Bearing fault self-detection system and method

Abstract

A method for performing self-detection of bearing defects (500) provides an algorithm (100) for processing condition monitoring data containing vibration harmonics of at least one bearing (105) coupled to a rotating shaft (109), the bearing having an inner and an outer ring. The algorithm is used to confirm, with a high degree of certainty, whether or not a bearing defect is present. Figure 1

Description

Title of the invention: System and method for self-detection of bearing fault Technical field of the invention

[0001] In a train where all axles are equipped with the same axlebox structure, bearings from different manufacturers may be installed, which, even if they have the same spatial envelope and the same "capacity", often have different bearing failure frequencies due to slight changes in internal geometry. Managing the choice of bearing to be installed on a given axlebox, for example using @ptitude Observer (@O) manufactured by Aktiebolaget SKF, whose principal place of business is located at 41550 Gothenburg, Sweden, and updating a database in a timely manner is costly in terms of man-hours and is prone to errors. State of the prior art

[0002] One type of system, manufactured by Aktiebolaget SKF, whose principal place of business is located at 41550 Gothenburg, Sweden, entitled "Insight Rail System" uses GPS linear speed and wheel diameter to determine shaft rpm values ​​for each measurement, however, during the life of a railway vehicle wheelset, wheel profiles experience fatigue wear, bumps, flattening and squatting (spalling) and therefore need to be reprofiled under certain conditions or at regular intervals. Reprofiling is done by turning or grinding and each time decreases the wheel diameter (both wheels on the same wheelset are always kept at the same diameter), and therefore, during the life of a wheelset, its diameter is typically reduced by more than 10% compared to new.Managing individual axle wheel diameters, initially (when Insight is installed) and after each re-profiling, and then updating the @ptitude Observer (@O) database in a timely manner is costly in terms of man-hours and is error-prone.

[0003] Frequently, an Insight Rail system, manufactured by Aktiebolaget SKF, is installed on a train whose bearings are in service and it is therefore possible that one or more bearings have already spalled to some extent, so it must detect bearing spalling without requiring a learning period. Unlike the @0 Protean systems manufactured by Aktiebolaget SKF, whose principal place of business is located at 41550 Gothenburg, Sweden, which both require a "learning" period with a "healthy" bearing to establish trend characteristics and / or alert thresholds. Summary of the invention

[0004] According to one or more non-limiting embodiments, a method is provided for performing bearing fault self-detection. The method comprises the steps of receiving condition monitoring data by a processor from one or more sensors, the condition monitoring data comprising vibration harmonics of at least one bearing coupled to a rotating shaft, the bearing having an inner ring and an outer ring, the method further comprising: a) receiving the vibration data previously or sequentially transformed into the frequency domain and finally provided as tables of magnitudes and frequencies, as well as an approximate shaft speed in rpm or rpm, b) applying a peak determination method which determines individual peaks from the background noise, and the actual peak amplitudes and their exact peak center frequencies,the exact frequency and amplitude of each possible peak being stored in an array, c) identifying all plausible integer suborders of each peak up to a predefined order number including the peak itself that lie within a specific range of target fundamental frequency ranges, one for each defined defect type and storing these peaks together with their harmonic number and theoretical fundamental as order arrays, d) grouping the theoretical fundamental frequencies that lie within each specific target range and are put into groups by one or more of many methods, the theoretical fundamental frequencies having a predefined group size limit and the groups having acceptable grouping characteristics by number of peaks therein and / or the narrowest group sizes being organized into groups and all others as outliers,and the groups stored in order tables, and e) identifying all peaks in the original list that may be plausible sideband components of the peaks identified in step d), which have theoretical fundamental frequencies that lie within each specific target range and are grouped by one or more of numerous methods, the theoretical fundamental frequencies having a predefined group size limit or tolerance and clusters having more than a predefined number of peaks therein being organized into groups and all others being outliers,and f) applying a clustering method to their basic delta frequencies having a predefined cluster size limit or tolerance and the clusters having acceptable clustering characteristics by number of peaks therein and / or the narrowest cluster sizes being organized into sideband groups and all others as outliers, the sidebands being groups stored in tables with reference to their central frequency peak at , from the order tables.

[0005] According to a second aspect, the method provides grouping shapes for each inner ring or outer ring defect (or other types of defects).

[0006] According to a third aspect, the method makes it possible to determine state indicators. Here, the most plausible group can then be selected by (a) correlation with weighted shape components, (b) presence or absence of components (c) RSS of group components (d) pattern recognition method on each group.

[0007] According to another aspect, the method allows creating and using status indicators and elements derived from a group from one to several ranges in machine learning models with predetermined supervised, semi-supervised and unsupervised models for the application class so that defects can be identified using the first measurements and / or continuously learning and developing models specific to this measurement point or to a specific application (multiple measurement points).

[0008] According to another aspect, the method further allows the amplitude and frequency tables to be converted into an envelope acceleration.

[0009] According to another aspect, the method further allows dividing each peak frequency by a number from 1 up to the number of harmonics to be included in turn to determine all plausible theoretical fundamental sub-frequencies.

[0010] According to another aspect, the method further allows subtracting from the frequency of each identified peak all other peak frequencies in turn and dividing the absolute value of the result by the number 1 up to the number of sidebands to be searched in order to identify all plausible sideband peaks.

[0011] According to another aspect, the method further allows to select only the group having (a) the most peaks and / or (b) the narrowest group size tolerance and / or (c) the best harmonic peak presence shape as being representative of the most likely symptom related to that defined fault type and being stored as group component tables.

[0012] According to another aspect, the method further allows to select only the sideband group having (a) the most peaks and / or (b) the narrowest group size tolerance and / or (c) the best shape of sideband peak presence as being representative of the most probable sidebands related to that defined defect type and being stored as tables of sideband components associated with that specific group.

[0013] According to another aspect, the method further allows applying a noise carpet filter to filter noise from the frequency spectrum to remove unwanted spectral noise, which is achieved by keeping only the components that are +10dB above the local spectral carpet level.

[0014] In another aspect, the method further identifies peaks from the noise-filtered spectrum, and then a quadratic peak interpolation process is performed on the unfiltered spectrum that identifies the exact frequency as orders of assumed shaft speed and amplitude of each possible peak and stores them in an array, the result creating normalized arrays of peak frequencies and peak amplitudes.

[0015] According to another aspect, the method further allows for the elimination of known mechanical frequency peaks using a module leaving a set of frequencies which comprise the bearing fault frequency components.

[0016] According to another aspect, the method further allows to determine, using a "fundamental frequency in band" function, whether each peak itself or an exact suborder thereof is in the frequency range of fundamental frequencies of interest (FunL -> FunH) and to create a new table of original fundamentals, harmonic numbers and indices, those whose fundamentals are grouped together within pre-specified or training tolerances are grouped as "orders" for that specific fundamental, then a "sides in band" function is applied, and for the selected group(s) for each order in that group each possible peak in the original table is checked to determine whether its delta frequency with respect to that order is or has a suborder in the sideband range.

