US 7,769,552 B2Grant

Method and apparatus for estimating induction motor rotor temperature

Issue Date:2010-08-03

•27 Claims

•18 Drawing Sheets

Abstract

A method and apparatus to provide continuous and reliable rotor temperature estimates for line-connected induction motors during steady-state and/or dynamic motor operations. Rotor temperature is calculated from voltage and current measurements without any temperature or speed sensors. First, complex space vectors are synthesized from voltage and current measurements. Second, the instantaneous rotor speed is detected by calculating the rotational speed of a single rotor slot harmonic component with respect to the rotational speed of the fundamental frequency component. Third, the positive sequence fundamental frequency components are extracted from complex space vectors. Fourth, the rotor time constant is estimated in a model-reference adaptive system based on a dynamic induction motor equivalent circuit model. Finally, the rotor temperature is calculated according to the linear relationship between the rotor temperature and the estimated rotor time constant. Real-time induction motor thermal protection is achieved through this continuous tracking of the rotor temperature.

Metadata

Assignee

Schneider Electric USA, Inc.

Inventors

Roy S. Colby
Zhi Gao
Larry A. Turner

Application Information

Application Number:US 12/268,819

Filing Date:2008-11-11

Priority Date:2008-05-16

Art Unit:2863

Classifications

IPC:

G01R27/00H02P23/00

Patent Drawings (18 sheets)

Description

Cross-Reference to Related Application

[0001] This application claims the benefit of U.S. Provisional Application Ser. No. 61/053,941, filed May 16, 2008, entitled “Methods and Apparatuses for Estimating Transient Slip, Rotor Slots, Motor Parameters, and Induction Motor Temperature.”

Field of the Invention

[0002] The present disclosure relates generally to the estimation of rotor temperatures in induction motors without using actual temperature or speed sensors. More particularly, this disclosure relates to the continuous estimation of the rotor temperature from measurements of voltage and current, for motors operated at steady-state and/or dynamic conditions.

Background

[0003] The accumulation of excessive heat on the rotor of an induction motor can cause severe thermal stress to the rotor structure, resulting in rotor conductor burnout or even total motor failure. Consequently, an accurate and reliable real-time tracking of the rotor temperature is needed to provide an adequate warning of imminent rotor over-heating. Furthermore, many line-connected induction motors are characterized by totally enclosed frames, as well as, small air gaps to increase their efficiencies. Therefore, the stator and rotor temperatures are highly correlated due to such motor designs, and the rotor temperature can often be used as an indicator of the stator winding temperature. For these motors, the continuous monitoring of the rotor temperature also provides a way to properly protect the motors against stator winding insulation deterioration due to over-heating. Thus, rotor temperature estimation is considered an integral part of induction motor condition monitoring, diagnosis, and protection.

[0004] There are typically four major approaches to obtaining the rotor temperature in an induction motor. The first approach makes direct measurements of rotor temperatures by way of temperature sensors, such as thermocouples, temperature-sensitive stick-ons, temperature-sensitive paint, and/or infrared cameras. These methods of measurement typically involve expensive instruments and dedicated wiring from the motors back to a motor control center. Therefore, this approach adds operating cost to line-connected induction motors.

[0005] The second approach utilizes a thermal model-based temperature estimator to calculate the rotor temperature. A lumped-parameter thermal circuit, typically with a single thermal capacitor and a single thermal resistor, is often used to emulate the thermal characteristic of a motor. For a class of motors with the same rating, the thermal capacitance and resistance are often pre-determined by plant operators or electrical installation engineers based on a set of known parameters, such as full load current, service factor, and trip class. Consequently, such thermal models cannot respond to changes in the motors' cooling capability, such as a broken cooling fan or a clogged motor casing. Therefore, this approach is incapable of giving an accurate estimate of the rotor temperature tailored to a specific motor's cooling capability.

[0006] The third approach typically applies to inverter-fed induction motors. By injecting high frequency signals into a motor, a magnetic field is created inside the motor. The electrical response waveforms are recorded, and the value of the rotor resistance or the rotor time constant is extracted. The rotor temperature information is then derived from the extracted rotor resistance and/or rotor time constant value. However, this approach requires an injection circuit and is usually impractical for line-connected induction motors.

[0007] The fourth approach derives the rotor temperature from the rotor resistance and/or the rotor time constant, based on the induction motor equivalent circuit. This technique requires the knowledge of rotor speed. This speed information is usually obtained from a mechanical speed sensor attached to the shaft of a line-connected motor. Because of the cost and fragile nature of such a speed sensor, and because of the difficulty of installing the sensor in many motor applications, speed-sensorless schemes based on induction motor magnetic saliency have been preferred. By sampling a single-phase current at steady-state motor operation, the rotor speed information is extracted from spectral estimation techniques such as fast Fourier transform. This speed information is then used in conjunction with the induction motor equivalent circuit to produce an estimate of the rotor resistance and subsequently the rotor temperature. However, accurate rotor temperature cannot be reliably obtained during dynamic motor operations, where motors are connected to time-varying loads such as reciprocating compressors or pumps. In addition, because most spectral estimation techniques require that the whole sequence of current samples, acquired at steady-state motor operation, be available for batch processing, such techniques can cause significant delay between the data acquisition and the final output of the temperature estimates.

[0008] What is needed is a cost-effective induction motor condition monitoring, diagnosis, and protection system with a highly accurate and reliable rotor temperature estimator that relies on measurements of voltage and/or current. What is also needed is an induction motor condition monitoring, diagnosis, and protection system with the ability to continuously track the rotor temperature during steady-state and/or dynamic motor operations. It is further needed to interleave the data acquisition process with the rotor temperature estimation process such that: (1) a set of voltage and current samples is acquired, (2) the corresponding rotor temperature is estimated, and (3) the estimation is completed before the next set of voltage and current samples is available.

Summary of the Invention

[0009] According to some aspects, a method and apparatus for estimating rotor temperature in induction motors are provided. The estimations are based on voltage and current measurements readily available at motor control centers or at motor terminals. Continuous tracking of the rotor temperature is provided during steady-state and/or dynamic motor operations. No additional sensors or injection circuits are needed. The method can be divided into five major components.

[0010] First, a complex voltage space vector is synthesized from voltage measurements. A complex current space vector is synthesized from current measurements. The voltage measurements can be either phase-to-neutral or line-to-line, and the current can be measured either from two phases or from all three phases.

[0011] Second, an instantaneous rotor speed is detected by calculating the rotational speed of a single rotor slot harmonic component, which is extracted from the complex current space vector, with respect to the rotational speed of a fundamental frequency component, which is extracted from the complex voltage space vector. This rotor speed measurement/estimation eliminates the need for a speed measurement from a tachometer, which adds cost and bulk to the motor.

[0012] Third, a positive sequence fundamental frequency voltage component is obtained from the complex voltage space vector. A positive sequence fundamental frequency current component is obtained from the complex current space vector. A computationally efficient discrete-time filtering technique can be used.

[0013] Fourth, a rotor time constant can be estimated through the use a model-reference adaptive system (“MRAS”). The MRAS is based on a dynamic induction motor equivalent circuit model. The MRAS adjusts the rotor time constant so that a flux-related vector from the rotor circuit equation is aligned with another flux-related vector from the stator circuit equation.

[0014] Fifth, a rotor temperature is calculated according to an inverse relationship between the rotor temperature and the rotor time constant. This relationship is derived from a linear relationship between the rotor temperature and rotor resistance.

[0015] The foregoing and additional aspects and embodiments of the present invention will be apparent to those of ordinary skill in the art in view of the detailed description of various embodiments and/or aspects, which is made with reference to the drawings, a brief description of which is provided next.

Brief Description of the Drawings

[0016] FIG. 1 is a block diagram illustrating an architecture of a method and apparatus for estimating an induction motor rotor temperature;

[0017] FIG. 2A is a block diagram of a complex voltage space vector synthesis for use in the architecture of FIG. 1;

[0018] FIG. 2B is a block diagram showing a complex current space vector synthesis for use in the architecture of FIG. 1;

[0019] FIG. 3A is a chart exemplifying a stator current harmonic spectrum computed from a complex current space vector for a 20-hp 4-pole induction motor with 40 rotor slots according to some aspects;

[0020] FIG. 3B is a chart exemplifying a stator current harmonic spectrum computed from a complex current space vector for a 7.5-hp 4-pole induction motor with 44 rotor slots according to some aspects;

[0021] FIG. 3C is a chart exemplifying a stator current harmonic spectrum computed from a complex current space vector for a 2-hp 6-pole induction motor with 33 rotor slots according to some aspects;

[0022] FIG. 4A is a block diagram illustrating methods of extracting a fundamental frequency from a complex voltage space vector according to some aspects;

[0023] FIG. 4B is a block diagram illustrating methods of extracting a rotor slot harmonic frequency from a complex current space vector according to some aspects;

[0024] FIG. 5 is a block diagram illustrating methods of estimating an instantaneous rotor speed according to some aspects;

[0025] FIG. 6 is a chart exemplifying results of an instantaneous rotor speed detected from voltage and current measurements for a 20-hp 4-pole induction motor with 40 rotor slots during a dynamic motor operation with random load oscillation according to some aspects;

[0026] FIG. 7A is a block diagram illustrating methods of extracting a positive sequence fundamental frequency voltage component from a complex voltage space vector according to some aspects;

[0027] FIG. 7B is a block diagram illustrating methods of extracting a positive sequence fundamental frequency current component from a complex current space vector according to some aspects;

[0028] FIG. 8 is a schematic of a dynamic induction motor equivalent circuit model that characterizes the motor's stator and rotor circuit equations;

[0029] FIG. 9 is a block diagram illustrating methods of estimating a rotor time constant according to some aspects;

[0030] FIG. 10 is a block diagram illustrating methods of calculating a rotor temperature according to some aspects;

[0031] FIG. 11A is a chart exemplifying a dynamic motor operation for a 20-hp 4-pole induction motor with 40 rotor slots according to some aspects;

[0032] FIG. 11B is a chart exemplifying an estimated rotor temperature and measured stator winding temperatures for a 20-hp 4-pole induction motor with 40 rotor slots during the dynamic motor operation in FIG. 11A;

[0033] FIG. 12 is a block diagram showing an architecture of a method and apparatus for estimating induction motor electrical parameters without speed sensors according to some aspects of the present concepts;

[0034] FIG. 13 is a schematic showing a dynamic induction motor equivalent circuit model that characterizes the motor's stator and rotor circuit equations according to some aspects;

[0035] FIG. 14 is a block diagram illustrating methods of determining various quantities used for estimating induction motor electrical parameters according to some aspects;

[0036] FIG. 15 is a flow chart in a unified modeling language (UML) defining a matrix condition and pseudo-inverse solution according to some aspects;

[0037] FIG. 16 is a flow diagram of methods for performing an efficient recursive method for estimating induction motor electrical parameters; and

[0038] FIG. 17 is a block diagram illustrating a method and apparatus for estimating induction motor electrical parameters, as well rotor temperature, according to some aspects of the present concepts.

Detailed Description of the Invention

[0039] Although the invention will be described in connection with certain aspects and/or embodiments, it will be understood that the invention is not limited to those particular aspects and/or embodiments. On the contrary, the invention is intended to cover all alternatives, modifications, and equivalent arrangements as may be included within the spirit and scope of the invention as defined by the appended claims.

[0040] Referring to FIG. 1, an architecture (100) for estimating rotor temperature in an induction motor 106 is illustrated according to some aspects of the present concepts. The architecture (100) includes a complex voltage space vector synthesis (108), a complex current space vector synthesis (110), a rotor speed estimation (120), a positive sequence fundamental frequency voltage component extraction (112), a positive sequence fundamental frequency current component extraction (118), a rotor time constant estimation (122), and a rotor temperature calculation (124).

Complex Space Vector Synthesis 108, 110

[0041] According to some aspects, voltage measurements (102) and current measurements (104) can be acquired in the form of samples from phase a, b and c at a sampling frequency f_sfor a line-connected induction motor (e.g., an induction motor 106) having a floating neutral point. The complex space vector, v_Cor i_C, synthesized from voltage measurements (102) or current measurements (104), is a sequence of complex numbers carrying frequency information between −f_s/2 and f_s/2. The positive half of the complex space vector frequency spectrum corresponds to the positive sequence components at various frequencies, while the negative half corresponds to the negative sequence components at various frequencies.

Complex Voltage Space Vector Synthesis

[0042] FIG. 2A illustrates a block diagram of the complex voltage space vector synthesis (108) of FIG. 1. The complex voltage space vector, v_C,n, (218) is calculated from the voltage measurements (102) by Equation 1,
v_C,n=⅔·(v_ao,n+α·v_bo,n+α²·v_co,n) Equation 1
where α=e^j2π/3; v_ao, v_bo, v_coare the voltage difference between phase a, b and c and an arbitrary voltage reference point o selected by a measurement device, respectively; the subscript n denotes the nth instant (or sample) in a discrete-time system.

