected at the given wind turbine. In this respect, the data analytics platform may derive such a rules-based model by identifying values or ranges of values (and/or other features indicative of rotor behavior) corresponding to wind turbines exhibiting “balanced” rotor behavior and then in turn identifying one or more deviations from these values or ranges of values (and/or other features) that correspond to wind turbines exhibiting a rotor imbalance issue, which may then be embodied into a set of rules that comprises the rules-based model for detecting rotor imbalance. For instance, one possible example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when the average value of its “harmonic mode” values for a given rotor frequency (e.g., its 1P.sub.x and 1P.sub.y values) over one week exceeds a first threshold or the standard deviation over two weeks exceeds a second threshold. Other manners of deriving a rules-based model for detecting rotor imbalance also exist; [0185] In some instances, asset data platform 102 may derive the model for detecting rotor imbalance based, at least in part, on comparing respective time-series sets of “harmonic mode” values of two or more neighboring wind turbines, such as two or more wind turbines that are geographically located proximate to one another and/or have experienced substantially similar meteorological conditions. Other examples are also possible; wherein the data for “balanced” or “normal” vibration is the nominal and the asset family are when neighboring turbines are utilized in the comparison of the harmonic modes when building the model for comparison against current values) “determining, via the controller, a multi-variate anomaly score based, at least in part, on the deviation scores” ([[0026] Carrying out the foregoing function for at least one “burst” within the given wind turbine's reference time-series vibration data may result in at least one corresponding set of “harmonic mode” values for the given wind turbine at a respective time point, where the respective time point comprises a multivariate vector that includes the derived set of “harmonic mode” values for the at least one “burst” associated with that time point. In some implementations, after deriving the at least one set of “harmonic mode” values, the data analytics platform may also perform a temporal “roll-up” operation as discussed before with respect to the “training” phase but using previously-derived sets of “harmonic mode” values for “bursts” from reference vibration data that preceded the at least one “burst” in time. [0027] In any case, the data analytics platform may be configured to execute the model defined in the “training” phase using the at least one set of “harmonic mode” values (either as initially derived or in rolled-up form) that was derived based on the given wind turbine's reference time-series vibration data, which results in an indication of whether the given wind turbine's multi-dimensional vibration data is indicative of a rotor imbalance issue. As noted above, in some implementations, executing the model for detecting rotor imbalance may take into account both (i) one or more sets of “harmonic mode” values derived based on the given wind turbine's reference time-series vibration data and (ii) other data that may be indicative of a rotor imbalance issue, such as other operating data for the given wind turbine, environmental data for the given wind turbine, and/or other data derived therefrom. [0184] For instance, one possible example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when the average value of its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) over a one-week period of time exceeds a first threshold or the standard deviation of its “harmonic mode” values for a given rotor frequency (e.g., its 1P.sub.x and 1P.sub.y values) over a two-week period of time exceeds a second threshold. Another example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) exceed a threshold, such as a threshold relative to the median value of “harmonic mode” values for the given rotor frequency for multiple wind turbines for that given time (e.g., 3 standard deviations, 0.03 m/s.sup.2, etc.); wherein the indication is an anomaly score based on the value being compared to the threshold for the rolled up harmonics) “determining, via the controller, a fault probability for the industrial asset based, at least in part, on the multi-variate anomaly score;” ([0187] As one possibility, the indication may take the form of a metric reflecting a likelihood that the given wind turbine has a rotor imbalance issue. For example, this metric may take the form of a probability metric reflecting a predicted probability of a rotor imbalance issue existing at the given wind turbine that may be represented on a scale of either 0 to 100 or 0.0 to 1.0) “implementing a control action based on the fault probability exceeding a fault threshold” ([0028] If the model outputs an indication that there does appear to be a rotor imbalance issue at the given wind turbine, the data analytics platform may generate an output (e.g., an alert or other notification) and provide the output to an individual responsible for overseeing the operation of the given wind turbine, such as by presenting a representation of an alert at a work station, mobile device, or the like. In some cases, the data analytics platform may be configured to perform various other functions as a result of executing the model for detecting rotor imbalance as well; wherein an alert is a control action).
While Beckerman is not explicit in that a “power spectral density” is utilized, such a feature is implied as the magnitude of the vibration signal at a plurality of frequencies is determined using a Fourier transform that converts the time-domain energy to frequency-domain energy to determine which harmonic mode has an energy that indicated an anomaly/fault in the form of an imbalance. These are just different names for the same concept of performing spectral analysis on sensor feedback data (Fourier transform) to determine where different peaks are localized on the basis of frequency information.
In regards to Claim 2, Beckerman further teaches “The method of claim 1, wherein determining the plurality of frequency-parameter pairings further comprises: receiving, via the controller, a plurality of time-series observations from at least one sensor of the industrial asset, the plurality of time-series observations corresponding to a parameter of the industrial asset” ([0053] The operating data that is captured and sent by asset 104A may take various forms. As one possibility, an asset's operating data may include sensor data that comprises time-series measurements for certain operating parameters of the asset, examples of which may include vibration, speed, velocity, acceleration, location, weight, temperature, pressure, friction, power usage, throttle position, fluid usage, fluid level, voltage, current, magnetic field, electric field, presence or absence of objects, current position of a component, and power generation, among many others. As another possibility, an asset's operating data may include abnormal-conditions data that indicates occurrences of discrete abnormal conditions at the asset, examples of which include fault codes that indicate the occurrence of certain faults at the asset (e.g., when an operating parameter exceeds a threshold), asset shutdown indicators, and/or other types of abnormal-condition indicators. As yet another possibility, an asset's operating data may include data that has been derived from the asset's sensor data and/or abnormal-conditions data, examples of which may include “roll-up” data (e.g., an average, mean, median, etc. of the raw measurements for an operating parameter over a given time window) and “features” data (e.g., data values that are derived based on the raw measurements of two or more of the asset's operating parameters). An asset's operating data may take various other forms as well) “converting, via the controller, the plurality of time-series observations into the least one power spectral density of the industrial asset; and” ([0168] As one possible approach, asset data platform 102 may take each respective “burst” within the historical time-series vibration data for a given wind turbine and then perform a frequency-domain transform (e.g., a Fourier, Laplace, Z, or Wavelet transform) of the particular “burst” in each dimension measured by the given wind turbine's multi-dimensional sensor (e.g., fore-to-aft and side-to-side). This function results in a representation of the particular “burst” in the frequency domain (for each dimension) that includes a set of peaks spaced out at different frequency ranges corresponding to different “harmonic modes” (e.g., integer multiples of the given wind turbine's rotor frequency, such as 1P, 3P, etc.).[0173] Asset data platform 102 carrying out the foregoing function for each respective “burst” within a given wind turbine's historical time-series vibration data may result in a time-series set of “harmonic mode” values for the given wind turbine, where each respective time point in the time-series set of “harmonic mode” values comprises a multivariate vector that includes the derived set of “harmonic mode” values for the respective “burst” associated with that time point. Viewed another way, this function may result in multiple, different time-series sets for the given wind turbine that each correspond to a particular type of “harmonic mode” value, such as a first time-series set of 1P.sub.x values, a second time-series set of 1P.sub.y values, a third time-series set of 3P.sub.x values, a fourth time-series set of 3P.sub.y values, etc.; wherein the converting of vibration power/energy/magnitude signals from the sensors into frequency domain is a power spectral density) “and identifying, via the controller, at least one frequency band of the plurality of frequency intervals at which the power spectral density of the industrial asset deviates from the corresponding power spectral density for the asset family at the at least one frequency band” ([0179] Returning to FIG. 5, at block 512, asset data platform 102 may be configured to derive a model for detecting rotor imbalance based on the respective time-series sets of “harmonic mode” values for the plurality of wind turbines. In other words, asset data platform 102 may be configured to evaluate the time-series sets of “harmonic mode” values for the plurality of wind turbines (either as initially derived or in rolled-up form) and thereby derive a model for determining whether a wind turbine's “harmonic mode” values are indicative of a rotor imbalance issue. In this respect, it is expected that the “harmonic mode” values at or near the rotor frequency (e.g., the 1P.sub.x and 1P.sub.y values) for the wind turbines will be most indicative of whether the turbines have a rotor imbalance issue, but it is possible that other “harmonic mode” values may be correlated to rotor imbalance as well. In practice, the model for detecting rotor imbalance may take a variety of forms, and asset data platform 102 may derive this model in a variety of manners. [0184] Another example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) exceed a threshold, such as a threshold relative to the median value of “harmonic mode” values for the given rotor frequency for multiple wind turbines for that given time (e.g., 3 standard deviations, 0.03 m/s.sup.2, etc.). Other manners of deriving a rules-based model for detecting rotor imbalance also exist; wherein the 1P harmonics are a particular frequency band and thus are the identified anomalous frequency band).
