Method to detect bearing faults in rotating machinery

Abstract

A method to detect a bearing fault in a rotating machine is provided. The method includes receiving current data from the rotating machine during operation at a sampling frequency, obtaining a current frequency spectrum for a sample of the received current data, reducing noise in the current frequency spectrum to create a reduced current frequency spectrum, performing a peak detection algorithm to detect peak amplitude values in the reduced frequency spectrum for a frequency range of the current frequency spectrum, determining a bearing fault from the peak amplitude values in the reduced current frequency spectrum, and responsive to the determination of the bearing fault, generating an alarm.

Description

1 ETN-092XINPCTP23-1363WO01METHOD TO DETECT BEARING FAULTS IN ROTATING MACHINERYCROSS-REFERENCE TO RELATED APPLICATION

[0001] The present invention claims the benefit of India Provisional Application Number 202411070202, filed on September 17, 2024, which is hereby incorporated by reference in its entirety, including any figures, tables, and drawings.BACKGROUND

[0002] In rotating machinery, bearings are an important part of the healthy functioning of the rotating machine. A high percentage of the failures of the rotating machine, e.g., an induction motor, are the result of a faulty bearing. Real-time monitoring and fault diagnosis of faulty bearings are of great importance for scheduling predictive maintenance which reduces downtime of the motor or other rotating machine. When faulty bearings are not detected in a reasonable amount of time, secondary failures, e.g., eccentric windings and other issues that prolong downtime, may lead to catastrophic failures or failures that result in expensive repairs.

[0003] Currently, bearing faults can be determined by different methods that use motor current data. One method calculates the bearing specific fault frequencies in the motor current data. For example, four characteristic frequencies of the ball bearings can be calculated using specific input parameters of the ball bearings, e.g., ball diameter, number of ball bearings, etc. The bearing fault is detected when the frequency peaks around the characteristic fault frequencies exceed a particular threshold. In another method, a healthy baseline motor current spectrum, specific to each rotating machine, is captured and stored when the motor is operating normally. Then, during further operation of the motor, the motor current spectrum is monitored and compared to the healthy baseline. When a difference between the baseline signal and the current operating signal exceeds a threshold, a faulty bearing is detected. With these methods, however, instantaneous detection of faulty bearings is not possible.BRIEF SUMMARY

[0004] Methods for detecting bearing faults in rotating machinery are disclosed. For descriptive and illustrative purposes of the disclosure, the methods describe detecting bearing faults in an induction motor, however, the described methods can be used to determine bearing faults in other types of rotating machinery as well. Methods that detect peak amplitude values in the current frequency spectrum are proposed that can instantaneously detect fault bearings2 ETN-092XINPCTP23-1363WO01 with minimal user input. Additionally, using the proposed methods, bearing roughness faults can also be detected. A bearing roughness fault can be characterized by a degraded bearing surface. Most methods currently in use today, one of which is described above, are directed towards single point defects that rely on locating and processing the characteristic bearing fault frequencies in the stator current. Methods such as these may not be suitable for detecting bearing roughness faults.

[0005] A computer-implemented method to detect a bearing fault in a rotating machine includes the steps of receiving current data from the rotating machine during operation at a sampling frequency, obtaining a current frequency spectrum for a sample of the received current data, reducing noise in the current frequency spectrum to create a reduced current frequency spectrum, performing a peak detection algorithm to detect peak amplitude values in the reduced frequency spectrum for a frequency range of the current frequency spectrum, determining a bearing fault from the peak amplitude values in the reduced current frequency spectrum, and responsive to the determination of the bearing fault, generating an alarm.

[0006] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.BRIEF DESCRIPTION OF THE DRAWINGS

[0007] FIG. 1 illustrates an operating environment for an AC induction motor system according to one embodiment.

[0008] FIG. 2 illustrates a flowchart of a method to detect a bearing fault in a rotating machine.

[0009] FIG. 3 illustrates a reduced current frequency spectrum depicting detected peak amplitudes.

