System and process for pattern matching bearing vibration diagnostics

Abstract

A system and method comprising: receiving vibration spectrum data from a plurality of different assets; determining a cluster for the plurality of assets based on shapes of the vibration spectrum data for each of the plurality of assets, assets grouped in the same cluster having vibration spectrum data of similar spectral shapes; determining one of a plurality of fault classifications for each of the clusters based on application of domain-derived pattern recognition rules to the vibration spectrum data; generating an output comprising an association of each of the plurality of assets with a fault classification of the cluster in which a particular asset is grouped; and saving a record of the output.

Description

Technical Field

[0001] The field of the present disclosure relates generally to system monitoring and diagnostics. More particularly, the present disclosure relates to systems, devices, and methods to detect and determine faults based on patterns in vibration spectrum data. Background Art

[0002] The working condition of various components of an asset may be monitored by a condition monitoring system (CMS), where alarms may be generated to indicate faults and other warnings regarding the operation of the asset. For some assets, such as, for example, wind turbines, a wind turbine vibration analyst may monitor the health of the turbine's gearbox and other bearings using a CMS that uses the bearing vibration data to detect abnormal spectral spikes and, therefore, generates alarms based on the detected spectral spikes. This conventional approach relies heavily on the availability and knowledge of the exact gearbox kinematic information for the wind turbine being monitored, and is rule-based, and uses limits or trend changes to warn of possible component wear or damage.

[0003] In general, some conventional vibration monitoring and analysis systems usually consist of two steps, where the first step is an automatic alarm system that stores or has access to physical information such as natural fault frequencies and rotational frequencies derived from kinematic information of the wind turbine such as the number of gear teeth / rolling elements. In the second step, the vibration analyst can review the vibration spectrum graph associated with the wind turbine to understand whether the diagnosis from the alarm system is correct. This second step is necessary because the alarm system usually only looks at the absolute amplitude value or trend change at a specific location (i.e., frequency) in the spectrum. By using precise kinematic information, traditional CMS systems may also miss known faults if the characteristics are even slightly different from the textbook (i.e., theoretical) characteristics. This is due to the rule-based nature of the system, which relies on spectral spikes at specific frequencies to indicate known types of faults. Although this approach can be applicable to problems where the fault characteristics are clearly defined and well-known, it often fails when new (i.e., novel) faults appear.

[0004] Thus, conventional systems may potentially miss novel faults and / or may incorrectly detect known faults that have characteristics that are even slightly different from a standard fault frequency, or that shift away from a standard fault frequency as wear or damage progresses.

[0005] Therefore, there is a need for methods and systems that support and facilitate a holistic view of the vibration spectrum of an asset, rather than specific amplitude values ​​at specific frequencies in the signal spectrum. Summary of the invention

[0006] In one aspect, an embodiment of the present disclosure relates to a method comprising: receiving vibration spectrum data from a plurality of different assets; determining a cluster for the plurality of assets based on a shape of the vibration spectrum data for each of the plurality of assets, assets grouped in the same cluster having vibration spectrum data of similar spectral shapes; determining one of a plurality of fault classifications for each of the clusters based on application of domain-derived pattern recognition rules to the vibration spectrum data; generating an output comprising an association of each of the plurality of assets with a fault classification of the cluster in which the particular asset is grouped; and saving a record of the output.

[0007] In other embodiments, a system including a spectrum pattern matching (or similar functional) module and a root cause identification (or similar functional) module may be implemented to perform at least some of the features of the methods and processes disclosed herein. In yet another example embodiment, a tangible medium may embody executable instructions that may be executed by a processor-enabled device or system to implement at least some aspects of the processes of the present disclosure.

[0008] Although the present disclosure describes the use of various methods and systems for detecting wind turbine bearing, shaft and transmission problems, the present disclosure is applicable to all rotating or reciprocating machinery, including but not limited to and across many different types of power generation, industrial and manufacturing industries. Illustrative examples of some of the equipment to which the methods and systems disclosed herein are applicable include but are not limited to power generation equipment, motors, pumps, gearboxes, bearings, shafts, gears, compressors, metal extrusion equipment, mining and metal production equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] These and other features, aspects and advantages of the present disclosure will become better understood when the following detailed description is read with reference to the accompanying drawings, in which like characters refer to like parts throughout the drawings, and in which:

[0010] Figure 1 is an illustrative schematic example of a wind turbine including a bearing mechanism therein;

[0011] Figure 2 is an illustrative example signal spectrum including a bearing outer race fault according to some embodiments herein;

[0012] Figure 3 is an illustrative example signal spectrum including a bearing inner race fault according to some embodiments herein;

[0013] Figure 4 is an illustrative example signal spectrum including a bearing ball spin fault according to some embodiments herein;

[0014] Figure 5A and Figure 5Bis a graph related to planetary ball failure;

[0015] Figure 6 is an illustrative example of a graph associated with a ring gear fault according to some embodiments herein;

[0016] Figure 7 is an illustrative example block diagram of a system according to some embodiments herein;

[0017] Figure 8 is an illustrative example flow chart of a process according to some embodiments herein;

[0018] Fig. 9 is an illustrative example flow chart of a fault detection and identification process according to some embodiments herein;

[0019] Fig.10 is an illustrative example flow chart of a signal normalization process according to some embodiments herein;

[0020] Fig.11 is an illustrative example flow chart of a signal clustering process according to some embodiments herein;

[0021] Fig.12 is an illustrative example flow chart of a process for determining a bearing inner race fault for a high speed shaft according to some embodiments herein;

[0022] Fig.13 is an illustrative example flow chart of a process for determining a bearing outer race fault for a high speed shaft according to some embodiments herein;

[0023] Fig.14 is an illustrative example flow chart of a process for determining a bearing ball spin fault for a high-speed shaft according to some embodiments herein;

[0024] Fig.15 is an illustrative example flow chart of a fault diagnosis process for a low speed shaft according to some embodiments herein; and

[0025] Fig.16 is an illustrative depiction of a block diagram of a system or device that can support implementation of some of the processes disclosed herein.