In regards to Claim 3, Beckerman further teaches “The method of claim 2, wherein the at least one power spectral density comprises a range of energy levels at each of the plurality of frequency intervals of the at least one power spectral density” ([0012] Next, for each wind turbine, the data analytics platform may be configured to break the wind turbine's respective set of historical time-series vibration data into a plurality of time segments, convert each respective time segment into the frequency domain, and then aggregate frequency-domain values for each respective time segment within one or more frequency-ranges of interest. This function may take various forms. [0183] As yet another possibility, the model for detecting rotor imbalance may take the form of a rules-based model that is configured to (i) receive one or more sets of “harmonic mode” values that have been derived based on multi-dimensional vibration data captured at a given wind turbine and perhaps also historical data related to the operation of the given wind turbine (e.g., rotor frequency, windspeed, ambient temperature, etc.) and (ii) output an indication of whether a rotor imbalance issue has been detected at the given wind turbine. In this respect, asset data platform 102 may derive such a rules-based model by identifying values or ranges of values (and/or other features indicative of rotor behavior) corresponding to wind turbines exhibiting “balanced” rotor behavior and then in turn identifying one or more deviations from these values or ranges of values (and/or other features) that correspond to wind turbines exhibiting a rotor imbalance issue, which may then be embodied into a set of rules that comprises the rules-based model for detecting rotor imbalance) “the range of energy levels being defined between a maximal energy level and a minimal energy level of the parameter at each frequency interval and being indicative of an energy level of the parameter at each frequency interval for a plurality of operating conditions of the industrial asset” ([0183] asset data platform 102 may derive such a rules-based model by identifying values or ranges of values (and/or other features indicative of rotor behavior) corresponding to wind turbines exhibiting “balanced” rotor behavior and then in turn identifying one or more deviations from these values or ranges of values (and/or other features) that correspond to wind turbines exhibiting a rotor imbalance issue, which may then be embodied into a set of rules that comprises the rules-based model for detecting rotor imbalance; wherein the range of values that are determined will naturally have a maximum and minimum value; [0184] For instance, one possible example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when the average value of its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) over a one-week period of time exceeds a first threshold or the standard deviation of its “harmonic mode” values for a given rotor frequency (e.g., its 1P.sub.x and 1P.sub.y values) over a two-week period of time exceeds a second threshold. Another example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) exceed a threshold, such as a threshold relative to the median value of “harmonic mode” values for the given rotor frequency for multiple wind turbines for that given time (e.g., 3 standard deviations, 0.03 m/s.sup.2, etc.). Other manners of deriving a rules-based model for detecting rotor imbalance also exist; wherein x and y imbalance are plurality of different parameters; [0192] In practice, asset data platform 102 receives the reference time-series vibration data that is from a particular reference time, such as a particular point in time or period of time in the past. For example, asset data platform 102 may periodically receive a batch of time-series vibration data corresponding to the given wind turbine's operation (or the whole park's operation) for a particular period of time (e.g., the past 10 minutes, the past 24 hours, etc.). In some implementations, asset data platform 102 may receive this reference vibration data in real-time, near real-time, or otherwise shortly after the given wind turbine's multi-dimensional sensor captures the data).
While Beckerman is not explicit that a maximum energy level and minimum energy level is determined, such a feature is implied because Beckerman discuss how the values of the vibration magnitude at each harmonic can be a ranges of values, and ranges of values inherently have a minimum and maximum value. The claim does not state what is done with that minimum and maximum, only that they exist.
In regards to Claim 6, Beckerman further teaches “The method of claim 2, wherein determining the plurality of frequency-parameter pairings further comprises: receiving, via the controller, a training data set comprising a first plurality of historical power spectral densities corresponding to a nominal population of industrial asset of the asset family and a second plurality of historical power spectral densities corresponding to a fault population of the asset family” ([0018] the data analytics platform may derive such a predictive model by applying a machine-learning technique (e.g., logistic regression, support vector machine, random forest, neural network, etc.) to training data that includes time-series sets of “harmonic mode” values for a plurality of wind turbines (perhaps in rolled-up form) that have been derived in the manner described above, perhaps along with labels indicating times when the plurality of wind turbines were known to have a rotor imbalance issue; wherein those data labeled as imbalance are the fault population, while those without are the nominal population) “wherein the first plurality of historical power spectral densities is indicative of a nominal operating condition for a plurality of parameters, and wherein the second plurality of historical power spectral densities is indicative of at least one fault condition for the plurality of parameters” ([0011] During the “training” phase, the data analytics platform may be configured to optionally filter each wind turbine's respective set of historical vibration data to exclude vibration data corresponding to times when the wind turbine may have been operating abnormally for a reason other than rotor imbalance. In other words, the data analytics platform may be configured to preprocess the data so that it derives the model based on historical vibration data that corresponds to the “normal” operation of the set of wind turbines. To that end, the data analytics platform may filter the respective sets of historical vibration data based on a variety of other data related to the historical operation of the plurality of wind turbines, such as other types of operating data (e.g., historical energy production data) and/or environmental data (e.g., historical weather data for the wind site). [0019] As another possibility, the model for detecting rotor imbalance may take the form of a predictive model that is configured to (i) receive one or more sets of “harmonic mode” values that have been derived based on multi-dimensional vibration data captured at a given wind turbine along with other types of data related to the operation of the given wind turbine (e.g., rotor frequency, windspeed, ambient temperature, etc.) and (ii) output an indication of whether the given wind turbine is predicted to have a rotor imbalance issue. In this respect, the data analytics platform may derive such a predictive model by applying a machine-learning technique (e.g., logistic regression, support vector machine, random forest, neural network, etc.) to training data that includes time-series sets of “harmonic mode” values for a plurality of wind turbines (perhaps in rolled-up form) that have been derived in the manner described above and historical data related to the operation of the plurality of wind turbines, perhaps along with labels indicating times when the plurality of wind turbines were known to have a rotor imbalance issue.; wherein those data labeled as imbalance are the fault population, while those without are the nominal population) “generating, via the controller, a fault-detection model configured to determine the plurality of frequency-parameter pairings which are indicative of the at least one fault condition, the plurality of frequency-parameter pairings being determined from a plurality of potential frequency-parameter pairings for the first and second pluralities of historical power spectral densities” ([0179] Returning to FIG. 5, at block 512, asset data platform 102 may be configured to derive a model for detecting rotor imbalance based on the respective time-series sets of “harmonic mode” values for the plurality of wind turbines. In other words, asset data platform 102 may be configured to evaluate the time-series sets of “harmonic mode” values for the plurality of wind turbines (either as initially derived or in rolled-up form) and thereby derive a model for determining whether a wind turbine's “harmonic mode” values are indicative of a rotor imbalance issue. In this respect, it is expected that the “harmonic mode” values at or near the rotor frequency (e.g., the 1P.sub.x and 1P.sub.y values) for the wind turbines will be most indicative of whether the turbines have a rotor imbalance issue, but it is possible that other “harmonic mode” values may be correlated to rotor imbalance as well. In practice, the model for detecting rotor imbalance may take a variety of forms, and asset data platform 102 may derive this model in a variety of manners. [0180] asset data platform 102 may derive such a predictive model by applying a machine-learning technique (e.g., a supervised or unsupervised machine-learning technique) to training data that includes time-series sets of “harmonic mode” values for a plurality of wind turbines (perhaps in rolled-up form) that have been derived in the manner described at blocks 502-510, perhaps along with labels indicating times when the plurality of wind turbines were known to have a rotor imbalance issue. In example implementations, the machine-learning technique may involve utilizing one or more of logistic regression, support vector machine, random forest, and/or a neural network, among other possibilities) “and training, via the controller, the fault-detection model via the training data set so as to determine the plurality of frequency-parameter pairings indicative of the at least one fault condition” ([0008] the disclosed process may involve (i) a “training” phase during which a model for detecting rotor imbalance (e.g., a predictive model or a set of rules) is derived based on historical vibration data for wind turbines and perhaps other data related to the operation of the wind turbine as well (e.g., other types of historical operating data, historical environmental data, historical maintenance data, etc.), [0010] As mentioned above, in one possible implementation of the “training” phase of the process for detecting rotor imbalance disclosed herein, a data analytics platform may be configured to analyze respective sets of historical time-series vibration data for a plurality of wind turbines and use that as a basis to derive a model for detecting rotor imbalance. For instance, the data analytics platform may receive respective sets of historical time-series vibration data captured by multi-dimensional sensors at a plurality of wind turbines, where each wind turbine's historical time-series vibration data may be indicative of vibrations at the wind turbine in multiple different dimensions (e.g., side-to-side vibrations corresponding to an x-dimension and fore-to-aft vibrations corresponding to a y-dimension). In operation, this historical time-series vibration data may take a variety of forms, such as a time-series of periodic “bursts” of vibration data captured at a high sample rate (e.g., 500-second windows of vibration data that includes 100 samples per second).
In regards to Claim 15, Beckerman teaches “The method of claim 1, wherein the industrial asset comprises a wind turbine” ([0008] To help address these and other problems, the present disclosure is directed to an improved technological process for detecting rotor imbalance at a wind turbine (e.g., a horizontal-axis wind turbine) using vibration data captured by a wind turbine's preinstalled multi-dimensional sensor, either alone or in combination with other data related to the operation of the wind turbine (e.g., other operating data, vibration data from one or more of the wind turbine's other preinstalled sensors, weather data, etc.)).