[0043] When phase-to-neutral voltage measurements are not available and only the line-to-line voltage measurements (208) are available, v_C,n(218) can also be calculated from the line-to-line voltage measurements (210, 212, 214) using Equation 2,
v_C,n=⅔·(v_ab,n−α²·v_bc,n), when v_aband v_bcare available;
v_C,n=⅔·(−v_ca,n+α·v_bc,n), when v_bcand v_caare available;
v_C,n=⅔·(−α·v_ab,n+α²·v_ca,n),when v_caand v_abare available Equation 2
where v_abis the voltage difference between phase a and phase b, v_bcis the voltage difference between phase b and phase c, v_cais the voltage difference between phase c and phase a.
Complex Current Space Vector Synthesis

[0044] FIG. 2B illustrates a block diagram of the complex current space vector synthesis (110). The complex current space vector, i_C,n, (238) is calculated from the current measurements (104) and Equation 3,
i_C,n=⅔·(i_a,n+α·i_b,n+α²·i_c,n) Equation 3
where i_a, i_b, i_care phase a, b and c current measurements, respectively.

[0045] When three phase measurements are not available and only current measurements from two phases are available, i_C,n(238) can also be calculated from such current measurements. Specifically, depending on which two phase measurements (228) are available, Equation 4 specifies three alternatives (230, 232, and 234) for determining i_C,n.
i_C,n=⅔·[(2+α)·i_a,n+(1+2α)·i_b,n], when i_aand i_bare available;
i_C,n=⅔·[(−1+α)·i_b,n+(−2−α)·i_c,n], when i_band i_care available;
i_C,n=⅔·[(1−α)·i_a,n+(−1−2α)·i_c,n], when i_cand i_aare available; Equation 4
Instantaneous Rotor Speed Detection

[0046] According to some aspects of the present disclosure, instantaneous rotor speed detection can include 3 subsystems: 1) fundamental frequency extraction (114) from the complex voltage space vector (218); 2) rotor slot harmonic frequency extraction (116) from the complex current space vector (238); and 3) rotor speed estimation (120). While the present disclosure illustrates examples relating to a particular instantaneous rotor speed detection scheme, other schemes are contemplated. For example, U.S. Provisional Application Ser. No. 61/053,941, filed May 16, 2008, entitled “Methods and Apparatuses for Estimating Transient Slip, Rotor Slots, Motor Parameters, and Induction Motor Temperature,” which is incorporated by reference herein, provides additional examples of schemes that can be used in place of the scheme described herein for detecting the instantaneous rotor speed. Alternately, rotor speed can be determined conventionally via a speed sensing device coupled to a rotor of the induction motor 106.

Fundamental Frequency Extraction

[0047] For most motors (e.g., induction motor 106) connected directly to regulated power distribution networks, the motor's instantaneous fundamental frequency, f_0,n, typically varies within a small range around the rated fundamental frequency f₀. Because of this varying frequency, an accurate estimate of a particular motor's instantaneous rotor speed requires a reliable estimate of the instantaneous fundamental frequency.

[0048] For example, FIG. 4A illustrates a block diagram of the fundamental frequency extraction block 114 of FIG. 1, according to some aspects. A voltage controlled oscillator (“VCO”) 406 generates a complex exponential signal x_nin the demodulation and mixes the positive sequence fundamental frequency component in the complex voltage space vector down to a complex baseband. A generalized linear-phase low-pass filter 410 produces a complex baseband signal z_nfrom a mixed complex voltage signal, y_n. A numerical differentiation block 412 extracts the discrete-time derivative of the phase from the complex baseband signal. The instantaneous fundamental frequency (414) can be calculated from the discrete-time derivative of the phase.

Voltage Controlled Oscillator

[0049] The VCO 406 reduces the complexity of the filter architecture in a subsequent stage. The VCO takes the rated fundamental frequency, f₀, (404) as its input and synthesizes the complex exponential signal, x_n, at sample n, as its output according to Equation 5:
x_n=exp(j·2πn·f₀/f_s) Equation 5
where j is the imaginary unit (e.g., j²=−1).

[0050] This synthesizing process is analogous to demodulating an FM signal. By multiplying the complex voltage space vector, v_C,n, (218) with the complex conjugate (408) of the VCO 406 output, x_n*, the positive sequence fundamental frequency voltage component is mixed down to the complex baseband, where y_nis the mixed complex voltage signal defined in Equation 6.
y_n=v_C,n·x_n* Equation 6
Generalized Linear-Phase Low-Pass Filter

[0051] According to some aspects, the generalized linear-phase low-pass filter 410 takes the mixed complex voltage signal y_nas its input and produces the complex baseband signal z_nas its output. The generalized linear-phase low-pass filter 410 can provide linear phase response over a frequency band and suppress out-of-band interference from other sources. For example, if the complex baseband bandwidth is denoted as ω_B, then signals between −ω_Band ω_Bin y_n, including the signal at zero frequency, are retained. Signals beyond this frequency band [−ω_Bω_B] are attenuated (e.g., filtered out). The generalized linear-phase low-pass filter 410 can preserve the shape of the mixed complex voltage signal, y_n. Specifically, the generalized linear-phase low-pass filter 410 introduces the same amount of time shift to the signals with frequencies between −ω_Band ω_BUsing a generalized linear-phase low-pass filter (e.g., 410) ensures that phase information in the mixed complex voltage signal, y_n, is maintained without distortion. After filtering the mixed complex voltage signal, y_n, the numerical differentiation block 412 is applied to the complex baseband signal, z_n, to calculate the signal's angular rotational speed.

[0052] According to some aspects, discrete-time finite impulse response (FIR) filters can be used when implementing a generalized linear-phase low-pass filter (e.g., 410). Carefully designed infinite impulse response (IIR) Bessel filters can also be used with approximately linear phase within the [−ω_Bω_B] range. Compared with their FIR counterparts, the IIR filters usually offer similar performance with smaller memory depth.

[0053] Commonly used filter design techniques, such as a Kaiser window design technique, offer useful solutions when selecting appropriate generalized linear-phase low-pass filters (e.g., generalized linear-phase low-pass filter 410). The desired cutoff frequency of the generalized linear-phase low-pass filter depends on the complex baseband bandwidth, ω_B. For most line-connected motors in regulated power distribution networks, the frequency difference between the instantaneous fundamental frequency and the rated fundamental frequency is usually less than 1 Hz. Therefore, the cutoff frequency of a generalized linear-phase low-pass filter 410 is often chosen to be a few Hertz for the purpose of fundamental frequency extraction.

[0054] In some aspects, a time shift can be introduced by the generalized linear-phase low-pass filter 410, which can cause a delay between the filter input and output. Because the thermal time constant of a typical motor ranges from several minutes to an hour, this filter delay, usually a few seconds at most, can be regarded as negligible in final rotor temperature calculation (e.g., 124).

[0055] The start-up transient caused by a zero initial state in the generalized linear-phase low-pass filter 410 can be addressed by (1) determining the filter's initial state in advance, and/or (2) discarding the output that corresponds to the start-up transient. This start-up transient is typically negligibly short as compared to the motor's thermal dynamics, thus, discarding the first few seconds of the filter's output does not adversely affect final rotor temperature estimation results.

Numerical Differentiation Block

[0056] If φ is the phase of the complex baseband signal, z_n, produced by the generalized linear-phase low-pass filter (410), (e.g., φ= custom character

z), then the angular speed of the complex baseband signal, ω [electrical rad/s], is related to the continuous-time derivative of the phase φ via Equation 7.
ω=dφ/dt Equation 7

[0057] Frequency deviation, δf, is defined as the difference between the instantaneous frequency and its prescribed nominal value. In the fundamental frequency extraction (114), this frequency deviation is related to the difference between the instantaneous fundamental frequency (414) and the rated fundamental frequency (404). The frequency deviation quantity is calculated from ω via Equation 8.
δf=ω/(2π) Equation 8

[0058] In a discrete-time system, the numerical differentiation block (412) approximates the continuous-time derivative of the phase given in Equations 7 and 8. If the sampling interval is denoted as T_s(T_s=1/f_s), three different formulas can be utilized to approximate the continuous-time derivative of the phase depending on the desired precision.

[0059] 1) A forward difference formula provides:
δf_n=(φ_n+1−φ_n)/(2π·T_s)=( custom character

z_n+1−

z_n)/(2π·T_s) Equation 9
where the approximation error in δf_nis proportional to T_s. A backward difference formula can be used in place of the forward difference formula to compute δf_n, which provides:
δf_n=(φ_n−φ_n−1)/(2π·T_s)=( custom character

z_n−

z_n−1)/(2π·T_s) Equation 10
where the approximation error in δf_nis proportional to T_s.

[0060] 2) A three-point formulas provides:
δf_n=(−3 custom character

z_n+4

z_n+1−

z_n+2)/(2π·2T_s) Equation 11a
δf_n=(− custom character

z_n−1+

z_n+1)/(2π·2T_s) Equation 11b
δf_n=( custom character

z_n−2−4

z_n−1+3

z_n)/(2π·2T_s) Equation 11c
where the approximation error in δf_nis proportional to T_s². For a sequence of complex numbers, Equation 11a is used to compute δf for the first sample in this sequence, while Equation 11c is used to compute δf for the last sample. For the rest samples in this sequence, Equation 11b is used to compute δf.

[0061] 3) A five-point formulas provides:
δf_n=(−25 custom character

z_n+48

z_n+1−36

z_n+2+16

z_n+3−3

z_n+4)/(2π·12T_s) Equation 12a
δf_n=( custom character

z_n−2−8

z_n−1+8

z_n+1−

z_n+2)/(2π·12T_s) Equation 12b
δf_n=(3 custom character

z_n−4−16

z_n−3+36

z_n−2−48

z_n−1+25

z_n)/(2π·12T_s) Equation 12c
where the approximation error in δf_nis proportional to T_s⁴. For a sequence of complex numbers, Equation 12a determines δf for the first two samples in this sequence, while Equation 12c determines δf for the last two samples. For the rest samples in the sequence, Equation 12b determines δf.

[0062] According to some aspects of the present concepts, a discrete-time differentiator can be used as an alternative to the numerical differentiation block (412) discussed above. Such a discrete-time differentiator is often designed from a linear-phase system using a Kaiser window.

Instantaneous Fundamental Frequency Calculation

[0063] The instantaneous fundamental frequency, f_0,n[Hz] (414), can be calculated by summing up the frequency deviation, output from the numerical differentiation block (412) with the rated fundamental frequency (404), as provided in Equation 13.
f_0,n=f₀+δf_n Equation 13
Dominant Rotor Slot Harmonic Frequency Extraction

[0064] For a squirrel-cage induction motor, a finite number of rotor slots produce steps in a waveform of rotor magnetomotive forces (MMFs) when excited by a positive sequence flux wave at a rated fundamental frequency from the stator. These steps in the rotor MMF affect the waveform of an air gap flux. In addition, the large reluctance of the rotor slot conductors and flux saturation due to high field concentration at the rotor slot openings produce spatial variations in, for example, an induction motor's air gap permeance. Because the air gap permeance interacts with the MMF from both the stator and the rotor to produce air gap flux, variations in the air gap permeance manifest themselves in the air gap flux. Furthermore, because the air gap flux is linked to stator windings, the variations in the air gap flux are reflected in a stator current harmonic spectrum.

[0065] A dominant rotor slot harmonic typically has the largest amplitude among all rotor slot harmonic components in the stator current harmonic spectrum. The dominant rotor slot harmonic frequency is related to the rotor design parameters via the relationship shown in Table I below:

[0066]

TABLE I

Relationship between the Dominant Rotor Slot
Harmonic Frequency and Rotor Design Parameters
Case	h = f_sh/f₀\|_slip=0	Condition

1	(kR/P − 1)	kR/P − 1 = 6m + 1
2	−(kR/P − 1)	kR/P − 1 = 6m − 1
3	NONE	kR/P − 1≠6m ± 1
4	(kR/P + 1)	kR/P + 1 = 6m + 1
5	−(kR/P + 1)	kR/P + 1 = 6m − 1
6	NONE	kR/P + 1≠6m ± 1

where f_sh[Hz] is the dominant rotor slot harmonic frequency; k=1, 2, 3, . . . , indicates the rotor MMF distribution harmonic order; m=1, 2, 3, . . . , is a positive integer; P is the number of pole-pairs; R is the number of rotor slots.