[0010] FIG. 4A illustrates a distribution plot of the number of peak amplitude values detected for a healthy inductor motor and a faulty induction motor.

[0011] FIG. 4B illustrates a distribution plot of a summation of peak amplitude values detected for a healthy induction motor and a fault induction motor.

[0012] FIG. 5 illustrates a flowchart of an implementation for determining a severity score and fault status.

[0013] FIG. 6 illustrates a schematic diagram illustrating components of a computing device.3 ETN-092XINPCTP23-1363WO01DETAILED DESCRIPTION

[0014] Methods for detecting bearing faults in rotating machinery are disclosed.

[0015] FIG. 1 illustrates an operating environment for an AC induction motor system according to one embodiment. Referring to FIG. 1, operating environment 100 includes AC motor drive 110, an induction motor 120 and a grid 130. The AC motor drive 110 can be a variable frequency drive (VFD) or a DOL soft starter. AC motor drive 110 is coupled to grid 130 that provides the AC motor drive 110 with a three-phase AC power input. AC motor drive 110 includes power unit / starter 140 and a control and monitoring unit 150. AC motor drive 110 receives the three phase AC input which is fed to power unit / starter 140. Power unit / starter 140 controls power to the induction motor 120. AC motor drive 110 also includes a control and monitoring unit 150 that receives current data from the induction motor 120. In one embodiment, as illustrated in FIG. 1, the control and monitoring unit 150 is integrated into AC motor drive 110. In other embodiments, control and monitoring unit 150 can be included in an edge device coupled to the AC motor drive or reside in a cloud environment. User interface 160 can be included for communication between the control and monitoring unit 150 and a user. The control and monitoring unit 150 includes a processor that implements an algorithm to detect bearing faults and alerts a user via user interface 160 when the bearing fault is detected.

[0016] Control and monitoring unit 150 receives motor current data in the form of current signals from the induction motor. Current data can be obtained from current sensors positioned to receive appropriate current signals for motor current analysis. In some cases, multiple sensors can be used to acquire current data for each phase of a three-phase induction motor. In some cases, one current sensor can be used to acquire one phase of current. In one embodiment, the current sensors acquire stator current data from the induction motor. The received motor current data can be used to obtain a motor current frequency spectrum by performing a FFT (Fast Fourier Transform) or other frequency-domain signal processing technique on the received motor current data. In graphical form, the motor current spectrum illustrates the individual spectral components of the current signal providing frequency information about the current signal.

[0017] FIG. 2 illustrates a flowchart of a method to detect a bearing fault in a rotating machine. A code of method 200 can be integrated into the control and monitoring unit 150. The control and monitoring unit 150 includes a processor that can implement the code to determine the presence of a bearing fault in an induction motor or other rotating machine.4 ETN-092XINPCTP23-1363WO01Referring to FIG. 2, method begins upon receiving (210) current data from the rotating machine during operation at a sampling frequency. The current data received can be a singlephase current, a ground current or a common mode current of the rotating machine. In an embodiment, the current data can be received from a current sensor to acquire a single phase of current from the stator of an induction motor at the sampling frequency. The sampling frequency will be chosen for the specific sensor utilized.

[0018] Method 200 further includes obtaining (220) a current frequency spectrum for a sample of the received current data. The current frequency spectrum can be obtained by performing a FFT, or other frequency-domain signal processing technique, on the received current data from the induction motor. In some cases, the FFT is performed up to a particular frequency. As bearing faults typically appear in a lower frequency range, higher frequency components are ignored. For example, frequencies above 500 Hz are ignored. As an example, a 10 second sample of a single-phase current signal can be sensed by a current sensor installed on an induction motor at a sampling frequency of 16 KHz.

[0019] In some cases, the current frequency spectrum can be normalized with respect to an amplitude of a supply frequency of the rotating machine. For example, in order to perform the normalization, the amplitude of the current signal in the frequency domain is divided with respect to the amplitude of the supply frequency. Normalization is performed to make the current frequency spectrum independent of the motor rating and loading.