In regards to Claim 16, Beckerman teaches “A system for controlling an industrial asset of an asset family, wherein the asset family comprises a plurality of industrial assets, the system comprising” ([0009] In practice, a data analytics platform may be configured to monitor and/or analyze the operation of wind turbines in a real-world environment (e.g., a set of wind turbines at a given wind site). [0028] the data analytics platform may generate an output (e.g., an alert or other notification) and provide the output to an individual responsible for overseeing the operation of the given wind turbine, such as by presenting a representation of an alert at a work station, mobile device, or the like. In some cases, the data analytics platform may be configured to perform various other functions as a result of executing the model for detecting rotor imbalance as well; wherein a set of wind turbines is a family of assets, while an individual turbine is the asset) “at least one sensor operably coupled to the industrial asset” ([0008] the present disclosure is directed to an improved technological process for detecting rotor imbalance at a wind turbine (e.g., a horizontal-axis wind turbine) using vibration data captured by a wind turbine's preinstalled multi-dimensional sensor, either alone or in combination with other data related to the operation of the wind turbine (e.g., other operating data, vibration data from one or more of the wind turbine's other preinstalled sensors, weather data, etc.)) “a controller communicatively coupled to the at least one sensor, the controller comprising at least one processor configured to perform a plurality of operations, the plurality of operations comprising” ([0085] FIG. 2 is a simplified block diagram illustrating some structural components that may be included in an example computing platform 200, which could serve as the asset data platform 102 in FIG. 1. In line with the discussion above, platform 200 may generally comprise one or more computer systems (e.g., one or more servers), and these one or more computer systems may collectively include at least a processor 202, data storage 204, and a communication interface 206, all of which may be communicatively linked by a communication link 208 that may take the form of a system bus, a communication network such as a public, private, or hybrid cloud, or some other connection mechanism. [0053] The operating data that is captured and sent by asset 104A may take various forms. As one possibility, an asset's operating data may include sensor data that comprises time-series measurements for certain operating parameters of the asset, examples of which may include vibration, speed, velocity, acceleration, location, weight, temperature, pressure, friction, power usage, throttle position, fluid usage, fluid level, voltage, current, magnetic field, electric field, presence or absence of objects, current position of a component, and power generation, among many others) “determining a plurality of frequency-parameter pairings corresponding to at least one power spectral density of the industrial asset, each frequency-parameter pairing comprising an energy-level distribution for a parameter of the industrial asset across a plurality of frequency intervals of a portion of the at least one power spectral density;” ([0167] At block 506, asset data platform 102 may be configured to, for each wind turbine, transform each of a plurality of time segments of the historical vibration data into a frequency-domain representation. In example implementations, this function may involve asset data platform 102, on a wind-turbine-by-wind-turbine basis, (i) breaking the given wind turbine's set of historical time-series vibration data into a plurality of time segments and (ii) converting each respective time segment into the frequency domain; wherein the frequency domain representation of vibration magnitude/power is a power spectral density; [0169] To illustrate, FIG. 7 provides a conceptual illustration of two bursts of the set of historical time-series vibration data from FIG. 6 transformed into a frequency-domain representation. In particular, data plots 700 and 710 depict a Fourier representation of the x-dimension vibration data (data plot 700) and y-dimension vibration data (data plot 710) of burst B.sub.1 from FIG. 6. Similarly, data plots 720 and 730 depict a Fourier representation of the x-dimension vibration data (data plot 720) and y-dimension vibration data (data plot 730) of burst B.sub.2 from FIG. 6. In practice, asset data platform 102 computes a similar Fourier transform for each of the N bursts of vibration data for the given wind turbine. Each data plot 700, 710, 720, and 730 includes a respective horizontal axis 701, 711, 721, and 731 corresponding to frequency and a respective vertical axis 702, 712, 722, and 732 corresponding to vibration magnitude (e.g., Fourier coefficient value). As shown in FIG. 7, data plots 700, 710, 720, and 730 are divided into multiple frequency-ranges of interest (e.g., a 1P, 3P, and 6P range), each of which corresponds to a different harmonic mode (e.g., integer multiples of the given wind turbine's rotor frequency); wherein each corresponding harmonic frequency to the magnitude in a particular vibration direction is a frequency-parameter pair) “determining a deviation score for each of the plurality of frequency-parameter pairings, wherein each of the deviation scores is indicative of a magnitude difference between the energy-level distribution of each frequency- parameter pairing and a corresponding energy-level distribution of a nominal frequency-parameter pairing of the asset family;” ([0020] As yet another possibility, the model for detecting rotor imbalance may take the form of a rules-based model that is configured to (i) receive one or more sets of “harmonic mode” values that have been derived based on multi-dimensional vibration data captured at a given wind turbine and perhaps also historical data related to the operation of the given wind turbine (e.g., rotor frequency, windspeed, ambient temperature, etc.) and (ii) output an indication of whether a rotor imbalance issue has been detected at the given wind turbine. In this respect, the data analytics platform may derive such a rules-based model by identifying values or ranges of values (and/or other features indicative of rotor behavior) corresponding to wind turbines exhibiting “balanced” rotor behavior and then in turn identifying one or more deviations from these values or ranges of values (and/or other features) that correspond to wind turbines exhibiting a rotor imbalance issue, which may then be embodied into a set of rules that comprises the rules-based model for detecting rotor imbalance. For instance, one possible example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when the average value of its “harmonic mode” values for a given rotor frequency (e.g., its 1P.sub.x and 1P.sub.y values) over one week exceeds a first threshold or the standard deviation over two weeks exceeds a second threshold. Other manners of deriving a rules-based model for detecting rotor imbalance also exist; [0185] In some instances, asset data platform 102 may derive the model for detecting rotor imbalance based, at least in part, on comparing respective time-series sets of “harmonic mode” values of two or more neighboring wind turbines, such as two or more wind turbines that are geographically located proximate to one another and/or have experienced substantially similar meteorological conditions. Other examples are also possible; wherein the data for “balanced” or “normal” vibration is the nominal and the asset family are when neighboring turbines are utilized in the comparison of the harmonic modes when building the model for comparison against current values) “determining a multi-variate anomaly score based, at least in part, on the deviation scores” ([0026] Carrying out the foregoing function for at least one “burst” within the given wind turbine's reference time-series vibration data may result in at least one corresponding set of “harmonic mode” values for the given wind turbine at a respective time point, where the respective time point comprises a multivariate vector that includes the derived set of “harmonic mode” values for the at least one “burst” associated with that time point. In some implementations, after deriving the at least one set of “harmonic mode” values, the data analytics platform may also perform a temporal “roll-up” operation as discussed before with respect to the “training” phase but using previously-derived sets of “harmonic mode” values for “bursts” from reference vibration data that preceded the at least one “burst” in time. [0027] In any case, the data analytics platform may be configured to execute the model defined in the “training” phase using the at least one set of “harmonic mode” values (either as initially derived or in rolled-up form) that was derived based on the given wind turbine's reference time-series vibration data, which results in an indication of whether the given wind turbine's multi-dimensional vibration data is indicative of a rotor imbalance issue. As noted above, in some implementations, executing the model for detecting rotor imbalance may take into account both (i) one or more sets of “harmonic mode” values derived based on the given wind turbine's reference time-series vibration data and (ii) other data that may be indicative of a rotor imbalance issue, such as other operating data for the given wind turbine, environmental data for the given wind turbine, and/or other data derived therefrom. [0184] For instance, one possible example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when the average value of its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) over a one-week period of time exceeds a first threshold or the standard deviation of its “harmonic mode” values for a given rotor frequency (e.g., its 1P.sub.x and 1P.sub.y values) over a two-week period of time exceeds a second threshold. Another example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) exceed a threshold, such as a threshold relative to the median value of “harmonic mode” values for the given rotor frequency for multiple wind turbines for that given time (e.g., 3 standard deviations, 0.03 m/s.sup.2, etc.); wherein the indication is an anomaly score based on the value being compared to the threshold for the rolled up harmonics) “determining, via the controller, a fault probability for the industrial asset based, at least in part, on the multi-variate anomaly score;” ([0187] As one possibility, the indication may take the form of a metric reflecting a likelihood that the given wind turbine has a rotor imbalance issue. For example, this metric may take the form of a probability metric reflecting a predicted probability of a rotor imbalance issue existing at the given wind turbine that may be represented on a scale of either 0 to 100 or 0.0 to 1.0) “implementing a control action based on the fault probability exceeding a fault threshold” ([0028] If the model outputs an indication that there does appear to be a rotor imbalance issue at the given wind turbine, the data analytics platform may generate an output (e.g., an alert or other notification) and provide the output to an individual responsible for overseeing the operation of the given wind turbine, such as by presenting a representation of an alert at a work station, mobile device, or the like. In some cases, the data analytics platform may be configured to perform various other functions as a result of executing the model for detecting rotor imbalance as well; wherein an alert is a control action).
While Beckerman is not explicit in that a “power spectral density” is utilized, such a feature is implied as the magnitude of the vibration signal at a plurality of frequencies is determined using a Fourier transform that converts the time-domain energy to frequency-domain energy to determine which harmonic mode has an energy that indicated an anomaly/fault in the form of an imbalance. These are just different names for the same concept of performing spectral analysis on sensor feedback data (Fourier transform) to determine where different peaks are localized on the basis of frequency information.
