EP4634676A1

EP4634676A1 - System and methods for detecting electrical imbalance and early-stage winding failure for industrial induction motors

Info

Publication number: EP4634676A1
Application number: EP23703341.0A
Authority: EP
Inventors: Marius RUTKEVICIUS; Steven C. ENGLEBRETSON; Stefan Rakuff; Rajib Mikail
Original assignee: ABB Schweiz AG
Current assignee: ABB Schweiz AG
Priority date: 2023-01-27
Filing date: 2023-01-27
Publication date: 2025-10-22
Also published as: WO2024157054A1; CN120584293A

Abstract

System and methods for detecting a phase voltage of a poly-phase motor are described. The method includes receiving, by a controller, a collection of readings from an auxiliary coil installed in the motor; converting, by the controller, the collection of readings to a plurality of principal components, wherein the principal components are based on eigen vectors and eigen values generated based on the spectrum of readings; and inputting the plurality of primary components into a trained machine learning model to obtain a predicted phase voltage of the motor.

Description

SYSTEM AND METHODS FOR DETECTING ELECTRICAL IMBALANCE AND EARLY-STAGE WINDING FAILURE FOR INDUSTRIAL INDUCTION MOTORS

FIELD

[0001] The present disclosure relates to detecting winding voltage and winding damage during operation of electric motors. In particular, the systems and methods are described herein to predict early-stage winding failures or voltage imbalances between different windings in electric motors as well as motor power supplies that are fed from the electrical grid.

BACKGROUND

[0002] Winding imbalances may be introduced in electric motors for a variety of reasons. A possible cause for imbalance of the 3-phase electrical supply in an industrial setting is a high number of single-phase industrial equipment that is fed by the same phase. This causes a voltage reduction in that phase while the other phases are unaffected. Another possible cause for voltage phase imbalance at the motor is turn-to-turn electrical short circuits in the stator windings of the motors that may cause performance degradation. This failure mode can occur when the dielectric coating (insulation) of the magnetic wire of the stator field windings is compromised so that neighboring turns become electrically connected, reducing the number of effective turns in that coil. The insulation of the magnetic wire can be compromised over time due to insulation breakdown, contamination, or vibration induced relative motion between windings that wears away the insulation. Motors exposed to frequent start-stop cycles might be most susceptible to this failure.

[0003] Electric machine design makes the machine relatively insensitive to small phase imbalances. However, at certain voltage imbalance levels the motor begins to be significantly affected, which may result in permanent damage, reduced lifetime, or unplanned immediate shutdowns. If not addressed early enough, worsening winding insulation breakdown can cause catastrophic motor failure if the short-circuit connects two different windings or any winding to ground.

SUMMARY

[0004] A first aspect of the present disclosure provides a method for detecting a phase voltage of a poly-phase motor. The method comprises: receiving, by a controller, a collection of readings from an auxiliary coil installed in the motor; converting, by the controller, the collection of readings to a plurality of principal components, wherein the principal components are based on eigen vectors and eigen values generated based on the spectrum of readings; and inputting the plurality of primary components into a trained machine learning model to obtain a predicted phase voltage of the motor

[0005] According to an implementation of the first aspect, the method further comprises providing an alert, based on comparing the predicted voltage to an expected voltage value.

[0006] According to an implementation of the first aspect, determining the plurality of principal components comprises scaling the collection of readings received from the auxiliary coil.

[0007] According to an implementation of the first aspect, converting the collection of readings to the plurality of principal component comprises detecting, by the controller, dominant frequency components of the collection of readings.

[0008] According to an implementation of the first aspect, training the machine learning model comprises: storing the plurality of principal components in a training database; and providing a label to the stored principal components and the predicted phase voltage.

[0009] According to an implementation of the first aspect, the label provided to the stored principal components and the predicted phase voltage comprises at least one of an operations load label and a measured voltage label.

[0010] According to an implementation of the first aspect, the operations load label indicates a power consumed by a load connected to the poly-phase motor.

[0011] According to an implementation of the first aspect, the predicted voltage indicates a supply voltage imbalance and/or value outside of nominal range measured during operation of the poly-phase motor.

[0012] According to an implementation of the first aspect, the method further comprises: computing a difference between the predicted voltage and measured voltage; and comparing the computed difference to a threshold to determine whether there is an imbalance present in the poly-phase motor.

[0013] According to an implementation of the first aspect, the method further comprises: computing a difference between the predicted voltage and measured voltage; and comparing the computed different to a threshold to determine a health of a rotor of the poly-phase motor. [0014] According to an implementation of the first aspect, wherein the method further comprises: computing a difference between the predicted voltage and measured voltage; and comparing the computed difference to a threshold to determine a deterioration of the stator winding insulation during operation of the poly-phase motor. [0015] A second aspect of the present disclosure provides a method for detecting a phase voltage of a poly-phase motor. The method comprises: receiving, by a controller, a collection of readings from an auxiliary coil installed in the motor; calculating, by the controller, root mean square value of the collection of readings from the auxiliary coil installed in the motor; and inputting the root mean square values into a trained machine learning model to obtain a predicted phase voltage of the motor.

[0016] A third aspect of the present disclosure provides a system for detecting a phase voltage of a poly-phase motor. The system comprises a controller configured to: receive a collection of readings from an auxiliary coil installed in the motor; convert the collection of readings to a plurality of principal components, wherein the principal components are based on eigen vectors and eigen values generated based on the collection of readings; input the plurality of primary components into a trained machine learning model to obtain a predicted phase voltage of the motor.

