Systems and methods for providing enhanced electrical characteristic analysis for fault detection
By using adaptive machine learning algorithms to capture and analyze electrical signals and thermal data of motors in real time, this method solves the problem of fault detection in motors such as variable frequency drive motors, and achieves efficient and sensitive fault detection and automatic threshold setting. It is applicable to variable frequency drive motors and other motors.
Patent Information
- Application Number
- CN202210245643.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2021-03-15
- Filing Date
- 2022-03-14
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2042-03-14
AI Technical Summary
Existing electrical characteristic analysis methods are difficult to effectively detect faults in electrical machines such as variable frequency drive motors, especially when the operating frequency changes, and it is difficult to automatically set fault thresholds.
An adaptive machine learning algorithm is used to capture electrical signals and thermal data of electric machines in real time, convert them into frequency domain data, use the frequency domain data and machine learning algorithm to determine fault thresholds, monitor and generate fault indicators and warnings in real time, and automatically adjust the thresholds to adapt to the specific characteristics of electric machines.
It improves the sensitivity of fault detection, reduces the complexity of manually setting thresholds, enables the application of a unified threshold across multiple electrical machines, adapts to changes in electrical machines, and reduces computational and data storage requirements.
Smart Images

Figure CN115078828B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to power systems, and more particularly to systems and methods for providing enhanced electrical signature analysis for fault detection. Background Technology
[0002] Current electrical characteristic analysis (ESA) methods may struggle to detect faults in certain types of electrical machinery, such as variable frequency drive (VFD) motors, due to their varying operating frequencies. Operators may also find it difficult to manually establish fault thresholds for each of the fault types associated with the machinery, such as bearing, mechanical, and stator faults. There are no guidelines or standards available for ESA to configure thresholds for these diverse fault types. Therefore, setting thresholds for each individual machinery type within a fleet of machines can be a cumbersome task. The established thresholds must be uniform, independent of the machine type, application, and field conditions. Furthermore, when variable speed drive (VDS) motors are involved, ESA must be more sensitive to fault detection based on motor condition deterioration and application. Consequently, current ESA methods may be inadequate and / or inefficient in detecting faults in VFD motors, as well as some other types of electrical machinery and certain types of oscillating loads on those machines. Summary of the Invention
[0003] The present invention provides a set of technical solutions, as follows.
[0004] Technical Solution 1. A system comprising: Electrical machinery; and A computer processor, operable to execute a set of computer-readable instructions; and A memory operable to store the set of computer-readable instructions operable to: Real-time electrical signal data and thermal data associated with the electromechanical device are captured during baseline mode; Convert the real-time electrical signal data into frequency domain data; Baseline clustering of one or more peak dB values and one or more root mean square dB values of the captured electrical and thermal data and a portion of the frequency domain data included in the band, using the frequency domain data in real time and adaptive or incremental machine learning algorithms; One or more fault thresholds are determined as a combination of clustering distances from baseline data clusters of the one or more peak decibel values and the one or more root mean square decibel values, and associated electrical and thermal data. The real-time electrical signals and thermal data are captured during monitoring mode; Based on the real-time electrical signal, thermal data, one or more peak decibel values from frequency domain data, and one or more root mean square decibel values, the fault index of the motor is determined as a combination of electrical, thermal, and frequency domain data clustering distances. The fault index of the motor is determined to be higher than one or more fault thresholds established during the baseline mode; Generate a warning when the fault indicator of the motor exceeds one or more fault thresholds; and Control actions are performed based on the determination that the fault index of the motor is higher than one or more fault thresholds.
[0005] Technical Solution 2. The system as described in Technical Solution 1, wherein the processor is further configured to determine the number of frequency values to be included in the band based on the amplitude of the frequency values.
[0006] Technical Solution 3. The system as described in Technical Solution 1, wherein the frequency domain data is divided into a low range of frequency values, a medium range of frequency values, and a high range of frequency values, and the processor is further configured to determine the number of frequency values to be included in the band based on the amplitude of the frequency values, wherein the number of frequency values included in the band is greater in the medium range of frequency values than in the low range of frequency values, and the number of frequency values included in the band is greater in the high range of frequency values than in the medium range of frequency values.
[0007] Technical Solution 4. The system as described in Technical Solution 1, wherein the processor is further configured to use the one or more peak decibel values and one or more root mean square decibel values, as well as the associated electrical and thermal data, to determine the covariance matrix.
[0008] Technical Solution 5. The system as described in Technical Solution 1, wherein the processor is further configured to perform outlier filtering on the frequency domain data included in the band based on at least one of the following: the amplitude of the frequency included in the band, the power supply device or load type of the motor, the application of the motor, or the presence of load oscillation.
[0009] Technical Solution 6. The system as described in Technical Solution 1, wherein determining one or more fault thresholds is further based on specific characteristics of the electric machine, such as nameplate information, design data, supply device or load conditions, the life and condition of the electric machine during the baseline period, or the frequency range of the fault under diagnosis.
[0010] Technical Solution 7. The system as described in Technical Solution 1, wherein the fault threshold is established as a normalized distance from a baseline cluster of one or more peak decibel values and one or more root mean square decibel values, and the associated electrical and thermal data, and wherein the fault threshold is applied to multiple different electrical machines in a team of electrical machines.
[0011] Technical Solution 8. The system as described in Technical Solution 1, wherein the machine learning algorithm includes any clustering-based technique.
[0012] Technical Solution 9. The system as described in Technical Solution 1, wherein the electric machine is a variable frequency drive (VFD) machine.
[0013] Technical Solution 10. The system as described in Technical Solution 1, wherein the baseline clustering for determining one or more peak decibel values and one or more root mean square decibel values, as well as the associated electrical and thermal data, is performed with historical data stored.