[0017] According to another aspect, the method further makes it possible to determine the tolerances learned by artificial intelligence.

[0018] According to another aspect, the method further makes it possible to associate, using a grouping order sub-algorithm, which frequency elements belong to a frequency group.

[0019] According to another aspect, the method further makes it possible to apply an OrdCIs sub-algorithm to determine an RSS of the orders of each group and order peak presence values ​​and an average fundamental frequency.

[0020] According to another aspect, the method further allows applying a SideCIs sub-algorithm to determine the RSS of the sidebands of each group and also sideband peak presence values ​​and an average sideband frequency.

[0021] According to another aspect, the method further allows applying a groupCIs sub-algorithm to determine the group RSS and the relative band percentage CI value.

[0022] According to another aspect, the method further allows the search range for the fundamental frequency to be, for a first iteration, centered around the expected frequency of a known mechanical vibration component and using a low number of harmonics, then by clustering the results the most plausible group is used to calculate a considerably narrower search range and the number of harmonics is increased for a second iteration, from which, with or without the additional clustering option, the identified peaks are eliminated from the peak tables before the iteration for fault frequencies.

[0023] According to another aspect, the method further allows a test for ignoring sides to comprise a) the group has at least 2 "first" sideband components, and / or b) the group has at least 1 "first" sideband component and 2 "second" sideband components.

[0024] In another aspect, the method further allows using a group scalar state indicator method to group shapes, each measurement being attributable to a specific node and / or step(s) and a timestamp as well as the estimated shaft speed (RPM or Hz), quantitative and qualitative scalar CIs derived from non-specific measurements are determined from two sources, firstly a waveform; such as, but not limited to, RMS (Root Mean Square), True Peak, True Peak-to-Peak, crest factor, kurtosis, skewness, % sinusoidal / harmonic activity, and secondly the spectrum; such as, but not limited to, RSS of various fixed bands as absolute and / or normalized to the RSS of the total spectrum, spectral kurtosis, spectral skewness, non-specific spectral harmonic indicator.

[0025] In another aspect, the method further allows using a search band scalar state indicator method to group shapes, each measurement having multiple search bands, to search for fault symptoms, each search band providing an absolute RSS value of the spectral components in that band based on the provided velocity, effectively a velocity tracking band, and a band RSS sub-algorithm being applied to calculate the RSS of the spectral band of interest as derived from the velocity and step ranges and the order and sideband numbers.

[0026] According to another aspect, the method further allows using a CI forming vector method to group shapes, wherein multiple CIs are extracted for each group identified in a measurement.

[0027] According to another aspect, the method further allows to use the most plausible group components to produce the relevant state indicators both quantitative and qualitative which, with their fundamental and sideband frequencies, are passed to (1) an existing "post-scan" logic with parameters derived from a design of experiments, (2) a "decision logic" with parameters designed or learned by AI / ML or (3) AI or ML with trending or "pattern recognition" methods.

[0028] According to another aspect, the method further allows that each peak can be associated only with one and only one group, and called "most plausible group", this group being selected inter alia by a) the most probable combination of harmonics in the group and / or b) the fundamental which is the rotational speed frequency or another mechanical frequency c) a lower fundamental frequency, and after the selection of the most plausible group, peaks associated with this group being eliminated from other groups.

[0029] According to another aspect, the method further allows the frequencies to be kept absolute to identify group shapes with static electrical fundamental frequencies and eliminate them from the peak tables to keep only the velocity tracking component, and when having multiple measurements with varying rotational speed, only those groups that are close to the expected static noise frequencies being selected, and if such groups were detected in M ​​out of N measurements, then these groups being considered as static noise and can be eliminated from further processing, and normal processing then being applied to the filtered data.

[0030] In another aspect, the method further allows the fundamental frequency ranges for group identification to start at a low frequency and be progressively increased in steps, with the range maximum not exceeding twice the range minimum and the next range step minimum not exceeding the previous maximum, and peaks that satisfy inclusions in the group being eliminated from the peak tables before the next range increment so as to eliminate harmonics of lower fundamental frequencies to prevent them from being mistakenly considered as belonging to a group having a higher fundamental frequency.

[0031] According to one or more non-limiting embodiments, additional features and advantages are realized using the techniques of the present disclosure. Other embodiments and aspects of the disclosure are described in detail herein. In order to better understand the disclosure with the advantages and features, reference should be made to the description and drawings. Brief description of the figures

[0032] The subject matter of the invention is particularly emphasized and clearly claimed in the claims at the end of the specification. The above-mentioned and other features and advantages of the embodiments will become apparent from the following detailed description taken in conjunction with the accompanying drawings in which:

[0033] [Fig.l] represents a flow diagram illustrating an example of an algorithm of base according to one or more embodiments.

[0034] [Fig.2] represents a schematic diagram of a basic system 200 for performing the function of the basic algorithm proposed according to one or more embodiments.

[0035] [Fig. 3] represents a flow diagram of a first application of the present invention, and

[0036] [Fig.4] represents a flow diagram of a second application of the present invention, and

[0037] [Fig.5] represents a process flow of the method steps required to perform the function of the algorithm according to the present invention. Detailed description of the invention

[0038] The present embodiments relate to functionality for a proposed method of self-detecting bearing spall defects on railway vehicle axlebox bearings without needing to know the exact bearing designation (or defect frequencies) or have a precise shaft speed. The method involves a "reverse scan" and outputs for multiple identified shapes are disclosed. With the right parameters, the algorithm is applicable to railway monitoring systems and other industries in which the exact bearing designation (or defect frequencies) or precise shaft speed is not known. A sub-algorithm or module typically comprises software that executes a specific set of instructions. However, those skilled in the art will appreciate that a specific set of instructions can also be performed by hardware or a combination of software and hardware.

[0039] The term "method" is used to describe the implementation of various functions throughout this document. A method may provide a number of different ways of performing the intended function and is not intended to be limiting.

[0040] [Fig. 1] depicts a flow diagram illustrating an algorithm 100 that formulates the functionality for a proposed method of self-detecting bearing spall defects on railway vehicle axlebox bearings without needing to know the exact bearing designation (or defect frequencies) or have a precise shaft speed. The self-detection algorithm may be executed on a processor 102 (see [Fig. 2]) or other computing framework comprising and / or utilizing any number and combination of computer devices and networks utilizing various communication technologies, as described herein. The computing framework may be readily scalable, extensible, and modular, with the ability to transition to different services or reconfigure certain features independently of others. The algorithm may be a computer-executable program or a product executable on computer.