[0020] Method 200 further includes reducing noise (230) in the current frequency spectrum to create a reduced current frequency spectrum. Noise reduction from the current frequency spectrum can be achieved through bucketing the frequencies and taking the maximum frequency from each bucket. For example, the current frequency spectrum can first be sorted into a number of frequency groups. Then, in each frequency group (i) a maximum amplitude value of the current frequency spectrum is determined. The maximum amplitude values from each frequency group creates the reduced frequency spectrum.

[0021] In some cases, undesired frequencies are also removed from the current frequency spectrum. The undesired frequency bands such as supply frequency and eccentric frequency and related harmonic components can be removed from the current frequency spectrum as these frequencies are not related to bearing faults. For example, the frequency bands removed can include the supply frequency + / - 3.5 Hz as well as any harmonic frequencies + / - 1 Hz and the eccentric frequency which appears at Fs+ / -(Fs / 2), where Fs is the supply frequency.5 ETN-092XINPCTP23-1363WO01

[0022] Method 200 further includes performing (240) a peak detection algorithm to detect peak amplitude values in the reduced current frequency spectrum for a frequency range of the current frequency spectrum. A frequency range of the current frequency spectrum can be chosen so that the first two frequency groups (i = 1 and i = 2) typically will not have peak amplitude values. Each amplitude value in the reduced current frequency spectrum starting from the third amplitude value, e.g., maximum amplitude of the third frequency group (z = 3) in the current frequency spectrum, is tested against a plurality of conditions. The conditions can be defined by Equations 1-3 shown below to determine if the amplitude value is a peak amplitude. In the context of this application, a peak amplitude can be defined as an amplitude in the current frequency spectrum that suddenly rises and then suddenly drops. The severity of the rise and drop can be quantified using the following equations:

[0023] Amplitude value j (maximum amplitude of zth frequency group) > T * median (all amplitudes in reduced current frequency spectrum) (1)

[0024] Amplitude value j I Amplitude value j-1 > A (2)

[0025] Amplitude value j / Amplitude value j-1 > B (3)

[0026] where T, A, and B are all constant values based on experimental data from the specific rotating machine used. For example, when a 30 kW induction motor was tested, the values of 8, 3, and 5 were chosen respectively through analyzing data from the motor operating at a chiller plant with both healthy and faulty bearings.

[0027] In order to be identified as a peak amplitude value, the amplitude value satisfies Equation (1) and either Equation (2) or Equation (3). The condition defined by Equation (1) checks if the amplitude is greater than a specific threshold. In some cases, the specific threshold is based on a median value of all the amplitude values in the reduced current frequency spectrum. The median value of all the amplitude values is chosen to evaluate this threshold as it is more independent of outliers. The median value is then multiplied by a constant T. The value of constant T is set to a horizontal boundary to ensure a severity of the peak amplitude value. The condition defined by Equation (2) checks if a first rise in amplitude value is at least three times that of the previous amplitude value ( / -I). The first rise in amplitude value can be calculated by dividing the amplitude value by a first previous amplitude value ( / -I). The condition defined by Equation (3) checks if a second rise in amplitude value is at least five times that of the second previous amplitude ( / -2). The second rise in amplitude value can be calculated by dividing the amplitude value by a second previous amplitude value (j-1).

[0028] FIG. 3 illustrates a reduced current frequency spectrum depicting detected peak amplitudes. Referring to FIG. 3, reduced current frequency spectrum 300 includes a plurality6 ETN-092XINPCTP23-1363WO01 of amplitudes for a frequency range. The peak frequency peak detection algorithm is used to detect peak amplitude values in the reduced current frequency spectrum for a frequency range. The median value for the illustrated reduced current frequency spectrum can be seen in FIG. 3 by line 310. The median 310 can be used for the specific threshold corresponding to the first condition, e.g., Equation (1). From the amplitudes 320 rising above the threshold, e.g., median 310, each amplitude value can be checked against conditions 2 and 3 (corresponding to Equations (2) and (3), respectively). If the amplitude value satisfies either condition 2 or condition 3, a peak amplitude value is detected. As an example, a couple of amplitude values satisfy conditions 2 or condition 3 and are shown as detected peak amplitude values 330.