[0014] Technical Solution 11. A method comprising: Real-time electrical signal and thermal data associated with the electromechanical device are captured during baseline mode; Convert the real-time electrical signal data into frequency domain data; Baseline clustering of one or more peak dB values and one or more root mean square dB values of the captured electrical and thermal data and a portion of the frequency domain data included in the band, using the frequency domain data in real time and adaptive or incremental machine learning algorithms; One or more fault thresholds are determined as a combination of clustering distances from baseline data clusters of the one or more peak decibel values and the one or more root mean square decibel values, and associated electrical and thermal data. The real-time electrical signals and thermal data are captured during monitoring mode; Based on the real-time electrical signal data, thermal data, one or more peak decibel values from frequency domain data, and one or more root mean square decibel values, the fault index of the motor is determined as a combination of electrical, thermal, and frequency domain data clustering distance. The fault index of the motor is determined to be higher than one or more fault thresholds established during the baseline mode; Generate a warning when the fault indicator of the motor exceeds one or more fault thresholds; and Control actions are performed based on the determination that the fault index of the motor is higher than one or more fault thresholds.
[0015] Technical Solution 12. The method of Technical Solution 11, further comprising determining the number of frequency values to be included in the band based on the amplitude of the frequency values.
[0016] Technical Solution 13. The method of Technical Solution 11, wherein the frequency domain data is divided into a low range of frequency values, a medium range of frequency values, and a high range of frequency values, and the method further includes determining the number of frequency values to be included in the band based on the magnitude of the frequency values, wherein the number of frequency values included in the band is greater in the medium range of frequency values than in the low range of frequency values, and the number of frequency values included in the band is greater in the high range of frequency values than in the medium range of frequency values.
[0017] Technical Solution 14. The method of Technical Solution 11 further includes using the one or more peak decibel values and one or more root mean square decibel values, as well as the associated electrical and thermal data, to determine a covariance matrix.
[0018] Technical Solution 15. The method of Technical Solution 11 further includes performing outlier filtering on the frequency domain data included in the band based on at least one of the following: the amplitude of the frequency included in the band, the power supply device or load type of the motor, the application of the motor, or the presence of load oscillation.
[0019] Technical Solution 16. The method according to Technical Solution 11, wherein determining one or more fault thresholds is further based on specific characteristics of the electric machine, such as nameplate information, design data, supply device or load conditions, the life and condition of the electric machine during the baseline period, or the frequency range of the fault under diagnosis.
[0020] Technical Solution 17. The method of Technical Solution 11, wherein the fault threshold is established as a normalized distance from a baseline cluster of one or more peak decibel values and one or more root mean square decibel values, and the associated electrical and thermal data, and wherein the fault threshold is applied to multiple different motors in a group of motors.
[0021] Technical Solution 18. The method as described in Technical Solution 11, wherein the machine learning algorithm includes any clustering-based technique.
[0022] Technical Solution 19. The method as described in Technical Solution 11, wherein the electric machine is a variable frequency drive (VFD) machine.
[0023] Technical Solution 20. The method of Technical Solution 11, wherein baseline clustering for determining one or more peak decibel values and one or more root mean square decibel values, along with associated electrical and thermal data, is performed while storing historical data. Attached Figure Description
[0024] A detailed description is illustrated with reference to the accompanying drawings. The drawings are provided for illustrative purposes only and depict only exemplary embodiments of the present disclosure. The drawings are provided to facilitate understanding of the present disclosure and should not be construed as limiting the breadth, scope, or applicability of the disclosure. In the drawings, the leftmost one or more numerals of the reference numerals identify the drawing in which the reference numeral first appears. The same reference numerals are used to indicate similar, but not necessarily identical or identical, components. However, different reference numerals may also be used to identify similar components. Various embodiments may utilize elements or components other than those illustrated in the drawings, and some elements and / or components may not be present in various embodiments. Depending on the context, the use of singular terms to describe an component or element may include a plural of such components or elements, and vice versa.
[0025] Figure 1 A schematic diagram depicting an example system according to one or more example embodiments of the present disclosure.
[0026] Figure 2 Depicts an example frequency domain plot according to one or more example embodiments of this disclosure.
[0027] Figure 3 Example flowcharts are depicted according to one or more example embodiments of this disclosure.
[0028] Figure 4 Example flowcharts are depicted according to one or more example embodiments of this disclosure.
[0029] Figure 5 An example cluster plot is depicted according to one or more example embodiments of this disclosure.
[0030] Figure 6 Example methods are described according to one or more example embodiments of this disclosure.
[0031] Figure 7 A schematic diagram depicting an example computing system and apparatus architecture according to one or more example embodiments of the present disclosure. Detailed Implementation
[0032] This disclosure relates in particular to enhanced electrical characteristic analysis (ESA) for fault detection in electrical machinery. The electrical machinery can be, for example, an induction motor or a synchronous motor, as well as any other type of electrical machinery. In some cases, enhanced ESA can be particularly beneficial for fault detection in variable frequency drive (VFD) motors (however, it can also be applied to other types of electrical machinery as well as those with oscillating loads). For certain types of electrical machinery (e.g., VFD motors), previous ESA methods may have difficulty in detecting faults because VFDs can operate at varying frequencies. Operators may also find it difficult to manually establish fault thresholds to be used for fault detection in such machines (e.g., bearing, mechanical, and stator faults, to name a few). Therefore, it may be advantageous to employ algorithms capable of adaptively learning the behavior of a particular electrical machine and automatically establishing fault thresholds for the machine without requiring manual input from the operator. The specific algorithms described herein for achieving this can utilize machine learning, which can be used to model the behavior of the electrical machine in real time and based on any properties specific to the electrical machine. In some instances, modeling the behavior of electrical machinery may involve creating one or more clusters of baseline data associated with the "healthy" operation of the machinery (e.g., the operation of the machinery when it has not experienced a fault). In such instances, the type of machine learning can include any clustering-based technique, such as Mahalanobis distance (MD). Figure 5 A more detailed illustration of clustering implemented using Mahalanobis distance can be provided. However, machine learning algorithms can typically be computationally intensive and may require large amounts of input data to produce output. The algorithm described in this paper, however, can be implemented on embedded devices in power systems (e.g., protective relays in power systems including electric machines) that may not be able to store large amounts of data. To illustrate this, the algorithm incrementally builds the machine learning model using real-time data rather than relying on stored historical data.