[0067] For many motors, Table I provides guidelines on the relationship between the dominant rotor slot harmonic frequency and the rotor design parameters. For example, for a 20-hp 4-pole induction motor with R=40, the condition associated with Case 1 in Table I, kR/P−1=6m+1, is satisfied when k=1 and m=3. Therefore, the dominant rotor slot harmonic frequency lies at (kR/P−1)·f₀=19·f₀when slip=0. FIG. 3A illustrates an exemplary stator current harmonic spectrum indicating that the motor's dominant rotor slot harmonic frequency (302) is 1113 Hz when the motor is connected to the power supply with a rated fundamental frequency of 60 Hz and operated at around rated load.

[0068] By way of another example, a 7.5-hp 4-pole induction motor with R=44 (e.g., the condition associated with Case 5 in Table I), where kR/P+1=6m−1, satisfies the condition when k=1 and m=4. Therefore, the dominant rotor slot harmonic frequency lies at −(kR/P+1)·f₀=−23·f₀when slip=0. FIG. 3B illustrates an exemplary stator current harmonic spectrum indicating that the motor's dominant rotor slot harmonic frequency (314) is −1357 Hz when the motor is connected to a power supply with a rated fundamental frequency of 60 Hz and operated at around rated load.

[0069] By way of yet another example with a 2-hp 6-pole induction motor with R=33, (e.g., the condition associated with Case 5 in Table I), where kR/P+1=6m−1, the condition is satisfied when k=2 and m=4. Therefore, the dominant rotor slot harmonic frequency lies at −(kR/P+1)·f₀=−23·f₀when slip=0. FIG. 3C illustrates an exemplary stator current harmonic spectrum indicating that the motor's dominant rotor slot harmonic frequency (324) is −1351 Hz when the motor is connected to a power supply with a rated fundamental frequency of 60 Hz and operated at around rated load.

[0070] FIG. 4B is a block diagram of the rotor slot harmonic frequency extraction (116) of FIG. 1. Similar to the fundamental frequency extraction (114) scheme described above, the rotor slot harmonic frequency extraction (116) scheme includes a VCO 426, a generalized linear-phase low-pass filter 430, a numerical differentiation block 432 and an instantaneous dominant rotor slot harmonic frequency calculation block 434.

Voltage Controlled Oscillator 426

[0071] In FIG. 4B, the Voltage Controlled Oscillator 426 takes a nominal dominant rotor slot harmonic frequency as its input and produces a complex exponential signal as its output. The nominal dominant rotor slot harmonic frequency, f_sh, (424) can be determined from a motor (e.g., induction motor 106) nameplate data plus rotor design parameters. The rotor design parameters can often be obtained from either the motor manufacturers or certain electric motor energy and management databases, such as the software tools provided by the U.S. Department of Energy. According to other aspects, the dominant rotor slot harmonic frequency can be determined by applying spectral estimation techniques to the stator current harmonic spectrum. A first approach relies on the knowledge of an approximate slip, usually derived from a linear relationship between the slip and the input power. A second approach typically requires steady-state motor operation during data acquisition.

[0072] Once the nominal dominant rotor slot harmonic frequency (424) is determined, the VCO 426 takes this nominal frequency (424) as its input and synthesizes a complex exponential signal, x_n, as its output, where x_nis defined in Equation 14.
x_n=exp(j·2πn·f_sh/f_s) Equation 14
Multiplying the complex current space vector, i_C,n, (238) with the complex conjugate (428) of the VCO 426 output, x_n*, mixes the dominant rotor slot harmonic down to a complex baseband.
Generalized Linear-Phase Low-Pass Filter

[0073] A filter with nonlinear phase can distort the shape of a signal, even when the filter's frequency-response magnitude is constant. Thus, according to some aspects, the generalized linear-phase low-pass filter 430 is used in the rotor slot harmonic frequency extraction (116). This filter 430 has a structure similar to its counterpart (e.g., filter 410) used in the fundamental frequency extraction (114). However, the cutoff frequency of the filter 430 can be chosen so that the filter 430 can pass a range of dominant rotor slot harmonic frequencies that corresponds to 50-150% of the motor's rated input power. The motor's (e.g., induction motor 106) rated input power is usually calculated from motor nameplate data, and the corresponding rotor slot harmonic frequencies can be approximated according to a linear relationship between the input power and the slip. For example, U.S. Provisional Application Ser. No. 61/053,941, filed May 16, 2008, entitled “Methods and Apparatuses for Estimating Transient Slip, Rotor Slots, Motor Parameters, and Induction Motor Temperature,” which is incorporated by reference herein, provides examples for approximating the corresponding rotor slot harmonic frequencies. Thus, the generalized linear-phase low-pass filter 430 is designed along the same rules outlined above in reference to the generalized linear-phase low-pass filter 410 but with a different cutoff frequency.

Numerical Differentiation Block

[0074] The structure of the numerical differentiation block 432 used in the rotor slot harmonic frequency extraction (116) is identical to its counterpart (412) used in the fundamental frequency extraction (114).

Instantaneous Dominant Rotor Slot Harmonic Frequency Calculation

[0075] The instantaneous dominant rotor slot harmonic frequency, f_sh,n[Hz], (434) is calculated by summing up frequency deviation and the nominal dominant rotor slot harmonic frequency (424) according to Equation 15.
f_sh,n=f_sh+δf_n Equation 15
Rotor Speed Estimation

[0076] FIG. 5 is a block diagram of the rotor speed estimation (120). The rotor speed, ωR [electrical rad/s], (520) can be calculated from the extracted instantaneous fundamental frequency (414) and the extracted instantaneous dominant rotor slot harmonic frequency (434) via Equation 16:
ω_R,n=2πP·(±f_sh,n−n_w·f_0,n)/(kR) Equation 16
where ‘+’ sign (504) corresponds to Cases 1 and 4 in Table I, and ‘−’ sign (506) corresponds to Cases 2 and 5 in Table I. Furthermore, n_wis the stator winding distribution harmonic order. Corresponding to Cases 1 and 2 in Table I, n_w=−1 (514). Corresponding to Cases 4 and 5 in Table I, n_w=+1 (516). For example, speed estimation for a motor with a stator current harmonic spectrum as illustrated in FIG. 3A, will use a ‘+’ sign (504) and n_w=−1 (514) in Equation 16. For another example, rotor speed estimation for a motor with a stator current harmonic spectrum as illustrated in FIG. 3B or FIG. 3C, will use a ‘−’ sign (506) and n_w=+1 (516) in Equation 16.

[0077] FIG. 6 is a chart 600 depicting an estimated instantaneous rotor speed (RPM) over time (604) from the voltage measurements (102) and current measurements (104) for a 20-hp 4-pole induction motor (e.g., induction motor 106) with 40 rotor slots, as outlined above. FIG. 6 compares the estimated instantaneous rotor speed with an instantaneous rotor speed detected from a speed sensing device, such as a tachometer mounted on the motor shaft to provide a voltage output that is proportional to the rotor speed. The tachometer output can be converted into rpm values and plotted over time (602), shown in FIG. 6, as a reference signal to compare and/or validate the rotor speed estimation (120) scheme outlined above.

[0078] While the instantaneous rotor speed, ω_R,n, can be detected by applying the rotor slot harmonic frequency extraction (116) scheme to a dominant rotor slot harmonic described above, the present disclosure also contemplates other variations of determining the instantaneous rotor speed, ω_R,n. For example, the instantaneous rotor speed, ω_R,n, can also be detected by applying a slightly modified scheme to other non-dominant rotor slot harmonics (e.g., rotor slot harmonics 304 and 312 shown in FIGS. 3A and 3B). In these other schemes, a nominal rotor slot harmonic frequency (e.g., 304 and 312) that corresponds to the non-dominant rotor slot harmonic is inputted into a VCO, and the value of n_wused in Equation 16 is an odd integer other than ±1.

Positive Sequence Fundamental Frequency Component Extraction

[0079] For the purpose of rotor time constant estimation (122), which occurs after the rotor speed estimation, the positive sequence fundamental frequency voltage and current components are typically extracted from the complex voltage and current space vectors, respectively.

Positive Sequence Fundamental Frequency Voltage Component Extraction

[0080] FIG. 7A is a block diagram of the positive sequence fundamental frequency voltage component extraction (112). It includes a VCO 706 and a discrete-time low-pass filter 702.

Voltage Controlled Oscillator

[0081] The VCO 706 used in the positive sequence fundamental frequency voltage component extraction (112) has a structure similar to the VCO 406. The complex conjugate (708) of the VCO output, x_n*, is used in the demodulation to mix the positive sequence fundamental frequency voltage component in the complex voltage space vector (218) down to a complex baseband.

Low-Pass Filter

[0082] The low-pass filter 702 can be either a discrete-time FIR or a discrete-time IIR low-pass filter. The cutoff frequency of this low-pass filter 702 is usually chosen to accommodate the frequency deviation, allowed by power distribution network regulators, around the rated fundamental frequency f₀(404). To simplify the implementation, the generalized linear-phase low-pass filter 410 used in the fundamental frequency extraction (114) can also be used.

[0083] Multiplying the low-pass filter output, z_n, with the VCO output, x_n, according to Equation 17, provides the positive sequence fundamental frequency voltage component, v_F,n, (704).
v_F,n=z_n·x_n Equation 17
Positive Sequence Fundamental Frequency Current Component Extraction

[0084] As shown in FIG. 7B, the positive sequence fundamental frequency current component extraction (118), of FIG. 1, has a structure that resembles the positive sequence fundamental frequency voltage component extraction (112). Similarly, the structure includes a rated fundamental frequency f₀(404) input, a VCO 726, a complex conjugate block (728), and a low-pass filter 712. The positive sequence fundamental frequency current component extraction (118) takes the complex current space vector, i_C,n, (238) as an input, and produces the positive sequence fundamental frequency current component, i_F,n, (714) as its output. In addition, the cutoff frequency of the low-pass filter 712 should be high enough to retain the variations in the load-dependent positive sequence fundamental frequency current component. For example, if a motor experiences a 5 Hz load oscillation when connected to, for example, a reciprocating compressor, then the low-pass filter cutoff frequency can be chosen to be a few Hertz higher than this load oscillation frequency to accommodate the load-dependent positive sequence fundamental frequency current component.

Rotor Time Constant Estimation

[0085] A rotor time constant (e.g., 924) can be estimated from a model-reference adaptive system, such as model-reference adaptive system 900 (“MRAS”) shown in FIG. 9. The MRAS 900 can be configured to adjust the rotor time constant (924) to align a flux-related vector calculated from a rotor circuit equation (930) with another flux-related vector calculated from a stator circuit equation (925).

Stator and Rotor Circuit Equations

[0086] FIG. 8 illustrates a dynamic induction motor equivalent circuit (e.g., 800) in a stationary reference frame.

[0087] According to FIG. 8, the stator circuit Equation 18 is:
pλ_F=(v_F−R_S·i_F)−σL_S·pi_F Equation 18

[0088] where the operator p represents the continuous-time derivative operator d/dt; λ_Fis a complex flux space vector; R_Sis stator resistance; σ is total leakage factor; L_Sis stator inductance.

[0089] According to FIG. 8, the rotor circuit Equation 19 is:
pλ_F=[(1−σ)L_S/τ_R]·i_F+(jω_R−1/τ_R)·λ_F Equation 19
where τ_Ris the rotor time constant. It is defined as the ratio between rotor inductance, L_R, and rotor resistance, R_R, i.e., τ_R=L_R/R_R. Instantaneous rotor speed is ω_R[rad/s]. The imaginary unit j rotates the complex quantities counter-clockwise by π/2.

[0090] The total leakage factor, σ, and the stator inductance, L_S, can be obtained from either motor manufacturer or standard no load and locked rotor tests according to the IEEE Standard 112 entitled “Standard Test Procedure for Polyphase Induction Motors and Generators”. In case the information from neither channel is available, the inductances are estimated according to the method outlined in the section entitled “Induction Motor Electrical Parameter Estimation.”