[0041] [Fig.2] shows a schematic diagram of a basic system 200 for performing the function of the proposed method which comprises a processor 102 and one or more sensors 104 connected to at least one bearing 105 mounted on a wheelset 108 of a railway vehicle.

[0042] In a sub-algorithm module 110, condition monitoring data comprising vibration harmonics of the at least one bearing 105 coupled to a rotating shaft 109 is received by the processor 102. The vibration data, previously or sequentially transformed into the frequency domain, is processed to obtain tables of magnitudes and frequencies. This includes an approximate shaft speed in rpm or rpm.

[0043] In a sub-algorithm module 120, a peak determination sub-algorithm / module is applied. The peak determination sub-algorithm determines individual peaks from the background noise, and the actual peak amplitudes and their exact peak center frequencies. The exact frequency and amplitude of each possible peak are stored in an array. Here, the frequency of each identified peak is divided by all other peak frequencies in turn. The absolute value of the result is divided by 1 up to the number of sidebands to be searched in order to identify all plausible sideband peaks.

[0044] Regardless of the format in which the condition monitoring data is provided, it is frequently converted to an envelope acceleration, which is normally done by an acquisition system and then to a spectrum preferably with peak-to-peak scaling and using the Hanning window before a Fast Fourier Transform (FFT). The FFT results are converted to "amplitude" and the spectrum truncated into number segments with a spectral FMax of (waveform sampling rate samples per second) / 2.56. For example, a waveform having 4096 samples acquired at 2560 samples per second will produce a spectrum of 1601 segments (including the zero segment) with an FMax of 1000 Hz (last segment).

[0045] To improve the identification of spectral peaks, it is advantageous to zero out all spectral components considered to be in or near the noise carpet and keep only those that are clearly above the carpet for the pattern scan correlation. A filter threshold of 10 dB and above gives good results in terms of eliminating unwanted noise. Here, a module is provided for calculating the local carpet level and eliminating the "noise" components for the noise-filtered spectrum. This is achieved by traversing the spectrum segment by segment, calculating the median value using a number of segments on either side of this specific segment corresponding to + and - 1 / 32 of the number of segments spectral. If this specific segment has a value greater than or equal to the median value plus 10 dB (or median value*3.16), its value is kept, otherwise its value is zero.

[0046] For each non-zero segment value in the noise-filtered spectrum, it is necessary to determine whether it or the interior of a segment spacing is representative of a "peak" and calculate its exact frequency and amplitude by the quadratic peak identification / interpolation method using the unfiltered spectrum to create tables of (a) peak frequencies (PkHz) and (b) peak amplitudes (PkMags).

[0047] In a sub-algorithm module 130, all plausible integer sub-orders of each peak up to a predefined order number including the peak itself that fall within a specific range of target fundamental frequency ranges are identified. Here, the frequency of each peak is divided by a number from 1 up to the number of harmonics to be included in turn to determine all plausible theoretical fundamental sub-frequencies. One peak is identified for each defined fault type. The peaks are stored along with their harmonic number and theoretical fundamental as order arrays.

[0048] Parameters relating to the fault characteristics 135 are stored in a separate database and are accessible via the sub-algorithm module 130. The fault information relates to each defined fault type, such as the BPFO Bail (roller) Pass Frequency Outer-race frequency, the frequency of passage of the balls (rollers) on the outer ring, and the BPFI Bail (roller) Pass Frequency Inner-race frequency, the frequency of passage of the balls (rollers) on the inner ring.

[0049] The fault fundamental frequency as an order of rolling speed is specific to the fault. Examples may be 6.23 and 11.34, etc. A typical / default fault fundamental range in + / -% is <1% to 33%. A typical / default fault sideband base delta frequency as an order of rolling speed is 0.37, 1.0, etc. Further, a typical / default fault sideband range in + / -% is <1 to 33. Furthermore, a typical / default fundamental (harmonic) order range to be included is 3 to 100+. Finally, a typical / default number of sidebands to be included (on either side) is 0 to 10. Therefore, all fault information disclosed in this paragraph is accessible through the sub-algorithm module 130 to identify faults.

[0050] In a sub-algorithm module 140, all peaks in the original list that can be plausible sideband components of the peaks identified in the sub-algorithm module 130 are identified. Peaks that can be plausible sideband components are identified by those having a delta frequency base in a specific range of the target sideband frequency ranges, one for each defined defect type, and up to a predefined order of sideband + / - delta to determine all plausible theoretical base delta sideband frequencies and store these plausible sideband peaks along with the sideband number. The theoretical delta sideband frequency and center peak ID index that fall within each of the target sideband ranges are stored as sideband arrays. All features regarding the defects 135 disclosed in this paragraph are accessible through the sub-algorithm module 140 to identify plausible peaks.

[0051] In a sub-algorithm module 145, a mechanical frequency elimination function is performed to eliminate mechanical noise. This option is used to eliminate wheel noise components interposed between the peak array module 120 and the fundamental peak processing modules 130 and 140. Only the group having (a) the most peaks and / or (b) narrowest group size (tolerance) and / or (c) best harmonic peak presence shape is selected as representative of the most likely symptom related to that defined fault type and is stored as group component arrays. The mechanical frequency elimination function can be manually "on" or "off" or even made dependent on the amount of wheel noise activity present.

[0052] The mechanical frequency elimination function option 145 obtains inputs from databases that contain parameters and characteristics that are common to all mechanical noise. A first input is received from a mechanical clustering parameter module 147. In the mechanical clustering parameter module 147, a default / typical range for the smallest acceptable cluster size in the number of peaks is in a range between 2 and 5. The mechanical order clustering limits as orders of rolling speed for a first iteration are approximately 0.01. Note: This default / typical range may vary in value and definition depending on the clustering method used since many types of clustering methods can be used such as DBSCAN, OPTICS, mixture of Gaussians etc.The mechanical order clustering limits as rolling speed orders for a second iteration are approximately 0.006 for the "point distance". The default / typical range for the second iteration may also vary in value and definition depending on the clustering method used. The mechanical sideband clustering limits as rolling speed orders are approximately 0.01. The default / typical range may vary in value and definition depending on the clustering method used.

[0053] A second input to the mechanical frequency elimination function 145 is received from a second mechanical frequency elimination module 149. Here, a default / typical mechanical fundamental frequency as the order of the bearing speed or shaft speed is 1. A typical / default fundamental fault range in + / - % is <1% to 33%. A typical / default fault sideband base delta frequency as the order of the bearing speed is 0.37, 1.0 etc. Further, a typical / default sideband fault range in + / - % is <1 to 33. A typical / default fundamental harmonic order range to be included in a first iteration is 3 to 11. Further, a typical / default fundamental harmonic order range to be included in a second iteration is 0 to up to 100+. Finally, a typical / default number of sidebands to be included (on each side) is 0 to 10.Therefore, all fault information disclosed in this paragraph is accessible through the mechanical frequency elimination function 145 to identify mechanical faults.