[0029] In some cases, the number of peak amplitude values and a sum of the peak amplitude values can be calculated from the detected peak amplitude values. FIG. 4A illustrates a first distribution plot 410 of the number of peak amplitude values detected for a healthy inductor motor having no bearing faults and a faulty induction motor having a bearing fault. FIG. 4B illustrates a second distribution plot 420 of a summation of peak amplitude values detected for a healthy induction motor and a faulty induction motor. From the distribution plots of FIG. 4A and FIG. 4B, it can be seen that the healthy induction motor has significantly fewer peak amplitude values and likewise the sum of the peak amplitude values is lower for the healthy induction motor than the induction motor having a bearing fault.

[0030] Method 200 further includes determining (250) a bearing fault from the peak amplitude values in the reduced current frequency spectrum. Machine learning and other artificial intelligence technologies continue to be developed to solve a variety of different problems. Machine learning is a field of computer science using statistical algorithms that can learn from data. The machine learning models can be high-level models that utilize deep learning techniques or can use various neural networks that are trained using a set of observations. A machine learning model can be trained using data, e.g., number of peak amplitude values detected and summation of peak amplitude values such as shown in FIG. 4A and FIG. 4B, for an induction motor (as well as other rotating machines) that is healthy, e.g., no faults as well as faulty, e.g., with a bearing fault. The machine learning model can be a binary classification model that uses as input the number of peak amplitude values detected and summation of peak amplitude values and outputs a classification of healthy or faulty. The machine learning model calculates probability thresholds from which an optimal threshold value can be chosen that is used for classification. The binary classification model can include logistic regression, random forest, or decision tree. Additionally, the binary classification model can include AdaBoost, Gradient boost, XG boost, or support vector machines. Thus, the7 ETN-092XINPCTP23-1363WO01 machine learning model from its classification determines a bearing fault, e.g., heathy, or no bearing fault, e.g., faulty, from the peak amplitude values. Method 200 further includes generating (260) an alarm when the bearing fault is detected.

[0031] In some cases, a bearing fault can occur intermittently in the current frequency spectrum due to the mechanical nature of the bearing fault. Generating an alarm for each occurrence of the intermittent fault can create alarm flooding and false alarms. Therefore, in some cases, the determination of bearing faults from a specific number of samples over time includes determining that a number of bearing faults are above a threshold before an alarm will be generated. In some cases, a severity score can be used as the threshold. In addition, the severity score can be used to categorize the severity of the fault based on the number of bearing faults that have occurred as determined from the received current data samples over a period of time.

[0032] FIG. 5 illustrates a flowchart of an implementation for determining a severity score and fault alert status. Referring to FIG. 5, the technique begins by receiving (502) the classification from the machine learning model that the current data sample is healthy or faulty. A severity score can be implemented as a counter that is incremented by a value depending on the classification of the sample. The severity score can be stored in memory of the control and monitoring unit 150 and retrieved (504) after the classification of the sample is received. The severity score can be reset to zero when the motor or rotating machine is not operating due to maintenance, for example. Responsive to the classification indicating a bearing fault (506), e.g., faulty, the severity score is incremented (510) by a first value. In an embodiment, the first value can be five, for example. Responsive to the classification not indicating a bearing fault, e.g., healthy, the severity score is decreased (508) by a second value. In an embodiment, the second value can be one, for example. The severity score then can be compared to a threshold value. In the implementation shown in FIG. 5, the severity is compared to two thresholds, a high threshold value (512) and a low threshold value (514). If the severity score is greater than or equal to the high threshold value (512), e.g., 900 as shown in FIG. 5, the fault alert status is set to critical fault (516). If the severity score is greater than or equal to the low threshold value (514), e.g., 30 as shown in FIG. 5, and below the high threshold value (512), the fault alert status is set to incipient fault (518). An alarm can be generated when the severity score is above the high threshold. In addition, an alarm can be generated when the severity score is above the low threshold. The fault alert status distinguishes a critical fault from an incipient fault. Otherwise, if the severity score is below both the high and low threshold values, the fault alert status is no fault. An alarm is not generated with a status of no fault.8 ETN-092XINPCTP23-1363WO01