[0017] According to an implementation of the third aspect, the controller is further configured to provide an alert, based on comparing the predicted voltage to an expected voltage value.

[0018] According to an implementation of the third aspect, the controller configured to convert the collection of readings to the plurality of principal component is further configured to detect dominant frequency components of the collection of readings.

[0019] According to an implementation of the third aspect, the controller configured to train the machine learning model is further configured to: store the plurality of principal components in a training database; and provide a label to the stored principal components and the predicted phase voltage.

[0020] According to an implementation of the third aspect, the label provided to the stored principal components and the predicted phase voltage comprises at least one of an operations load label and a measured voltage label.

[0021] According to an implementation of the third aspect, the controller is further configured to: predict a load connected to the poly-phase motor; and compare a load rating associated with the poly-phase with the motor predicted load; and based on determining whether the predicted load exceeds the load rating on the poly-phase motor, provide an alert. [0022] According to an implementation of the third aspect, the predicted load is used to calculate energy usage of the poly-phase motor.

[0023] According to an implementation of the third aspect, wherein the energy usage of the poly-phase motor is used to determine motor efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Subject matter of the present disclosure will be described in even greater detail below based on the exemplary figures. All features described and/or illustrated herein can be used alone or combined in different combinations. The features and advantages of various embodiments will become apparent by reading the following detailed description with reference to the attached drawings, which illustrate the following:

[0025] FIG. 1 illustrates a simplified block diagram depicting a system for detecting winding voltage imbalance in a motor during operation, according to one or more examples of the present disclosure;

[0026] FIGS. 2A-2C depict different configurations of auxiliary coil installed on the stator to detect winding voltage imbalance in a motor during operation, according to one or more examples of the present disclosure;

[0027] FIG. 3 is a block diagram of an exemplary sensor device associated with an auxiliary coil, according to one or more examples of the present disclosure;

[0028] FIGS. 4A-4C depict graphs that show data collected from a multi-coil auxiliary coil assembly installed in a stator of a motor, according to one or more examples of the present disclosure;

[0029] FIGS. 5A-5C depict graphs that shows the performance of a multiple-output regressor using RMS values of only a single coil of an auxiliary coil assembly attached to a stator of a motor, according to one or more examples of the present disclosure;

[0030] FIGS. 6A-6C depict plots generated by analyzing the spectral information from an induced voltage signal, according to one or more examples of the present disclosure;

[0031] FIG. 7 illustrates an exemplary process for detecting voltage phase imbalance in a poly-phase motor, according to one or more examples of the present disclosure; and

[0032] FIG. 8 illustrates an exemplary process for training a machine learning model to detect winding voltage imbalance in a poly-phase motor, according to one or more examples of the present disclosure.

DETAILED DESCRIPTION

[0033] The present disclosure describes detecting imbalanced voltages at the stator windings of an electrical AC power motor that may be present in an industrial setting. Unbalanced voltages at the windings reduce the power factor, efficiency, and longevity of motors and generators. Industrial electrical machines are typically designed to be powered by a specific single phase or 3-phase AC voltage with magnitudes of 110V, 220 V, 230 V, 400 V, 460V, etc. While this voltage may be nearly perfect at the electric substation, voltage variation may be encountered at the final electric load location (electric outlets at industrial, residential, or other location).

[0034] Voltage imbalance detection of the motor windings is highly desirable to trigger actions to correct the imbalance, whether internal or external to the motor. Motor winding voltage monitoring can be readily accomplished for motors powered by variable frequency drives (VFD) if the drive is equipped to do so. While VFDs can monitor and regulate motor supply voltages of the different stator windings they come at a cost premium. Additionally, VFDs may also require specific environmental operating conditions. Furthermore, many applications of a generator do not even require a variable speed option, and so a VFD cannot be practically used in such cases.

[0035] For direct-on-line (DOL) motors operated without VFDs, winding voltage monitoring can be accomplished by potentially expensive health monitoring equipment. A typical motor health monitoring device only measures the applied motor voltages at the motor terminals. In the early stages of winding failure, the terminal voltages may still be mostly balanced, so this traditional method can fail to detect impending motor failures. Many types of motor health monitoring devices for DOL motors must be connected on a temporary basis by a trained technician. In these cases, rather than continuously, motor monitoring is carried out only at certain maintenance intervals or after an issue is believed to have occurred. The added expense associated with this monitoring method is the primary reason that prevents many customers from winding voltage monitoring.

[0036] The present disclosure provides a low-cost and permanent monitoring solution for early winding failure detection of electric motors. The invention is realized as an integral motor component and does not incur additional installation cost for the customer. The invention is also considerably lower cost than VFDs or typical motor health monitoring devices connected either permanently or by technicians on a non-permanent basis. Customers can leverage the present disclosure to reduce unplanned downtime.