[0054] In a sub-algorithm module 150, peaks that have theoretical fundamental frequencies that fall within each specific target range are clustered by one or more of many methods. Theoretical fundamental frequencies having a predefined cluster size limit and clusters having more than a predefined number of peaks therein are organized into groups and all others as outliers, and the groups are stored in order tables. A "group" in a measurement consists of spectral components (identified peaks) related by being nearly exact orders of a fundamental in the fundamental range of interest, whether one of the peaks or theoretical and other spectral components all having a delta frequency relative to an "order" that is nearly exactly the same and within the sideband range of interest.

[0055] In a sub-algorithm module 155, fault clustering parameters common to all defined faults are stored in a separate database and are accessible through the sub-algorithm module 150. A smallest acceptable cluster size in number of peaks is in a range between 2 and 5. The limits for clustering fault orders as rolling speed orders are approximately 0.005. The limits for clustering fault sidebands as rolling speed orders are approximately 0.01. All features regarding the fault clustering parameters 155 disclosed in this paragraph are accessible through the sub-algorithm module 150 to identify plausible peaks.

[0056] In a sub-algorithm module 157, a decision as to the routing of the group orders may or may not be optimized. If the group orders are to be optimized in a sub-algorithm module 159, only the group having (a) the most peaks and / or (b) the narrowest group size (tolerance) and / or (c) the best harmonic peak presence shape is selected as representative of the most likely symptom related to that defined fault type and is stored as group component arrays.

[0057] During optimization in sub-algorithm module 159, unwanted groups are eliminated from a list of potential groups stored in the arrays. The groups to be eliminated are generally a subgroup of another group having a lower or higher fundamental frequency, such that, but not limited to (a) the majority of the order peaks in the group are present as orders in another group of lower fundamental frequency, (b) the peak orders are all multiples of either one, 2, 3 or more, (c) a majority of the first few orders of the fundamental must exist as peaks, generally 2 of the first 3 or 3 of the first 5, etc.

[0058] Further, during cluster optimization 159, peaks of orders that are in more than one cluster are identified. This is determined by one or more probability-based rules or models evaluating the relationship between the peak and other peaks in the cluster, the cluster from which it most likely originates, and its elimination from other clusters. This includes, but is not limited to, whether (a) the specific fundamental frequency of the peak is closer to the median value of the fundamental frequency of one cluster than the other, or (b) the order of the peak in one cluster is lower than its order in the other.

[0059] In a sub-algorithm module 160, it is determined which sidebands of the sideband arrays relate to the center frequency components that are present in a group component array and have delta sideband frequencies within the target sideband frequency range for that group defect definition. A clustering sub-algorithm is applied to their basic delta frequencies with a predefined group size limit. Clusters having more than a predefined number of sideband peaks in them are organized into sideband groups and all others as outliers. The sideband groups are stored in arrays with reference to their respective center frequency peak from the order arrays.All features regarding the fault clustering parameters 155 disclosed above, which are stored in a separate database, are accessible through the sub-algorithm module 160 in order to identify plausible peaks.

[0060] In a sub-algorithm module 167, a decision as to whether or not to route the group sidebands may be optimized. If the sidebands of groups are to be optimized in a sub-algorithm module 169, only the sideband group having (a) the most peaks and / or (b) the narrowest group size (tolerance) and / or (c) the best sideband peak presence shape is selected as representative of the most likely sidebands related to that specific defined defect type and is stored as arrays of sideband components associated with that specific group.

[0061] During optimization 169, it is determined whether each sideband peak is also a peak present in another group through one or more probability-based rules or models. The rules include, but are not limited to, (a) the peak is a sideband component in one group and an order component in the other (b) the peak is positioned at a lower sideband delta frequency multiple in one group than in the other, (c) the sideband peak base delta frequency is closer to the median value of one group than the other group, (d) the sideband peak is part of a larger group in one group than in the other.

[0062] Further, during sideband optimization, it is determined whether each sideband peak in a group is likely to be truly related to that group by one or more probability-based rules or models. The rules include, but are not limited to, (a) whether its central peak is an order peak in the same group, (b) if the delta frequencies of the sideband group are all multiples of the base delta frequency of any of 2, 3 or more, then the sidebands belong to another base delta frequency and if they are outside the defined sideband range or are deemed unlikely, eliminating the sideband peaks from the group.

[0063] After the group orders have been optimized 159 and the group sidebands have been optimized 169 (or not) in their respective sub-algorithm modules, the two outputs are combined into an output containing the combined groups represented in a sub-algorithm module 175. The output of the combined optimized groups provides the skilled person with the necessary information to automatically detect bearing defects. That is, the basic algorithm separates the wheat from the chaff so to speak by providing related spectral component shapes from each defined range of interest. At this point in the disclosure, the essential function of the present invention has been disclosed.

[0064] Some applications of the base algorithm 100 will now be disclosed. [Fig. 3] shows an application 300 of the base algorithm 100 whose outputs are provided as input to a block 310. In block 310, group scalar state indicator (SSI) measurements are first identified. Each scalar SSI measurement is attributable to a specific node and / or step and a timestamp as well than at the estimated shaft speed (rpm or Hz). Quantitative and qualitative scalar CIs derived from non-specific measurements can be determined easily from two sources.

[0065] First, the waveform, such as, but not limited to, RMS (Root Mean Square), True Peak, True Peak-to-Peak, crest factor, flattening, asymmetry, % sinusoidal / harmonic activity. Second, the spectrum, such as, but not limited to, RSS of various fixed bands as absolute and / or normalized to the RSS of the total spectrum, spectral flattening, spectral asymmetry, non-specific spectral harmonic indicator. RSS is the acronym for Root Sum Squared and corresponds to a calculation of the Square Root of the Sum of the Squares of the Spectral Amplitudes.

[0066] In addition, search band scalar IC measurements are identified. Each search band scalar IC measurement may have multiple search bands, although currently only one is used to search for both BPFO and BPFI fault symptoms. If roller faults were to be added to the monitoring list, then one search band for BSF (rolling element fault frequency) symptoms and another for FTF (cage fault frequency) symptoms would need to be included. Each search band provides an absolute RSS value of the spectral components in that band based on the provided speed, effectively a speed tracking band. It stamps measurement date / time, RPM, and bandRSS. Measurement ICs based on waveforms and spectral components may also be included.

[0067] Furthermore, for each group identified in a measurement, several CIs are extracted. The "specified band" is derived from the theoretically possible lowest frequency component (i.e., 1st order minus 3 sidebands) to the highest (i.e., last order plus 3 sidebands) over all fundamental and sideband frequency ranges as derived in a BANDrss function. The BANDrss function allows the RSS of the spectral band of interest to be calculated as a derivative of the speed and step ranges and the order and sideband numbers.