In regards to Claim 17, Beckerman teaches “The system of claim 16, wherein determining the plurality of frequency-parameter pairings further comprises: receiving a plurality of time-series observations from the at least one sensor, the plurality of time-series observations corresponding to a parameter of the industrial asset” ([0053] The operating data that is captured and sent by asset 104A may take various forms. As one possibility, an asset's operating data may include sensor data that comprises time-series measurements for certain operating parameters of the asset, examples of which may include vibration, speed, velocity, acceleration, location, weight, temperature, pressure, friction, power usage, throttle position, fluid usage, fluid level, voltage, current, magnetic field, electric field, presence or absence of objects, current position of a component, and power generation, among many others. As another possibility, an asset's operating data may include abnormal-conditions data that indicates occurrences of discrete abnormal conditions at the asset, examples of which include fault codes that indicate the occurrence of certain faults at the asset (e.g., when an operating parameter exceeds a threshold), asset shutdown indicators, and/or other types of abnormal-condition indicators. As yet another possibility, an asset's operating data may include data that has been derived from the asset's sensor data and/or abnormal-conditions data, examples of which may include “roll-up” data (e.g., an average, mean, median, etc. of the raw measurements for an operating parameter over a given time window) and “features” data (e.g., data values that are derived based on the raw measurements of two or more of the asset's operating parameters). An asset's operating data may take various other forms as well) “converting the plurality of time-series observations into the least one power spectral density of the industrial asset, wherein the at least one power spectral density comprises a range of energy levels at each of the plurality of frequency intervals of the at least one power spectral density, the range of energy levels being defined between a maximal energy level and a minimal energy level of the parameter at each frequency interval and being indicative of an energy level of the parameter at each frequency interval for a plurality of operating conditions of the industrial asset” (Fig. 6-8, Fig. 8 shows ranges of values at each frequency interval and inherently has maximum and minimum values and [0162] To illustrate, FIG. 6 provides a conceptual illustration of a given wind turbine's set of historical time-series vibration data from the given wind turbine's multi-dimensional sensor that was obtained by asset data platform 102. As shown, this conceptual illustration includes a data plot 600 of historical vibration measurements from the multi-dimensional sensor's x-dimension and a data plot 610 of historical vibration measurements from the multi-dimensional sensor's y-dimension. Each data plot 600 and 610 includes a respective horizontal axis 601 and 611 corresponding to time and a respective vertical axis 602 and 612 corresponding to vibration magnitude (e.g., in units of acceleration or power). In this example, the given wind turbine's set of historical time-series vibration data takes the form of a plurality (i.e., N) “bursts” of vibration data captured at a high sample rate, where each “burst” corresponds to a given window of time. [0168] As one possible approach, asset data platform 102 may take each respective “burst” within the historical time-series vibration data for a given wind turbine and then perform a frequency-domain transform (e.g., a Fourier, Laplace, Z, or Wavelet transform) of the particular “burst” in each dimension measured by the given wind turbine's multi-dimensional sensor (e.g., fore-to-aft and side-to-side). This function results in a representation of the particular “burst” in the frequency domain (for each dimension) that includes a set of peaks spaced out at different frequency ranges corresponding to different “harmonic modes” (e.g., integer multiples of the given wind turbine's rotor frequency, such as 1P, 3P, etc.); [0173] Asset data platform 102 carrying out the foregoing function for each respective “burst” within a given wind turbine's historical time-series vibration data may result in a time-series set of “harmonic mode” values for the given wind turbine, where each respective time point in the time-series set of “harmonic mode” values comprises a multivariate vector that includes the derived set of “harmonic mode” values for the respective “burst” associated with that time point. Viewed another way, this function may result in multiple, different time-series sets for the given wind turbine that each correspond to a particular type of “harmonic mode” value, such as a first time-series set of 1P.sub.x values, a second time-series set of 1P.sub.y values, a third time-series set of 3P.sub.x values, a fourth time-series set of 3P.sub.y values, etc. Wherein the converting of vibration power/energy/magnitude signals from the sensors into frequency domain is a power spectral density; [0183] asset data platform 102 may derive such a rules-based model by identifying values or ranges of values (and/or other features indicative of rotor behavior) corresponding to wind turbines exhibiting “balanced” rotor behavior and then in turn identifying one or more deviations from these values or ranges of values (and/or other features) that correspond to wind turbines exhibiting a rotor imbalance issue, which may then be embodied into a set of rules that comprises the rules-based model for detecting rotor imbalance; wherein the range of values that are determined will naturally have a maximum and minimum value) “and identifying at least one frequency band of the plurality of frequency intervals at which the power spectral density of the industrial asset deviates from the corresponding power spectral density for the asset family at the at least one frequency band” ([0179] Returning to FIG. 5, at block 512, asset data platform 102 may be configured to derive a model for detecting rotor imbalance based on the respective time-series sets of “harmonic mode” values for the plurality of wind turbines. In other words, asset data platform 102 may be configured to evaluate the time-series sets of “harmonic mode” values for the plurality of wind turbines (either as initially derived or in rolled-up form) and thereby derive a model for determining whether a wind turbine's “harmonic mode” values are indicative of a rotor imbalance issue. In this respect, it is expected that the “harmonic mode” values at or near the rotor frequency (e.g., the 1P.sub.x and 1P.sub.y values) for the wind turbines will be most indicative of whether the turbines have a rotor imbalance issue, but it is possible that other “harmonic mode” values may be correlated to rotor imbalance as well. In practice, the model for detecting rotor imbalance may take a variety of forms, and asset data platform 102 may derive this model in a variety of manners. [0184] Another example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) exceed a threshold, such as a threshold relative to the median value of “harmonic mode” values for the given rotor frequency for multiple wind turbines for that given time (e.g., 3 standard deviations, 0.03 m/s.sup.2, etc.). Other manners of deriving a rules-based model for detecting rotor imbalance also exist; wherein the 1P harmonics are a particular frequency band and thus are the identified anomalous frequency band).