[0026] Unless otherwise indicated, the drawings provided herein are intended to illustrate features of embodiments of the present disclosure. These features are believed to be applicable to many systems including one or more embodiments of the present disclosure. As such, the drawings are not intended to include all conventional features known to those of ordinary skill in the art required to practice the embodiments disclosed herein. DETAILED DESCRIPTION

[0027] In the following description and claims, reference is made to a number of terms that have the following meanings.

[0028] The singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise.

[0029] "Optional" or "optionally" means that the subsequently described event or circumstance may or may not occur, and that the description includes instances where the event occurs and instances where it does not.

[0030] The present disclosure relates to processes and systems to detect and diagnose faults and other conditions based on the pattern or spectral shape of a vibration spectrum from an asset. Figure 1 1 is an illustrative schematic illustration of a wind turbine 100 that can be monitored to determine the health of a wind turbine generator. Wind turbine generator (or simply turbine) 100 illustrates a class of assets to which the processes and systems disclosed herein may be applied. In some aspects, the present disclosure, including the processes and systems disclosed herein, may be embodied and implemented in other applications, environments, and use cases besides the specific examples set forth below.

[0031] about Figure 1 , the turbine 100 may be monitored to determine the operating condition or health of various components of the turbine. Figure 1 In the case of the turbine illustrated in , the drive train bearings of the turbine 100 may be the components of interest. In other assets, other components may be the body(ies) of interest. The turbine 100 includes a rotor 105 and a drive train including three (3) main bearings. The bearings include: a main bearing 120, which is located closest to the rotor 105; a generator bearing 115 including an inboard bearing 125 and an outboard bearing 135, which is coupled to and rotates a generator (not shown) that generates energy; and a drive train bearing, which is located at 110 between the main bearing and the generator bearing. Figure 1 The drive train bearings in the example include four (4) different sets of bearings on the low speed shaft 140 and the high speed shaft 145, which are collectively referred to as the drive train bearings. The four sets of drive train bearings operate to increase the revolutions per minute (rpm) from about 15-20 rpm at the rotor to about 1440 rpm at the generator side. The drive train bearings can be divided into three (3) stages, including: (1) a low speed stage (LSS) 150; (2) an intermediate stage, which includes a low speed intermediate stage (LSIS) 155 and a high speed intermediate stage (HSIS) 160; and (3) a high speed stage (HSS) 165. Sensors such as accelerometers, proximity probes, or acoustic sensors can be deployed in the drive train gearbox to capture vibrations generated by different stages of the drive train gearbox. In some embodiments, although sensors can be placed in different stages of the drive train, some vibrations from one or more of the stages can infiltrate vibration data acquired by another stage.

[0032] The data captured by the sensors in the drive train may be primarily waveform data of vibrations, displacements, or acoustic signatures generated by drive train components (e.g., high-speed shaft and low-speed shaft bearings) and derivatives of those waveforms. In some aspects, the primary data types that may be captured may include asynchronous waveforms and synchronous waveforms. Asynchronous waveforms are a type of vibration signal that is recorded over a fixed amount of time. Figure 1 The example asset (ie, the turbine) itself operates at variable speed, so fixed time recording means that the duration does not need to be a multiple of the period of any particular bearing (ie, asynchronous).

[0033] As mentioned herein, a synchronous waveform is a waveform recorded within a certain number of revolutions of a bearing and may not be directly stored. Instead, a derived spectrogram associated with the waveform may be stored for a certain period of time (e.g., every 4 hours). The direct FFT (Fast Fourier Transform) of a waveform is referred to as a high-resolution spectrum, while a finer and more detailed spectrum derived via a digital signal processing (DSP) post-envelope is referred to as a synchronous (or sync) envelope spectrum. Both the high-resolution spectrum and the synchronous envelope spectrum graphs can capture and reveal clearly visible faults including sidebands (if any), while the high-resolution spectrum can display fault features in a graph with a higher background noise or a graph containing features that are not directly related to component wear or damage.

[0034] In some applications, signatures for high-speed level faults may be more evident in the synchronous envelope spectrum data. In some aspects, some faults may exhibit their signatures in the synchronous envelope spectrum data before any anomalies are manifested in the high-resolution spectrum data during the typical life cycle of a gearbox bearing. In some embodiments herein, the synchronous envelope spectrum data may be used in an HSS fault detection algorithm (discussed in more detail below).

[0035] Regarding the spectrum graphs, a brief discussion is presented to better understand the frequency ranges they represent. Figure 1 In the example of a gearbox, where the HSS shaft 145 has a rpm of about 1440, the HSS shaft will have a rotational frequency of 24 Hz. This fundamental frequency is referred to as the 1X of the HSS. Every rotating component inside the gearbox has a 1X fundamental frequency. For example, the HSIS shaft also rotates and has a 1X fundamental frequency related to the HSS 1X by the following relationship:

[0036] .

[0037] Therefore, if the gear ratio of the HSIS to the HSS is 3, the fundamental frequency value will be and .

[0038] Note that different faults may have their own 1X values. In some aspects, the 1X value of a fault may be viewed as the frequency at which a particular fault generates noise within a system. For example, a crack in the outer race of a bearing may cause a slight disturbance / noise or vibration signature each time a ball traverses the crack. As such, the outer race fault 1X fundamental frequency (also referred to as the bearing ball passing frequency (BPFO) 1X of the outer race) would be the frequency of this generated noise.