High-Pass Filter

[0091] Incorrect initial values produce dc offset errors in the flux vectors when computed according to Equations 18 and 19. One or more high-pass filters (e.g., a high-pass filter 906) can be used in the MRAS 900 to remove such errors by filtering the outputs of both the stator and rotor circuit equations (925 and 930). Substituting the expression pλ_Fwith (p+ω_c)γ_F, where ω_cis the high-pass filter cutoff frequency, and γ_Fis an auxiliary vector denoting the frequency contents in λF that are above the high-pass filter cutoff frequency, transforms the Equations 18 and 19 to Equations 20 and 21, respectively,
γ_FS=(v_F−R_S·i_F)/(p+ω_c)−σL_S·pi_F/(p+ω_c) Equation 20
γ_FR=[(1−σ)L_S/τ_R]·i_F/(p+ω_c)+(jω_R−1/τ_R)·γ_FR/p Equation 21
where the expression, γ_FS, is a flux-related vector calculated from the stator circuit equation (925) according to Equation 20; and the expression, γ_FR, is a flux-related vector calculated from the rotor circuit equation (930) according to Equation 21. High-pass filter (906) has a form of p/(p+ω_c), and low-pass filters 904, 914 are in the form of 1/(p+ω_c). Additionally, according to some aspects, both the high-pass filter and the low-pass filters have the same cutoff frequency.
Rotor Time Constant Adaptation

[0092] The MRAS 900 adapts the rotor time constant based on the magnitude of the cross product between γ_FSand γ_FR, as shown in Equation 22:
ε_F=γ_FS×γ_FR=real(γ_FS)·imag(γ_FR)−imag(γ_FS)·real(γ_FR) Equation 22
where ε_Fdenotes the magnitude of the cross product; the symbol × denotes the cross product operation; real(·) denotes the real part of a complex quantity; and imag(·) denotes the imaginary part of a complex quantity. Rotor time constant adaptation (935) uses γ_FSfrom the stator circuit equation (925) as a reference vector and adjusts the rotor time constant, τ_R, so that ε_Fapproaches 0. This indicates that the flux-related vector, γ_FR, is aligned with the reference vector γ_FS. As shown in FIG. 9, a proportional-integral block is used in the rotor time constant adaptation (935) to correlate ε_Fto γ_Rusing Equation 23:
τ_R=(K_p+K_i/s)·ε_F Equation 23
where 1/s denotes an integrator (922).

[0093] The tuning of K_p(918) and K_i(920) values depends on knowledge of induction motor electrical parameters, as well as a motor's operating condition. According to some aspects, commonly used methods for tuning proportional-integral controllers in feedback control systems, such as the trial and error method or the Ziegler-Nichols method, provide a good starting point for fine-tuning. For most line-connected induction motor applications, small positive values of K_p(918) and K_i(920) suffice. Large positive values of K_p(918) and K_i(920) can decrease the rise time and increase the overshoot, but they can also introduce instability when tuned improperly.

[0094] According to some aspects of the present concepts, a positive value, typically between 0.1 and 1 second, is set as the initial value of τ_Rbefore the adaptation (935) is started. Such a practice avoids the “division by zero” error arising from the implementation of 1/τ_Rand the subsequent instability in MRAS operation.

[0095] For most line-connected induction motors, thermal behavior is characterized by a slow thermal process with a large thermal time constant. Therefore, when tuned according to the tuning methods discussed above, the response time of the MRAS 900 is much smaller than the motor's thermal time constant. As a result, the MRAS output is regarded as the real-time value of the rotor time constant (924).

Rotor Temperature Calculation

[0096] According to some aspects of the present disclosure, a rotor temperature calculation can be based on an inverse relationship between a rotor temperature and the estimated rotor time constant (924), determined in block (122) of FIG. 1. This relationship is derived from a linear relationship between the rotor temperature and a rotor resistance.

[0097] Although rotor resistance is a function of both rotor temperature and slip frequency, the slip frequency of a line-connected induction motor (e.g., induction motor 106) only varies within a small range during normal motor operation. In addition, deep rotor bars or double cage rotors are rarely used in such motors. Therefore, the skin effect of the rotor resistance is not pronounced. Given that the rotor resistance is either independent of the slip, or properly compensated for based on known slip-resistance characteristics, the change in the rotor resistance is caused primarily by the change in the rotor temperature.

[0098] Assuming at time n₀, that the rotor resistance is R_R,n0, and the corresponding rotor temperature is θ_R,n0[° C.]; and at time n, the rotor resistance is R_R,n, and the corresponding rotor temperature is θ_R,n, then the relationship between the rotor resistance and the rotor temperature can be defined as in Equation 24:
R_R,n/R_R,n0=(θ_R,n+k)/(θ_R,n0+k) Equation 24
where k is a temperature coefficient related to the inferred temperature for zero resistance. According to some aspects, a temperature coefficient k is 234.5 for motors having 100% International Annealed Copper Standard conductivity copper. According to other aspects, a temperature coefficient k is 225 for motors having aluminum with a volume conductivity of 62%. It is contemplated that other suitable values of k are used for other motor materials.

[0099] The baseline rotor temperature, θ_R,n0, can be either specified by the user, or determined from the motor's ambient temperature, a motor's ambient temperature is often a known quantity. Because thermal processes are usually slow, according to some aspects, this ambient temperature is taken as the baseline rotor temperature, θ_R,n0, during the first several seconds after the motor is energized from a cold state. The baseline rotor time constant, τ_R,n0, can be obtained from any of the following method 1) user input; 2) τ_R,0according to the method described in “Induction Motor Electrical Parameter Estimation” described below. 3) use the averaged rotor time constant produced by the rotor time constant estimation (122) during the first several seconds after the motor is energized and the rotor time constant estimation (122) produces stabilized outputs. This rotor time constant and temperature pair, (τ_R,n0, θ_R,n0), provides baseline information for the subsequent calculation of the rotor temperature at time n.

[0100] According to some aspects, a comparison of the rotor time constant, τ_R,n(924) at time n, which is obtained at the same motor load level, with the baseline rotor time constant, τ_R,n0, at time n₀, yields the rotor temperature θ_R,n, at time n, which is defined in Equation 25.
θ_R,n=(θ_R,n0+k)·τ_R,n0/τ_R,n−k Equation 25

[0101] For example, FIG. 10 illustrates a flowchart (1000) of a rotor temperature calculation. Assuming that the ambient temperature is used as the baseline rotor temperature θ_R,n0, and that τ_R,n0is obtained using the third method outlined above. Each time a new rotor time constant value, τ_R,n, is calculated from current and voltage measurements along the lines outlined above, it is compared with the baseline value, τ_R,n0, and the corresponding new rotor temperature, θ_R,n, is computed using Equation 25 (1006).

[0102] FIG. 11A illustrates an exemplary chart 1100 of a dynamic motor's operation (i.e., power and speed) during an experiment using a 20-hp 4-pole induction motor with 40 rotor slots. During the experiment, a dc dynamometer can be used as a load. Additionally, a 5 Hz oscillation can be superimposed on the rated load of the motor to emulate, for example, a reciprocating compressor load. The amplitude of the oscillating load can be set to approximately 5% of the motor's rated load. After about 58 seconds, the oscillating load is disconnected and the motor continues to operate at the rated load until the end of the experiment. The upper plot of FIG. 11A illustrates the instantaneous input power over time (1102) calculated from the motor terminal voltage and current measurements (e.g., 102 and 104), and the lower plot illustrates the instantaneous rotor speed over time (1104), which is estimated from the voltage and current measurements (e.g., 102 and 104) as described above.

[0103] FIG. 11B depicts graphical results (1110) of calculating a rotor temperature. According to some aspects, the estimated rotor temperature for the tested a 20-hp 4-pole induction motor with 40 rotor slots is calculated in accordance with the present disclosure, and the calculated rotor temperature is plotted over time (1112). Additionally, rotor temperatures are measured from two J-type thermocouples that are instrumented in the stator windings, and the temperature readings from those thermocouples are also plotted over time (1114 and 1116) in the same figure for comparison purposes. As illustrated in FIG. 11B, the estimated rotor temperature (1112) converges to the temperatures measured by the two thermocouples after about 20 seconds of rotor time constant adaptation. Thus, the estimated rotor temperature outputted from, for example, rotor temperature calculation block (124) tracks the temperature measured by the two thermocouples continuously and accurately using the voltage and current measurements (e.g., 102 and 104) sampled from, for example, the induction motor 106 of FIG. 1.

[0104] Referring to FIG. 12, an architecture 1200 for estimating electrical parameters of an induction motor 106 is illustrated according to some aspects of the present disclosure. The architecture 1200 includes a complex voltage space vector synthesis block 108, a complex current space vector synthesis block 110, a rotor speed estimation block 120, a positive sequence fundamental frequency voltage component extraction block 112, a positive sequence fundamental frequency current component extraction block 118, and an induction motor electrical parameter estimation block 1204.

Induction Motor Electrical Parameter Estimation

[0105] According to some aspects, induction motor electrical parameters can be determined from the motor nameplate data, as well as voltage and current measurements (e.g., 102 and 104) acquired at motor control centers and/or at motor terminals. According to some aspects, an induction motor electrical parameter estimation algorithm can be formulated from a dynamic induction motor equivalent circuit model.

[0106] For example, FIG. 13 depicts a dynamic induction motor equivalent circuit model 1300 in a stationary reference frame. The complex stator flux linkage, λ_S, is related to the positive sequence fundamental frequency voltage component, v_F, and the positive sequence fundamental frequency current component, i_F, as defined in Equation 26:
pλ_S=v_F−R_S·i_F Equation 26
where p represents the continuous-time derivative operator d/dt; and R_Sis the stator resistance. The complex stator flux linkage, λ_S, is also related to the instantaneous rotor speed, ω_R[rad/s], via:
pλ_S=σL_S·pi_F+(L_S/τ_R−jω_R·σL_S)·i_F+(jω_R−1/τ_R)·λ_S Equation 27
where L_Sis the stator inductance; σ is the total leakage factor; τ_Ris the rotor time constant. The rotor time constant is defined as the ratio between the rotor inductance, L_R, and the rotor resistance, R_R, i.e., τ_R=L_R/R_R. The imaginary unit j rotates the complex space vector counter-clockwise by π/2.
Stator Resistance

[0107] An estimate of stator resistance is conveniently produced based on the motor nameplate data. This estimate is calculated as a ratio of a nameplate voltage, v₀, and a nameplate current, i₀, scaled by a scalar constant, α_R.

[0108]

\begin{matrix} R_{S} = α_{R} \cdot (\frac{v_{0}}{i_{0}}) ❘_{α_{R} = 1.155 e - 2} Ω & Equation 28 \end{matrix}

where the nameplate voltage is typically a line-to-line root mean square rated voltage, and the nameplate current is typically a single phase root mean square rated current.
Induction Motor Electrical Parameters

[0109] The process of estimating induction motor electrical parameters is illustrated as a compound process, as two distinct and independent methods, each expressing relative strengths and advantages. A first method is an Efficient Recursive Method, described in the section below entitled Efficient Recursive Method. A second method is a Maximum Diversity Method, described in the section below entitled Maximum Diversity Method. The induction motor electrical parameters, L_S,m, σ_m, and τ_R,0,m, determined by either the Efficient Recursive Method or the Maximum Diversity Method, are equally valid, and the selection of the specific method to employ is application dependent.

[0110] The Efficient Recursive Method supports the definition and evaluation of a source matrix and a source vector with a variable row dimension, with a recursive solution. The Maximum Diversity Method defines a source matrix and a source vector with fixed dimensions, opportunistically replacing rows when new information is available, to ensure that the condition of derived matrix, and the accuracy of the solution vector are optimized.

Induction Motor Equivalent Circuit Model Equations

[0111] The stator inductance, the total leakage factor and the rotor time constant are estimated according to the dynamic induction motor equivalent circuit 1300 shown in FIG. 13 by substituting Equation 26 to Equation 27, which yields Equation 29:
v_F−R_S·i_F=σL_S·pi_F+(L_S/τ_R−jω_R·σL_S)·i_F+(jω_R−1/τ_R)λ_S Equation 29
Rearranging equation 29 yields:
σL_S·(pi_F−jω_R·i_F)=−(L_S/τ_R)·i_F+(1/τ_R)λ_S+[(v_F−R_S·i_F)−jω_R·λ_S] Equation 30
Dividing both sides of Equation 30 by σL_Sand then taking the continuous-time derivative of both sides yield:
p²i_F−jω_R·(pi_F)=[L_S/(σL_Sτ_R)]·(−pi_F)+[1/(σL_Sτ_R)]·(pλ_S)+[1/(σL_S)]·[p(v_F−R_S·i_F)−jω_R·pλ_S] Equation 31
where p²i_Fdenotes the second-order continuous-time derivative of the positive sequence fundamental frequency current component (e.g., p²i_F=d²(i_F)/dt²). Equation 30 can be simplified by replacing pλ_Swith (v_F−R_S·i_F) according to Equation (27), which yields:
p²i_F−jω_R·(pi_F)=[L_S/(σL_Sτ_R)]·(−pi_F)+[1/(σL_Sτ_R)]·(v_F−R_S·i_F)+[1/(σL_S)]·[p(v_F−R_S·i_F)−jω_R·(v_F−R_S·i_F)] Equation 32
Equation 32 can be written in a matrix form by defining the following quantities:
y=p²i_F−jω_R·(pi_F) Equation 33
{right arrow over (u)}=[−pi_F, v_F−R_S·i_F, p(v_F−R_S·i_F)−jω_R·(v_F−R_S·i_F)] Equation 34
{right arrow over (ξ)}=[L_S/(σL_Sτ_R), 1/(σL_Sτ_R), 1/(σL_S)]^T Equation 35
where the superscript T in Equation 35 denotes the matrix transpose. Note that it is not a complex conjugate transpose.