In regards to Claim 18, Beckerman further teaches “The system of claim 16, wherein determining the plurality of frequency-parameter pairings further comprises: receiving a training data set comprising a first plurality of historical power spectral densities corresponding to a nominal population of the asset family and a second plurality of historical power spectral densities corresponding to a fault population of the asset family” ([0018] the data analytics platform may derive such a predictive model by applying a machine-learning technique (e.g., logistic regression, support vector machine, random forest, neural network, etc.) to training data that includes time-series sets of “harmonic mode” values for a plurality of wind turbines (perhaps in rolled-up form) that have been derived in the manner described above, perhaps along with labels indicating times when the plurality of wind turbines were known to have a rotor imbalance issue; wherein those data labeled as imbalance are the fault population, while those without are the nominal population) “wherein the first plurality of historical power spectral densities is indicative of a nominal operating condition for a plurality of parameters, and wherein the second plurality of historical power spectral densities is indicative of at least one fault condition for the plurality of parameters” ([0011] During the “training” phase, the data analytics platform may be configured to optionally filter each wind turbine's respective set of historical vibration data to exclude vibration data corresponding to times when the wind turbine may have been operating abnormally for a reason other than rotor imbalance. In other words, the data analytics platform may be configured to preprocess the data so that it derives the model based on historical vibration data that corresponds to the “normal” operation of the set of wind turbines. To that end, the data analytics platform may filter the respective sets of historical vibration data based on a variety of other data related to the historical operation of the plurality of wind turbines, such as other types of operating data (e.g., historical energy production data) and/or environmental data (e.g., historical weather data for the wind site). [0019] As another possibility, the model for detecting rotor imbalance may take the form of a predictive model that is configured to (i) receive one or more sets of “harmonic mode” values that have been derived based on multi-dimensional vibration data captured at a given wind turbine along with other types of data related to the operation of the given wind turbine (e.g., rotor frequency, windspeed, ambient temperature, etc.) and (ii) output an indication of whether the given wind turbine is predicted to have a rotor imbalance issue. In this respect, the data analytics platform may derive such a predictive model by applying a machine-learning technique (e.g., logistic regression, support vector machine, random forest, neural network, etc.) to training data that includes time-series sets of “harmonic mode” values for a plurality of wind turbines (perhaps in rolled-up form) that have been derived in the manner described above and historical data related to the operation of the plurality of wind turbines, perhaps along with labels indicating times when the plurality of wind turbines were known to have a rotor imbalance issue.; wherein those data labeled as imbalance are the fault population, while those without are the nominal population) “generating, via the controller, a fault-detection model configured to determine the plurality of frequency-parameter pairings which are indicative of the at least one fault condition, the plurality of frequency-parameter pairings being determined from a plurality of potential frequency-parameter pairings for the first and second pluralities of historical power spectral densities” ([0179] Returning to FIG. 5, at block 512, asset data platform 102 may be configured to derive a model for detecting rotor imbalance based on the respective time-series sets of “harmonic mode” values for the plurality of wind turbines. In other words, asset data platform 102 may be configured to evaluate the time-series sets of “harmonic mode” values for the plurality of wind turbines (either as initially derived or in rolled-up form) and thereby derive a model for determining whether a wind turbine's “harmonic mode” values are indicative of a rotor imbalance issue. In this respect, it is expected that the “harmonic mode” values at or near the rotor frequency (e.g., the 1P.sub.x and 1P.sub.y values) for the wind turbines will be most indicative of whether the turbines have a rotor imbalance issue, but it is possible that other “harmonic mode” values may be correlated to rotor imbalance as well. In practice, the model for detecting rotor imbalance may take a variety of forms, and asset data platform 102 may derive this model in a variety of manners. [0180] asset data platform 102 may derive such a predictive model by applying a machine-learning technique (e.g., a supervised or unsupervised machine-learning technique) to training data that includes time-series sets of “harmonic mode” values for a plurality of wind turbines (perhaps in rolled-up form) that have been derived in the manner described at blocks 502-510, perhaps along with labels indicating times when the plurality of wind turbines were known to have a rotor imbalance issue. In example implementations, the machine-learning technique may involve utilizing one or more of logistic regression, support vector machine, random forest, and/or a neural network, among other possibilities) “and training, via the controller, the fault-detection model via the training data set so as to determine the plurality of frequency-parameter pairings indicative of the at least one fault condition” ([0008] the disclosed process may involve (i) a “training” phase during which a model for detecting rotor imbalance (e.g., a predictive model or a set of rules) is derived based on historical vibration data for wind turbines and perhaps other data related to the operation of the wind turbine as well (e.g., other types of historical operating data, historical environmental data, historical maintenance data, etc.), [0010] As mentioned above, in one possible implementation of the “training” phase of the process for detecting rotor imbalance disclosed herein, a data analytics platform may be configured to analyze respective sets of historical time-series vibration data for a plurality of wind turbines and use that as a basis to derive a model for detecting rotor imbalance. For instance, the data analytics platform may receive respective sets of historical time-series vibration data captured by multi-dimensional sensors at a plurality of wind turbines, where each wind turbine's historical time-series vibration data may be indicative of vibrations at the wind turbine in multiple different dimensions (e.g., side-to-side vibrations corresponding to an x-dimension and fore-to-aft vibrations corresponding to a y-dimension). In operation, this historical time-series vibration data may take a variety of forms, such as a time-series of periodic “bursts” of vibration data captured at a high sample rate (e.g., 500-second windows of vibration data that includes 100 samples per second).