[0033] In some embodiments, the algorithm may more specifically relate to a method for diagnosing electrical machine faults by deriving machine fault indicators (e.g., peak and energy decibel (dB) values) based on data captured from the electrical machine (e.g., spectrum, temperature data from different subsystems of the machine, and calculated electrical data, such as current imbalance, rate of change in current frequency, power factor, etc.). The data may be captured by a protective relay (or any intelligent electronic or edge device on which the algorithm is implemented). The data may be captured in the time domain but may be converted to the frequency domain for analysis. In some instances, a Fast Fourier Transform (FFT) or any other suitable method may be used to perform the data conversion from the time domain to the frequency domain. Once the data from the electrical machine has been converted to the frequency domain, the algorithm may involve performing baselineization to establish one or more baseline thresholds for different types of faults the electrical machine may experience (as described above, these baselines may include, for example,...). Figure 5 (Data clustering as depicted in the text). Baseline formation can be performed incrementally or adaptively based on real-time data capture. Once baseline formation is complete, the algorithm can switch to monitoring mode, in which data generated by the motor can be compared with one or more baseline thresholds established during baseline formation to identify the presence of a certain type of fault in the motor.
[0034] During the baseline mode of operation, the health behavior of the motor can be characterized for specific operating states of the motor. Operating states may refer to, for example, load, frequency, voltage / current imbalance, etc. Data can be captured during the monitoring mode of operation, as described below, under the same or similar operating states to allow comparison between data obtained during the baseline mode and the monitoring mode of operation. Additionally, during the baseline mode of operation, fault thresholds can be established (e.g., fault thresholds can be established for different types of faults that can occur in the motor). Learning health behavior may refer to acquiring data that characterizes the behavior of a healthy motor. Once the algorithm learns the health behavior of the motor, it can calculate one or more fault thresholds based on the learned healthy motor data. As mentioned above, the fault threshold for a given type of fault can be based on, for example... Figure 5 The diagram illustrates the clustering of baseline data. The calculated fault threshold(s) can be adapted to a specific electric machine. Calculating the fault threshold(s) may involve using data from the electric machine's frequency domain to obtain the peak and energy (which can be referred to as the area under a frequency plot within the boundaries of the frequency band) in decibels (dB) for a given type of fault within an established sumband. Figure 2(See further detailed illustrations), along with other associated electrical and thermal data. The band can refer to a range of frequency domain data that can be sampled to identify peak and energy dB values for a given type of fault. That is, different types of faults can typically be associated with certain frequencies and / or frequency ranges, and the band can be a range of data points that can encompass certain frequencies or frequency ranges. Bands can be established to include a sufficiently large frequency range to capture a sufficient number of data points around the frequencies or frequency ranges typically associated with the fault type, but not large enough to capture other data points that may be unrelated to the fault type. In some cases, the size of the band (e.g., the resolution of data capture) can vary for a given type of fault. See also... Figure 2 To describe in more detail.
[0035] In some embodiments, outliers and covariance can also be established based on [the data set]. Outliers can refer to extreme data points, and establishing such outliers can be used to eliminate extreme data points or any incorrectly collected points, so that any statistical calculations on the data are unaffected by outliers. Covariance helps to check whether data points are increasing or decreasing in the same direction, such that they change according to the failure degradation process. These data points that change together indicate better clustering.
[0036] In some embodiments, the algorithm may also involve an operational monitoring mode that occurs once a baselined mode of operation has been completed for a given electrical machine. During the monitoring mode, one or more fault indicators may be calculated and compared to one or more fault thresholds established during the baselined mode. That is, the protective relay may continue to capture data from the electrical machine in real time, and the real-time data may continue to be used to determine peak and energy dB values in the frequency domain. Based on this comparison, it may be determined whether one or more faults exist in the electrical machine. For example, if a fault indicator associated with a given type of fault is greater than a threshold established for that type of fault, a fault may be determined to exist. When a fault indicator exceeds the established threshold, an indication of the fault may be provided and / or an alarm may be triggered. In some cases, a fault may only be determined to exist if a fault indicator consistently exceeds a threshold (e.g., if the fault indication exceeds the threshold a given number of times or for a given time length).
[0037] In some embodiments, if a fault is identified in a given electrical machine during monitoring mode, action may be taken. First example actions may include providing notification to the operation in the form of an alarm, warning, or any other type of notification. Second example actions may include performing changes to the power system (e.g., shutting down and / or turning the electrical machine on) or performing any other actions concerning the electrical machine and / or any other power assets in the system, such as repair, maintenance actions, or planning for future maintenance actions.
[0038] In some embodiments, the algorithm described herein may include several benefits. For example, because the algorithm characterizes the behavior of a particular electric machine, it can have high fault detection sensitivity compared to conventional methods used for VFD machines or machines with load oscillations. The algorithm also allows operators to avoid setting up any complex pickup, alarm, and delay settings and can be adapted to the specific electric machine to which it is deployed. The algorithm can be computationally simple and incrementally constructed, thus mitigating the large data requirements associated with machine learning algorithms (which can incrementally build models using live data captured from the electric machines). The algorithm also allows fault thresholds to be established across a group of electric machines (i.e., thresholds applicable to multiple different electric machines can be established). That is, machine-specific thresholds can be calculated, but the range or boundaries of thresholds indicating fault maturity or component degradation can also be common across a group of machines. The algorithm can also be automated to relearn healthy electric machine behavior when the hardware used for the electric machine is changed or repaired. The completion of the learning process can also be determined by the algorithm itself, whereas conventional algorithms may determine completion based on the amount of time elapsed for the threshold.
[0039] Turn to the attached diagram. Figure 1 The illustration depicts an example system 100 according to an embodiment of this disclosure. System 100 may be a power system including one or more power assets. The one or more power assets may include at least a protection relay 102, a power supply device 104, an electrical machine 106, and / or a load 108. The protection relay may be a smart protection relay, providing high-performance protection, high-density input and output (I / O), extensive programmable logic, and flexible configuration capabilities. Utilizing protection and control logic, such a protection relay 102 may allow for simplified coordination with upstream and downstream disconnecting devices (e.g., one or more electrical machines 106 within system 100). The protection relay 102 itself may also have the capability to monitor and capture data generated by one or more electrical machines 106. However, it should be noted that in some cases, the algorithm may be implemented on a remote server rather than on a local device such as the protection relay 102. That is, data may be captured and transmitted to a remote server for analysis. In some cases, data may be captured in the time domain and converted to the frequency domain for analysis (illustrated as frequency domain plot 110, which can be referenced). Figure 2 (Description in more detail). The power supply device 104 may be a power supply device connected to the electric motor 102, which can be used to provide power to the electric motor 102. The electric motor 106 may be, for example, an induction motor or a synchronous motor, and any other type of electric motor. In some cases, the electric motor may specifically involve a variable frequency drive (VFD) motor. The load 108 may be any system powered by the electric motor.