[0112] Consequently, Equation 32 can be transformed into matrix format as defined in Equation 36.
y={right arrow over (u)}·{right arrow over (ξ)} Equation 36
Calculation of y and u

[0113] Given a sequence of samples of positive sequence fundamental frequency voltage and current components, it is desirable to use a numerical differentiation block to compute first-order continuous-time derivative of the positive sequence fundamental frequency voltage and current components, pv_Fand pi_F, respectively; as well as second-order continuous-time derivative of the positive sequence fundamental frequency current component, p²i_F. The structure of the numerical differentiation block used here is described in the section entitled Numerical Differentiation Block. FIG. 14 shows the temporal alignment that needs to be carefully considered in the implementation.

Temporal Alignment

[0114] As shown in FIG. 14, two numerical differentiation blocks 1406, 1410 on the right are cascaded to compute second-order continuous-time derivative of the positive sequence fundamental frequency current component (e.g., p²i_F). When Kaiser window-based discrete-time differentiators are used, the latency introduced by the differentiator must be compensated. The “Latency Compensation—Time Delay Δn_L” blocks 1404, 1408 in FIG. 14 ensure the temporal alignment among all signals for subsequent calculations. The time delay, Δn_L, is typically obtained by evaluating a differentiator's architecture during its design stage.

Least-Squares Solution

[0115] According to some aspects, for motor operation that covers more than one load level, one y_iand one {right arrow over (u)}_ican be formed for each load level. Consequently, a vector, {right arrow over (y)}, can be obtained for m load levels along with a matrix, {right arrow over (U)}, where {right arrow over (y)} is defined according to Equation 37 and {right arrow over (U)} is defined according to Equation 38.
{right arrow over (y)}=[y₁, y₂, . . . , y_i, . . . , y_m]^T Equation 37
{right arrow over (U)}=[{right arrow over (u)}₁^T, {right arrow over (u)}₂^T, . . . , {right arrow over (u)}_i^T, . . . , {right arrow over (u)}_m^T]^T Equation 38

[0116] Consequently, Equation 36 can be expanded for motor operation with m load levels according to Equation 39.
{right arrow over (y)}={right arrow over (U)}·{right arrow over (ξ)} Equation 39

[0117] According to some aspects, {right arrow over (y)}_mand {right arrow over (U)}_mcan be constructed by taking the real and/or imaginary part of components in {right arrow over (y)} and {right arrow over (U)} separately as defined in Equations 40a,b,c and 41a,b,c.
{right arrow over (y)}_m=[real(y₁), imag(y₁), real(y₂), imag(y₂), . . . , real(y_i), imag(y_i), . . . , real(y_m), imag(y_m)]^T Equation 40a
{right arrow over (U)}_m=[real({right arrow over (u)}₁^T), imag({right arrow over (u)}₁^T), real({right arrow over (u)}₂^T), imag({right arrow over (u)}₂^T), . . . , real({right arrow over (u)}_i^T), imag({right arrow over (u)}_i^T), . . . , real({right arrow over (u)}_m^T), imag({right arrow over (u)}_m^T)]^T Equation 41a
{right arrow over (y)}_m=[real(y₁), real(y₂), . . . , real(y_i), . . . , real(y_m)]^T Equation 40b
{right arrow over (U)}_m=[real({right arrow over (u)}₁^T), real({right arrow over (u)}₂^T), . . . , real({right arrow over (u)}_i^T), . . . , real({right arrow over (u)}_m^T)]^T Equation 41b
{right arrow over (y)}_m=[imag(y₁), imag(y₂), . . . , imag(y_i), . . . , imag({right arrow over (y)}_m)]^T Equation 40c
{right arrow over (U)}_m=[imag({right arrow over (u)}₁^T), imag({right arrow over (u)}₂^T), . . . , imag({right arrow over (u)}_i^T), . . . , imag({right arrow over (u)}_m^T)]^T Equation 41c

[0118] At least two distinct load levels are required if Equations 40a and 41a are used to construct {right arrow over (y)}_mand {right arrow over (U)}_m, (e.g., m≧2). In case either Equations 40b and 41b, or Equations 40c and 41c, are used to construct {right arrow over (y)}_mand {right arrow over (U)}_m, then at least three distinct load levels are required (e.g., m≧3).

[0119] A least-squares solution can be obtained from Equation 39 via a pseudo-inverse solution as defined in Equation 42.
{right arrow over (ξ)}_m=({right arrow over (U)}_m^T{right arrow over (U)}_m)⁻¹·({right arrow over (U)}_m^T{right arrow over (y)}_m) Equation 42

[0120] Finally, based on the definitions of Equation 35, {right arrow over (ξ)}_m(1)=L_S/(σL_Sτ_R), {right arrow over (ξ)}_m(2)=1/(σL_Sτ_R), {right arrow over (ξ)}_m(3)=1/(σL_S). Therefore, the stator inductance, the total leakage factor, and the rotor time constant can be estimated from the least-squares solution produced by Equation 42, where the stator inductance, Ls, is defined in Equation 43; the total leakage factor, σ, is defined in Equation 44; and the rotor time constant, τ_R, is defined in Equation 45.
L_S,m={right arrow over (ξ)}_m(1)/{right arrow over (ξ)}_m(2) Equation 43
σ_m={right arrow over (ξ)}_m(2)/[{right arrow over (ξ)}_m(1){right arrow over (ξ)}_m(3)] Equation 44
τ_R,0,m={right arrow over (ξ)}_m(3)/{right arrow over (ξ)}_m(2) Equation 45
This τ_R,0,mcan be the same as an initial rotor time constant, τ_R,0, outputted by the induction motor electrical parameter estimation block (1204).

[0121] As discussed above, the induction motor electrical parameter estimation given in Equations 43-45 relies on the least-squares solution of {right arrow over (ξ)}_min Equation 42. Thus, according to some aspects, matrix {right arrow over (U)}_m^T{right arrow over (U)}_mis checked to not be singular before performing the matrix inverse, ({right arrow over (U)}_m^T{right arrow over (U)}_m)⁻¹. Instead of performing a conventional singular value decomposition for this purpose, readily available quantities can be used more efficiently, such as the complex input power, as indices to ensure a nonsingular {right arrow over (U)}_m^T{right arrow over (U)}_m. The complex input power is defined in the “Initial Synthesis” section below.

Efficient Recursive Method

[0122] An efficient recursive method is described in this section as an efficient way to reduce the computation of the pseudo-inverse in Equation 42 each time a new set of qualified data becomes available, as shown in the flow chart of FIG. 16. Unlike the Maximum Diversity Method (e.g., 1500 in FIG. 15), which operates on the vector, {right arrow over (y)}, and the matrix, {right arrow over (U)}, whose number of rows is fixed, the Efficient Recursive Method has no such restrictions and can be applied to a vector, {right arrow over (y)}, and a matrix, {right arrow over (U)}, with a growing number of rows.

[0123] The efficient recursive method first determines if vector {right arrow over (y)}_mand matrix {right arrow over (U)}_mare empty (1604) (e.g., whether they contain any values). If vector {right arrow over (y)}_mand matrix {right arrow over (U)}_mare empty, then the appropriate {right arrow over (y)}_mand {right arrow over (U)}_mare automatically constructed (1608). It is possible to construct vector {right arrow over (y)}_mand matrix {right arrow over (U)}_maccording to Equations 40b and 41b, where {right arrow over (U)}_m^T·{right arrow over (U)}_mand {right arrow over (U)}_m^T·{right arrow over (y)}_mare assumed to have been computed in accordance with a Least-Squares Solution obtained from Equation 42 (1612). If the vector {right arrow over (y)}_mand the matrix {right arrow over (U)}_mare not empty, then an intermediate index, s_IN, can be defined based on the complex input power or a quantity derived from the complex input power. There are altogether m intermediate indices related to {right arrow over (y)}_mand {right arrow over (U)}_m, (e.g., s_IN,k), where k is a discrete index (1≦k≦m), and each k corresponds to one distinct row in {right arrow over (y)}_mand {right arrow over (U)}_m.

[0124] Whenever a new set of voltage and current measurements (e.g., 102 and 104) becomes available, an intermediate index, s_IN,NEW, can be computed for this new data set. This intermediate index can then be compared to the existing m intermediate indices associated with m rows of {right arrow over (y)}_mand {right arrow over (U)}_m, and m non-negative distances are calculated between s_IN,NEWand s_IN,k(1≦k≦m) according to:
Δs_IN,k=∥s_IN,NEW−s_IN,k∥, 1≦k≦m Equation 46
where the ∥·∥ denotes either an absolute value or an Euclidean norm.

[0125] Taking a minimum value of m distances from Equation 46 yields:
d_IN=min([Δs_IN,1, Δs_IN,2, . . . , Δs_IN,k, . . . , Δs_IN,m]^T) Equation 47
A diversity index, d_IN, is computed for each new data set. This diversity index is subsequently compared to a threshold, predetermined from the motor nameplate data. If the diversity index exceeds the threshold, then the new data set is regarded as a qualified data set (1606). This comparison procedure ensures that the matrix {right arrow over (U)} will not become ill-conditioned when solving Equation 42.

[0126] Each time a qualified new data set is available, the corresponding y and {right arrow over (u)}, denoted by a new quantity {right arrow over (y)}_m+1and a new vector {right arrow over (u)}_m+1, are appended to {right arrow over (y)}_mand {right arrow over (U)}_mrespectively (1608):
{right arrow over (y)}_m+1=[{right arrow over (y)}_m^T, real(y_m+1)]^T Equation 48
{right arrow over (U)}_m+1=[{right arrow over (U)}_m^T, real({right arrow over (u)}_m+1^T)]^T Equation 49

[0127] Consequently, the matrices in equation 42 are updated via:
{right arrow over (U)}_m+1^T·{right arrow over (U)}_m+1=[{right arrow over (U)}_m^T,{right arrow over (u)}_m+1^{T]·[{right arrow over (U)}}_m^T,{right arrow over (u)}_m+1^T]^T={right arrow over (U)}_m^T·{right arrow over (U)}_m+{right arrow over (u)}_m+1^T·{right arrow over (u)}_m+1 Equation 50
{right arrow over (U)}_m+1^T·{right arrow over (y)}_m+1=[{right arrow over (U)}_m^T, {right arrow over (u)}_m+1^T]·[{right arrow over (y)}_m^T, y_m+1]^T={right arrow over (U)}_m^T·{right arrow over (y)}_m+{right arrow over (u)}_m+1^T·y_m+1 Equation 51

[0128] Once enough rows are accumulated (e.g., m≧3), (1610) the estimated electrical parameters become:
{right arrow over (ξ)}_m+1=({right arrow over (U)}_m+1^T{right arrow over (U)}_m+1)⁻¹·({right arrow over (U)}_m+1^T{right arrow over (y)}_m+1)=({right arrow over (U)}_m^T·{right arrow over (U)}_m+{right arrow over (u)}_m+1^T·{right arrow over (u)}_m+1)⁻¹·({right arrow over (U)}_m^T·{right arrow over (y)}_m+{right arrow over (u)}_m+1^T·y_m+1) Equation 52

[0129] Since the matrices {right arrow over (U)}_m^T·{right arrow over (U)}_mand {right arrow over (U)}_m^T·{right arrow over (y)}_mhave previously been computed with m≧3, it is now only necessary to compute {right arrow over (u)}_m+1^T·{right arrow over (u)}_m+1and {right arrow over (u)}_m+1^T·y_m+1and subsequently to inverse a 3 by 3 matrix to solve for the least-squares solution of {right arrow over (ξ)}_m+1(1612). This procedure helps significantly reduce the computation time in the electrical parameter estimation each time a qualified new data set becomes available. By doing so, the induction motor electrical parameters computed from {right arrow over (ξ)}_m+1, following the same manner as Equations 43-45, benefit from the knowledge brought by the qualified new quantity y_m+1and the qualified new vector {right arrow over (u)}_m+1. After the vector {right arrow over (y)} and the matrix {right arrow over (U)} contain enough rows and sufficiently qualified (e.g., diverse) data sets to support the least-squares solution of {right arrow over (ξ)}, the induction motor electrical parameter estimation can be ceased (1614).