[0039] In some respects, Figure 1 The frequency spectrum of each stage of the drive train of an example wind turbine may generally represent a frequency range where the maximum value is a multiple of a certain shaft frequency. For example, the 3rd stage synchronous envelope spectrum data for a particular model of turbine may represent frequencies up to a maximum value of about 50X HSS. Note that both this value and the 1X HSS value are related to the actual revolutions per minute (RPM) of the shaft at a given point in time and may vary depending on the specific subject asset. Similarly, the 2nd stage maximum frequency may typically be about 50 or 100X LSIS.

[0040] For the detection of faults in the drive train bearings of wind turbines Figure 1 For example, the faults of primary concern may include three fault locations in the bearing. In particular, the fault locations may include the outer race, inner race, and rolling elements of the bearing itself. For a specific bearing design, the outer race, inner race, and rolling elements have natural frequencies. In the event of a fault in any of those locations, the resulting vibration signal, as captured by a sensor (e.g., an accelerometer), will exhibit those natural frequencies. For example, if a bearing has a defect in its outer race, a vibration pulse will be generated and recorded each time one of the rolling elements touches the defective location. The frequency of the pulse will be the same as the natural fault frequency of the outer race (referred to herein as the ball passing frequency or BPFO of the outer race). Similarly, the bearing fault frequency (i.e., IRBP or ball passing of the inner race) defect for the inner race and the bearing fault frequency (i.e., ball spin or BS) defect for the rolling elements are referred to as the ball passing frequency (BPFI) and ball spin frequency (BSF) faults of the inner race.

[0041] In some scenarios, if fault frequency information for a particular asset is available and known, analysis of the amplitude of the fault frequency information can be used to determine whether a particular type of fault exists in the asset's bearing. However, there are several limitations to this approach based on knowledge of the fault frequency information. First, the fault frequency information may not be readily available, if at all. The fault frequency may be known through mathematical formulas that relate design parameters such as, for example, the number of rolling elements, pitch angle, roller diameter, pitch diameter, etc., to specific defect frequencies (e.g., BPFI, BPFO, and BSF). However, the design parameters vary greatly from one bearing to another and may not be known or even available to vibration engineers or other entities.

[0042] Second, the fault frequency may not be present in the spectrum for a particular asset (i.e., the "faulty" spectrum). In theory, natural fault frequencies (e.g., BPFO, BPFI, etc.) should exist where the corresponding fault exists in the bearing. However, in practice this may not always be the case. For example, a frequency close to BPFO is excited in a bearing that actually has an ORBP problem. As demonstrated by this example, in practical (i.e., real-world) applications, over-reliance on theoretical fault frequencies can be detrimental because faults characterized by amplitude spikes may actually occur at frequencies different from the expected / theoretical frequency and may not be seen when looking for faults at the expected / theoretical frequency. Deviations from the actual fault signature are typically caused by secondary damage to the bearing in and around the location of the initial fault, causing sideband energy or frequency shifts in the signature produced when the rolling element encounters the defect.

[0043] These limitations, which rely on known fault frequencies, often necessitate that vibration engineers review the spectral data for even common failure modes (e.g., failures in high-speed stages). Such manual review by expert(s) can be costly and / or result in lost productivity.

[0044] Some previous techniques, processes and systems include a vibration engineer reviewing a report where some of the subject turbines are flagged as objects of concern. For each flagged turbine, the vibration engineer may review different spectrum plots for that turbine to determine if a suspected fault actually exists. This type of review may include a review of the spectrum DEI (Dynamic Energy Index) numbers, in addition to visual inspection of spikes and alignment with natural fault frequencies, where this effort is necessary because these previous types of systems rely on threshold algorithms that look for increases in amplitude at specific locations (i.e., frequencies).

[0045] Some embodiments herein may use an asset (e.g., gearbox) agnostic algorithm to determine defects based on basic defect signatures (i.e., "patterns") in vibration spectrum data. An overview of some embodiments will be explained in the context of bearing signatures for ORBP, IRBP, BS, and other defects.

[0046] For an outer race defect in a bearing, a shock pulse will be recorded each time a rolling element of the bearing crosses the defective location. For these types of generated pulses, the ORBP frequency (BPFO) and its harmonics are present in the resulting spectrum. Figure 2 A frequency domain spectrum 200 is illustrated. Moving from left to right, the range of frequencies analyzed increases. The synchronized envelope spectrum data for this example is shown in Figure 1 205 is shown.

[0047] For inner race defects in bearings (i.e., IRBP signatures), the signature of the inner race fault is more complex than the ORBP signature. In some embodiments and constructions, the inner race of the bearing typically rotates with the shaft, and therefore the inner race defect location moves in and out of the bearing loading zone. As such, the impact pulses generated when the bearing is in the loading zone are more pronounced than the pulses generated when the bearing moves out of the loading zone. In this way, the BPFI pulse signal is amplitude modulated by the rotational frequency of the shaft (referred to as Axis 1X). These factors result in the presence of a spike in amplitude at the BPFI frequency (and its harmonics) and the presence of one or two sidebands separated by the Axis 1X frequency. In Figure 3 An illustrative IRBP fault signature is depicted in graph 300 of FIG. 1 , where the frequency domain is shown at 305 and the range of frequencies analyzed increases from left to right. Figure 3 In the example of , the 1X frequency is shown at 320, and the BPFI defect is shown at 325, and its harmonic frequencies (eg, at 330) are shown. As can be seen, the BPFI defect includes sidebands that are 1X away from the peak center amplitude of the BPFI defect.