[0068] In a block 320, the group CIs are calculated. The group CIs are calculated in three steps (functions). A function OrdCIs determines the RSS of the orders of each group (Orss) and order peak presence values ​​(OPreT, Opre3) and the average fundamental frequency (OFunF). A function SideCIs determines the RSS of the sidebands of each group (Srss) and also sideband peak presence values ​​(SPreT, SPre2) and the average sideband delta frequency (SbndF). A function groupCIs determines the group RSS value (Grss) and the relative CI value of band percentage (Gpob).

[0069] Furthermore, the group ICs can then be applied to artificial intelligence and machine learning models. First, predetermined supervised models for the application class can be applied so that faults can be identified using the first measurements. Continuously learning and developing models specific to this measurement point or a specific application (multiple measurement points) that can take into account changing noise factors and new operating conditions can be applied.

[0070] A first approach uses a learning process performed on classified labeled data sets covering a wide range of fleets / applications and then applying the preconfigured models to all fleets / applications using bearings that meet the criteria. This AI / ML (Artificial Intelligence / Machine Learning) approach is considered to be able to provide immediate fault detection in a few measurements even if the node is installed on a bearing that is already faulty.

[0071] A second approach uses data-driven anomaly detection based on each node / bearing in isolation. While this method of detecting a transition from a healthy bearing state to a faulty state is considered reliable, there are concerns about installing a node on a bearing that is already faulty.

[0072] A combination of the first and second methods mentioned above followed by continuous optimization of the second method provides an algorithm according to the present invention which can be used to confirm with a high degree of certainty whether a bearing defect is present or not.

[0073] Training an AI / ML model with labeled data may be performed using one or both of the following approaches with respect to the labeled data. 1) Each individual measurement among one or more data sets is labeled as to whether it individually exhibits symptoms of a specific fault. 2) The labeling is performed at a data set level representative of a tier exhibiting a specific fault.

[0074] Another application of the basic algorithm 100 will now be disclosed. [Fig. 4] shows an application 400 of the basic algorithm 100 whose outputs are provided as input to a block 410. Here, group shapes define the following for each bearing fault (inner or outer): fundamental component, none or some harmonic components, none or some sideband components having a delta frequency other than its fundamental (usually another fundamental). For the bearing fault shapes, three factors (for shape construction) are defined:

[0075] harmonics as a number of orders (including the fundamental)

[0076] the number of side bands

[0077] the fundamental frequency and mechanical frequency identifier for the sidebands.

[0078] The advantage of treating all group forms together is that many of them will have components that are also present in another group's form but in reality should only be allowed to be 'present' in one group, therefore a decision process is required to determine which group this component fits best into and remove it from the other. In many cases, the less suitable group will likely be 'destroyed' to the point where it can no longer be considered a group and eliminated from the evaluation process.

[0079] Furthermore, the group shapes can then be applied to artificial intelligence and machine learning models. Here, predetermined learned shapes for the application class (not only in the railway field) so that shapes can be identified using the first measurements as defects or other mechanical sources. Continuously learning and developing shape models specific to this measurement point or a specific application (multiple measurement points) that can take into account changing noise factors and new operating conditions can be applied.

[0080] A method for performing self-detection of bearing fault 500 will now be disclosed in association with Figures 1-5.

[0081] In a first step 510, the method makes it possible to receive condition monitoring data by a processor, from one or more sensors. The condition monitoring data comprises vibration harmonics of at least one bearing coupled to a rotating shaft. The bearing having an inner ring and an outer ring.

[0082] In a second step 520, the method makes it possible to receive vibration data previously or sequentially transformed into the frequency domain and finally provided as tables of amplitudes and frequencies, as well as an approximate shaft speed in rpm or rpm.

[0083] In a further step 530, the method applies a peak determination method that determines individual peaks from the background noise, and the actual peak amplitudes and their exact peak center frequencies. The exact frequency and amplitude of each possible peak are stored in an array.

[0084] In a subsequent step 540, the method enables the identification of all plausible integer sub-orders of each peak up to a predefined order number including the peak itself that lie within a specific range of target fundamental frequency ranges, one for each defined defect type and the storage of these peaks as well as their harmonic number and their theoretical fundamental as order tables.

[0085] In a step 550, the method enables clustering of theoretical fundamental frequencies that fall within each specific target range and are clustered by one or more of numerous methods, the theoretical fundamental frequencies having a predefined cluster size limit and clusters having acceptable clustering characteristics by number of peaks therein and / or the narrowest cluster sizes being organized into clusters and all others as outliers, and the clusters stored in order tables.

[0086] In a further step 560, the method enables identification of all peaks in the original list that may be plausible sideband components of the peaks identified in step 550, which have theoretical fundamental frequencies that fall within each specific target range and are clustered by one or more of numerous methods. The theoretical fundamental frequencies having a predefined cluster size limit or tolerance and clusters having more than a predefined number of peaks therein are organized into groups and all others as outliers, and the groups are stored in order tables.

[0087] In a subsequent step 570, the method allows the application of a clustering method to the basic delta frequencies having a predefined cluster size limit or tolerance. Clusters that have acceptable clustering characteristics by number of peaks therein and / or the narrowest cluster sizes are organized into sideband groups and all others as outliers. The sideband groups are stored in arrays with reference to their respective center frequency peak from the order arrays.

[0088] In a step 580, the method provides cluster shapes for each inner ring or outer ring bearing defect (or other types of defects). Three shape construction factors are defined by 1) identification of harmonics as a number of orders, 2) number of sidebands, and 3) fundamental frequency range and frequency range for the sidebands, one fundamental component, none or some harmonic component(s), none or some sideband component(s) having a delta frequency other than its fundamental comprise the possible cluster shapes.

[0089] In a step 590, the method further allows applying the grouped shapes against the predetermined learned shapes for the application class so that shapes can be identified using the first measurements as defects or other mechanical sources. This is achieved by continuously learning and developing the shape models specific to that measurement point or to a specific application (multiple measuring points) capable of taking into account changing noise factors and new operating conditions.

[0090] In another step, the method allows to determine state indicators. Here, the most plausible group can be selected by (a) correlation with weighted shape components, (b) presence or absence of components (c) RSS of group components (d) pattern recognition method on each group.

[0091] In a further step, the method comprises creating and using status indicators and characteristics derived from a group from one to several ranges in machine learning models with predetermined supervised, semi-supervised and unsupervised models for the application class so that faults can be identified using the first measurements and / or continuously learning and developing models specific to this measurement point or to a specific application (multiple measurement points) capable of taking into account changing noise factors and new operating conditions, and identifying faults with a high degree of certainty on the basis of the algorithm coded using a compiler or a computing resource with a high-level language, c) by means of machine learning,and using an artificial intelligence algorithm to “learn” what defects exist.