[0130] In case {right arrow over (y)}_mand {right arrow over (U)}_mare constructed according to either Equations 40a and 41a, or Equations 40c and 41c, comparably efficient equations (e.g., Equations 48-52) can be executed to produce estimation results. As shown in Equation 34, the structure of the vector {right arrow over (u)} and consequently the matrix {right arrow over (U)} makes {right arrow over (U)}_m^T{right arrow over (U)}_ma well-conditioned matrix once the conditions described above are satisfied. Therefore, a robust and efficient estimate of induction motor electrical parameters can be obtained (e.g., Equations 43-45) based on the above description. A default transition to the terminus can be defined, and the stator inductance, the total leakage factor, and the rotor time constant are produced as the output (1616).

Maximum Diversity Method

[0131] The Maximum Diversity Method (1500) defines a source matrix and a source vector with fixed dimensions, opportunistically replacing rows when new information is available, to ensure that the condition of derived matrix, and the accuracy of the solution vector is optimized.

[0132] Control flow presents an alternative system view, where the nature of the relationships between entities and processes are defined in terms of order and conditions of operation, in FIG. 15.

[0133] The epoch of control flow transitions to the Initial Synthesis process (1502).

[0134] The Initial Synthesis process (1502) constructs the initial representations of the source matrix, the source vector, and the diversity set. The diversity set is employed as an indirect means to estimate the condition of a matrix derived from the source matrix, ensuring robust and accurate subsequent definition of a solution vector. Minimum distance and mean distance are calculated, with respect to members of the initial diversity set. The Initial Synthesis process (1502) initializes the relevant values precisely once. In subsequent invocations, the process performs no actions. The Initial Synthesis process (1502) transitions to the Candidate Diversity Sets process.

[0135] The Candidate Diversity Sets process (1504) iteratively synthesizes candidate diversity sets, or potential replacements to the existing diversity set, formed by substituting a new value for each member of the existing set. For each candidate diversity set, a minimum distance is evaluated, and the maximum minimum distance is compared to the minimum distance of the diversity set. If the maximum candidate minimum distance is greater than the minimum distance of the diversity set, of if the distances are equal and the candidate mean distance is greater than the mean distance of the diversity set, the source matrix, source vector, diversity set, minimum distance and mean distance associated with a selected candidate diversity set are modified, and the Candidate Diversity Set process transitions to the Matrix Condition process. If the diversity set remains unchanged, the Candidate Diversity Sets process transitions to itself at the next sequential increment of the synchronous temporal index.

[0136] The Matrix Condition process (1506) evaluates the condition of the matrix {right arrow over (U)}_m^T{right arrow over (U)}_mby either direct or indirect means. An indirect means of evaluating the condition of the matrix is defined through a process of maximizing the minimum distance between any two members of a diversity set. The basis for the diversity set is selected such that the minimum distance, or diversity of the set, corresponds, positively or inversely, to the condition of the matrix {right arrow over (U)}_m^T{right arrow over (U)}_m. The application of the diversity set is an economic means to evaluate the condition, which would otherwise be computationally prohibitive and impractical in many environments. If the condition number is below a specified threshold, in direct evaluation, or a minimum distance is above a specified threshold, in indirect evaluation, the Matrix Condition process transitions to the Pseudo-Inverse Solution process. If the condition is insufficient to support a solution, the Matrix Condition process transitions to the Candidate Diversity Sets process at the next sequential increment of the synchronous temporal index.

[0137] The Pseudo-Inverse Solution process (1508) applies the Moore-Penrose pseudo-inverse method to extract a solution vector from the source matrix and the source vector, facilitating the direct definition of stator inductance, total leakage factor, and initial rotor time constant from the solution vector. The Pseudo-Inverse Solution process (1508) transitions to the terminus of control flow.

[0138] It is possible, and recommended, that the process of solving the induction motor electrical parameters not be terminated after the first available solution is extracted, but rather iteratively refined through continued application of the techniques described in the Maximum Diversity Method process. As new information is extracted through passive observation of motor operation, the resulting solution vector will become more robust and accurate, and can be used with greater confidence. If computational complexity is a significant concern, the synchronous temporal index can be accessed with a greater than unity stride, effectively downsampling the basis for the diversity set, and reducing the bandwidth required by a linear factor equal to the inverse of the stride. This is generally a reasonable compromise, as potential basis values extracted from motor electrical signals, including complex input power and complex impedance, cannot change instantaneously and are generally quasi-stationary for periods which significantly exceed the sampling frequency.

[0139] A practical temporal stride value can be selected such that the effective sampling frequency is reduced to no less than two times the rated fundamental frequency. For example, in a system with a sampling frequency of 5 kHz, and a rated fundamental frequency equal to 60 Hz, a stride of less than or equal to 40 is recommended.

[0140] The source matrix {right arrow over (U)}_mis composed of N_Drows formed from independent observations of complex fundamental voltage, complex fundamental current and rotor speed signals. The complex fundamental voltage can be the same as the positive sequence fundamental frequency voltage component (e.g., 704). The complex fundamental current can be the same as the positive sequence fundamental frequency current component (e.g., 714). Asynchronous temporal index _mis applied as a means to enumerate successive asynchronous estimates, or opportunistic updates. The row dimension, N_D, must be greater than or equal to 3 to support the application of the Moore-Penrose pseudo-inverse method, based on the dimension of the solution vector, {right arrow over (ξ)}_m. A practical dimension, N_D, typically defines 10 or more observations.

[0141] The rows of the source matrix are independent, though not necessarily sufficiently diverse to support definition of a solution vector. If the rows of the source matrix are representative of too narrow a range of operating conditions, the square and potentially invertible matrix {right arrow over (U)}_m^T{right arrow over (U)}_mcan be ill-conditioned, or nearly singular. The matrix {right arrow over (U)}_m^T{right arrow over (U)}_mmust be sufficiently well-conditioned to support a solution.

[0142] The source matrix {right arrow over (U)}_mand source vector {right arrow over (y)}_mare defined and allowed to evolve opportunistically, with each observation of a diversity basis determined to increase diversity of the rows of {right arrow over (U)}_m, and correspondingly decrease the condition number of the matrix {right arrow over (U)}_m^T{right arrow over (U)}_m. When the matrix {right arrow over (U)}_m^T{right arrow over (U)}_mis sufficiently well-conditioned, the solution vector is extracted. A single estimate of the solution vector can be sufficient, or the solution can be improved with subsequent iterations of the source matrix in an application-dependent manner.

Initial Synthesis

[0143] An initial matrix {right arrow over (U)}₀is formed by application of Equation 34 to N_Dsequential samples of complex fundamental voltage and complex fundamental current, in Equation 53.

[0144]

\begin{matrix} {\overset{->}{U}}_{0} =_{REAL} [\begin{matrix} - {pi}_{F, n - N_{D} + 1} & v_{F, n - N_{D} + 1} - R_{S} \cdot i_{F, n - N_{D} + 1} & p (v_{F, n - N_{D} + 1} - R_{S} \cdot i_{F, n - N_{D} + 1}) - j \cdot ω_{R, n - N_{D} + 1} \cdot (v_{F, n - N_{D} + 1} - R_{S} \cdot i_{F, n - N_{D} + 1}) \\ - {pi}_{F, n - N_{D} + 2} & v_{F, n - N_{D} + 2} - R_{S} \cdot i_{F, n - N_{D} + 2} & p (v_{F, n - N_{D} + 2} - R_{S} \cdot i_{F, n - N_{D} + 2}) - j \cdot ω_{R, n - N_{D} + 2} \cdot (v_{F, n - N_{D} + 2} - R_{S} \cdot i_{F, n - N_{D} + 2}) \\ ⋮ \\ - {pi}_{F, n} & v_{F, n} - R_{S} \cdot i_{F, n} & p (v_{F, n} - R_{S} \cdot i_{F, n}) - j \cdot ω_{R, n} \cdot (v_{F, n} - R_{S} \cdot i_{F, n}) \end{matrix}] ❘_{N_{D} \geq 3} & Equation 53 \end{matrix}

[0145] An initial source vector y₀is formed once by application of Equation 33 to N_Dsequential samples of complex fundamental current and rotor speed, in Equation 54.

[0146]

\begin{matrix} {\overset{->}{y}}_{0} =_{REAL} [\begin{matrix} p^{2} i_{F, n - N_{D} + 1} - j \cdot ω_{R, n - N_{D} + 1} \cdot {pi}_{F, n - N_{D} + 1} \\ p^{2} i_{F, n - N_{D} + 2} - j \cdot ω_{R, n - N_{D} + 2} \cdot {pi}_{F, n - N_{D} + 2} \\ ⋮ \\ p^{2} i_{F, n} - j \cdot ω_{R, n} \cdot {pi}_{F, n} \end{matrix}] ❘_{N_{D} \geq 3} & Equation 54 \end{matrix}

[0147] Source matrix and source vector definitions correspond to real projections, as defined in Equation 41b and 40b. Imaginary projections, or interlaced real and imaginary projections are valid alternatives, as defined in Equation 41c, 40c and Equation 41a and 40a, respectively.

[0148] Complex input power, ρ_F,n, is defined in Equation 55,

[0149]

\begin{matrix} ρ_{F, n} = \frac{3^{0.5} \cdot v_{F, n} \cdot i_{F, n}^{*}}{2 \cdot P_{F}} & Equation 55 \end{matrix}

where P_Fis a rated power factor from the motor nameplate data, and the superscript * denotes the complex conjugate of a complex number.

[0150] An initial diversity set, {right arrow over (D)}₀, is formed once with N_Dsequential samples of complex input power at sample indices corresponding to row synthesis in the initial matrix {right arrow over (U)}₀, in Equation 56:
{right arrow over (D)}₀={ρ_F,n−N_D₊₁ρ_F,n−N_D₊₂. . . ρ_F,n}|_N_D_≧3 Equation 56

[0151] The diversity of the source matrix {right arrow over (U)}₀is indirectly specified in the diversity set, employing a basis of the complex input power as a matter of convenience. Complex input power, complex input impedance, and the real or imaginary projections of these signals can be used as an alternative basis to specify diversity.

[0152] The distance, δ_m,j,k, is an Euclidean distance between any two members of the diversity set, at independent indices j and k, in Equation 57:

[0153]

\begin{matrix} δ_{m, j, k} = \langle D_{m, j} - D_{m, k} \rangle ❘_{\underset{\underset{N \geq 3}{j \neq k}}{j, k} : [0, N_{D} - 1]} & Equation 57 \end{matrix}

[0154] A minimum distance, d_m, is the minimum Euclidian distance between all unique combination of members of the diversity set, {right arrow over (D)}_m, in Equation 58:

[0155]

\begin{matrix} d_{m} =_{MIN} (δ_{m, j, k}) ❘_{\underset{\underset{N \geq 3}{j \neq k}}{j, k} : [0, N_{D} - 1]} & Equation 58 \end{matrix}

[0156] The mean distance, d′_m, is the mean Euclidean distance between each member pair of the diversity set {right arrow over (D)}_m, in Equation 59:

[0157]

\begin{matrix} {\overline{d}}_{m} = \frac{1}{\sum_{c = 1}^{N_{D} - 1} c} \cdot \sum_{j = 0}^{N_{D} - 2} \sum_{k = j + 1}^{N_{D} - 1} δ_{m, j, k} ❘_{N_{D} \geq 3} & Equation 59 \end{matrix}

[0158] According to some aspects, an initial minimum distance, d₀, and mean distance, d′₀, are calculated once for the initial diversity set, {right arrow over (D)}₀, and retained for comparison to values extracted from candidate diversity sets.