In regards to Claim 19, Beckerman further teaches “The system of claim 18, wherein determining the fault probability for the industrial asset further comprises: determining a nominal distribution score for each industrial asset of the nominal population for each of the plurality of frequency-parameter pairings, the nominal distribution score being indicative of a distribution of the nominal deviation scores for each industrial asset of the nominal population within the nominal score range for each of the plurality of frequency-parameter pairings;” ([0176] asset data platform 102 may perform a temporal “roll-up” operation on the time-series set of “harmonic mode” values for the given wind turbine using a lookback window approach. One implementation of a lookback window approach may involve asset data platform 102 performing the following operations for each of a plurality of reference time points in the given wind turbine's time-series set of “harmonic mode” values: (i) obtaining the “harmonic mode” values for time points within a predetermined window of time preceding the reference time point (e.g., a two-week lookback window), (ii) applying a statistical aggregation (e.g., averaging, identifying the median value, etc.) to the obtained “harmonic mode” values for the time points within the predetermined window of time preceding the reference time point as well as the “harmonic mode” values for the reference time point, and (iii) assigning the resulting, rolled-up “harmonic mode” values to the reference time point (e.g., a rolled-up 1P.sub.x value, 1P.sub.y value, 3P.sub.x value, 3P.sub.y value, etc.).[0183] asset data platform 102 may derive such a rules-based model by identifying values or ranges of values (and/or other features indicative of rotor behavior) corresponding to wind turbines exhibiting “balanced” rotor behavior and then in turn identifying one or more deviations from these values or ranges of values (and/or other features) that correspond to wind turbines exhibiting a rotor imbalance issue, which may then be embodied into a set of rules that comprises the rules-based model for detecting rotor imbalance [0184] For instance, one possible example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when the average value of its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) over a one-week period of time exceeds a first threshold or the standard deviation of its “harmonic mode” values for a given rotor frequency (e.g., its 1P.sub.x and 1P.sub.y values) over a two-week period of time exceeds a second threshold. Another example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) exceed a threshold, such as a threshold relative to the median value of “harmonic mode” values for the given rotor frequency for multiple wind turbines for that given time (e.g., 3 standard deviations, 0.03 m/s.sup.2, etc.). Other manners of deriving a rules-based model for detecting rotor imbalance also exist; wherein a standard deviation) “determining a multi-variate nominal distribution score for each industrial asset of the nominal population based, at least in part, on the nominal distribution score for each of the plurality of frequency-parameter pairings” ([0163] asset data platform 102 may be configured to preprocess the historical vibration data so that asset data platform 102 defines the model based on historical vibration data that corresponds to the “normal” operation of the set of wind turbines. In practice, asset data platform 102 may filter each wind turbine's respective set of historical vibration data in a variety of manners; [0184] one possible example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when the average value of its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) over a one-week period of time exceeds a first threshold or the standard deviation of its “harmonic mode” values for a given rotor frequency (e.g., its 1P.sub.x and 1P.sub.y values) over a two-week period of time exceeds a second threshold. Another example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) exceed a threshold, such as a threshold relative to the median value of “harmonic mode” values for the given rotor frequency for multiple wind turbines for that given time (e.g., 3 standard deviations, 0.03 m/s.sup.2, etc.) wherein the statistical analysis used for determine standard deviations, averages, etc. for each frequency parameter pair 1Px, 1Py, 3Px, etc. to determine where thresholds are for imbalance determinations are nominal scores from nominal distributions of the normal operations) “implementing a probabilistic model to determine a multi-variate distribution of the industrial assets of the nominal population based on the corresponding multi- variate nominal distribution scores” ([0019] the data analytics platform may derive such a predictive model by applying a machine-learning technique (e.g., logistic regression, support vector machine, random forest, neural network, etc.) to training data that includes time-series sets of “harmonic mode” values for a plurality of wind turbines (perhaps in rolled-up form) that have been derived in the manner described above and historical data related to the operation of the plurality of wind turbines, perhaps along with labels indicating times when the plurality of wind turbines were known to have a rotor imbalance issue; wherein support vector machines; [0187] As one possibility, the indication may take the form of a metric reflecting a likelihood that the given wind turbine has a rotor imbalance issue. For example, this metric may take the form of a probability metric reflecting a predicted probability of a rotor imbalance issue existing at the given wind turbine that may be represented on a scale of either 0 to 100 or 0.0 to 1.0. However, the metric reflecting a likelihood that the given wind turbine has a rotor imbalance issue may take various other forms as well) “and determining a fault-probability profile for the asset family based on the probabilistic model” ([0045] an asset may have various other characteristics that more specifically define the type of asset, examples of which may include the asset's brand, make, model, vintage, and/or software version, among other possibilities. In this respect, depending on the implementation, the assets monitored by asset data platform 102 may either be of the same type or various different types. Additionally yet, the assets monitored by asset data platform 102 may be arranged into one or more “sites” (or “parks”/“farms”) of assets, which refers to any group or two or more assets that are related to one another in some manner (regardless of whether such assets are of the same type). [0187] As one possibility, the indication may take the form of a metric reflecting a likelihood that the given wind turbine has a rotor imbalance issue. For example, this metric may take the form of a probability metric reflecting a predicted probability of a rotor imbalance issue existing at the given wind turbine that may be represented on a scale of either 0 to 100 or 0.0 to 1.0. However, the metric reflecting a likelihood that the given wind turbine has a rotor imbalance issue may take various other forms as well).
Claims 4-5 are rejected under 35 U.S.C. 103 as being unpatentable over Beckerman as applied to claim 2 above, and further in view of Potter et al. (US 20180347548, hereinafter Potter).