[0048] For ball spin defects in bearings (i.e., BS features), the characteristic pattern of rolling element defects can be similar to IRBP defects. Figure 4 As illustrated in , the BS feature 405 in the spectrum 400 is manifested by the presence of ball spin frequencies (BSFs) 410, 415, and 420 and one or two sidebands separated by a cage 1X frequency (denoted as FT), which may be about 0.3 to about 0.5 of the shaft 1X frequency.

[0049] In some instances, there may be other fault signatures for which there are no existing detection algorithms. Some such examples may include some signatures associated with the low speed stages of the turbine drive train that are only visible in data derived from stage 1 of the drive train.

[0050] For example, a planetary bearing fault may be characterized by a unique "haystack" pattern in the region between the gear mesh harmonics. Figures 2 to 4 Unlike the synchronous envelope spectrum in the example of , where any kind of high-amplitude spike can be a potential indicator of a fault, Figure 5A and Figure 5B The spikes in the low-speed spectrum graph depicted in can occur under normal or non-fault conditions. Figure 5A and Figure 5B In the curve diagram, Figure 5A There is a confirmed planetary bearing (PB) fault at 505 and 510, and Figure 5B There is no such problem. The markers (eg, 515 and 520) indicate the natural frequencies present in the level 1 low speed spectrum for the subject drive train. Note that PB faults are not limited to the low speed level.

[0051] Another low speed stage (LSS) fault is a ring gear fault, which exhibits a distinctive “sawtooth” pattern in the Level 1 high resolution data. Figure 6 An illustrative example of a level 1 spectral graph 600 for a wind turbine exhibiting a ring gear fault is included.

[0052] In some embodiments, the present disclosure is directed to a system (including software and hardware components) that receives wind turbine gearbox accelerometer data and uses pattern matching techniques to detect assets that display similar spectral shapes to the accelerometer vibration spectrum. Figure 7 is an illustrative block diagram of some systems for performing some of the processes disclosed herein. Vibration spectrum data 705 is received by system 700. The system may include a pattern matching module 710 that receives wind turbine gearbox accelerometer data and uses pattern matching techniques to identify and determine (e.g., based on mathematical algorithm calculations) assets that display similar shapes in their accelerometer vibration spectra. System 700 may also include a root cause detection module 715. The root cause detection module 715 is operable to determine at least one of several known root causes of (multiple) faults that may affect an asset (e.g., a turbine) and manifest in the vibration spectrum data. System 700 is operable to identify and determine the exact cause for a known fault based on the application of domain-derived pattern recognition rules to vibration spectrum data 705, and to identify unexpected or novel faults based on pattern matching techniques based on vibration spectrum data. An output record including a fault classification for each of the assets may be generated by a report generation module 720 based on inputs from modules 710 and 715. The generated report may be saved and / or used in further processing (e.g., analysis, etc.) operations.

[0053] In some embodiments, the assets may each be Figure 1However, the systems and processes disclosed herein are not necessarily limited to the analysis of vibration spectrum data and the detection and identification of faults in the drive train of a wind turbine.

[0054] Figure 8 is an overview of a process for detecting and identifying faults in assets based on pattern matching techniques for vibration spectrum data for one or more assets. The process of flowchart 800 outlines a process that looks at the shape of the vibration spectrum data of an asset as a whole, rather than, for example, looking for amplitudes at specific frequencies.

[0055] At operation 805, vibration spectrum data is received from a plurality of assets. The vibration spectrum data may be received from, for example, a plurality of wind turbines deployed in a "wind farm." The data received at operation 805 may include one or more files, wherein vibration spectrum data for each of the subject assets can be identified in the data received at operation 805.

[0056] At operation 810, a pattern matching module or other system or device with similar functionality obtains vibration spectrum data from each wind turbine asset. In some embodiments, operation 810 may also include other processing aspects, such as, but not limited to, averaging the received data for each asset within a certain time frame (i.e., time period). In some embodiments, operation 810 may use a clustering algorithm or a combination of clustering algorithms to complete the pairing of assets. In some embodiments, one or more features may be extracted from the received vibration spectrum data by operation 810 to reduce the dimensionality of the data on a per-asset basis.

[0057] At operation 815, the root cause detection process may use a clustering algorithm on the vibration spectrum (e.g., lower dimension) data to identify assets (e.g., turbines) that share similar spectral shapes. Assets that are determined to have similar spectral shapes are grouped together. Groupings of assets with similar spectral shapes (i.e., patterns) are referred to herein as "clusters." In some aspects, assets in the same cluster are similar in spectral shape and, therefore, have similar faults or health conditions / states.

[0058] In some embodiments, the granularity of the clustering results of operation 810 may be too high for intended or practical use. For example, in some instances, turbines that are only facing upwind or downwind HSIS IRBP faults may be grouped together, while a vibration engineer may only be interested in faults at the summary fault level of the HSIS IRBP (i.e., not including upwind or downwind differences).

[0059] Continuing with operation 815 , the root cause identification module or other system(s) and device(s) having similar functionality may be operable to identify the exact cause of the problem or fault characterized by the spectral shape (ie, signature) in the cluster of differently shaped vibration signals.

[0060] In some aspects, the non-parametric clustering algorithm of operation 810 may be good at grouping assets based on the spectral shape of the vibration spectrum data of individual assets, and will not be able to identify the exact nature of the fault represented in the cluster (i.e., group). In some embodiments, the (multiple) algorithms used to determine clusters based on identifying vibration spectrum data with similar spectral shapes may include at least one of a hierarchical clustering algorithm, a k-means algorithm, a nearest neighbor algorithm, and a combination of clustering methods. As used herein, the hierarchical clustering algorithm assigns each asset to its own cluster, and then continuously merges the closest cluster pairs to each other. This process can be repeated until the desired number of clusters is found. As used herein, the k-means algorithm consists of randomly selecting several spectra as cluster centroids. Then, based on their distance from the centroid, each of the other spectra is assigned to these clusters. The centroid is updated, and the process is repeated until the algorithm converges (i.e., the cluster assignment or centroid does not change).