Candidate Diversity Sets

[0159] At each successive index, _n, a sample of complex input power, or alternative diversity basis, can be evaluated to determine if it should replace an existing member of the diversity set. Alternatively, a temporal index stride greater than unity can be selected to conserve bandwidth, and it is generally not possible to have instantaneous changes in the basis for the diversity set. Candidate diversity sets, {right arrow over (D)}_m,c, are synthesized by temporarily replacing an existing member at index c with the new sample, forming ND unique candidate diversity sets, in Equation 60:

[0160]

\begin{matrix} {\overset{->}{D}}_{m, c} = {{\overset{->}{D}}_{m} : D_{m, j} = ρ_{F, n}} ❘_{\underset{\underset{N_{D} \geq 3}{c = j}}{c, j} : [0, N_{D} - 1]} & Equation 60 \end{matrix}

[0161] The Candidate distance, δ_m,c,j,k, is computed for any two members of each candidate diversity set, {right arrow over (D)}_m,c, at independent indices _jand _k, in Equation 61:

[0162]

\begin{matrix} δ_{m, c, j, k} = \langle D_{m, c, j} - D_{m, c, k} \rangle ❘_{\underset{\underset{N_{D} \geq 3}{j \neq k}}{c, j, k} : [0, N_{D} - 1]} & Equation 61 \end{matrix}

[0163] The Candidate minimum distance, d_m,c, is the distance between the two most proximate members of a candidate diversity set, {right arrow over (D)}_m,c, in Equation 62:

[0164]

\begin{matrix} d_{m, c} =_{MIN} (δ_{m, c, j, k}) ❘_{\underset{\underset{N_{D} \geq 3}{j \neq k}}{c, j, k} : [0, N_{D} - 1]} & Equation 62 \end{matrix}

[0165] The maximum candidate minimum distance, d_m,z, is identified at set index _zin the N_Dcandidate diversity sets, {right arrow over (D)}_m,c, in Equation 63:

[0166]

\begin{matrix} d_{m, z} =_{MAX} (d_{m, c}) |_{\underset{N_{D} \geq 3}{c, z : [0, N_{D} - 1]}} & Equation 63 \end{matrix}

[0167] In some aspects, if the maximum candidate minimum distance, d_m,z, is greater than or equal to the existing minimum distance, d_m, the mean candidate distance, d′_m,z, is evaluated corresponding to the candidate diversity set, {right arrow over (D)}_m,z, with set index _z, in Equation 64:

[0168]

\begin{matrix} {\overline{d}}_{m, z} = (\frac{1}{\sum_{c = 1}^{N_{D} - 1} c}) \cdot \sum_{j = 0}^{N_{D} - 2} \sum_{k = j + 1}^{N_{D} - 1} δ_{m, c, j, k} ❘_{\underset{N_{D} \geq 3}{c, z} : [0, N_{D} - 1]} & Equation 64 \end{matrix}

[0169] In some aspects, if the maximum candidate minimum distance, d_m,z, is greater than the existing minimum distance, d_m, or if the maximum candidate minimum distance is equal to the existing minimum distance and the candidate mean distance, d′_m,z, is greater than the existing mean distance, d′_m, it is implicit that the condition of the matrix {right arrow over (U)}_m^T{right arrow over (U)}_mwould be improved by updating the matrix {right arrow over (U)}_m+1, by synthesizing a new vector to replace the existing row at index _z, in Equation 65.

[0170]

\begin{matrix} {\overset{->}{U}}_{m + 1} = {\begin{matrix} {\overset{->}{U}}_{m} : {\overset{->}{U}}_{m, z} = \\ [\begin{matrix} - {pi}_{F, n} v_{F, n} - R_{S} \cdot \\ i_{F, n} p (v_{F, n} - R_{S} \cdot i_{F, n}) - \\ j \cdot ω_{R, n} \cdot (v_{F, n} - R_{S} \cdot i_{F, n}) \end{matrix}] ❘_{\underset{d_{m, z} = d_{m}, {\overline{d}}_{m, z} > {\overline{d}}_{m}}{d_{m, z} > d_{m}}} \end{matrix}} & Equation 65 \end{matrix}

[0171] Source vector {right arrow over (y)}_m+1is equal to the previous vector, {right arrow over (y)}_m, modified by synthesizing a new value to replace the existing row at index _z, in Equation 66:

[0172]

\begin{matrix} {\overset{->}{y}}_{m + 1} = {\begin{matrix} {\overset{->}{y}}_{m} : y_{m, z} = p^{2} i_{F, n} - \\ j \cdot ω_{R, n} \cdot {pi}_{F, n} \end{matrix} ❘_{\underset{d_{m, z} = d_{m}, {\overline{d}}_{m, z} > {\overline{d}}_{m}}{d_{m, z} > d_{m}}}} & Equation 66 \end{matrix}

[0173] Diversity set, {right arrow over (D)}_m+1, is equal to the candidate diversity set, {right arrow over (D)}_m,z, corresponding to the basis sample substitution at index _z, in Equation 67:

[0174]

\begin{matrix} {\overset{->}{D}}_{m + 1} = {\overset{->}{D}}_{m, z} ❘_{\underset{d_{m, z} = d_{m}, {\overline{d}}_{m, z} > {\overline{d}}_{m}}{d_{m, z} > d_{m}}} & Equation 67 \end{matrix}

[0175] The minimum distance, d_m+1, is equal to the candidate diversity, d_m,z, at index _z, in Equation 68:

[0176]

\begin{matrix} d_{m + 1} = d_{m, z} ❘_{\underset{d_{m, z} = d_{m}, {\overline{d}}_{m, z} > {\overline{d}}_{m}}{d_{m, z} > d_{m}}} & Equation 68 \end{matrix}

[0177] The mean distance, d′_m+1, is equal to the candidate mean distance, d′_m,z, at index _z, in Equation 69:

[0178]

\begin{matrix} {\overline{d}}_{m + 1} = {\overline{d}}_{m, z} ❘_{\underset{d_{m, z} = d_{m}, {\overline{d}}_{m, z} > {\overline{d}}_{m}}{d_{m, z} > d_{m}}} & Equation 69 \end{matrix}

[0179] In some aspects, if the candidate diversity sets, {right arrow over (D)}_m,c, do not produce a candidate, {right arrow over (D)}_m,z, which is more diverse than the existing diversity set, {right arrow over (D)}_m, then the process of synthesis and evaluation of candidate diversity sets can be repeated at each subsequent sample index until the opportunity to improve the diversity of the set is observed.

Matrix Condition

[0180] For each new definition of the diversity set, the matrix {right arrow over (U)}_m^T{right arrow over (U)}_mcan be evaluated to determine if the matrix condition is sufficient to support the definition of the solution vector. The condition of the matrix {right arrow over (U)}_m^T{right arrow over (U)}_mdetermines the sensitivity of the solution vector to perturbations in the matrix. Sensitivity analysis can be performed through direct or indirect means, though direct analysis is often computationally impractical.

[0181] Condition number, c_m, a direct measure of the sensitivity of the solution vector {right arrow over (ξ)}_mto perturbations in the matrix {right arrow over (U)}_m^T{right arrow over (U)}_m, is defined as the magnitude of the ratio of the maximum and minimum eigenvalues of the matrix, in Equation 70.

[0182]

\begin{matrix} c_{m} = \langle \frac{λ_{m, x}}{λ_{m, y}} \rangle ❘_{λ_{x, m} \geq λ_{m, j} \geq λ_{m, y}} & Equation 70 \end{matrix}

[0183] The Eigenvalues, {right arrow over (λ)}_m, of the matrix {right arrow over (U)}_m^T{right arrow over (U)}_mcan be extracted by solving a characteristic polynomial associated with the matrix. Eigenvalues are the roots of the characteristic polynomial, in Equation 71.
_DET({right arrow over (U)}_m^T·{right arrow over (U)}_m−{right arrow over (λ)}_m·{right arrow over (I)})=0 Equation 71

[0184] According to some aspects, the matrix {right arrow over (U)}_m^T{right arrow over (U)}_mis relatively small, with dimension [3, 3], thus, eigenvalues can be extracted with reasonable efficiency through a direct solution to a third order characteristic polynomial, expressed in terms of matrix determinant, _DET, and trace, _TR, operations, in Equation 72.

[0185]

\begin{matrix} - {\overset{->}{λ}}_{m}^{3} + {\overset{->}{λ}}_{m}^{2} \cdot_{TR} ({\overset{->}{U}}_{m}^{T} \cdot {\overset{->}{U}}_{m}) + \frac{{\overset{->}{λ}}_{m}}{2} \cdot [_{TR} ({({\overset{->}{U}}_{m}^{T} \cdot {\overset{->}{U}}_{m})}^{2}) - {}_{TR}{({\overset{->}{U}}_{m}^{T} \cdot {\overset{->}{U}}_{m})}^{2}] +_{DET} ({\overline{U}}_{m}^{T} \cdot {\overset{->}{U}}_{m}) = 0 & Equation 72 \end{matrix}

[0186] Condition number, cm, of the matrix {right arrow over (U)}_m^T{right arrow over (U)}_mapproaches unity when the matrix is well-conditioned, and has eigenvalues which are nearly equal, and approaches infinity when the matrix is nearly singular, and has asymmetric eigenvalues.

[0187] A static maximum condition number, c_x, can be defined to determine if the matrix {right arrow over (U)}_m^T{right arrow over (U)}_mis sufficiently well-conditioned to extract a solution vector. The maximum condition number, c_L, is defined through sensitivity analysis of the elements of the solution vector, based on application-specific requirements.

[0188] Practical variants of the algorithm include evaluation of the condition number, c_m, after several successive iterations to the matrix {right arrow over (U)}_m, or only if the minimum distance d_m, exceeds a threshold, d_y. According to some aspects, alternative eigenvalue estimation methods can be applied to reduce computational complexity. Numerical stability analysis can also be addressed.

[0189] An efficient, practical, indirect method of estimating the condition of the matrix {right arrow over (U)}_m^T{right arrow over (U)}_mn can omit synthesis of condition number, cm, entirely. The basis for the diversity set, {right arrow over (D)}_m, is selected based on the assumption that an increase in the minimum distance, d_m, of the set yields a decrease in the condition number of the related matrix {right arrow over (U)}_m^T{right arrow over (U)}_m. While the distance and the condition number correlate, it is possible that their relationship is not linear.

[0190] In some aspects, in lieu of a condition number synthesis, a static minimum distance, d_y, can be defined to indirectly determine if the matrix {right arrow over (U)}_m^T{right arrow over (U)}_mis sufficiently well-conditioned to support a solution. In these aspects, a minimum distance threshold, d_y, is defined experimentally to result in a solution that is accurate enough to support the needs of a specific application.

Pseudo-Inverse Solution

[0191] According to some aspects, when the condition of the matrix {right arrow over (U)}_m^T{right arrow over (U)}_mis found to be sufficient in the Matrix Condition process (1506), the solution vector, {right arrow over (ξ)}_mis evaluated through application of the Moore-Penrose pseudo-inverse technique, in Equation 73.
{right arrow over (ξ)}_m=({right arrow over (U)}_m^T·{right arrow over (U)}_m)⁻¹·({right arrow over (U)}_m^T·{right arrow over (y)}_m) Equation 73

[0192] The stator inductance, L_S,m, the total leakage factor, σm, and the initial rotor time constant, τ_R,0,m, are directly evaluated from the solution vector according to Equations 43-45, respectively.

[0193] FIG. 17 illustrates an architecture (1700) used to track an induction motor rotor temperature in accordance with some aspects of the present disclosure. Once the induction motor electrical parameters, R_S, L_S, σ and τ_R,0, are accurately and reliably estimated (1204), as described above, the parameters are fed into the rotor time constant estimation block (122). The quantity, τ_R,0, is used as the initial rotor time constant value in a model-reference adaptive system inside the rotor time constant estimation block (122). Furthermore, this τ_R,0can also be used as an approximate indicator and be associated with the baseline rotor time constant and temperature pair (τ_R,n0, θ_R,n0), which is used to calculate the rotor temperature, θ_R,n, (124) from the rotor time constant, τ_R,n. The quantity τ_R,nbeing the output of the rotor time constant estimation block (122).

[0194] While particular aspects, embodiments, and applications of the present invention have been illustrated and described, it is to be understood that the invention is not limited to the precise construction and compositions disclosed herein and that various modifications, changes, and variations be apparent from the foregoing descriptions without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

What is claimed is:

1. A method for estimating a temperature of a rotor of an induction motor through analysis of voltage and current signals associated with the motor, comprising:

measuring the voltage signal associated with the induction motor;

determining a representation of the measured voltage signal;

measuring the current signal associated with the induction motor;

determining a representation of the measured current signal;

determining a speed of the rotor;

determining an estimate of a rotor time constant according to a function that includes a fundamental frequency voltage component extracted from the representation of the measured voltage signal, a fundamental frequency current component extracted from the representation of the measured current signal, and the rotor speed;

determining an estimate of the rotor temperature based on an inverse relationship between a known rotor temperature and the estimated rotor time constant; and

storing the estimated rotor temperature.

2. The method of claim 1, wherein the representation of the measured voltage signal is determined by calculating a complex voltage space vector from the voltage measurements of the voltage signal, and the representation of the measured current signal is determined by calculating a complex current space vector from the current measurements of the current signal.

3. The method of claim 2, wherein the measured voltage signal is a three-phase voltage signal and the voltage measurements are measured from the voltage differences between at least one of the three phases of the three-phase voltage signal and either every other phase of the three phases or a voltage reference point.

4. The method of claim 2, wherein the measured current signal is a three-phase current signal and the current measurements are measured from at least two of the three phases of the three-phase current signal.

5. The method of claim 1, wherein the induction motor is a line-connected motor.

6. The method of claim 1, wherein the determining the speed of the rotor is carried out according to a function that includes a fundamental frequency extracted from the representation of the measured voltage signal, a rotor slot harmonic frequency extracted from the representation of the measured current signal, a stator winding distribution harmonic order, a rotor magnetomotive force distribution harmonic order, a number of rotor slots of the motor, and a number of pole-pairs of the motor.