[0033] FIG. 6 illustrates a schematic diagram illustrating components of a computing device that may be used in certain implementations described herein. The computing device can be representative of the control and monitoring unit as described herein. Referring to FIG. 6, computing device 600 can represent a server, a personal computer, a reader, a mobile device, a personal digital assistant, a wearable computer, a smart phone, a tablet, a laptop computer, a hybrid computer, a desktop computer, or a smart television. Accordingly, more or fewer elements described with respect to computing device 600 may be incorporated to implement a particular computing device.

[0034] The computing device 600 can include at least one processor 610, a memory 620, software 630 that includes operating system 640 and application 650, network interface 660, and user interface 670. Processor 610 processes data according to instructions of software 630. The instructions of application 650 may be loaded into computing device 600 and run on or in association with the operating system 640. Application 650 can include the methods as described. Memory 620 may comprise any computer readable storage media readable by processor 610 and capable of storing software including application 650. As used herein, computer readable storage media does not consist of carrier waves or propagating signals.

[0035] Computing device 600 can further include a user interface system 670, which may include input / output (I / O) devices and components that enable communication between a user and the computing device 600. Computing device 600 may also include a network interface 660 that allows the system to communicate with other computing devices, including server computing devices and other client devices, over a network.

[0036] Accordingly, embodiments are described herein that include the following examples.

[0037] Clause 1. A computer-implemented method to detect a bearing fault in a rotating machine, the method comprising: receiving current data from the rotating machine during operation at a sampling frequency; obtaining a current frequency spectrum for a sample of the received current data; reducing noise in the current frequency spectrum to create a reduced current frequency spectrum; performing a peak detection algorithm to detect peak amplitude values in the reduced current frequency spectrum for a frequency range of the current frequency spectrum; determining a bearing fault from the peak amplitude values in the reduced current frequency spectrum; and responsive to the determination of the bearing fault, generating an alarm.9 ETN-092XINPCTP23-1363WO01

[0038] Clause 2. The computer-implemented method of clause 1, wherein obtaining a current frequency spectrum includes performing a FFT (fast Fourier transform) on the sample of the received current data.

[0039] Clause 3. The computer-implemented method of clause 1 or 2, wherein the current frequency spectrum is normalized with respect to an amplitude of a supply frequency of the rotating machine.

[0040] Clause 4. The computer-implemented method of any preceding clause, further comprising removing undesired frequencies from the current frequency spectrum.

[0041] Clause 5. The computer-implemented method of any preceding clause, wherein reducing noise in the current frequency spectrum includes: sorting the current frequency spectrum into a plurality of frequency groups, and determining a maximum amplitude value of the current frequency spectrum for each frequency group (i), wherein the reduced current frequency spectrum comprises the maximum amplitude value for each frequency group (i).

[0042] Clause 6. The computer-implemented method of clause 5, wherein performing a peak detection algorithm to detect peak amplitude values in the reduced current frequency spectrum for a frequency range of the current frequency spectrum includes: for each amplitude value of the reduced current frequency spectrum starting from a third amplitude value in the reduced current frequency spectrum: determining that the amplitude value is greater than a first threshold value; and in response to the determination that the amplitude value is greater than the first threshold value: calculating a first amplitude rise by dividing the amplitude value by a first previous amplitude value; comparing the first amplitude rise to a second threshold value; responsive to the first amplitude rise being greater than the second threshold value, determining that the amplitude value is a peak amplitude value; responsive to the first amplitude rise not being greater than the second threshold value: calculating a second amplitude rise by dividing the amplitude value by a second previous amplitude value; and responsive to the second amplitude rise being greater than a third threshold value: determining that the amplitude value is a peak amplitude value.