[0061] In some embodiments, the root cause identification algorithm(s) at operation 815 may employ a data-based, gearbox-agnostic approach to detect specific faults in the cluster. In some instances, the process for detecting or determining specific faults at operation 815 may be determined in a gearbox-agnostic manner as much as possible. In some embodiments, a variety of different algorithms may be used to detect abnormal spikes, sidebands, and other characteristic features in a specific frequency spectrum, and generate fault diagnostics based on different patterns (e.g., spikes and sidebands at a certain location, specific haystack patterns, presence of asynchronous harmonics, sawtooth patterns, etc.) or features, such as, for example, inner race / outer race / ball spin issues in higher speed stages and planetary bearings and other faults in lower speed stages.

[0062] Operation 820 may include a final output that combines the outputs of the pattern matching module and the root cause identification module to generate a unified view including fault determinations for multiple assets that may outperform strict clustering or strict rule-based approaches.

[0063] In some embodiments, the output of operation 820 may be presented via a user interface front end, which may include text, graphics, and other visual representations that provide a mechanism for a user to view and analyze the spectrum of an asset individually and / or in comparison to other assets that may exhibit similar spectral patterns.

[0064] Fig. 9The overall fault detection and determination process 900 according to some embodiments herein is related. At operation 905, input vibration signals for multiple assets are received. At operation 910, the received vibration spectrum data can be pre-processed to prepare for further processing. Some (pre)processing techniques of operation 910 can include, for example, correction of signal slope and / or normalization of the received signal.

[0065] At operation 915, initial clustering of assets is performed based on the spectral shapes of the associated vibration signals of the assets. Operation 915 may include performing a hierarchical clustering function. Operation 920 may further merge clusters based on, for example, the centroid of the cluster being less than a certain threshold (e.g., <1σ) away from the average centroid distance for the cluster.

[0066] Having grouped the different assets into clusters based on their spectral shapes at operations 915 and 920, operation 925 operates to identify whether the signals in each of the merged clusters indicate a fault in the asset based on the centroid of the merged cluster.

[0067] Operation 930 determines, for each cluster, the distance of each asset (e.g., turbine) in the cluster from the centroid of the cluster. If the asset is relatively far from a mean based on some threshold (e.g., >0.75σ) as determined at operation 935, process 900 proceeds to operation 940 where the asset can be classified, for example, by a rules engine, for the condition of the asset because the particular asset under consideration is sufficiently different from the centroid asset in its cluster to not be classified as the same.

[0068] If the asset under consideration at operation 935 is not relatively far from the mean based on some threshold (eg, <0.75σ), process 900 proceeds to operation 945 where the asset may be classified as being the same as the centroid asset in its cluster.

[0069] Fig.10 According to some embodiments of the present invention, Fig. 9 1000 is an example of a flow chart of a signal normalization function performed in the execution of operation 910 of FIG. At operation 1005, an input vibration signal is received, and at operation 1010, a signal RMS (root mean square) and standard deviation (σ) values ​​are calculated. In addition, based on the relationship shown in FIG. 1000, a normalized signal (e.g., NormSpectra ).

[0070] At operation 1020, a linear regression model is executed to calculate the slope of the signal. The value of the linear regression output is further subtracted from the normalized value calculated at operation 1015. At operation 1025, the RMS energy of the normalized signal is calculated. All values ​​less than the RMS value (calculated at operation 1010) are set to zero (e.g., finalSpectra ).

[0071] As indicated by feedback loop 1030, process 1000 is repeated for all signals for each asset being evaluated.

[0072] Fig.11 According to some embodiments of the present invention, Fig. 9 1100 is an example of a flow chart of a signal clustering function performed in the execution of operations 915 and 920 of FIG. At operation 1105, an input vibration signal is received, and at operation 1110, a hierarchical clustering for all assets is performed. Operation 1115 is used to calculate and normalize the centroids for all clusters. At operation 1120, a calculation is performed to calculate the average distance between all centroids and the standard deviation of the distances between centroids.

[0073] At operation 1125, a cluster is selected ( n ) as a reference cluster, and operation 1130 operates to determine whether the clusters under consideration are close enough (ie, similar) to be merged together. Operation 1130 determines whether the clusters n Is the centroid far from the cluster? n+1 If the determination at operation 1130 is "yes," process 1100 proceeds to operation 1135, where cluster n and n+1 After merging, the next cluster to be checked is updated, and the centroid of the newly merged cluster is calculated.

[0074] If the determination at operation 1130 is "no," process 1100 proceeds to operation 1140 where the cluster n and n+1 No merging is done, and the next cluster to be checked is updated. Thereafter, operations 1125-1140 are repeated until no further merging of clusters is possible.

[0075] Figure 12 to Figure 14 Each involves a different process or rules engine algorithm to determine specific high speed shaft (HSS) faults that may be manifested in vibration spectrum data of a wind turbine asset. Fig.12 Relates to classifications used to determine the fault state of an asset (e.g. Fig. 91200 is an example of a "rule engine" for detecting an IRBP type fault in an HSS. In particular, the flowchart 1200 is an example process for detecting an IRBP type fault in an HSS. At operation 1205, an input vibration signal is received, and at operation 1210, the received vibration signal is normalized. Continuing to operation 1215, the harmonic series in the normalized signal are identified, and at operation 1220, for each harmonic peak, an exponential right scan of the signal 150 is performed to identify the first 5 peaks. At operation 1225, for each peak identified, the signal is checked to see if the peak exists at the same distance away from the harmonic peak to find the sideband width.