7. The method of claim 6, wherein the representation of the measured current signal is determined by calculating a complex current space vector from the current measurements of the current signal, and wherein the rotor slot harmonic frequency corresponds to a dominant rotor slot harmonic frequency extracted from the complex current space vector.

8. The method of claim 6, wherein the function for determining the rotor speed is represented by

ω_{R} = \frac{2 π P \cdot (\pm f_{sh} - n_{w} \cdot f_{0})}{(kR)},

where ω_Rrepresents the rotor speed, P represents the number of pole-pairs, f_shrepresents the rotor slot harmonic frequency; n_wrepresents the stator winding distribution harmonic order, f₀represents the fundamental frequency; k represents the rotor magnetomotive force distribution harmonic order; and R represents the number of rotor slots.

9. The method of claim 7, wherein the representation of the measured voltage signal is determined by calculating a complex voltage space vector from the voltage measurements of the voltage signal, and wherein the fundamental frequency is extracted from the complex voltage space vector by:

generating a complex exponential signal in a voltage controlled oscillator evaluated at a rated fundamental frequency associated with the motor;

multiplying the complex voltage space vector with a complex conjugate of the complex exponential signal to produce a mixed complex voltage signal;

filtering the mixed complex voltage signal to produce a first complex baseband signal; and

extracting a time derivative of the phase of the first complex baseband signal to produce the fundamental frequency as a function of the extracted time derivative.

10. The method of claim 9, wherein the filtering is carried out by a low-pass filter that has an associated phase response over a frequency band that at least approximates a linear phase.

11. The method of claim 10, wherein the low pass filter includes a Finite Impulse Response (FIR) filter or in an Infinite Impulse Response (IIR) filter.

12. The method of claim 6, wherein the representation of the measured current signal is determined by calculating a complex current space vector from the current measurements of the current signal, and wherein the rotor slot harmonic frequency is extracted from complex current space vector by:

determining a nominal rotor slot harmonic frequency;

synthesizing, in a voltage controlled oscillator, a complex exponential signal evaluated at the nominal rotor slot harmonic frequency;

multiplying the complex current space vector with a complex conjugate of the complex exponential signal to produce a mixed complex current signal;

filtering the mixed complex current signal to produce a second complex baseband signal; and

extracting a time derivative of the phase of the second complex baseband signal to produce the rotor slot harmonic frequency as a function of the extracted time derivative.

13. The method of claim 7, wherein the dominant rotor slot harmonic frequency has the largest amplitude among a plurality of observable rotor slot harmonic components in a stator current harmonic spectrum.

14. The method of claim 13, wherein the dominant rotor slot harmonic frequency is extracted according to a relationship between dominant rotor slot harmonic frequency and rotor design parameters.

15. The method of claim 1, wherein the determining the rotor speed is carried out by converting an output of a speed sensing device coupled to a shaft of the motor to the rotor speed.

16. The method of claim 1, wherein the rotor time constant is determined by substantially aligning a first flux-related vector calculated from a rotor circuit equation with a second flux-related vector calculated from a stator circuit equation.

17. The method of claim 16, wherein the stator circuit equation is expressed as a time derivative of a second complex flux space vector and includes the fundamental frequency current component, the fundamental frequency voltage component, a stator resistance, a total leakage factor, and a stator inductance.

18. The method of claim 16, wherein the rotor circuit equation is expressed as a time derivative of a first complex flux space vector and includes a ratio between (a) a rotor inductance and (b) a rotor resistance.

19. The method of claim 18, further comprising:

filtering an output of the rotor circuit equation to produce the first flux-related vector; and

filtering an output of the stator circuit equation to produce the second flux-related vector.

20. The method of claim 19, wherein the rotor time constant is further determined by adapting the rotor time constant according to a function that includes a cross product between the first flux-related vector and the second flux-related vector such that the magnitude of the cross product is minimized.

21. The method of claim 20, wherein the rotor time constant is further determined by correlating the magnitude of the cross product to the rotor time constant based on a tuned proportional-integral block.

22. The method of claim 1, wherein the inverse relationship is represented by a function that includes the known rotor temperature, at least two rotor time constants determined at different times, and a temperature coefficient based on an inferred temperature for zero resistance.

23. The method of claim 22, wherein the inverse relationship is represented by:

θ_R,n=(θ_R,n0+k)·τ_R,n0/τ_R,n−k, where θ_R,nrepresents the rotor temperature at time n, θ_R,n0represents the known rotor temperature at time n₀, τ_R,nrepresents the rotor time constant at time n, τ_R,n0represents a baseline rotor time constant at time n₀, and k represents the temperature coefficient.

24. The method of claim 1, wherein the inverse relationship is derived from a linear relationship based on at least two rotor resistances at different times, a temperature coefficient based on an inferred temperature for zero resistance, and at least two rotor temperatures at the different times.

25. The method of claim 1, wherein the fundamental frequency voltage component is extracted from the representation of the measured voltage signal by:

generating a complex exponential signal in a voltage controlled oscillator evaluated at a rated fundamental frequency associated with the motor;

calculating a mixed complex voltage signal according to a function that includes the product of the complex voltage space vector and the complex conjugate of the complex exponential signal;

filtering the mixed complex voltage signal to produce a complex baseband signal; and

multiplying the complex baseband signal by the complex exponential signal to produce the fundamental frequency voltage component.

26. The method of claim 25, wherein the filtering is carried out via a finite impulse response low-pass filter, or an infinite impulse response low-pass filter, or a generalized linear-phase low-pass filter.

27. The method of claim 1, wherein the fundamental frequency current component is extracted from the representation of the measured current signal by:

generating a complex exponential signal in a voltage controlled oscillator evaluated at a rated fundamental frequency associated with the motor;

calculating a mixed complex current signal according to a function that includes the product of the complex current space vector and the complex conjugate of the complex exponential signal;

filtering the mixed complex current signal to produce a complex baseband signal; and

multiplying the complex baseband signal by the complex exponential signal to produce the fundamental frequency current component.

Patent Citations (27)

Patent	Date	Inventor	Cited By
US4114077(A)	1978-09-01	Oates et al.	Applicant
US4441065(A)	1984-04-01	Bayer et al.	Applicant
US4884023(A)	1989-11-01	Schmidt et al.	Applicant
US4914386(A)	1990-04-01	Zocholl	Applicant
US5232285(A)	1993-08-01	Mannik	Applicant
US5257863(A)	1993-11-01	Chu et al.	Applicant
US5436784(A)	1995-07-01	Schweitzer, III et al.	Applicant
US5477162(A)	1995-12-01	Heikkilä	Applicant
US5559419(A)	1996-09-01	Jansen et al.	Applicant
US5565752(A)	1996-10-01	Jansen et al.	Applicant
US5585709(A)	1996-12-01	Jansen et al.	Applicant
US5689169(A)	1997-11-01	Kerkman et al.	Applicant
US5729113(A)	1998-03-01	Jansen et al.	Applicant
US5861728(A)	1999-01-01	Tazawa et al.	Applicant
US5965995(A)	1999-10-01	Seibel et al.	Applicant
US6008618(A)	1999-12-01	Bose et al.	Examiner
US6042265(A)	2000-03-01	Kliman et al.	Applicant
US6281659(B1)	2001-08-01	Giuseppe	Applicant
US6388420(B1)	2002-05-01	Jansen et al.	Applicant
US6452360(B1)	2002-09-01	Colby	Applicant
US6636012(B2)	2003-10-01	Royak et al.	Applicant
US6774664(B2)	2004-08-01	Godbersen	Applicant
US6862538(B2)	2005-03-01	El-Ibiary	Applicant
US7135830(B2)	2006-11-01	El-Ibiary	Applicant
US7184902(B2)	2007-02-01	El-Ibiary	Applicant
US7389409(B2)	2008-06-01	Swanson et al.	Applicant
US2009/0284211(A1)	2009-11-01	Gao et al.	Examiner

Non-Patent Literature (19)

C. Schauder, “Adaptive speed identification for vector control of induction motors without rotational transducers,” IEEE Trans. Ind. Appl., vol. 28, No. 5, pp. 1054-1061, Sep./Oct. 1992.Applicant
A. Ferrah, K. J. Bradley, and G. M. Asher, “An FFT-based novel approach to noninvasive speed measurement in induction motor drives,” IEEE Trans. Instrum. Meas., vol. 41, No. 6, pp. 797-802, Dec. 1992.Applicant
P. L. Jansen and R. D. Lorenz, “Transducerless position and velocity estimation in induction and salient AC machines,” IEEE Trans. Ind. Appl., vol. 31, No. 2, pp. 240-247, Mar./Apr. 1995.Applicant
K. D. Hurst and T. G. Habetler, “Sensorless speed measurement using current harmonic spectral estimation in induction machine drives,” IEEE Trans. Power Electron., vol. 11, No. 1, pp. 66-73, Jan. 1996.Applicant
R. Blasco-Gimenez, G. M. Asher, M. Sumner, and K. J. Bradley, “Performance of FFT-rotor slot harmonic speed detector for sensorless induction motor drives,” IEE Proc. B, Electr. Power Appl., vol. 143, No. 3, pp. 258-268, May 1996.Applicant
R. Blasco-Gimenez, K. D. Hurst, and T. G. Habetler, “Comments on ‘Sensorless speed measurement using current harmonic spectral estimation in induction machine drives’ [and reply],” IEEE Trans. Power Electron., vol. 12, No. 5, pp. 938-940, Sep. 1997.Applicant
K. D. Hurst and T. G. Habetler, “A comparison of spectrum estimation techniques for sensorless speed detection in induction machines,” IEEE Trans. Ind. Appl., vol. 33, No. 4, pp. 1-8, Jul./Aug. 1997.Applicant
K. D. Hurst, T. G. Habetler, G. Griva, F. Profumo, and P. L. Jansen, “A self-tuning closed-loop flux observer for sensorless torque control of standard induction machines,” IEEE Trans. Power Electron., vol. 12, No. 5, pp. 807-815, Sep. 1997.Applicant
J. Holtz, “Sensorless position control of induction motors—An emerging technology,” IEEE Trans. Ind. Electron., vol. 45, No. 6, pp. 840-851, Dec. 1998.Applicant
A. Bellini, G. Franceschini, E. Lorenzani, C. Tassoni, and F. Filippetti, “Sensorless speed detection in induction machines,” in Proc. Int. Conf. Electrical Machines (ICEM), Bruges, Belgium, Aug. 25-28, 2002.Applicant
A. M. Salamah, S. J. Finney, and B. W. Williams, “Three-phase phase-lock loop for distorted utilities,” IET Electr. Power Appl., vol. 1, No. 6, pp. 937-945, Nov. 2007.Applicant
Z. Gao, T. G. Habetler, R. G. Harley, and R. S. Colby, “A sensorless rotor temperature estimator for induction machines based on a current harmonic spectral estimation scheme,” IEEE Trans. Ind. Electron., vol. 55, No. 1, pp. 407-416, Jan. 2008.Applicant
M. Blodt, P. Granjon, B. Raison, and G. Rostaing, “Models for bearing damage detection in induction motors using stator current monitoring,” IEEE Trans. Ind. Electron., vol. 55, No. 4, pp. 1813-1822, Apr. 2008.Applicant
Guan-Chyun Hsieh and James C. Hung, “Phase-Locked Loop Techniques—A Survey”, IEE Transactions on Industrial Electronics, vol. 43, No. 6, Dec. 1996.Applicant
A. V. Oppenheim, R. W. Schafer, and J. R. Buck, Discrete-Time Signal Processing, 2nd ed. Upper Saddle River, NJ: Prentice-Hall, 1999, pp. 482-485.Applicant
J. Stephan, M. Bodson, and J. Chiasson, “Real-time estimation of the parameters and fluxes of induction motors”, IEEE Trans. Ind. Appl., vol. 30, No. 3, May/Jun. 1994.Applicant
L. A. S. Ribeiro, C. R. Jacobina, A. M. N. Lima, and A. C. Oliveira, “Parameter sensitivity of MRAC models employed in IFO-controlled AC motor drive,” IEEE Trans. Ind. Electron., vol. 44, No. 4, pp. 536-545, Aug. 1997.Applicant
M. Cirrincione, M. Pucci, G. Cirrincione, and G.-A. Capolino, “A new experimental application of least-squares techniques for the estimation of the induction motor parameters,” IEEE Trans. Ind. Appl., vol. 39, No. 5, pp. 1247-1256, Sep./Oct. 2003.Applicant
IEEE Standard Test Procedure for Polyphase Induction Motors and Generators, IEEE Standard 112-2004, Nov. 2004.Applicant

Abstract