Systems and methods for improved fault diagnostics of electrical machines under dynamic load oscillations

Patent metadata

Field	Value
Publication number	US-12149196-B2
Application number	US-202117452838-A
Country	US
Kind code	B2
Filing date	Oct 29, 2021
Priority date	Oct 29, 2021
Publication date	Nov 19, 2024
Grant date	Nov 19, 2024

Abstract

Systems and methods are disclosed for improved fault diagnostics of electrical machines under dynamic load oscillations. The systems and methods may rely on one or more different algorithms for performing such fault diagnostics. One example, algorithm may involve determining a ratio of an instantaneous real power and a reactive power of the motor.

First claim

That which is claimed is:

1. A method comprising:
receiving, by a processor, operational data associated with a motor;
determining, by the processor, using one or more algorithms, and based on the operational data, processed data comprising a first ratio of an instantaneous real power and a reactive power of the motor;
determining, by the processor, using a frequency domain transform of the processed data, a frequency domain representation of the processed data;
calculating, by the processor, a fault value of the frequency domain representation, the fault value corresponding to a first type of motor failure;
determining, by the processor and based on the fault value, a baseline of operation for the motor, wherein the baseline is associated with detecting a future fault in the motor;
monitoring, by the processor, a second ratio of a first energy around a broken rotor fault frequency to a second energy around the broken rotor fault frequency in the frequency domain;
monitoring, by the processor, a difference between a first angle of a first vector, resulting from a sum of first frequency components around the broken rotor fault frequency, and a second angle of a second vector, resulting from a second sum of second frequency components around the broken rotor fault frequency;
determining, by the processor, based on the second ratio and the difference, a deviation from the baseline, wherein the deviation is above a threshold amount; and
causing to send, by the processor and based on determining the deviation, an alert indicative of a fault in the motor.

2. The method of claim 1 , wherein the one or more algorithms compute at least one multi-dimensional feature-set from the operational data and the processed data, wherein the feature-set includes operational features including at least: loading level, supply voltage unbalance, supply frequency and signal powers at one or more frequencies from the processed data, and/or transformed functions of frequency-based features.

3. The method of claim 2 , wherein the feature-set further includes a set of physics-based features specific to broken rotor faults, the physics-based features being calculated by:
estimating a speed of the motor;
calculating a fault frequency;
calculating energy in a fault frequency band of the processed data; and
calculating an angle at fault frequency of the processed data.

4. The method in claim 1 , wherein the processed data includes at least one of: the instantaneous real power, the instantaneous reactive power, d-axis and q-axis currents of the motor, real and reactive apparent impedance, amplitude and phase demodulated current signal and/or a square of one or more three-phase supply currents.

5. The method in claim 1 , wherein the one or more algorithms are run during a monitoring phase and an alerting phase, wherein during the monitoring phase, a multi-dimensional baseline cluster is formed form by stacking feature-sets, and wherein, in the alerting phase, a test cluster is formed from the feature-sets calculated from the operational data.

6. The method of claim 5 , wherein determining the baseline of operation for the motor further comprises:
extracting random sub-clusters from the multi-dimensional baseline cluster;
estimating statistical deviation between the multi-dimensional baseline cluster and the sub-clusters;
determining a mean and a standard deviation of these statistical deviation;
determining a threshold value based on the mean and the standard deviation; and
determining a moving average of one or more fault parameters over a time period.

7. The method of claim 5 , wherein the alerting phase further comprises;
determining a history of multi-dimensional feature-sets; and
applying a moving window filter to the history to assemble the test cluster, wherein the moving window filter comprises at least a rectangular window.

8. The method in claim 1 wherein the deviation is calculated using machine learning techniques by estimating an overlap between the baseline and test clusters using divergence methods including at least: KL Divergence, JS Divergence, and statistical distance scores including at least: Mahalanobis Distance and Silhouette Score.

9. The method of claim 1 , wherein the one or more algorithms also include at least: calculating an active parks vector and a reactive parks vector.

10. The method of claim 1 , wherein determining the first ratio of the instantaneous real power and the reactive power of the motor further comprises:
calculating energy in a frequency band corresponding to a broken rotor fault from the frequency domain transformations, wherein the energy comprises the first energy and the second energy.

11. A system comprising:
a computer processor operable to execute a set of computer-executable instructions; and
memory operable to store the set of computer-executable instructions operable to:
receive operational data associated with a motor;
determine, using one or more algorithms and based on the operational data, processed data comprising a ratio of an instantaneous real power and a reactive power of the motor;
determine, using a frequency domain transform of the processed data, a frequency domain representation of the processed data;
calculate a fault value of the frequency domain representation, the fault value corresponding to a first type of motor failure;
determine, based on the fault value, a baseline of operation for the motor, wherein the baseline is associated with detecting a future fault in the motor;
monitor a second ratio of a first energy around a broken rotor fault frequency to a second energy around the broken rotor fault frequency in the frequency domain;
monitor a difference between a first angle of a first vector, resulting from a sum of first frequency components around the broken rotor fault frequency, and a second angle of a second vector, resulting from a second sum of second frequency components around the broken rotor fault frequency;
determine, based on the second ratio and the difference, a deviation from the baseline, wherein the deviation is above a threshold amount; and
cause to send, based on determining the deviation, an alert indicative of a fault in the motor.

12. The system of claim 11 , wherein the one or more algorithms compute at least one multi-dimensional feature-set from the operational data and the processed data, wherein the feature-set includes operational features including at least: loading level, supply voltage unbalance, supply frequency and signal powers at one or more frequencies from the processed data, and/or transformed functions of frequency-based features included in the operational features.

13. The system of claim 12 , wherein the feature-set further includes a set of physics-based features specific to broken rotor faults, the physics-based features being calculated by:
estimating a speed of the motor;
calculating a fault frequency;
calculating energy in a fault frequency band of the processed data; and
calculating an angle at fault frequency of the processed data.

14. The system in claim 11 , wherein the processed data further comprises at least one of: the instantaneous real power, instantaneous reactive power, d-axis and q-axis currents of the motor, real and reactive apparent impedance, amplitude and phase demodulated current signal and/or a square of one or more three-phase supply currents.

15. The system in claim 11 , wherein the one or more algorithms are run during a monitoring phase and an alerting phase, wherein during the monitoring phase, a multi-dimensional baseline cluster is formed form by stacki

Assignees

• General Electric Technology Gmbh
• Ge Infrastructure Technology Llc

Inventors

• Shahid Ali
• Mohan Sumitha
• Mukherjee Rupam
• Tiwari Arvind Kumar
• Pamulaparthy Balakrishna

Classifications

• G01R31/343 Primary

  in operation · CPC title

• G01R31/346

  Testing of armature or field windings · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F17/142

  Fast Fourier transforms, e.g. using a Cooley-Tukey type algorithm · CPC title

• G06F17/16

  Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)} · CPC title