[0037] FIG. 1 illustrates a simplified block diagram depicting a system for detecting winding voltage imbalance in a motor during operation, according to one or more examples of the present disclosure. FIG. 1 depicts a poly-phase motor 102 that includes a rotor 104 and a stator 106. The rotor 104 and stator 106 may have a number of different windings that are connected to the different phases of the voltage supply. The poly-phase motor 102 may be connected to load 120. Auxiliary coil assembly 108 is attached to stator 106 of the poly-phase motor 102. The auxiliary coil assembly 108 is positioned such that the rotating magnetic field created by the currents in the windings of the rotor 104 and stator 106 induces voltages in the auxiliary coil 122. The magnitude and phase angle of the induced voltages in the auxiliary coil spectrum correspond to the voltages in the different windings of rotor 104 and stator 106 connected to the poly-phase motor 102. The induced voltage of the auxiliary coil 122 is a function of the voltage in the winding of stator 106,. By analyzing the voltage signal captured by the auxiliary coil 122, it is possible to approximate the voltage in the windings of rotor 104 and stator 106. The calculated voltages in the windings of stator 106 can be used to identify issues within the motor winding insulation, as well as provide information about the status of the applied external power supply. The calculated voltages in the windings of stator 106 can also be used to estimate the operating torque and power of the motor. The voltage signal from auxiliary coil assembly 108 may also be used to identify faults associated with rotor 104.

[0038] In some embodiments, the currents in the stator windings induce voltages in the auxiliary coil assembly. The stator windings most aligned to a particular coil of the auxiliary coil assembly will induce the highest voltage in the auxiliary coil 122, and the stator windings least aligned with the auxiliary coil assembly will induce the lowest voltage in that the auxiliary coil 122. The induced voltages will be superimposed to form a total induced voltage in that the auxiliary coil 122.

[0039] If the auxiliary coil 122 is comprised of two or more coils, and if those coils are placed at different angular positions around the stator, there will be a magnitude and phase difference between the total induced voltages in those coils. A change in the voltage of one of the motor stator windings or rotor induced voltages (i.e., if the motor becomes unbalanced) will also cause a difference in the phase between the coils of the auxiliary coil assembly. The magnitudes of the induced voltages in the coils of the auxiliary coil assembly will also be affected.

[0040] An imbalance of the stator winding voltages or rotor induced voltages (e.g., from winding failure or grid problems) may be detected uniquely from a phase change of the induced voltages in the coils of the auxiliary coil assembly 122, and it is distinguishable from a motor torque load change. The auxiliary coil assembly 108 includes sensor device 110 that measures the voltage induced in the coil of auxiliary coil assembly 108 by the rotating magnetic field of the motor stator windings of poly-phase motor 102. Auxiliary coil assembly 108 also includes an analog-to-digital (A/D) converter 112 to sample the voltage signal measured by sensor device 110 of the auxiliary coil assembly 108. A typical sampling rate may be 3, 6, or 10 kHz. In some embodiments, the A/D converter 112 may be a low power microcontroller with an integrated A/D converter and a processor for signal conditioning and interpretation. The A/D convertor 112 provides the sampled voltage from the auxiliary coil 108 to controller 114. The sensor device 110 of FIG. 3 is described in more detail below.

[0041] Controller 114 transfers the sampled voltage from the auxiliary coil 108 to memory 118 for storage. In some embodiments, the controller 114 may process the sampled voltage and subsequently provide the sampled voltage to a machine learning model 116 The machine learning model 116 uses the sampled voltage to predict voltage phase imbalance in poly-phase motor 102 during operation.

[0042] Machine learning model 116 is trained to predict the occurrence of winding voltage imbalance in poly-phase motor 102 based on information received from the polyphase motor 102 and auxiliary coil assembly 108. In some embodiments, the machine learning model 16 may be a single output regressor model that outputs either a predicted voltage imbalance of the stator windings or a predicted load 120 of poly-phase motor 102 using the information obtained from the sensor device 110 connected to auxiliary coil assembly 108. In some other embodiments, the machine learning model 116 may be a multiple output regressor model that outputs a predicted value of a load 120 connected to poly-phase motor 102 in addition to the predicted winding voltage imbalance.

[0043] In such embodiments, this information is useful to detect the winding voltage deviation from nominal values in the poly-phase motor 102 created by electrical supply or by motor stator or rotor faults.

[0044] The machine learning model 116 may be programmed with more than one algorithm to predict the winding voltage imbalance of poly-phase motor 102. For example, the machine learning model 116 may be configured to predict the winding voltage imbalance of poly-phase motor 102 using the RMS values of the voltage induced in the auxiliary coil. In other examples, the machine learning model 116 may be configured to predict the phase imbalance of poly-phase motor 102 using the spectra of the induced voltage of the auxiliary coils and a data reduction technique known as principal component analysis. Both these methods are explained in greater detail below.

[0045] FIGS. 2A-2C depict different configurations of auxiliary coils installed on the stator to detect winding voltage imbalance in a poly-phase motor during operation, according to one or more examples of the present disclosure. FIG. 2A shows a single-coil auxiliary coil assembly 108 associated with stator 106 of the poly-phase motor 102. In the single-coil auxiliary coil assembly 108, the coil span is such that the coil side starts from one particular winding slot and extends over the other winding slots at least once. In this configuration, voltage and current in any of the windings of stator 106 induces related voltage in the auxiliary coil 122 of auxiliary coil assembly 108. This allows the auxiliary coil 122 to be used to detect issues in any of the windings of the poly-phase motor 102 using only a singlecoil auxiliary coil assembly. In FIG. 2A, the stator 106 winding has 4 poles and 3 windings. In particular, the 3 windings include the U-winding 202, the V-winding 204, and W-winding 206. The coil of auxiliary coil assembly 108 at least partially covers each of the three windings, and is therefore able to pick up voltage changes in all three windings. The auxiliary coil is connected to sensor device 110 that measures the voltage induced in the auxiliary coil 108.