[0092] In another step, the method comprises converting the magnitude and frequency arrays into an envelope acceleration via a data acquisition system, and / or Hanning windowing is performed on the data prior to the Fast Fourier Transform (FFT), and / or a Fast Fourier Transform or Real Fourier Transform (RealFT) is performed on the data to generate spectral data, and / or the magnitude of the Fast Fourier Transform data is calculated from the real and imaginary spectral components, "magnitude" = SQRT(RealA2 + ImagA2), and / or the spectrum is truncated.

[0093] In a next step, the method divides each peak frequency in step 540 by a number from 1 to the number of harmonics to be included in turn to determine all plausible theoretical fundamental sub-frequencies.

[0094] According to another aspect, the method makes it possible to subtract from the frequency of each peak identified in step 530 all the other peak frequencies in turn and to divide the absolute value of the result by the number 1 up to the number of sidebands to be searched in order to identify all the plausible sideband peaks.

[0095] In a next step, the method makes it possible to select at step 560 only the group having (a) the most peaks and / or (b) the narrowest group dimension tolerance and / or (c) the best harmonic peak presence shape as being representative of the most probable symptom related to this defined fault type and being stored as group component tables.

[0096] In a further step, the method allows to select in step 570 only the sideband group having (a) the most peaks and / or (b) the narrowest group size tolerance and / or (c) the best sideband peak presence shape as being representative of the most probable sidebands related to that specific defined defect type and being stored as tables of sideband components associated with that specific group.

[0097] In one step, the method applies a noise carpet filter to filter noise from the frequency spectrum to remove unwanted spectral noise, which is achieved by keeping only the components that are +10dB above the local spectral carpet level, and the others are reduced to zero.

[0098] In a next step, the method identifies peaks from the noise-filtered spectrum. Then a quadratic peak interpolation process is performed on the unfiltered spectrum, and it identifies the exact frequency as orders of presumed shaft speed and amplitude of each possible peak, and stores them in an array. The result creates normalized arrays of peak frequencies and peak amplitudes.

[0099] In another step, the method eliminates known mechanical frequency peaks using a module leaving a set of frequencies that include bearing fault frequency components such as the inner ring and outer ring of the bearing and mechanical peaks result from the test apparatus such as shaft speed rather than the bearing. Then a "fundamental frequency in band" sub-algorithm is invoked twice to select array peaks that may have a sub-order and thus a theoretical fundamental frequency within a specified frequency band.

[0100] In a further step, the method determines, using a "fundamental frequency in band" function, whether each peak itself or an exact suborder thereof lies within the frequency range of fundamental frequencies of interest (FunL -> FunH) and creates a new table of original fundamentals, harmonic numbers and indices. Those whose fundamentals are grouped together within learned pre-specified tolerances are grouped as "orders" for that specific fundamental. Then, a "sides in band" function is applied. For the selected group(s) for each order in that group each possible peak in the original table is checked to determine whether its delta frequency with respect to that order is or has a suborder in the sideband range.

[0101] In another step, the method makes it possible to determine the tolerances learned by artificial intelligence.

[0102] In another step, the method makes it possible to associate, using the grouping order sub-algorithm, which frequency elements belong to a group of frequencies. A dimensional clustering method performs cluster outlier removal. Groupings by specific tolerances can be obtained by (i) segment size reduction histograms for one or more clusters, (ii) size reduction clustering by outlier removal for a single cluster, (iii) multiple clusters for multiple clusters. The outputs of the Clusterorders and / or Remove-Subs subalgorithms are arrays of individual order components sorted into an identifiable cluster(s).

[0103] In another step, the method makes it possible to apply an OrdCIs sub-algorithm to determine an RSS of the orders of each group and order peak presence values ​​and an average fundamental frequency. A SideCIs sub-algorithm is then applied to determine the RSS of the sidebands of each group and also sideband peak presence values ​​and an average sideband frequency. A groupCIs sub-algorithm is applied to determine the group RSS and the relative band percentage CI value.

[0104] In a further step, the method allows the search range for the fundamental frequency to be, for a first iteration, centered around the expected frequency of a known mechanical vibration component and using a low number of harmonics, then by clustering the results the most plausible group is used to calculate a considerably narrower search range and the number of harmonics is increased for a second iteration, from which, with or without the additional clustering option, the identified peaks are eliminated from the peak tables before iteration steps 510, 520, 530, 540, 550, 560 and 570 for the fault frequencies.

[0105] In a further step, the method allows the test for ignoring sides to include a) the cluster has at least 2 "first" sideband components, and / or b) the cluster has at least 1 "first" sideband component and 2 "second" sideband components. The outputs of the ClusterSides and / or IgnoreSides functions are arrays of individual sideband components sorted into the same cluster(s) of the order component as that which is their "center frequency".

[0106] In another step, the method allows using a group scalar state indicator method to group shapes. Here, each measurement is attributable to a specific node and / or bearing(s) and a timestamp as well as the estimated shaft speed (RPM or Hz), quantitative and qualitative scalar CIs derived from non-specific measurements are determined from two sources, firstly a waveform; such as, but not limited to, RMS, True Peak, True Peak-to-Peak, crest factor, kurtosis, skewness, % sine / harmonic activity, and secondly the spectrum; such as, but not limited to, RSS of various fixed bands as absolute and / or normalized to the RSS of the total spectrum, spectral flattening, spectral asymmetry, non-specific spectral harmonic indicator.

[0107] In another step, the method allows using a search band scalar state indicator method to group shapes. Here, each measurement has multiple search bands, to search for fault symptoms. Each search band provides an absolute RSS value of the spectral components in that band based on the provided velocity, effectively a velocity tracking band. A band BANDrss sub-algorithm is applied to calculate the RSS of the spectral band of interest as derived from the velocity and step ranges and the order and sideband numbers.

[0108] In another step, the method allows using a state indicator vector method to group shapes. For each group identified in a measurement, several CIs are extracted.

[0109] In a further step, the method allows to use the most plausible group components to produce the relevant state indicators both quantitative and qualitative which, with their fundamental and sideband frequencies, are passed to (1) a "post-scan" logic with parameters derived from a design of experiments, (2) a "decision logic" with parameters designed or learned by AI / AA or (3) AI or AA with trend or "pattern recognition" methods.

[0110] In another step, the method allows step 560 to select each peak that can be associated with only one and only one group, and called "most plausible group". This group is selected among others by a) the most probable combination of harmonics in the group and / or b) the fundamental which is the rotational speed frequency or another mechanical frequency c) a lower fundamental frequency. After the selection of the most plausible group, peaks associated with this group are eliminated from other groups.