[0040] Figure 2 Example frequency domain plot 200 is depicted. Frequency domain plot 200 can illustrate examples of data received from an electrical machine that has been converted to the frequency domain (e.g., power or current amplitude at various fault frequencies). The x-axis of frequency domain plot 200 can include frequency values (which may be Hertz (Hz)), and the y-axis of frequency domain plot 200 can include amplitude (which may be decibels (dB)). That is, the amplitude of different frequency values can be depicted in frequency domain plot 200. As mentioned above, the data can be generated by a protective relay or similar device (e.g., reference...). Figure 1 The described protective relay 102 captures the data. The frequency domain plot 200 may include a fundamental frequency 203, which may be the fundamental frequency of the electrical machine from which it captures input data (e.g., electrical machine 106 or other electrical machines). In the example frequency domain plot 200 depicted in the figure, the fundamental frequency of the electrical machine from which it captures input data may be 60 Hz. The plot may also depict one or more additional fault frequencies (shown as -Fe2, -Fe1, +Fe1, +Fe2, etc.), which may be frequencies at which certain types of faults can be identified by the amplitude of the plot at that particular frequency. Although not depicted in the figure, frequency domain data may be divided into one or more frequency bands, where each frequency band includes a range of frequency values. In some cases, frequency bands may include a "low frequency" band, a "medium frequency" band, and a "high frequency" band. For example, a low frequency band may include frequencies of 100 Hz or below, a medium frequency band may include frequencies between 100 Hz and 800 Hz, and a high frequency band may include frequencies above 800 Hz (this is merely illustrative). However, any other number of frequency bands including any frequency range may also be used. In some cases, different types of faults that an electrical machine may experience can fall within different frequency bands. For example, broken rotor bars and / or mechanical faults can typically fall within the low-frequency band, bearing faults within the mid-frequency band, and stator faults within the high-frequency band. These are merely illustrative examples, and any and / or any other fault type mentioned can fall within any other frequency band.
[0041] In some embodiments, the frequency domain plot 200 may also depict a sum band 202 including a lower limit 204 and an upper limit 206. The sum band 202 may refer to a range of frequency domain data that can be sampled to identify peak and energy dB values for a given type of fault. That is, different types of faults may typically be associated with certain frequencies and / or frequency ranges, and the sum band 202 may be a range of data points that may contain certain frequencies or frequency ranges. The sum band 202 may be established to include a sufficiently large frequency range to capture a sufficient number of data points around the frequencies or frequency ranges typically associated with the fault type, but not large enough to capture other data points that may be unrelated to the fault type. In some instances, the size of the sum band (e.g., the position of the lower limit 204 and upper limit 206 relative to each other) may depend on the frequency band. In other words, the data resolution captured by the sum band may be dynamic and varies based on the frequency band in which the sum band is created. In some cases, the size of the sum band 202 (the captured data resolution) may be increased for higher frequencies. As a non-limiting example, the data resolution in the mid-frequency band may be twice that in the low-frequency band. Continuing with this non-limiting example, the data resolution in the high-frequency band can be twice that in the mid-frequency band. That is, more data points can be included in the mid-frequency band 202 compared to the low-frequency band 202, and more data points can be included in the high-frequency band 202 compared to the mid-frequency band 202.
[0042] In some embodiments, outliers and covariance can also be established based on [the data set]. Outliers can refer to extreme data points, and establishing such outliers can be used to eliminate extreme data points or any incorrectly collected points, so that any statistical calculations on the data are not affected by outliers. Covariance helps to check whether data points are increasing or decreasing in the same direction, so that they change according to the failure degradation process. These data points that change together indicate better clustering.
[0043] Figure 3 Example flowchart 300 is depicted. Flowchart 300 can be a high-level illustration of the operations involved in the electrical machine fault detection algorithm described herein. In some cases, flowchart 300 can be provided to fault analysis algorithm 306 (which can be localized to the power system via devices such as those related to fault analysis algorithms). Figure 1 The described protective relay 102, implemented in a remote device or any other type of device, supplies one or more inputs to initiate operation. In some cases, these one or more inputs may include those from an electrical machine (such as...). Figure 1The described electric machine 106 (or any other electric machine or multiple electric machines) captures data 302, and motor-specific data 304, such as motor supply type and / or motor application. Fault analysis algorithm 306 may involve a baselineization mode 308 and / or a monitoring mode 316 of operation. In some cases, the baselineization mode 308 of operation may occur before the monitoring mode 316 of operation. The baselineization mode 308 of operation may involve operation 310. Operation 310 may include peak and energy dB calculations. Peak and energy dB calculations may be performed in the frequency domain on data captured from the electric machine in real time. After operation 310, flowchart 300 may proceed to operation 312. Operation 312 may involve an adaptive learning period. The adaptive learning period may, for example, include performing baselineization to establish one or more baseline thresholds for different types of faults the electric machine may experience (as described above, these baselines may include, for example...). Figure 5 (Data clustering as depicted). Baselineization can be performed incrementally or adaptively based on real-time data capture. Additionally, operation 312 may involve establishing outliers. Operation 312 may involve adaptively determining what the learning period should be based on the collected data, and may also involve determining the outlier band of the data, within which only data is acceptable for computation during the learning period or baseline mode. Following operation 312, flowchart 300 may proceed to operation 314, which may involve establishing a baseline data cluster that can be used to create a fault threshold. This fault threshold can then be used during monitoring mode 316 of the operation to identify whether the machine is experiencing different types of faults. The same outlier band identified during baseline mode can also be used to filter data during monitoring mode.