[0043] Clause 7. The computer-implemented method of clause 6, further comprising calculating a median value from all the amplitude values in the reduced current frequency spectrum, wherein the first threshold value is based on the median value.

[0044] Clause 8. The computer-implemented method of clause 6 or 7, wherein the second threshold value is three times the first previous amplitude value of the reduced current frequency spectrum.10 ETN-092XINPCTP23-1363WO01

[0045] Clause 9. The computer-implemented method of any of clauses 6-8, wherein the third threshold value is five times the second previous amplitude value of the reduced current frequency spectrum.

[0046] Clause 10. The computer-implemented method of any of clauses 6-9, further comprising determining a number of peak amplitude values and calculating a sum of the peak amplitude values.

[0047] Clause 11. The computer-implemented method of clause 10, wherein determining a bearing fault from the peak amplitude values in the current frequency includes utilizing a machine learning model that uses as input the number of peak amplitude values and the sum of peak amplitude values to classify the sample as healthy or faulty, and wherein when the sample is determined to be faulty, determining a bearing fault.

[0048] Clause 12. The computer-implemented method of clause 11, wherein the machine learning model is a binary classification model.

[0049] Clause 13. The computer-implemented method of clause 12, wherein the binary classification model comprises logistic regression, random forest, or decision tree.

[0050] Clause 14. The computer-implemented method of clause 12 or 13, wherein determining a bearing fault from the peak amplitude values in the reduced current frequency spectrum further includes calculating a severity score to categorize a severity of the bearing fault.

[0051] Clause 15. The computer-implemented method of clause 14, wherein responsive to the binary classification model classifying the sample as faulty: incrementing the severity score by a first value; responsive to the severity score being above or equal to a high threshold, setting a fault alarm status to critical fault and determining a bearing fault; and responsive to the severity fault being above or equal to a low threshold and below the high threshold, setting a fault alarm status to incipient fault and determining a bearing fault.

[0052] Clause 16. The computer-implemented method of clause 14 or 15, wherein responsive to the binary classification model classifying the sample as healthy: decreasing the severity score by a second value; responsive to the severity score being above or equal to a high threshold, setting a fault alarm status to critical fault and determining a bearing fault; and responsive to the severity fault being above or equal to a low threshold below the high threshold, setting a fault alarm status to incipient fault and determining a bearing fault.

[0053] Clause 17. The computer-implemented method of any preceding clause, wherein the rotating machine is an induction motor.11 ETN-092XINPCTP23-1363WO01

[0054] Clause 18. The computer-implemented method of clause 17, wherein an AC motor drive of the induction motor is coupled to one of a variable frequency drive and a DOL soft starter.

[0055] Clause 19. The computer-implemented method of any preceding clause, wherein the current data includes one of single-phase stator current, ground current, and common mode current.

[0056] Clause 20. The computer-implemented method of any preceding clause, wherein the bearing fault is one of a single point defect or a bearing roughness fault.

[0057] Although the subj ect matter has been described in language specific to structural features and / or acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as examples of implementing the claims and other equivalent features and acts that would be recognized by one skilled in the art are intended to be within the scope of the claims.

Claims

12 ETN-092XINPCTP23-1363WO01CLAIMSWhat is claimed is:

1. A computer-implemented method to detect a bearing fault in a rotating machine, the method comprising: receiving current data from the rotating machine during operation at a sampling frequency; obtaining a current frequency spectrum for a sample of the received current data; reducing noise in the current frequency spectrum to create a reduced current frequency spectrum; performing a peak detection algorithm to detect peak amplitude values in the reduced current frequency spectrum for a frequency range of the current frequency spectrum; determining a bearing fault from the peak amplitude values in the reduced current frequency spectrum; and responsive to the determination of the bearing fault, generating an alarm.

2. The computer-implemented method of claim 1, wherein obtaining a current frequency spectrum includes performing a FFT (fast Fourier transform) on the sample of the received current data.

3. The computer-implemented method of claim 1, wherein the current frequency spectrum is normalized with respect to an amplitude of a supply frequency of the rotating machine.