[0046] FIGS. 2B and 2C show auxiliary coils arranged in three or nine winding configurations respectively, according to one or more examples of the present disclosure. In some embodiments, the auxiliary coil assemblies 108 are comprised of insulated conductor loops that are inserted (“embedded”) in specific stator slots near the winding of stator 106 of the poly-phase motor 102. In particular, FIG. 2B depicts a 3-coil auxiliary coil assembly 108, and FIG. 2C depicts a 9-coil auxiliary coil assembly 108. In some cases, the more coils in the configurations of the auxiliary coil assembly 108, the more sensitive the auxiliary coil assembly 108 is to detecting winding voltage imbalance and the machine learning algorithm is more reliable in detecting winding voltage imbalance in a poly-phase motor 102. The winding and configurations of auxiliary coils are disclosed in more detail in P.C.T.

Application No. PCT/IB2022/054404 which is incorporated by reference, herein in its entirety.

[0047] In some embodiments, the auxiliary coil assembly 108 is comprised of a flexible printed circuit board (PCB) with a polyimide or similar high-temperature polymer substrate. The electrically conductive coils are formed by wave-wound, concentrically nested, or other conductor loops affixed to the flexible polymer substrate. Most flexible PCB coils contain a single layer of conductors, although multi-layer flexible PCBs may also be possible. The flexible PCB can be curved and bonded to the tips of the stator teeth using adhesives. The flexible PCB can be rectangular or have any other shape to cover certain areas of the stator. From induced voltages in the auxiliary coil at the airgap, it can detect onset of winding issues on any phase as well as irregularities in the voltage supply.

[0048] FIG. 3 is a block diagram of an exemplary sensor device associated with an auxiliary coil, according to one or more examples of the present disclosure. The sensor device 110 includes a processor 304, such as a central processing unit (CPU), and/or logic, that executes computer executable instructions for performing the functions, processes, and/or methods described herein. The computer executable instructions are locally stored and accessed from a non-transitory computer readable medium, such as data storage 306, which may be a hard drive or flash drive. Random-access memory (RAM) 308 is the main memory for loading and processing instructions executed by the processor 304. The communication segment 312 may connect to a wired network or cellular network and to a local area network or wide area network. The sensor device 110 may also include a detector 302 that detects the voltages induced in the auxiliary coil 108. The sensor device 110 may also include an indicator 310 that generates a message or an alert when either a detected voltage by the auxiliary coil, or a predicted voltage phase imbalance is outside of a certain threshold. A bus may connect the processor 304, RAM 308, data storage 306, and/or the communication segment 312. The components within the sensor device 110 may use the bus to communicate with each other. The components within the sensor device 110 are merely exemplary and might not be inclusive of every component within the sensor device 110. Additionally, and/or alternatively, the sensor device 110 may further include components that might not be included. For instance, the sensor device 110 might not include a communication segment 312.

[0049] FIGS. 4A-4C depict graphs that show data collected from a particular coils of a multi-coil auxiliary coil assembly installed in a stator of a motor, according to one or more examples of the present disclosure. For example, the auxiliary coil assembly 108 may have three coils, A, B, and C each corresponding to windings U, V, W of the poly-phase motor 102. FIG. 4A shows data collected from coils C of the auxiliary coil assembly 108, while the poly-phase motor 102 is run with a load 120. Winding W of the poly-phase motor 102 is powered by variable voltages, including 270 V, 256.5 V, and 283.5 V using a controllable three-phase power supply. The sampled auxiliary coil voltages induced in coil C of the auxiliary coil assembly 108 in each case are shown in FIG. 4A. In particular, graph 400 of FIG. 4A plots the voltage induced in the auxiliary coil assembly 108 on the y-axis against time on the x-axis. Graph 400 includes three curves 402, 404, and 406. Curve 402 represents the auxiliary voltage induced in coils C of auxiliary coil assembly 108 when the poly-phase motor 102 is powered using 283.5 V. Curve 404 represents the auxiliary voltage induced in coil C of auxiliary coil assembly 108 when the poly-phase motor 102 is powered using 270 V. Curve 406 represents the auxiliary voltage induced in coils C of auxiliary coil assembly 108 when the poly-phase motor 102 is powered using 256.5 V. The curves were sampled at different times, and the phase relationship between the three curves 402, 404, and 406 is selected to better differentiate the curves. [0050] Processor 304 of the sensor device 110 calculates the peak and RMS voltages of the sampled voltage signals of voltage induced in the coils of the auxiliary coil assembly 108 of the motor as shown in FIG. 4A In particular, the RMS values for each of the curves 402, 404, and 406, are shown in Figure 4B. Point 452 depicts the RMS voltage of signal 406 in graph 400 of FIG. 4A. Point 454 depicts the RMS voltage of signal 404 in graph 400 of FIG. 4A. Point 456 depicts the RMS voltage of signal 402 in graph 400 of FIG. 4A.