[0044] In some embodiments, after the baseline mode 308 of operation, flowchart 300 may begin the monitoring mode 316 of operation. The monitoring mode of operation may begin with operation 318, which may also involve peak and energy dB calculations. Similar to the baseline mode 308 of operation, the peak and energy dB calculations in the monitoring mode 316 of operation can be performed in real time in the frequency domain on data captured from the electromechanical system. After operation 318, flowchart 300 may proceed to operation 320, which may involve filtering the data acquired during the monitoring mode 316 of operation (e.g., using outlier bands identified during the baseline mode). After operation 320, flowchart 300 may proceed to operation 322. Operation 322 may involve determining the distance between the data clusters acquired through real-time data capture during the monitoring mode 316 of operation and the baseline clusters (and corresponding thresholds) established during the baseline mode operation 308. This distance may be, for example, Mahalanobis distance (MD) (or may be based on any other machine learning-based clustering technique), such as... Figure 5(As illustrated in the figure). Based on this distance, it can be determined whether a fault exists in the electrical machine (for example, this distance can be used to determine whether the data clustering obtained during the monitoring mode 316 of operation exceeds a threshold based on the data clustering obtained during the baseline mode 308 of operation).
[0045] Figure 4 An example flowchart 400 is depicted according to an embodiment of the present disclosure. Flowchart 400 may be available in... Figure 3 The flowchart 400 is an extension of the flowchart 300 depicted in the diagram. Flowchart 400 may illustrate one or more operations performed during baseline mode 401 and monitoring mode 402. Baseline mode 401 may occur to capture data during the "health" operation of the machine, which can be used to establish thresholds that can subsequently be used during monitoring mode 402 to identify the presence of any faults in the machine. References may be used. Figure 1 The described protective relay 102 is used to collect data. The data captured during baseline mode 401 and monitoring mode 402 can be real-time data and can be used to build data clusters using machine learning-based clustering techniques (e.g., Mahalanobis distance (MD)).
[0046] In some embodiments, flowchart 400 may begin at operation 404. Operation 404 may involve capturing input data. Input data may include, for example, a spectrum, temperature data from different subsystems of the machine, and calculated electrical data, such as current imbalance, rate of change in current frequency, power factor, etc. Flowchart 400 may then proceed to operation 406, which may involve applying a data filter. After operation 406, flowchart 400 may proceed to state 408. State 408 may involve determining whether baselineization is complete. If baselineization is determined to be complete in state 408, flowchart 400 may proceed to operation 420, which may be part of monitoring mode 402. If baselineization is determined not to be complete in state 408, flowchart 400 may proceed to operation 410, which may be part of baselineization mode 401. In some cases, baseline mode completion of the operation may be determined based on the minimum number of data samples being collected or based on user configuration settings. Operation 410 may involve incrementally updating machine learning model parameters using the current input data. For example, the machine learning model can be incrementally updated using new (current) data samples until the baseline period is completed or until a minimum number of data samples are collected. After operation 410, flowchart 400 can proceed to operation 412. Operation 412 may involve storing machine learning model parameters. Machine learning model parameters may include machine learning model characteristics that can be learned based on baseline data and model type, and may indirectly represent baseline clustering details. After operation 412, flowchart 400 can proceed to state 414. State 414 may involve determining whether baselineization is complete. If it is determined in state 414 that baselineization is complete, flowchart 400 can proceed to operation 416. If it is determined in state 414 that baselineization is not complete, flowchart 400 can return to operation 404. Operation 416 may involve establishing and / or storing fault thresholds.
[0047] In some embodiments, once baselineization mode 401 is completed and condition 408 is met, flowchart 400 may proceed to operation 420, which may be an operation in monitoring mode 402. That is, if baselineization is determined to be complete in condition 408, flowchart 400 may transition from baselineization mode 401 to monitoring mode 402. Operation 420 may involve calculating a fault metric for the current input point. Operation 420 may be associated with operation 418, which may involve retrieving machine learning model parameters from storage. Machine learning model characteristics may be derived from baseline data stored in a database, and the model may be retrieved for application to monitoring mode data. From operation 420, flowchart may proceed to condition 422. Condition 422 may involve determining whether the fault metric is greater than a fault threshold. In some cases, if the fault metric is greater than the threshold on a consistent basis (e.g., if the fault metric is greater than the threshold a given number of times or for a given time length), then only the presence of a fault may be determined. Condition 422 may also be associated with operation 419, which may involve retrieving the fault threshold from storage. If condition 422 is met, flowchart 400 proceeds to operation 426. If condition 422 is not met, flowchart 400 proceeds to operation 424. Operation 426 may involve triggering an alarm. Operation 424 may involve resetting an alarm. Operation 426 may also involve any other action besides triggering an alarm. A second example action may include performing changes to the power system (e.g., turning off and / or on the electrical machinery), or performing any other action concerning the electrical machinery and / or any other power assets in the system.
[0048] Figure 5 Example clustering plot 500 is depicted. That is, this plot depicts an example of how faults can be detected based on clustering of baseline data generated by machine learning clustering techniques (e.g., Mahalanobis distance). Clustering plot 500 may depict a first cluster 502 of baseline data. The baseline data may be data obtained during the baselineization mode of the algorithm. That is, the baseline data may be associated with the healthy operation of the electrical machine at frequencies typically associated with a given type of electrical machine fault. Based on the baseline data, a threshold 504 may be established to determine when a specific type of fault occurs in the electrical machine. Plot 500 may also depict a second cluster 508 and a third cluster 510. The second cluster 508 and the third cluster 510 may be data clusters captured during the monitoring mode of the algorithm. As depicted, the second cluster 508 may be within the threshold 504, but the third cluster 510 may be outside the threshold. Therefore, the second cluster 508 may indicate the healthy operation of the electrical machine for a given type of fault, and the third cluster 510 may indicate the specific type of fault occurring in the electrical machine. Figure 5In the examples provided, machine learning can be used to build clusters, so Mahalanobis distance (or any other similar method) can be used to determine whether the clusters of data captured during monitoring mode are within or outside the threshold 504. In some embodiments, the Mahalanobis distance can be determined using Equation 1 provided below.
[0049] (Equation 1) in MD n It can be the Mahalanobis distance for the square of a given sample n. energy n It could be the energy at sample n. peak n It could be the peak value at sample n. It can be the average of the baseline energy, as calculated using Equation 2. It can be the average of the baseline peak values, as calculated using Equation 3, and C -1 It can be the inverse of the covariance matrix, as can be calculated using Equation 4-5.