[0076] At operation 1230, if greater than 50% of the harmonic peaks exhibit peaks at equal distances from the harmonic peaks, then the signal is classified as indicating an IRBP fault at operation 1235. If less than 50% of the harmonic peaks do not exhibit peaks at equal distances from the harmonic peaks at 1230, then process 1200 continues to operation 1240 where the most common sideband width is identified and for the first three harmonic peaks, a check is performed to determine if a sideband exists at the identified sideband width.

[0077] If, as determined at operation 1245, less than 2 harmonic peaks show sidebands at the identified sideband width, then at operation 1255 the signal is classified as indicating an IRBP fault. Otherwise, at operation 1250 the signal is classified as indicating a "healthy" asset.

[0078] Fig.13 Relates to classifications used to determine the fault state of an asset (e.g. Fig. 9 1300 is an example of a "rule engine" for detecting an ORBP type fault in an HSS. In particular, flowchart 1300 is an example process for detecting an ORBP type fault in an HSS. At operation 1305, an input vibration signal is received, and at operation 1310, the received vibration signal is normalized. Operation 1315 operates to filter all indices below 300. Operation 1320 identifies the index of the highest peak in the filtered spectrum from operation 1315.

[0079] At operation 1325, the first 5 harmonics of the highest peak are calculated to form a manifest. If at operation 1330 the number of harmonics without sidebands is not greater than or equal to 2 (i.e., less than 2), then at operation 1335 the asset associated with the signal is considered "healthy". If operation 1330 determines that the number of harmonics without sidebands is greater than or equal to 2, then at operation 1340 a determination is made as to whether there are more than 3 consecutive peaks greater than 0 in the manifest (from operation 1325). If there are more than 3 consecutive peaks greater than 0 in the manifest, then at operation 1345 the signal is classified as indicating an ORBP fault. Otherwise, at operation 1350 the asset associated with the signal is considered "healthy".

[0080] Fig.14 is a classification used to determine the fault condition of an asset (e.g. Fig. 9 1400 of an example flow chart of a "rule engine" for detecting a BS-type failure in an HSS. In particular, flow chart 1400 is an example process for detecting a BS-type failure in an HSS. At operation 1405, an input normalized vibration signal is received, and at operation 1410, the received vibration signal is checked to identify indices where the peaks have non-zero values. The indices identified as having non-zero values ​​are saved to "nonZeroList". For the indices identified by "nonZeroList", operation 1415 identifies groups of peaks that are less than 4 indices apart from each other and saves them to "subList".

[0081] At operation 1420, "subList" is checked to identify locations where the slope of the signal changes (i.e., inflection points). In addition, only those inflection points where the slope changes less than 0.2 (i.e., identifying triangles that characterize BS failures) are retained for further processing. At operation 1425, the number of triangles present in the signal ("nTriangles") is calculated, and process 1400 proceeds to operation 1430.

[0082] Referring to operation 1435 , an input vibration signal is received, and at operation 1440 , if the position of the highest peak in the signal is greater than 10 and less than 100, and the first three multiples of the highest peak are non-zero, “BSFlag” is set to 1, otherwise “BSFlag” is set to 0. Process 1400 proceeds from operation 1440 to operation 1430 .

[0083] At operation 1430, it is determined whether "BSFlag" = 1 and "nTriangles" ≥ 5. If "BSFlag" = 1 and "nTriangles" ≥ 5, the signal is identified as indicating that the asset has a BS failure. Otherwise, at operation 1450, the asset associated with the signal is considered "healthy".

[0084] Fig.15 A rules engine or algorithm is provided for determining specific low speed shaft (LSS) faults that may be present in vibration spectrum data of a wind turbine asset. Fig.15 Two (2) processes are included, where process 1501 is performed during the first pass of the signal, and process 1502 is performed during the second pass of the signal. Referring to process 1501 and starting at operation 1505, the amplitude ratio is calculated, such as Fig.15 At operation 1510, spectrum amplitudes having an amplitude ratio greater than a threshold are classified as high amplitude peaks. At operation 1515, high amplitude peaks that are close to each other are grouped together.

[0085] At operation 1520, the grouped high amplitude peaks are filtered based on the number of peaks in the group, where any group with less than 4 peaks is filtered out. Fig.15 At operation 1530, low energy groups are also filtered out, where a low energy group in this example is a group with a "group ratio" less than 1. At operation 1535, statistics are calculated for the asset under consideration. The statistics may include the number of groups in the spectrum, the average energy of each group, the average group ratio, and other indicators.

[0086] Referring to process 1502 and beginning at operation 1540, the assets of this example are grouped based on common gear mesh frequencies and harmonic signatures. At operation 1545, and for each gearbox group, the following values ​​are calculated: the number of haystack modes in the spectrum of each turbine in the group ("nGroup"), the average number of haystack modes across the group ("AvgGroup"), and the number of turbines in each group ("Size").

[0087] At operation 1550, if the "Size" is greater than 10, the standard deviation of the "nGroups" in the gearbox is calculated and saved (e.g., "σ") at operation 1560. At operation 1565, the turbines in the group are classified based on the value of "nGroups". For example, for turbines with "nGroups" less than 1 "σ", the turbine is classified as "healthy"; for turbines with "nGroups" greater than 1 "σ" and less than 2 "σ", the turbine is classified as "potentially faulty"; and for turbines with "nGroups" greater than 2 "σ", the turbine is classified as "faulty".

[0088] At operation 1550, if "Size" is less than 10, then the average value ("AvgGroup") and the average value ("σ") across the larger group are calculated at operation 1552. At operation 1555, the turbines in the larger group are classified based on the value of "nGroups". For example, for a turbine with "nGroups" less than 2 [average value ("AvgGroup")], it is classified as "healthy", and for a turbine with "nGroups" greater than 2 [average value ("σ")], it is classified as "faulty".