[0111] In another step, the method keeps the frequencies (i.e. in Hz or rpm) in order to identify group shapes with static fundamental frequencies (not a speed tracking frequency, e.g. electrical) and eliminates them from the peak tables to keep only the speed tracking component, because when its frequency is close to the expected bearing fault frequencies it can be confusing. When having multiple measurements with varying rotational speed, only those groups that are close to the expected static noise frequencies are selected. If such groups were detected in M ​​out of N measurements, then these groups are considered as static noise and can be eliminated from further processing. Then, normal processing is applied to the filtered data.

[0112] In another step, the method allows for progressively increasing fundamental frequency ranges for group identification that start at a low frequency and are progressively increased in steps, the range maximum not exceeding twice the range minimum and the next range step minimum not exceeding the previous maximum, and peaks that satisfy inclusions in the group are eliminated from the peak tables before the next range increment so as to eliminate harmonics of lower fundamental frequencies to prevent them from being mistakenly considered as belonging to a group having a higher fundamental frequency.

[0113] The present invention may be a system, a method, and / or a computer program product at any possible level of integration of technical detail. The computer program product may include a computer-readable storage medium (or media) having computer-readable program instructions thereon for causing a processor to implement aspects of the present invention.

[0114] Aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams may be implemented via computer-readable program instructions.

[0115] These computer-readable program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which are executed through the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / actions specified in the flowchart and / or block diagram block(s).These computer-readable program instructions may also be stored in a computer-readable storage medium that can cause a computer, programmable data processing apparatus, and / or other devices to operate in a particular manner, such that the computer-readable storage medium on which instructions are stored includes an article of manufacture having instructions that implement aspects of the function / action specified in the flowchart and / or block diagram block(s).

[0116] The computer-readable program instructions may also be loaded onto a computer, other programmable data processing device, or other device for causing a series of functional steps to be performed on the computer, other programmable apparatus or other device to produce a computer-implemented process, such that the instructions that are executed on the computer, other programmable apparatus or other device implement the functions / actions specified in the flowchart and / or block diagram block(s).

[0117] The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which includes one or more functions that may be executed to implement the specified logical function(s). In some alternative implementations, the functions in the blocks may occur in a different order than shown in the figures. For example, two blocks shown in succession may, in fact, be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order, depending on the functionality involved.

[0118] It will also be understood that each block of the block diagrams and / or flowchart illustration, and combinations of blocks in the block diagrams and / or flowchart illustration may be implemented by special-purpose hardware-based systems that perform the specified functions or actions or implement combinations of special-purpose hardware and computer instructions.

[0119] Furthermore, the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. As used herein, the singular forms "a," "an," "the," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It is also to be understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of recited features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or components and / or groups thereof.

[0120] The descriptions of the various embodiments herein have been presented for purposes of illustration but are not intended to be exhaustive or limited to the disclosed embodiments. Numerous modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the disclosed embodiments. The terminology used herein has been chosen to best explain the principles of the embodiments, practical application or technical improvement over technologies on the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. Claims A method for performing self-detection of bearing fault (500) comprising: receiving condition monitoring data by a processor (102) from one or more sensors (104), the condition monitoring data comprising vibration harmonics of at least one bearing (105) coupled to a rotating shaft (109), the bearing having an inner ring and an outer ring, the method further comprising: a. receiving (110) vibration data previously or sequentially transformed into the frequency domain and finally provided as tables of magnitudes and frequencies, as well as an approximate shaft speed in rpm or rpm, b) applying a peak determination method (120) which determines individual peaks from the background noise, and the actual peak amplitudes and their exact peak center frequencies, the exact frequency and amplitude of each possible peak being stored in an array, c) identifying (130) all plausible integer suborders of each peak up to a predefined order number including the peak itself that lie within a specific range of target fundamental frequency ranges, one peak for each defined defect type and storing these peaks together with their harmonic number and theoretical fundamental as order tables, d) grouping (150) the theoretical fundamental frequencies that lie within each specific target range and are put into groups, the theoretical fundamental frequencies having a predefined group size limit and the groupings having acceptable grouping characteristics by number of peaks therein and / or the narrowest group sizes being organized into groups and all others as outliers, and the groups stored in order tables, and e) identifying (140) all peaks in the original list that may be plausible sideband components of the peaks identified in step d), which have theoretical fundamental frequencies that fall within each specific target range and are grouped, the theoretical fundamental frequencies having a predefined group size limit or tolerance and the groups having more than a predefined number of peaks therein being organized into groups and all others being outliers, and the groups stored in order tables and f) applying a grouping method (160) to their basic delta frequencies having a predefined group size limit or tolerance and the groups having acceptable grouping characteristics by number of peaks therein and / or the narrowest group sizes being organized into sideband groups and all others as outliers, the sidebands being groups stored in tables with reference to their respective center frequency peak from the order tables,each base delta frequency being a predetermined frequency associated with a fault according to the order of the bearing rolling speed.,

2. The method of claim 1, further comprising grouping shapes (155) for each inner ring or outer ring bearing defect (or other types of defects), three shape construction factors being defined by 1) identification of harmonics as a number of orders, 2) number of sidebands, and 3) fundamental frequency range and frequency range for the sidebands, wherein a fundamental component, no or some harmonic component(s), no or some sideband component(s) having a delta frequency other than its fundamental comprise the possible grouped shapes.

3. The method of claim 1, further comprising determining state indicators, selecting a most plausible cluster by (a) correlation with weighted shape components, (b) presence or absence of components, (c) RSS of cluster components (d) a pattern recognition process on each cluster.

4. The method of claim 2, further comprising creating and using cluster-derived status indicators and characteristics from one to more ranges in machine learning models with predetermined supervised, semi-supervised, and unsupervised models for the application class such that defects can be identified using the first measurements and / or the learning and development models. continuous specific to that measurement point or to a specific application (multiple measurement points) capable of taking into account changing noise factors and new operating conditions, and the identification of faults with a high degree of certainty based on an algorithm coded using a compiler or a computing resource with a high-level language, through machine learning, and using an artificial intelligence algorithm to "learn" which faults exist.

5. The method of claim 1, further comprising converting the magnitude and frequency arrays into an envelope acceleration via a data acquisition system, and / or Hanning windowing is performed on the data prior to a Fast Fourier Transform (FFT), and / or a Fast Fourier Transform or a Real Fourier Transform (RealFT) is performed on the data to generate spectral data, and / or the magnitude of the Fast Fourier Transform data is calculated from the real and imaginary spectral components, "magnitude" = SQRT(RealA2 + ImagA2), and / or the spectrum is truncated, Real and Imag being respectively real and imaginary spectral components obtained by the Fast Fourier Transform.

6. The method of claim 1, further comprising dividing each peak frequency in step c) by a number ranging from 1 to the number of harmonics to determine all plausible theoretical sub-fundamental frequencies.