In regards to Claim 4, Beckerman teaches the method as incorporated by claim 2 above. Beckerman further teaches “The method of claim 2, wherein the identifying at least one frequency band further comprises: identifying, via the controller, a first frequency band of the power spectral density corresponding to the parameter at which the power spectral density of the industrial asset deviates from the corresponding power spectral density for the asset family at the first frequency band” ([0179] Returning to FIG. 5, at block 512, asset data platform 102 may be configured to derive a model for detecting rotor imbalance based on the respective time-series sets of “harmonic mode” values for the plurality of wind turbines. In other words, asset data platform 102 may be configured to evaluate the time-series sets of “harmonic mode” values for the plurality of wind turbines (either as initially derived or in rolled-up form) and thereby derive a model for determining whether a wind turbine's “harmonic mode” values are indicative of a rotor imbalance issue. In this respect, it is expected that the “harmonic mode” values at or near the rotor frequency (e.g., the 1P.sub.x and 1P.sub.y values) for the wind turbines will be most indicative of whether the turbines have a rotor imbalance issue, but it is possible that other “harmonic mode” values may be correlated to rotor imbalance as well. In practice, the model for detecting rotor imbalance may take a variety of forms, and asset data platform 102 may derive this model in a variety of manners. [0184] Another example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) exceed a threshold, such as a threshold relative to the median value of “harmonic mode” values for the given rotor frequency for multiple wind turbines for that given time (e.g., 3 standard deviations, 0.03 m/s.sup.2, etc.). Other manners of deriving a rules-based model for detecting rotor imbalance also exist…asset data platform 102 may derive the model for detecting rotor imbalance based, at least in part, on comparing respective time-series sets of “harmonic mode” values of two or more neighboring wind turbines, such as two or more wind turbines that are geographically located proximate to one another and/or have experienced substantially similar meteorological conditions. Other examples are also possible; wherein the 1P harmonics are a particular frequency band and thus are the identified anomalous frequency band).
Beckerman fails to teach “and identifying, via the controller, a second frequency band of the power spectral density corresponding to the parameter at which the power spectral density of the industrial asset deviates from the corresponding power spectral density for the asset family at the second frequency band”. While Beckerman teaches that multiple frequencies bands that correspond with multiple parameters of directional vibration are used as thresholds for building a model to determine when a particular frequency band causes rotor imbalance (fault/anomaly), it fails to explicitly recite that once a frequency is found, another frequency that causes the fault/anomaly is determined.
Potter teaches “and identifying, via the controller, a second frequency band of the power spectral density corresponding to the parameter at which the power spectral density of the industrial asset deviates from the corresponding power spectral density for the asset family at the second frequency band” ([0016] The processor(s) is configured to perform one or more operations, including but not limited to identifying a plurality of harmonics in the vibration data that are indicative of bearing damage, eliminating harmonics within a specified frequency proximity to areas of high energy content representative of normal gearbox operation, determining at least one of increases in energy or a variance of the energy within each of the remaining harmonics, calculating a damage factor of the bearing as a function of a least one of the increases in energy or the variance of the energy within each of the remaining harmonics, and comparing the damage factor to a predetermined damage threshold, wherein a damage factor exceeding the predetermined damage threshold is indicative of a damaged bearing. It should also be understood that the system may include any of the additional features and/or may be configured to implement any of the steps described herein.[0052] With the harmonics within a specified frequency proximity to areas of high energy content representative of normal gearbox operation being eliminated, the controller 26 can then further analyze the remaining harmonics 80 to assess bearing damage. For example, still referring to FIG. 5, the controller 26 is configured to evaluate increases in energy for a given harmonic 80. [0053] More specifically, the controller 26 can determine the increases in the energy within each of the remaining harmonics 80 by summing the square of each individual data point of the frequency signals within a given harmonic 80 to obtain a sum of the squares value. In addition, the controller 26 can determine the variance of the energy within each of the remaining harmonics 80 by calculating a mean for each of the remaining harmonics 80 and then, utilizing the mean, calculating the variance for each of the remaining harmonics 80).
It would have been obvious to a person having ordinary skill in the art before the effective file date of the claimed invention to have modified the method of identifying a frequency range for which a harmonic is causing an anomaly, and which has data on multiple harmonic frequency bands/ranges, with the use of the system that analyzes each of the harmonics individually so that each harmonic that contributes to a fault/anomaly is identified on the basis of that harmonic’s energy in accordance with potential damage as taught by Potter, because it would gain the obvious benefit of having information on which of a plurality of harmonics led to the abnormality/fault determination as their contribution to the overall factor. Furthermore, by incorporating these features, Beckerman would gain the stated benefit of Potter, namely “[0004] Detection of damaged bearings in a wind turbine is essential in minimizing unplanned downtime of the turbine and increasing turbine availability” and thus both Beckerman and Potter are related in that they seek to avoid unplanned downtime/maintenance for a turbine. By combining these elements, it can be considered taking the known method of identifying a first frequency band of a spectral analysis as a harmonic mode for indicating when a particular turbine is abnormal/faulting in comparison with the neighboring assets, and incorporating the known technique of identifying a second frequency band at which a harmonic energy is indicative of abnormal/faulty operation in a known way that achieves predictable results.
In regards to Claim 5, Beckerman teaches the method as incorporated by claim 2 above. Beckerman further teaches “The method of claim 2, wherein the parameter of the industrial asset is a first parameter of the industrial asset, wherein the at least one power spectral density comprises a first power spectral density corresponding to the first parameter and a second power spectral density corresponding to a second parameter of the industrial asset, and wherein identifying at least one frequency band further comprises” ([0013] As one possible implementation of this function, the data analytics platform may take each respective “burst” within the historical time-series vibration data for a given wind turbine and then perform a Fourier transform of the particular “burst” in each dimension measured by the given wind turbine's multi-dimensional sensor (e.g., fore-to-aft and side-to-side). This function results in a representation of the particular “burst” in the frequency domain (for each dimension) that includes a set of peaks spaced out at different frequency ranges corresponding to different “harmonic modes” (e.g., integer multiples of the given wind turbine's rotor frequency, such as 1P, 3P, etc.). Once in the frequency domain, the data analytics platform may apply a statistical aggregation (e.g., averaging, identifying the median value, etc.) to the Fourier coefficient values of the particular “burst” within each respective frequency range and for each dimension to derive a set of “harmonic mode” values for the particular “burst” (e.g., a 1P.sub.x value, 3P.sub.x value, etc. for the side-to-side dimension and a 1P.sub.y value, 3P.sub.y value, etc. for the fore-to-aft dimension); wherein the x and y or fore to aft and side to side are different parameters) “identifying, via the controller, a first frequency band of the first power spectral density at which the first power spectral density deviates from the corresponding power spectral density for the asset family at the first frequency band” ([0179] Returning to FIG. 5, at block 512, asset data platform 102 may be configured to derive a model for detecting rotor imbalance based on the respective time-series sets of “harmonic mode” values for the plurality of wind turbines. In other words, asset data platform 102 may be configured to evaluate the time-series sets of “harmonic mode” values for the plurality of wind turbines (either as initially derived or in rolled-up form) and thereby derive a model for determining whether a wind turbine's “harmonic mode” values are indicative of a rotor imbalance issue. In this respect, it is expected that the “harmonic mode” values at or near the rotor frequency (e.g., the 1P.sub.x and 1P.sub.y values) for the wind turbines will be most indicative of whether the turbines have a rotor imbalance issue, but it is possible that other “harmonic mode” values may be correlated to rotor imbalance as well. In practice, the model for detecting rotor imbalance may take a variety of forms, and asset data platform 102 may derive this model in a variety of manners. [0184] Another example of a rules-based model may be configured to detect a rotor imbalance at a given wind turbine when its “harmonic mode” values for a given rotor frequency (e.g., its 1P, and 1P.sub.y values) exceed a threshold, such as a threshold relative to the median value of “harmonic mode” values for the given rotor frequency for multiple wind turbines for that given time (e.g., 3 standard deviations, 0.03 m/s.sup.2, etc.). Other manners of deriving a rules-based model for detecting rotor imbalance also exist; wherein the 1P harmonics are a particular frequency band and thus are the identified anomalous frequency band).