4. The computer-implemented method of claim 1, further comprising removing undesired frequencies from the current frequency spectrum.

5. The computer-implemented method of claim 1, wherein reducing noise in the current frequency spectrum includes: sorting the current frequency spectrum into a plurality of frequency groups, and determining a maximum amplitude value of the current frequency spectrum for each frequency group (i), wherein the reduced current frequency spectrum comprises the maximum amplitude value for each frequency group (i).13 ETN-092XINPCTP23-1363WO016. The computer-implemented method of claim 5, wherein performing a peak detection algorithm to detect peak amplitude values in the reduced current frequency spectrum for a frequency range of the current frequency spectrum includes: for each amplitude value of the reduced current frequency spectrum starting from a third amplitude value in the reduced current frequency spectrum: determining that the amplitude value is greater than a first threshold value; and in response to the determination that the amplitude value is greater than the first threshold value: calculating a first amplitude rise by dividing the amplitude value by a first previous amplitude value; comparing the first amplitude rise to a second threshold value; responsive to the first amplitude rise being greater than the second threshold value, determining that the amplitude value is a peak amplitude value; responsive to the first amplitude rise not being greater than the second threshold value: calculating a second amplitude rise by dividing the amplitude value by a second previous amplitude value; and responsive to the second amplitude rise being greater than a third threshold value: determining that the amplitude value is a peak amplitude value.

7. The computer-implemented method of claim 6, further comprising calculating a median value from all the amplitude values in the reduced current frequency spectrum, wherein the first threshold value is based on the median value.

8. The computer-implemented method of claim 6, wherein the second threshold value is three times the first previous amplitude value of the reduced current frequency spectrum.

9. The computer-implemented method of claim 6, wherein the third threshold value is five times the second previous amplitude value of the reduced current frequency spectrum.

10. The computer-implemented method of claim 6, further comprising determining a number of peak amplitude values and calculating a sum of the peak amplitude values.14 ETN-092XINPCTP23-1363WO0111. The computer-implemented method of claim 10, wherein determining a bearing fault from the peak amplitude values in the current frequency includes utilizing a machine learning model that uses as input the number of peak amplitude values and the sum of peak amplitude values to classify the sample as healthy or faulty, and wherein when the sample is determined to be faulty, determining a bearing fault.

12. The computer- implemented method of claim 11, wherein the machine learning model is a binary classification model.

13. The computer-implemented method of claim 12, wherein the binary classification model comprises logistic regression, random forest, or decision tree.

14. The computer-implemented method of claim 12, wherein determining a bearing fault from the peak amplitude values in the reduced current frequency spectrum further includes calculating a severity score to categorize a severity of the bearing fault.

15. The computer- implemented method of claim 14, wherein responsive to the binary classification model classifying the sample as faulty: incrementing the severity score by a first value; responsive to the severity score being above or equal to a high threshold, setting a fault alarm status to critical fault and determining a bearing fault; and responsive to the severity fault being above or equal to a low threshold and below the high threshold, setting a fault alarm status to incipient fault and determining a bearing fault.

16. The computer-implemented method of claim 14, wherein responsive to the binary classification model classifying the sample as healthy: decreasing the severity score by a second value; responsive to the severity score being above or equal to a high threshold, setting a fault alarm status to critical fault and determining a bearing fault; and responsive to the severity fault being above or equal to a low threshold below the high threshold, setting a fault alarm status to incipient fault and determining a bearing fault.

17. The computer-implemented method of claim 1, wherein the rotating machine is an induction motor.15 ETN-092XINPCTP23-1363WO0118. The computer-implemented method of claim 17, wherein an AC motor drive of the induction motor is coupled to one of a variable frequency drive and a DOL soft starter.

19. The computer-implemented method of claim 1, wherein the current data includes one of single-phase stator current, ground current, and common mode current.

20. The computer- implemented method of claim 1, wherein the bearing fault is one of a single point defect or a bearing roughness fault.

Images