[0050] (Equation 2) (Equation 3) (Equation 4) (Equation 5) Figure 6 An example method 600 according to an example embodiment of this disclosure is depicted. Figure 6At box 602, method 600 may include capturing real-time electrical signal and thermal data associated with the electric machine during a baseline mode. Box 604 of method 600 may include converting the real-time electrical signal data into frequency domain data. Baseline clusters are determined in real-time using the frequency domain data and adaptive or incremental machine learning algorithms for one or more peak dB values and one or more root mean square dB values of the captured electrical and thermal data, and a portion of the frequency domain data included in the band. Box 606 of method 600 may determine one or more fault thresholds as a combination of cluster distances from the baseline data clusters derived from one or more peak dB values and one or more root mean square dB values, and the associated electrical and thermal data. Box 608 of method 600 may capture real-time electrical signal and thermal data during a monitoring mode. Box 610 of method 600 may determine fault indicators of the electric machine as a combination of electrical, thermal, and frequency domain data cluster distances based on the real-time electrical signal data, thermal data, one or more peak dB values from the frequency domain data, and one or more root mean square dB values. Block 612 of method 600 determines that a motor's fault indicator is higher than one or more fault thresholds established during baseline mode. Block 614 of method 600 generates a warning that the motor's fault indicator is higher than one or more fault thresholds. Block 616 of method 600 performs a control action based on the determination that the motor's fault indicator is higher than one or more fault thresholds.
[0051] In some embodiments, method 600 may further include determining the number of frequency values to be included in the band based on the amplitude of the frequency values. Method 600 may further include determining a covariance matrix using one or more peak decibel values and one or more root mean square decibel values, along with associated electrical and thermal data. Method 600 may further include performing outlier filtering on the frequency domain data included in the band based on at least one of the following: the amplitude of the frequencies included in the band, the type of power supply or load of the electrical machine, the application of the electrical machine, or the presence of load oscillations.
[0052] In some embodiments, frequency domain data is divided into a low range, a medium range, and a high range of frequency values, and the method further includes determining the number of frequency values to be included in the band based on the amplitude of the frequency values, wherein the number of frequency values included in the band is greater in the medium range of frequency values compared to the low range of frequency values, and greater in the high range of frequency values compared to the medium range of frequency values. In some embodiments, determining one or more fault thresholds is also based on machine-specific characteristics, such as nameplate information, design data, supply or load status, the life and condition of the machine during a baseline period, or the frequency zone of a fault under diagnosis. In some embodiments, the fault threshold is established as a normalized distance from a baseline cluster of one or more peak decibel values and one or more root mean square decibel values, and associated electrical and thermal data, and wherein the fault threshold is applied to multiple different machines in a team of machines. In some embodiments, the machine learning algorithm includes any clustering-based technique. In some embodiments, the machine is a variable frequency drive (VFD) machine. In some embodiments, determining the baseline cluster of one or more peak decibel values and one or more root mean square decibel values, and the associated electrical and thermal data, is performed with historical data stored.
[0053] exist Figure 2 and Figure 4-5 The operations described and depicted in the illustrative process flow can be implemented or performed in any suitable order as desired in the various exemplary embodiments of this disclosure. Additionally, in some exemplary embodiments, at least a portion of the operations can be performed in parallel. Furthermore, in some exemplary embodiments, operations can be performed in a more... Figure 2 and Figure 4-5 The operations described in the text are fewer, more, or different.
[0054] Figure 7The illustration shows an example computing system and apparatus 700 according to one or more embodiments of the present disclosure. The computing apparatus 700 may represent any number of elements described herein (e.g., protection relay 102) or any other elements described herein. The computing apparatus 700 may include one or more processors 702 that execute instructions stored in one or more memory devices (referred to as memory 704). Instructions may be, for example, instructions for implementing functionality described as being implemented by one or more modules and systems disclosed above, or instructions for implementing one or more of the methods disclosed above. The one or more processors 702 may be embodied, for example, in a CPU, multiple CPUs, a GPU, multiple GPUs, a TPU, multiple TPUs, a multi-core processor, a combination thereof, etc. In some embodiments, the one or more processors 702 may be arranged in a single processing device. In other embodiments, the one or more processors 702 may be distributed across two or more processing devices (e.g., multiple CPUs, multiple GPUs, a combination thereof, etc.). Processors may be implemented as a combination of processing circuitry systems or computing processing units (such as CPUs, GPUs, or a combination of both). Therefore, for the sake of illustration, a processor can refer to a single-core processor; a single processor with software multithreading capabilities; a multi-core processor; a multi-core processor with software multithreading capabilities; a multi-core processor with hardware multithreading technology; a parallel processing (or computing) platform; and a parallel computing platform with distributed shared memory. Additionally, or as another example, a processor can refer to an integrated circuit (IC), an ASIC, a digital signal processor (DSP), an FPGA, a PLC, a complex programmable logic device (CPLD), discrete gate or transistor logic, discrete hardware components, or any combination thereof designed or otherwise configured (e.g., manufactured) to perform the functions described herein.
[0055] One or more processors 702 can access memory 704 via a communication architecture 706 (e.g., a system bus). The communication architecture 706 may be adapted to a specific arrangement (localized or distributed) and type of one or more processors 702. In some embodiments, the communication architecture 706 may include one or more bus architectures, such as a memory bus or memory controller; a peripheral bus; an accelerated graphics port; a processor or local bus; combinations thereof, etc. For illustration, such architectures may include an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MCA) bus, an Enhanced ISA (EISA) bus, a Video Electronics Standards Association (VESA) local bus, an Accelerated Graphics Port (AGP) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-express bus, a Personal Computer Memory Card International Association (PCMCIA) bus, a Universal Serial Bus (USB), etc.
[0056] The memory components or memory devices disclosed herein may be embodied in volatile or non-volatile memory, or may include both volatile and non-volatile memory. Furthermore, the memory components or memory devices may be removable or non-removable, and / or internal or external to a computing device or component. Examples of various types of non-transient storage media may include hard disk drives, zip drives, CD-ROMs, digital versatile discs (DVDs) or other optical storage devices, magnetic tape cassettes, magnetic tapes, disk storage devices or other magnetic storage devices, flash memory cards or other types of memory cards, cassette tapes, or any other non-transient media suitable for retaining desired information and accessible by a computing device.