7. The method of claim 1, further comprising subtracting the frequency of each peak determined in step b) from all other peak frequencies in turn and dividing the absolute value of the result by a number ranging from 1 to the number of sidebands to be searched for in order to identify all plausible sideband peaks.

8. The method of claim 1, further comprising selecting (560) in step e) only the group having (a) the most peaks and / or (b) the narrowest group size tolerance and / or (c) the best harmonic peak presence shape as being representative of the most likely symptom related to that defined fault type and being stored as group component tables.

9. The method of claim 1, further comprising selecting (160) in step f) only the group of sidebands stored in the arrays having (a) the most peaks and / or (b) the dimension tolerance of narrowest group and / or (c) the best shape of sideband peak presence as being representative of the most likely sidebands related to that specific defined defect type and being stored as tables of sideband components associated with that specific group.

10. The method of claim 1, further comprising applying a noise carpet filter to filter noise from the frequency spectrum of the previously or sequentially transformed vibration data into the frequency domain to eliminate unwanted spectral noise by keeping only those components that are +10dB above a predetermined local spectral carpet level, and the others being reduced to zero.

11. The method of claim 10, further comprising identifying peaks from the noise-filtered spectrum, then a quadratic peak interpolation process is performed on the unfiltered spectrum, and it identifies the exact frequency as orders of presumed shaft speed and amplitude of each possible peak, and stores them in an array, and the result creating normalized arrays of peak frequencies and peak amplitudes.

12. The method of claim 1, further comprising removing (145) known mechanical frequency peaks using a module leaving a set of frequencies that include the bearing fault frequency components such as the inner race and outer race of the bearing, the mechanical peaks resulting from a test apparatus rather than the bearing, and a "fundamental frequency in band" sub-algorithm being invoked twice to select array peaks that may have a sub-order and thus a theoretical fundamental frequency within a specified frequency band.

13. The method of claim 9, further comprising determining, using a "fundamental frequency in band" function, whether each peak itself or an exact suborder thereof lies within a predetermined frequency range of fundamental frequencies of interest (FunL -> FunH) and creating a new table of fundamentals, harmonic numbers and original indices, those whose fundamentals are grouped together within prespecified or learning tolerances being grouped as "orders" for that specific fundamental, an "in band sides" function being applied, and for the selected group(s) for each order in this group each possible peak in the original array being checked to determine whether its delta frequency with respect to that order is or has a suborder in the sideband range.

14. The method of claim 1, further comprising determining the tolerances learned by artificial intelligence.

15. The method of claim 1, further comprising associating with a clustering orders sub-algorithm which theoretical fundamental frequencies belong to a cluster, a dimensional cluster method performing cluster outlier removal, wherein the predefined cluster dimension tolerance clusterings of the theoretical fundamental frequencies are obtainable by (i) segment size reduction histograms for one or more clusters, (ii) size reduction clustering by outlier removal for a single cluster, (iii) multiple clusters for multiple clusters, and wherein the outputs of the Clusterorders and / or Remove-Subs sub-algorithms are arrays of individual order components sorted into an identifiable cluster / clusters.

16. The method of claim 13, further comprising applying (320) an OrdCIs sub-algorithm to determine an RSS of the orders of each group and order peak presence values ​​representative of the number of order peaks and an average fundamental frequency, a SideCIs sub-algorithm being applied to determine the RSS of the sidebands of each group and also sideband peak presence values ​​representative of the number of sideband order peaks and an average sideband frequency, and a groupCIs sub-algorithm being applied to determine the group RSS and the relative band percentage CI value.

17. A method according to claim 15, wherein a search range for the fundamental frequency is, for a first iteration, centered around the expected frequency of a known mechanical vibration component and using a low number of harmonics, then by clustering the results the most plausible group is used to calculate a considerably narrower search range and the number of harmonics is increased for a second iteration, from which, with or without the additional clustering option, the identified peaks are eliminated from the peak tables before the iteration of claim 1 for the fault frequencies.

18. A method according to claim 1, comprising a test for ignoring sides comprising a) the group has at least 2 "first" sideband components, and / or b) the group has at least 1 "first" sideband component and 2 "second" sideband components, and the outputs of the ClusterSides and / or IgnoreSides functions being arrays of individual sideband components sorted into the same cluster(s) of the order component as that which is their "center frequency", the ClusterSides and / or IgnoreSides functions performing steps a) and b respectively.

19. The method of claim 2, further comprising using a group scalar state indicator method to group shapes, each measurement being attributable to a specific node and / or bearing and a timestamp as well as the estimated shaft speed (RPM or Hz), quantitative and qualitative scalar state indicator measurements derived from non-specific measurements being determined from two sources, firstly a waveform; such as, but not limited to, RMS (Root Mean Square), True Peak, True Peak-to-Peak, crest factor, kurtosis, skewness, % sine / harmonic activity, and secondly the spectrum; such as, but not limited to, RSS of various fixed bands as absolute and / or normalized to the RSS of the total spectrum, spectral flattening, spectral asymmetry, non-specific spectral harmonic indicator.

20. The method of claim 17, further comprising using a search band scalar state indicator method to group shapes, each measurement having multiple search bands, to search for defect symptoms, each search band providing an absolute RSS value of the spectral components in that band based on the provided velocity, effectively a velocity tracking band, and a band RSS sub-algorithm being applied to calculate the RSS of the spectral band of interest as derived from the velocity and step ranges and the order and sideband numbers.

21. The method of claim 17, further comprising using a vector method forming state indicator metrics to group shapes, wherein multiple state indicator metrics are extracted for each group identified in a metric.

22. The method of claim 18, further comprising using the most plausible group components to produce indicators relevant state data both quantitative and qualitative which, with their fundamental and sideband frequencies, are passed to (1) an existing “post-scan” logic with parameters derived from a design of experiments, (2) a “decision logic” with parameters designed or learned by AI / ML (Artificial Intelligence / Machine Learning) or (3) AI or ML with trending or “pattern recognition” processes.

23. Method according to claim 1, each peak being able to be associated only with one and only one group, called "most plausible group", this group being selected inter alia by a) the most probable combination of harmonics in the group and / or b) the fundamental which is the rotational speed frequency or another mechanical frequency c) a lower fundamental frequency, and after the selection of the most plausible group, peaks associated with this group being eliminated from other groups.

24. A method according to claim 1, wherein the fundamental frequency ranges for identifying groups comprising the groupings according to e) start at a low frequency and are gradually increased in steps, the range maximum not exceeding twice the range minimum and the next range step minimum not exceeding the previous maximum, and peaks that satisfy inclusions in the group are eliminated from the peak tables before the next range increment so as to eliminate harmonics of lower fundamental frequencies to prevent them from being mistakenly considered as belonging to a group having a higher fundamental frequency.

Images