[0051] FIG. 4C shows a data plot generated from the three different coils of the auxiliary coil assembly 108, while the poly-phase motor is run with varying loads. In particular, plot 476 depicts the peak voltage induced in coils A of the auxiliary coil assembly 108. Plot 478 depicts the peak voltage induced in coil B of the auxiliary coil assembly 108, and plot 480 depicts the peak voltage generated in coils C of the auxiliary coil assembly 108. In each of the plots, winding voltage imbalance is introduced using different values of inductance in series with one winding of the motor stator 106. The inductance values are varied by connecting impedances of various values to the poly-phase motor 102. For example, inductance values of OmH, 5mH, and lOmH are used in series with the poly-phase motor 102. The different introduced inductance values are plotted on the x-axis, and the induced peak voltage is depicted on the y-axis. The poly-phase motor 102 is run with different loads 120 of 0 horsepower, 4.5 horsepower, and 8.3 horsepower. From the different plots, it is deduced that the RMS voltages induced in the auxiliary coil 122 generally decrease with increasing motor loads and with increasing phase imbalance. In such embodiments, we cannot differentiate between imbalance and load changes using peak or RMS values alone (see paragraph 26). A larger RMS voltage decrease is measured with winding voltage imbalance, and a smaller decrease is measured with increased motor load. As seen in FIG. 4C, in each of the coils A, B, and C, RMS voltage values are nearly linearly related to the load torque (i.e., power) or the winding voltage imbalance. Furthermore, when motor load and winding voltage imbalance are present simultaneously, an even higher combined drop in RMS voltage is measured. Using RMS values of voltage induced in the auxiliary coil 108 to determine winding voltage imbalance is useful when processing and memory capabilities of the processor 304 associated with the sensor device 110 are limited, although it should be noted that the usefulness of using RMS values only is limited (i.e., we cannot differentiate a load change from a winding voltage imbalance).

[0052] The measured RMS values of the coils of the auxiliary coil assembly 108 are stored in memory and paired with known torque values or known winding voltage imbalances. The known torque and known winding voltage imbalances are referred to as ‘labels’. These data pairs are used to form training dataset and stored in memory 108. The training dataset is used to train machine learning model 116 so that the machine learning model 116 is able to predict the voltage phase imbalance of poly-phase motor 102. In cases where the machine learning model 116 is a single output model, only a single label is provided to the training dataset. This label may be used to identify measured voltage phase imbalance of poly-phase motor 102. This label is used to train the machine learning model 116 to predict voltage-phase imbalance in the poly-phase motor 102.

[0053] In cases where the machine learning model 116 is a multiple output model, two or more labels may be provided to the training dataset. These labels may be used to identify the measured load 120 and measured voltage phase imbalance of the poly-phase motor 102. These labels are used to train the machine learning model 116 to predict voltage-phase imbalance in the poly-phase motor 102 and also the load 120 connected to the poly-phase motor 102. The measured values of load and operational parameters of the poly-phase motor 102 during operation are compared to predicted values of load and other operational parameters for the poly-phase motor 102 that are generated by the machine learning model 116. This comparison is used to train the machine learning model 116. The machine learning model 116 is used to then predict the winding voltage imbalance in the polyphase motor 102. In case the predicted winding voltage imbalance is outside a certain threshold, an alert is generated to prompt the custodians of the motor to take remedial action.

[0054] In order to accurately predict the winding voltage imbalance in a motor, data from at least 2 coils of an auxiliary coil assembly installed in a stator of the motor should be used. The angular separation of the coils provides the additional information necessary to infer load power and voltage imbalance in the windings. An angular separation of the coils results in the following characteristics:

• A change in torque load, i. empower, scales the voltages of the coils equally.

• A change in winding voltage imbalance in one or more windings affects some coils more than others.

[0055] FIG. 5A shows the performance of a multiple-output regressor using RMS values of only a single coils of an auxiliary coil assembly attached to a stator of a motor. Graph 502 of FIG. 5 A plots actual power of a motor on the x-axis and the plots the predicted power on the y-axis. Similarly graph 504 of FIG. 5 A plots predicted voltage phase imbalance on the y- axis and the actual winding voltage imbalance on the x-axis. As we can see from graphs 502 and 504 of FIG. 5A, when RMS values of only a single coils of auxiliary coil assembly 108 are used, the load power and the winding voltage imbalance values can not be predicted with confidence.

[0056] FIG. 5B shows the performance of a multiple-output regressor using RMS values of two coils of an auxiliary coil assembly attached to a stator of a motor. Graph 552 of FIG. 5B plots actual power of a motor on the x-axis and the predicted power on the y-axis. Similarly graph 554 of FIG. 5B plots predicted voltage phase imbalance on the y-axis and the actual winding voltage imbalance on the x-axis. As we can see from graphs 552 and 554 of FIG. 5B, when RMS values of two coils of auxiliary coil assembly 108 are used, the load power and the winding voltage imbalance values are predicted with more certainty than when the RMS values of only a single coils of auxiliary coil assembly 108 are used.

[0057] FIG. 5C shows the performance of a multiple-output regressor using RMS values of three coils of an auxiliary coil assembly attached to a stator of a motor. Graph 572 of FIG. 5C plots actual power of a motor on the x-axis and the predicted power on the y-axis. Similarly graph 574 of FIG. 5C plots predicted winding voltage imbalance on the y-axis and the actual winding voltage imbalance on the x-axis. As we can see from graphs 572 and 574 of FIG. 5C, when RMS values of three coils of auxiliary coil assembly 108 are used, the load power and the winding voltage imbalance values are predicted with a confidence.