[0089] In some embodiments, the processes and systems herein detect faults and fault types directly based on high-resolution spectra, and can significantly save time and manual effort. In some instances, the techniques disclosed herein can, for example, double the productivity of CMS experts (e.g., more turbines per analyst).

[0090] In some embodiments, the processes and systems herein provide a mechanism to analyze and determine faults in vibration spectrum data from a wide variety of assets, including assets for which fault frequency information is unknown and / or unavailable to an analyst. In some embodiments, the processes disclosed herein may be automatically performed, at least in part, in response to one or more events or actions.

[0091] Some embodiments herein provide signature-based, asset (eg, gearbox) independent / agnostic detection processes and systems that enable fault detection even when specific natural fault frequencies do not exist and / or are unknown.

[0092] Fig.16 1 is an illustrative block diagram of an example of an apparatus 1600 according to some embodiments. The apparatus 1600 may include a computing device and may execute program instructions to perform any of the functions described herein. The apparatus 1600 may include implementations of servers, dedicated processor-enabled devices, user entity devices, and other systems, including a cloud server embodiment of at least a portion of the system disclosed herein. According to some embodiments, the apparatus 1600 may include other elements not shown.

[0093] The apparatus 1600 includes a processor 1605 operatively coupled to a communication device 1615 for communicating with other systems, a data storage device 1630, one or more input devices 1610 for receiving input from other systems and entities, one or more output devices 1620, and a memory 1625. The communication device 1615 can facilitate communication with other systems and components (such as other external computing assets and data). The input device(s) 1610 can include, for example, a keyboard, a keypad, a mouse or other pointing device, a microphone, a knob or switch, an infrared (IR) port, a docking station, and / or a touch screen. The input device(s) 1610 can be used, for example, to input information into the apparatus 1600. The output device(s) 1620 can include, for example, a display (e.g., a screen), a speaker, and / or a printer.

[0094] Data storage device 1630 may include any suitable permanent storage device, including magnetic storage devices (e.g., magnetic tape, hard drives, and flash memory), solid-state storage devices, optical storage devices, read-only memory (ROM) devices, random access memory (RAM), storage class memory (SCM), or any other combination of fast access memories.

[0095] The fault rule engine 1635 may include program instructions executed by the processor 1605 to cause the device 1600 to perform any one or more of the processes described herein, including but not limited to Figures 8 to 15 The embodiments are not limited to a single device performing these processes.

[0096] Data 1640 (cached or complete database) may be stored in volatile memory such as memory 1625. Data storage device 1630 may also store data and other program code used to provide additional functionality and / or necessary for the operation of device 1600, such as device drivers, operating system files, etc. Data 1650 may include data related to the asset, which may be used in fault identification herein.

[0097] Although specific features of various embodiments of the present disclosure may be shown in some drawings but not in other drawings, this is only for convenience. According to the principles of the present disclosure, any feature of the drawings may be combined with any feature of any other drawings to reference and / or claim protection.

[0098] This written description uses examples to disclose embodiments (including the best mode) and also to enable any person skilled in the art to practice the embodiments (including making and using any devices or systems and performing any incorporated methods). The patentable scope of the present disclosure is defined by the claims and may include other examples that occur to those skilled in the art. Such other examples are intended to be within the scope of the claims if they have structural elements that do not differ from the literal language of the claims, or if they include equivalent structural elements that are insubstantially different from the literal language of the claims.

Claims

1. A computer-implemented method, the method comprising: receiving vibration spectrum data from a plurality of different assets; Preprocessing the received vibration spectrum data includes: receiving an input vibration signal, calculating a root mean square RMS and a standard deviation σ value of the signal, calculating a standardized signal, executing a linear regression model to calculate a slope of the signal, subtracting a value output by the linear regression from the standardized signal, calculating an RMS energy of the standardized signal, and setting all values ​​less than the RMS value to 0; Determining clusters for the plurality of assets based on shapes of the vibration spectrum data for each of the plurality of assets, the assets grouped in the same cluster having vibration spectrum data of similar spectrum shapes, comprising: performing initial clustering of the assets based on spectrum shapes of associated vibration signals of the assets, merging clusters based on a distance of a centroid of a cluster being less than a certain threshold away from an average centroid of the cluster; Determining one of a plurality of fault classifications for each of the clusters based on application of a domain-derived pattern recognition rule to the vibration spectrum data includes: identifying whether a signal in each of the merged clusters indicates a fault in the asset based on a centroid of the merged clusters; generating an output comprising an association of each of the plurality of assets with the fault classification of the cluster in which the particular asset is grouped; and saving a record of the output to a data storage device; The step of identifying whether the signal in each of the merged clusters indicates a fault in the asset based on the centroid of the merged cluster comprises: determining, for each cluster, a distance of each asset in the cluster from a centroid of the cluster; If the asset is relatively far from the average based on a certain threshold, the asset is classified by the rule engine as to the condition of the asset; An asset is classified as being the same as the centroid asset in its cluster if it is not relatively far from the mean based on a certain threshold; The rule engine further includes any one of (1)-(3): (1) receiving an input vibration signal; Standardizing the received vibration signal; Identify harmonic series in a standardized signal; For each harmonic peak, perform an exponential rightward sweep of the signal 150 to identify the first five peaks; For each peak identified, check the signal to see if the peak exists at the same distance away from the harmonic peak to find the sideband width; If more than 50% of the harmonic peaks exhibit peaks at equal distances from the harmonic peaks, the signal is classified as indicating an IRBP fault; if less than 50% of the harmonic peaks do not exhibit peaks at equal distances from the harmonic peaks then the most common sideband width is identified and for the first 3 harmonic peaks a check is performed to determine if a sideband is present at the identified sideband width, if less than 2 harmonic peaks show a sideband at the identified sideband width and the signal is classified as indicating an IRBP fault, otherwise the signal is classified as indicating a healthy asset; (2) receiving an input vibration signal; Standardizing the received vibration signal; Filter all indices below 300; an index identifying the highest peak in the filtered spectrum; Calculate the first 5 harmonics of the highest peak to form a list; If the number of harmonics without sidebands is less than 2 then the asset associated with the signal is considered healthy, if the number of harmonics without sidebands is greater than or equal to 2 then a determination is made as to whether there are more than 3 consecutive peaks greater than 0 in the list; If there are more than 3 consecutive peaks greater than 0 in the list then the signal is classified as indicating an ORBP failure, otherwise the asset associated with the signal is considered healthy; (3) Receiving an input standardized vibration signal; examining the received vibration signal to identify indices where peaks have non-zero values; Indices identified as having non-zero values ​​are saved to nonZeroList; For the indices identified by nonZeroList, identify the peak groups that are less than 4 indices away from each other and save them to subList; The subList is checked to identify where the slope of the signal changes, and only those inflection points where the slope change is less than 0.2 are retained for further processing; Calculate the number of triangles nTriangles present in the signal; receiving an input vibration signal; If the position of the highest peak in the signal is greater than 10 and less than 100, and the first three multiples of the highest peak are non-zero, then BSFlag is set to 1, otherwise BSFlag is set to 0; It is determined whether BSFlag=1 and nTriangles≥5, if BSFlag=1 and nTriangles≥5 then the signal is identified as indicating that the asset has a BS failure, otherwise the asset associated with the signal is considered healthy.