Beckerman fails to teach “and identifying, via the controller, a second frequency band of the second power spectral density at which the second power spectral density deviates from the corresponding power spectral density for the asset family at the second frequency band”. While Beckerman teaches that multiple frequencies bands that correspond with multiple parameters of directional vibration are used as thresholds for building a model to determine when a particular frequency band causes rotor imbalance (fault/anomaly), it fails to explicitly recite that once a frequency is found, another frequency that causes the fault/anomaly is determined.
Potter teaches “and identifying, via the controller, a second frequency band of the power spectral density corresponding to the parameter at which the power spectral density of the industrial asset deviates from the corresponding power spectral density for the asset family at the second frequency band” ([0016] The processor(s) is configured to perform one or more operations, including but not limited to identifying a plurality of harmonics in the vibration data that are indicative of bearing damage, eliminating harmonics within a specified frequency proximity to areas of high energy content representative of normal gearbox operation, determining at least one of increases in energy or a variance of the energy within each of the remaining harmonics, calculating a damage factor of the bearing as a function of a least one of the increases in energy or the variance of the energy within each of the remaining harmonics, and comparing the damage factor to a predetermined damage threshold, wherein a damage factor exceeding the predetermined damage threshold is indicative of a damaged bearing. It should also be understood that the system may include any of the additional features and/or may be configured to implement any of the steps described herein.[0052] With the harmonics within a specified frequency proximity to areas of high energy content representative of normal gearbox operation being eliminated, the controller 26 can then further analyze the remaining harmonics 80 to assess bearing damage. For example, still referring to FIG. 5, the controller 26 is configured to evaluate increases in energy for a given harmonic 80. [0053] More specifically, the controller 26 can determine the increases in the energy within each of the remaining harmonics 80 by summing the square of each individual data point of the frequency signals within a given harmonic 80 to obtain a sum of the squares value. In addition, the controller 26 can determine the variance of the energy within each of the remaining harmonics 80 by calculating a mean for each of the remaining harmonics 80 and then, utilizing the mean, calculating the variance for each of the remaining harmonics 80).
It would have been obvious to a person having ordinary skill in the art before the effective file date of the claimed invention to have modified the method of identifying a frequency range for which a harmonic is causing an anomaly, and which has data on multiple harmonic frequency bands/ranges, with the use of the system that analyzes each of the harmonics individually so that each harmonic that contributes to a fault/anomaly is identified on the basis of that harmonic’s energy in accordance with potential damage as taught by Potter, because it would gain the obvious benefit of having information on which of a plurality of harmonics led to the abnormality/fault determination as their contribution to the overall factor. Furthermore, by incorporating these features, Beckerman would gain the stated benefit of Potter, namely “[0004] Detection of damaged bearings in a wind turbine is essential in minimizing unplanned downtime of the turbine and increasing turbine availability” and thus both Beckerman and Potter are related in that they seek to avoid unplanned downtime/maintenance for a turbine. By combining these elements, it can be considered taking the known method of identifying a first frequency band of a spectral analysis as a harmonic mode for indicating when a particular turbine is abnormal/faulting in comparison with the neighboring assets, and incorporating the known technique of identifying a second frequency band at which a harmonic energy is indicative of abnormal/faulty operation in a known way that achieves predictable results.
Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over Beckerman as applied to claim 19 above, and further in view of Yeung et al. (US 20220155771, hereinafter Yeung).
In regards to Claim 20, Beckerman teaches the system as incorporated by claim 19 above.
Beckerman fails to teach “The system of claim 19, further comprising: fitting a receiver-operating-characteristic curve (ROC-curve) to a mean distance of a distribution of the industrial assets of the fault population relative to the multi-variate distribution of the industrial assets of the nominal population as indicated by the fault-probability profile for the asset family, wherein the ROC-curve corresponds to the fault threshold”.
Yeung teaches “The system of claim 19, further comprising: fitting a receiver-operating-characteristic curve (ROC-curve) to a mean distance of a distribution of the industrial assets of the fault population relative to the multi-variate distribution of the industrial assets of the nominal population as indicated by the fault-probability profile for the asset family, wherein the ROC-curve corresponds to the fault threshold” ([0136] For training, in one non-limiting embodiment, 30,000 time steps of data under healthy mode can be used. [0137] The baseline LSTM model can achieve satisfactory AD&HM results with the proposed dataset. For example, the raw prediction of LSTM.sub.1 can be visualized. FIG. 16 showcases five streams of model outputs against their respective ground truth signals. LSTM.sub.1 predicts the change in the upcoming signal timely. Especially remarkable are its good prediction of the transition between stationary and rotating plant, and the periodic signals, regarding both frequency and amplitude, as the plant rotates. By way of further example, LSTM.sub.1 can be applied for actual anomaly detection. For the data sequence whose macro health mode is shown in graph (A) of FIG. 18, the mean-squared-error (MSE) between LSTM.sub.1 output and the incoming signal is shown in graph (B) of FIG. 18. The distance threshold d.sub.th can be selected based, at least in part, on the receiver operating characteristic (ROC) curve for LSTM, shown in FIG. 17; wherein a healthy mode is a nominal population while anomalous modes are a fault population).
It would have been obvious to a person having ordinary skill in the art before the effective file date of the claimed invention to have modified the system for determining anomalies/faults using a machine learning model trained on sets of nominal harmonic information and abnormal harmonic information at each harmonic frequency range as taught by Beckerman, to utilize the method of Yeung of using an ROC curve for distinguishing the threshold between healthy/nominal and abnormal/faulty conditions on the basis of average distances as taught by Yeung, because it would gain the stated benefit of Yeung, namely an improved efficiency in identifying anomalies in periodic/rotational signals ([0002]). By combining these elements, it can be considered taking the known method of using an ROC curve to determine a threshold on the basis of healthy/anormal datasets, and using it to improve the known anomaly detection system of Beckerman in a similar way that ends in a predictable result.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Li et al. (US 20210036656) – teaches an arc-fault determination method from frequency analysis of multi-variate parameters around frequency bands
Cole (US 10551274) – teaches a multi-variable spectral analysis method for determining abnormal and faulty leaks on the basis of multi-dimensional sensor data
Okanohara et al. (US 20180365089) – teaches an abnormality score determination from multi-variate signals using training data that reflects normal operation
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JONATHAN M SKRZYCKI whose telephone number is (571)272-0933. The examiner can normally be reached M-Th 7:30-3:30.
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, Ken Lo can be reached at 571-272-9774. 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.
/JONATHAN MICHAEL SKRZYCKI/Examiner, Art Unit 2116