[0057] For illustration, non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) used as external cache memory. For illustration and not limitation, RAM is available in many forms, such as synchronous RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct memory bus RAM (DRRAM). The memory devices or memories disclosed in the operating or computing environment described herein are intended to include one or more of these and / or any other suitable types of memory. In addition to storing executable instructions, memory 704 may also retain data.
[0058] Each computing device 700 may also include a mass storage device 708 accessible by one or more processors 702 via a communication architecture 706. The mass storage device 708 may include machine-accessible instructions (e.g., computer-readable instructions and / or computer-executable instructions). In some embodiments, the machine-accessible instructions may be encoded in the mass storage device 708 and may be arranged in components that can be built (e.g., linked and compiled) and retained in the mass storage device 708 in a computer-executable form, or arranged in one or more other machine-accessible non-transient storage media included in the computing device 700. Such components may embody or constitute one or more of the various modules disclosed herein. Such a module is illustrated as module 714. Additionally, protocols such as Modbus, DNP, IEC 60870, IEC 61850, Profibus, Fieldbus, etc., may be used in conjunction with the systems and methods described herein.
[0059] The execution of the asset monitoring and diagnostic module 714 by one or more processors 702, individually or in combination, can enable the computing device 700 to perform any of the operations described herein (e.g., regarding...). Figure 3-4 and Figure 6 The described operation, and any other operations.
[0060] Each computing device 700 may also include one or more input / output interface devices 710 (referred to as I / O interfaces 710) that may allow or otherwise facilitate communication between external devices and computing device 700. For example, I / O interfaces 710 may be used to receive data and / or instructions from external computing devices and to send data and / or instructions to external computing devices.
[0061] The computing device 700 also includes one or more network interface devices 712 (referred to as network interfaces 712), which may allow or otherwise facilitate functional coupling between the computing device 700 and one or more external devices. Functionally coupling the computing device 700 to an external device may include establishing a wired or wireless connection between the computing device 700 and the external device. The network interfaces 712 may include one or more antennas and a communication processing device that may allow wireless communication between the computing device 700 and another external device. For example, between a vehicle and a smart infrastructure system, or between two smart infrastructure systems. Such a communication processing device may process data according to a defined protocol of one or more radio technologies. Radio technologies may include, for example, 3G, LTE, LTE-Advanced, 5G, IEEE 802.11, IEEE 802.16, Bluetooth, ZigBee, Near Field Communication (NFC), etc. The communication processing device may also process data according to other protocols, such as vehicle-to-infrastructure (V2I) communication, vehicle-to-vehicle (V2V) communication, etc. One or more network interfaces 712 can also be used to facilitate peer-to-peer self-organizing network connections as described herein.
[0062] As used herein, the terms “environment,” “system,” “unit,” “module,” “architecture,” “interface,” “component,” etc., refer to entities related to a computer or to an operating device having one or more defined functionalities. The terms “environment,” “system,” “module,” “component,” “architecture,” “interface,” and “unit” are used interchangeably and can collectively refer to functional elements. Such entities can be hardware, a combination of hardware and software, software, or software in execution. As an example, modules can be embodied in processes running on a processor, processors, objects, executable portions of software, threads of execution, programs, and / or computing devices. As another example, both software applications executing on a computing device and computing devices can embody modules. As yet another example, one or more modules can reside within a process and / or thread of execution. Modules can reside on a single computing device or be distributed among two or more computing devices. As disclosed herein, modules can be executed from various computer-readable non-transient storage media having various data structures stored thereon. Modules can communicate via local and / or remote processes based on signals (analog or digital) having one or more data packets (e.g., data from a component that interacts with a local system, another component in a distributed system, and / or interacts with other systems via signals across a network such as a wide area network).
[0063] As yet another example, a module may be embodied in or may include a device having defined functionality provided by mechanical components operated by an electrical or electronic circuitry system controlled by a software or firmware application executed by a processor. Such a processor may be internal or external to the device and may execute at least a portion of the software or firmware application. Furthermore, in another example, a module may be embodied in or may include a device providing defined functionality through electronic components without mechanical components. The electronic components may include a processor for executing software or firmware that at least partially allows or otherwise facilitates the functionality of the electronic components.
[0064] In some embodiments, modules may communicate via local and / or remote processes based on signals (analog or digital) having one or more data packets (e.g., data from a component that interacts with a local system, another component in a distributed system, and / or interacts with other systems via signals across a network such as a wide area network). Additionally, or in other embodiments, modules may communicate or be otherwise coupled via thermal, mechanical, electrical, and / or electromechanical coupling mechanisms (such as conduits, connectors, combinations thereof). Interfaces may include input / output (I / O) components and associated processors, applications, and / or other programming components.
[0065] Furthermore, in this specification and accompanying drawings, terms such as “store,” “storage device,” “data repository,” “data storage device,” “memory,” “repository,” and substantially any other information storage component related to the operation and functionality of the components of this disclosure refer to a memory component, an entity embodied in one or more memory devices, or a component forming a memory device. It should be noted that the memory components or memory devices described herein embody or include non-transitory computer storage media that can be read or otherwise accessed by a computing device. Such media can be implemented using any method or technique for storing information such as machine-accessible instructions (e.g., computer-readable instructions), information structures, program modules, or other information objects.
[0066] Unless otherwise specifically stated or otherwise understood in the context in which they are used, conditional languages such as “can,” “may,” “may,” or “possibly” are generally intended to convey, among other things, that certain implementations may include certain features, elements, and / or operations, while other implementations do not. Therefore, such conditional languages are generally not intended to imply that features, elements, and / or operations are required in any way for one or more implementations, or that one or more implementations necessarily include logic for determining, with or without user input or prompting, whether such features, elements, and / or operations are included in or to be performed in any particular implementation.
[0067] The contents described in this specification and accompanying drawings include examples of systems, apparatuses, technologies, and computer program products that, individually or in combination, allow for the automatic provision of updates to vehicle profile packages. It is certainly impossible to describe every conceivable combination of components and / or methods in order to describe the various elements of this disclosure, but it will be appreciated that many further combinations and arrangements of the disclosed elements are possible. Therefore, it will be apparent that various modifications can be made to this disclosure without departing from its scope. Additionally or as an alternative, other embodiments of this disclosure may become apparent from consideration of the specification and accompanying drawings and from the practice of this disclosure as presented herein. It is intended that the examples presented in the specification and accompanying drawings be considered illustrative rather than limiting in all respects. Although specific terminology is used herein, it is used only in a general and descriptive sense and not for limiting purposes.