2. The method according to claim 1, characterized in that The plurality of fault classifications include at least one of: healthy, one or more known fault types, and unknown.

3. The method according to claim 2, characterized in that The one or more known fault types include at least one of an inner race ball passing fault, an outer race ball passing fault, a ball spin fault, a planetary bearing fault, and a ring gear fault.

4. The method according to claim 1, characterized in that: The vibration spectrum data is received from two different levels for at least one of the plurality of different assets.

5. The method according to claim 4, characterized in that Further comprising, for each of the two different levels of the at least one of the plurality of different assets: determining clusters for the plurality of assets based on a shape of the vibration spectrum data; determining one of a plurality of fault classifications for each of the clusters based on application of domain-derived pattern recognition rules to the vibration spectrum data; as well as An output is generated that includes an association of each of the plurality of assets with the fault classification of the cluster in which the particular asset is grouped.

6. The method according to claim 1, characterized in that For each of the plurality of assets, the received vibration spectrum data is averaged over a period of time.

7. The method according to claim 6, characterized in that The received vibration spectrum data includes a plurality of spectrums for each of the plurality of assets.

8. The method according to claim 1, characterized in that The determination of the cluster for the plurality of assets is accomplished by executing an algorithm formulated to identify vibration spectrum data having similar spectral shapes.

9. The method according to claim 8, characterized in that The algorithm is one of at least one of a hierarchical clustering algorithm, a k-means algorithm, a nearest neighbor algorithm, and a combination based on clustering methods.

10. The method according to claim 1, characterized in that Further comprising determining at least one fault in the vibration spectrum data, wherein the determining of the identification of the at least one fault in the vibration spectrum data is accomplished by executing an algorithm formulated to detect the at least one fault.

11. The method according to claim 1, characterized in that: Further comprising extracting at least one feature from the received spectral data for the plurality of assets on a per-asset basis to reduce the dimensionality of the spectral data.

12. The method according to claim 11, characterized in that The determination of the clusters for the plurality of assets is performed on the spectral data of reduced dimensionality.

13. The method according to claim 1, characterized in that The plurality of assets are each a wind turbine.

14. The method according to claim 13, characterized in that The wind turbine includes at least one of a high speed shaft and a low speed shaft, and the determining of the cluster of the plurality of assets and the determining of the identification of at least one fault in the vibration spectrum data are performed independently for each of the at least one high speed shaft and the low speed shaft.

15. A system for performing the method according to any one of claims 1 to 14, comprising: a memory storing processor executable instructions; as well as one or more processors configured to execute the processor-executable instructions to: receiving vibration spectrum data from a plurality of different assets; determining a cluster for the plurality of assets based on shapes of the vibration spectrum data for each of the plurality of assets, assets grouped in the same cluster having vibration spectrum data of similar spectrum shapes; determining one of a plurality of fault classifications for each of the clusters based on application of domain-derived pattern recognition rules to the vibration spectrum data; generating an output comprising an association of each of the plurality of assets with the fault classification of the cluster in which the particular asset is grouped; as well as A record of the output is stored in a data storage device.

16. The system according to claim 15, characterized in that The plurality of fault classifications include at least one of: healthy, one or more known fault types, and unknown.

17. The system according to claim 16, characterized in that The one or more known fault types include at least one of an inner race ball passing fault, an outer race ball passing fault, a ball spin fault, a planetary bearing fault, and a ring gear fault.

18. The system according to claim 15, characterized in that The vibration spectrum data is received from two different levels for at least one of the plurality of different assets.

19. The system according to claim 18, characterized in that Further comprising, for each of the two different levels of the at least one of the plurality of different assets: determining clusters for the plurality of assets based on a shape of the vibration spectrum data; determining one of a plurality of fault classifications for each of the clusters based on application of domain-derived pattern recognition rules to the vibration spectrum data; as well as An output is generated that includes an association of each of the plurality of assets with the fault classification of the cluster in which the particular asset is grouped.

20. The system according to claim 15, characterized in that For each of the plurality of assets, the received vibration spectrum data is averaged over a period of time.

Images