[0058] As explained above, using RMS values to predict the winding voltage imbalance of a poly-phase motor 102 does not yield the most accurate results. While RMS values are linearly related to the winding voltage imbalance, RMS values are also linearly related to the load 120 applied to the poly-phase motor 102. It is difficult to confidently differentiate between winding voltage imbalance and load power influencing the change in RMS values. [0059] Rather than using only RMS or peak values of the induced voltage signals, it is possible to predict the winding voltage imbalance in a poly-phase motor 102 by analyzing the spectral information contained in an induced voltage signal. This analysis can be performed on data obtained from a single coils of auxiliary coil assembly 108. FIGS. 6A-6C depict using spectral information of an induced voltage signal to predict winding voltage imbalance, according to one or more examples of the present disclosure.

[0060] FIG. 6A depicts a plot generated by analyzing the spectral information from an induced voltage signal. In particular, graph 600 shown in FIG. 6 A plots the magnitude in units of Vs (Volt-seconds) on y-axis 602 and frequency on the x-axis 604. The voltage from the coil was obtained with a sampling rate of about 10 kHz. This divides the induced voltage signal into 2048 discrete frequency points ranging from 0 to about 5 kHz. Subsequently, principal component analysis (PCA) using singular value decomposition (SVD) is used to reduce the data from 2048 features or dimensions (corresponding to the 2048 discrete frequencies) to only a few features or dimensions (typically less than 20). Each of the principal components are composed by generating eigenvalues and eigenvectors based on the 2048 discrete frequency points. Once the principal components are generated, only the most important principal components (i.e., those with the largest explained variances - see later) are retained for the statistical analysis.

[0061] This is done by only retaining those principal components of the data that are most impacted by a load or imbalance change. The principal components that get affected a lot by the change in load power and imbalance have “large explained variances” while the principal components that are mostly unaffected tend to have “small explained variances.” After the PCA data reduction step, the machine learning model 116 is fitted to the reduced order data set. In particular, a support vector machine (SVM) multiple-output regressor model is fitted (trained) to known input-output pairs of data (i.e., the training data set). The data from auxiliary coil assembly 108 may be centered and scaled. Data scaling can be done with a min-max scaler or a scaling to zero mean and unity variance. The machine learning model 116 may comprise classifiers, logistic regressors, neural networks, decision trees, or random forests. In some embodiments, the fitting (training) of the model can be done offline, meaning not on the processor 304 of the sensor device 110. Once fitted, the model can be deployed to the processor 304.

[0062] To collect the training data for the machine learning model 116, the poly-phase motor 102 is operated at different known loads and with different known voltage phase imbalances that can be achieved by adding series inductances. During motor operation, data is continuously collected from the coils of the auxiliary coil assembly 108. This produces a family of DFTs where each DFT corresponds to a particular pair of imbalance and load power values.

[0063] The training dataset is used to train machine learning model 116 so that the machine learning model 116 is able to predict the winding voltage imbalance of poly-phase motor 102. The training dataset is provided with labels. In cases where the machine learning model 116 is a single output model, only a single label is provided to the training dataset. This label may be used to identify measured winding voltage imbalance of poly-phase motor 102. This label is used to train the machine learning model 116 to predict winding voltage imbalance in the poly-phase motor 102.

[0064] In cases where the machine learning model 116 is a multiple output model, two or more labels may be provided to the training dataset. These labels may be used to identify the measured load 120 and measured winding voltage imbalance of the poly-phase motor 102. These labels are used to train the machine learning model 116 to predict winding voltage imbalance in the poly-phase motor 102 and also the load 120 connected to the poly-phase motor 102.

[0065] Once the machine learning model 116 is fitted and trained with the training data, the machine learning model 116 is used to predict both the winding voltage imbalance and the load 120 running on the poly-phase motor 102 from the signal received from a single coil or multiple coils of the auxiliary coil assembly.

[0066] FIG. 6B depicts an exemplary performance plot of a multiple-output regressor to infer both the added series inductance (i.e., the resulting imbalance of the winding voltages) and the load power provided to the poly-phase motor 102. In this embodiment, the machine learning model 116 is a multiple output model that estimates both the load and the winding voltage imbalance of the poly-phase motor 102. In graph 652, using the training data of the PCA components provided, the machine learning model 116 is able to predict the load (i.e., torque, power) of the poly-phase motor 102. In graph 654, using the training data of the PCA components provided, the machine learning model 116 is able to predict the winding voltage imbalance of the poly-phase motor 102.

[0067] FIG. 6C shows example performance plots of single-output regressors that use the voltage signals from either coil A, B or C of auxiliary coil assembly 108. In such embodiments, the machine learning model 116 is a single output model and only estimates the winding voltage imbalance and not the load 120 associated with the poly-phase motor 102. The poly-phase motor 102 was operated at different load powers and different balanced/imbalanced conditions (achieved with 5 mH and 10 mH series inductors as above). Only the inductance values of the series inductors were used for training, and the load power values were ignored. The performance plots in FIG. 6C were obtained using new test data. Due to the power variation there is a lot of spread in the predictions of the winding voltage imbalance for each of the coils as shown in FIGS. 678, 680, and 682.