Claims
1. A system comprising: Electrical machinery; as well as A computer processor, operable to execute a set of computer-readable instructions; as well as A memory operable to store the set of computer-readable instructions operable to: Real-time electrical signal data and thermal data associated with the electromechanical device are captured during baseline mode; Convert the real-time electrical signal data into frequency domain data; Baseline clustering of one or more peak dB values and one or more root mean square dB values of the captured electrical and thermal data and a portion of the frequency domain data included in the band, using the frequency domain data in real time and adaptive or incremental machine learning algorithms; One or more fault thresholds are determined as a combination of clustering distances from baseline data clusters of the one or more peak decibel values and the one or more root mean square decibel values, and associated electrical and thermal data. The real-time electrical signals and thermal data are captured during monitoring mode; Based on the real-time electrical signal, thermal data, one or more peak decibel values from frequency domain data, and one or more root mean square decibel values, the fault index of the motor is determined as a combination of electrical, thermal, and frequency domain data clustering distances. The fault index of the motor is determined to be higher than one or more fault thresholds established during the baseline mode; Generate a warning when the fault indicator of the motor is higher than one or more fault thresholds; as well as The control action is executed based on the determination that the fault index of the motor is higher than one or more fault thresholds. The fault threshold is established as a normalized distance from a baseline cluster to one or more peak decibel values and one or more root mean square decibel values, along with the associated electrical and thermal data, and wherein the fault threshold is applied to multiple different electrical machines in a team of electrical machines.
2. The system as claimed in claim 1, wherein, The processor is further configured to determine the number of frequency values to be included in the band based on the magnitude of the frequency values.
3. The system as described in claim 1, wherein, The frequency domain data is divided into a low range of frequency values, a medium range of frequency values, and a high range of frequency values, and the processor is further configured to determine the number of frequency values to be included in the band based on the amplitude of the frequency values, wherein the number of frequency values included in the band is greater in the medium range of frequency values than in the low range of frequency values, and the number of frequency values included in the band is greater in the high range of frequency values than in the medium range of frequency values.
4. The system as claimed in claim 1, wherein, The processor is further configured to determine the covariance matrix using the one or more peak decibel values and one or more root mean square decibel values, as well as the associated electrical and thermal data.
5. The system as claimed in claim 1, wherein, The processor is further configured to perform outlier filtering on the frequency domain data included in the band based on at least one of the following: the amplitude of the frequency included in the band, the power supply or load type of the motor, the application of the motor, or the presence of load oscillations.
6. The system of claim 1, wherein, Determining one or more fault thresholds is further based on the specific characteristics of the electromechanical machine. The specific characteristics of the electric machine include nameplate information, design data, supply device or load conditions, the lifespan and condition of the machine during the baseline period, or the frequency range of the fault under diagnosis.
7. The system as claimed in claim 1, wherein, The machine learning algorithm includes any clustering-based technique.
8. The system of claim 1, wherein, The electric machine is a variable frequency drive (VFD) machine.
9. The system as claimed in claim 1, wherein, The baseline clustering, which determines one or more peak decibel values and one or more root mean square decibel values, along with the associated electrical and thermal data, is performed with historical data stored.
10. A method comprising: Real-time electrical signal data and thermal data associated with the electromechanical device are captured during baseline mode; Convert the real-time electrical signal data into frequency domain data; Baseline clustering of one or more peak dB values and one or more root mean square dB values of the captured electrical and thermal data and a portion of the frequency domain data included in the band, using the frequency domain data in real time and adaptive or incremental machine learning algorithms; One or more fault thresholds are determined as a combination of clustering distances from baseline data clusters of the one or more peak decibel values and the one or more root mean square decibel values, and associated electrical and thermal data. The real-time electrical signals and thermal data are captured during monitoring mode; Based on the real-time electrical signal data, thermal data, one or more peak decibel values from frequency domain data, and one or more root mean square decibel values, the fault index of the motor is determined as a combination of electrical, thermal, and frequency domain data clustering distance. The fault index of the motor is determined to be higher than one or more fault thresholds established during the baseline mode; Generate a warning when the fault indicator of the motor is higher than one or more fault thresholds; as well as Control actions are executed based on the determination that the fault index of the motor is higher than one or more fault thresholds. The fault threshold is established as a normalized distance from a baseline cluster of one or more peak decibel values and one or more root mean square decibel values, along with the associated electrical and thermal data, and the fault threshold is applied to multiple different motors in a set of motors.
11. The method of claim 10, further comprising determining the number of frequency values to be included in the band based on the amplitude of the frequency values.
12. The method of claim 10, wherein, The frequency domain data is divided into a low range of frequency values, a medium range of frequency values, and a high range of frequency values, and the method further includes determining the number of frequency values to be included in the band based on the magnitude of the frequency values, wherein the number of frequency values included in the band is greater in the medium range of frequency values than in the low range of frequency values, and the number of frequency values included in the band is greater in the high range of frequency values than in the medium range of frequency values.
13. The method of claim 10, further comprising using the one or more peak decibel values and one or more root mean square decibel values, along with the associated electrical and thermal data, to determine a covariance matrix.
14. The method of claim 10, further comprising performing outlier filtering on the frequency domain data included in the band based on at least one of the following: the amplitude of the frequency included in the band, the power supply or load type of the motor, the application of the motor, or the presence of load oscillations.
15. The method of claim 10, wherein, Determining one or more fault thresholds is further based on the specific characteristics of the electromechanical machine. The specific characteristics of the electric machine include nameplate information, design data, supply device or load conditions, the lifespan and condition of the machine during the baseline period, or the frequency range of the fault under diagnosis.
16. The method of claim 10, wherein, The machine learning algorithm includes any clustering-based technique.
17. The method of claim 10, wherein, The electric machine is a variable frequency drive (VFD) machine.
18. The method of claim 10, wherein, Baseline clustering is performed to determine one or more peak dB values and one or more root mean square dB values, along with associated electrical and thermal data stored in historical data.
Citation Information
Patent Citations
Autonomous procedure for monitoring and diagnostics of machine based on electrical signature analysis
US20180246506A1