[0068] FIG. 7 illustrates an exemplary process for detecting winding voltage imbalance in a poly-phase motor, according to one or more examples of the present disclosure. The process 700 may be performed by the processor 304 of sensor device shown in FIG. 3. However, it will be recognized that any of the following blocks may be performed in any suitable order and that the process 700 may be performed in any environment and by any suitable computing device and/or controller. For instance, the process 700 may also be performed by the controller 114 shown in FIG. 1. [0069] At block 702, processor 304 receives a collection of readings from an auxiliary coil installed in the motor. For example, the collection of readings may be measured by the detector 302 of sensor device 110 associated with auxiliary coil 108 that is attached to stator 106 of poly-phase motor 102. In some embodiments, collection of readings can be voltage data over time, or such data processed, i.e. via a Fast Fourier Transform.

[0070] At block 704, processor 304 converts the collection of readings to a plurality of principal components, wherein the principal components are based on eigenvectors and eigenvalues generated based on the collection of readings. For example, principal component analysis (PCA) using singular value decomposition (SVD) is used to reduce the spectral data from auxiliary coil 108 from 2048 dimensions (corresponding to the 2048 discrete frequencies) to only a few dimensions (typically less than 20). Each of the principal components are composed by generating eigen values and eigen vectors based on the 2048 discrete frequency points. Once the principal components are generated, only the most important principal components are retained for the statistical analysis.

[0071] At block 706, processor 304 inputs the plurality of primary components into a trained machine learning model to obtain a predicted winding voltage imbalance of the motor, wherein the predicted winding voltage imbalance indicates a reduction in voltage in at least one of a plurality of phases of operation of the motor. For example, the machine learning model 116 may be used to predict a winding voltage imbalance of poly-phase motor 102.

[0072] At block 708, processor 304 provides an alert, based on comparing the predicted winding voltage imbalance to an expected winding voltage imbalance value. In some examples, the measured values of load and operational parameters of the poly-phase motor 102 during operation are compared to predicted values of load and other operational parameters for the poly-phase motor 102 that are generated by the machine learning model 116. The alert may be a visual alert to a technician to provide necessary repairs to the polyphase motor 102. In some examples, the alert may be an audio alert also provided to a technician to provide necessary repairs to the poly-phase motor 102. In some other cases, the alert may be sent to a central computing system, which may decide any remedial action that needs to be taken.

[0073] FIG. 8 illustrates an exemplary process for training a machine learning model to detect winding voltage imbalance in a poly-phase motor, according to one or more examples of the present disclosure. The process 800 may be performed by the processor 304 of sensor device shown in FIG. 3. However, it will be recognized that any of the following blocks may be performed in any suitable order and that the process 800 may be performed in any environment and by any suitable computing device and/or controller. For instance, the process 800 may also be performed by the controller 114 shown in FIG. 1.

[0074] At block 802, the controller 114 runs poly-phase motor 102 at different loads and voltage phase imbalances. In some embodiments, the voltage phase imbalances may be achieved by series connection of an inductance with one of the motor phases as described earlier. These different operating conditions are important to generate a comprehensive training dataset for training machine learning model 116. The data from each of the different running conditions is collected from auxiliary coil 108 using sensor device 110.

[0075] At block 804, the processor 114 computes discrete Fourier transforms (DFTs) on the data. This is performed using a sampling rate of about 10 kHz on the induced voltage signal. In the particular example, the spectra contain 2048 discrete frequency points ranging from 0 to about 5 kHz.

[0076] At block 806, the controller 114 scales the collected data. The scaling that is used is a feature scaling process, where each frequency is a feature in the sense of machine learning. One scaling option is to scale a feature such that all its values fall between -1 and +1. Another option, which is the one we used, is to scale a feature such that the means are zero (removing the offset) and the variance is 1.

[0077] At block 808, the controller 1114 computes primary components based on the principal component analysis (PCA) using singular value decomposition (SVD) is used to reduce the data from 2048 dimensions (corresponding to the 2048 discrete frequencies) to only a few dimensions (typically less than 20). Each of the principal components are composed by generating eigen values and eigen vectors based on the 2048 discrete frequency points.

[0078] At block 810, the processor 114 retains only the most important principal components for statistical analysis. For example, this is done by only retaining those principal components of the data that are most impacted by a load or imbalance change. The principal components that get affected a lot by the change in load power and imbalance have “large explained variances” while the principal components that are mostly unaffected tend to have “small explained variances.”

[0079] At block 812, the processor 114 fits the machine learning model 116 with the reduced order data set for training. For example, a support vector machine (SVM) multipleoutput regressor model is fitted (trained) to known input-output pairs of data. [0080] At block 814, the processor 114 deploys the machine learning model 116 for prediction of voltage phase imbalance of poly-phase motor 102.

[0081] At block 816, the processor 114 receives incoming data for operation of a polyphase motor 102. The incoming data may be received from sensor device 110 associated with auxiliary coil 108 attached to stator 106 of poly-phase motor 102.

[0082] At block 818, processor 114 computes DFTs based on the data collected from the sensor device 110. This may be done, for example, by using a sampling rate of about 10 kHz on the induced voltage signal. This divides the induced voltage signal into 2048 discrete frequency points ranging from 0 to about 5 kHz.

[0083] At block 820, processor 114 scales the data. In some cases, before the data is scaled, the dataset is reduced using primary component analysis. The primary component analysis is used to reduce the 2048 discrete frequency points to the most relevant data points that are the most affected by changes in winding voltage imbalance or load operations.

[0084] At block 822, the processor 114 executes the machine learning model with the obtained scaled primary components to obtain a predicted winding voltage imbalance. This predicted winding voltage imbalance is compared to a threshold, and if the winding voltage imbalance is outside of the threshold, an alert is generated for a technician to review and provide remedial care to the poly-phase motor 102. In some embodiments, if machine learning model 116 is a multiple output regressor model, then the machine learning model 116 also predicts the load 120 connected to poly-phase motor 102.

[0085] While subject matter of the present disclosure has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive. Any statement made herein characterizing the invention is also to be considered illustrative or exemplary and not restrictive as the invention is defined by the claims. It will be understood that changes and modifications may be made, by those of ordinary skill in the art, within the scope of the following claims, which may include any combination of features from different embodiments described above.

[0086] The terms used in the claims should be construed to have the broadest reasonable interpretation consistent with the foregoing description. For example, the use of the article “a” or “the” in introducing an element should not be interpreted as being exclusive of a plurality of elements. Likewise, the recitation of “or” should be interpreted as being inclusive, such that the recitation of “A or B” is not exclusive of “A and B,” unless it is clear from the context or the foregoing description that only one of A and B is intended. Further, the recitation of “at least one of A, B and C” should be interpreted as one or more of a group of elements consisting of A, B and C, and should not be interpreted as requiring at least one of each of the listed elements A, B and C, regardless of whether A, B and C are related as categories or otherwise. Moreover, the recitation of “A, B and/or C” or “at least one of A, B or C” should be interpreted as including any singular entity from the listed elements, e.g., A, any subset from the listed elements, e g., A and B, or the entire list of elements A, B and C.

Claims

CLAIMS What is claimed is:

1. A method for detecting a phase voltage of a poly-phase motor, the method comprising: receiving a collection of readings from an auxiliary coil installed in the motor; converting the collection of readings to a plurality of principal components, wherein the principal components are based on eigen vectors and eigen values generated based on the collection of readings; and inputting the plurality of primary components into a trained machine learning model to obtain a predicted phase voltage of the motor.

2. The method of claim 1, wherein the method further comprises providing an alert, based on comparing the predicted phase voltage to an expected voltage value.

3. The method of claim 1, wherein determining the plurality of principal components comprises scaling the collection of readings received from the auxiliary coil.

4. The method of claim 1, wherein converting the collection of readings to the plurality of principal components comprises detecting dominant frequency components of the collection of readings.

5. The method of claim 1, wherein training the machine learning model comprises: storing the plurality of principal components in a training database; and providing a label to the stored principal components and the predicted phase voltage.

6. The method of claim 5, wherein the label provided to the stored principal components and the predicted phase voltage comprises at least one of an operations load label and a measured voltage label.

7. The method of claim 6, wherein the operations load label indicates a power consumed by a load connected to the poly-phase motor.

8. The method of claim 7, wherein the predicted voltage indicates a supply voltage imbalance and/or value outside of nominal range measured during operation of the poly-phase motor.

9. The method of claim 1, wherein the method further comprises: computing a difference between the predicted phase voltage and a measured voltage; and comparing the computed difference to a threshold to determine whether there is an imbalance present in the poly-phase motor.

10. The method of claim 1, wherein the method further comprises: computing a difference between the predicted phase voltage and an expected voltage; and comparing the computed difference to a threshold to determine a health of a rotor of the poly-phase motor.

11. The method of claim 1, wherein the method further comprises: computing a difference between the predicted phase voltage and a measured voltage; and comparing the computed difference to a threshold to determine an early stator winding degradation during operation of the poly -phase motor.

12. A method for detecting a phase voltage of a poly-phase motor, the method comprising: receiving a collection of readings from an auxiliary coil installed in the motor; calculating root mean square values of the collection of readings from the auxiliary coil installed in the motor; and inputting the root mean square values into a trained machine learning model to obtain a predicted phase voltage of the motor.

13. A system for detecting a phase voltage of a poly -phase motor, the system comprising: an auxiliary coil installed in the poly-phase motor; and a controller configured to: receive a collection of readings from the auxiliary coil installed in the motor; convert the collection of readings to a plurality of principal components, wherein the principal components are based on eigen vectors and eigen values generated based on the collection of readings; input the plurality of primary components into a trained machine learning model to obtain a predicted phase voltage of the motor.

14. The system of claim 13, wherein the controller is further configured to provide an alert, based on comparing the predicted phase voltage to an expected voltage value.

15. The system of claim 13, wherein the controller is further configured to detect dominant frequency components of the collection of readings.

16. The system of claim 13, wherein the controller trains the machine learning model by: storing the plurality of principal components in a training database; and providing a label to the stored principal components and the predicted phase voltage.

17. The system of claim 16, wherein the label provided to the stored principal components and the predicted phase voltage comprises at least one of an operations load label and a measured voltage label.

18. The system of claim 13, wherein the controller is further configured to: predict a load connected to the poly-phase motor; and compare a load rating associated with the poly-phase with the motor predicted load; and based on determining whether the predicted load exceeds the load rating on the polyphase motor, provide an alert.

19. The system of claim 18, wherein the predicted load is used to calculate energy usage of the poly-phase motor.

20. The system of claim 19, wherein the energy usage of the poly-phase motor is used to determine motor efficiency.