Electronic overload relay for mains-fed induction motors

Abstract

An induction motor having a rotor and a stator is protected during running overloads by connecting the motor to an overload protection relay that can be tripped to interrupt power to the motor in the event of an overload, tracking the stator winding temperature of the motor during running overloads with a hybrid thermal model by online adjustment in the hybrid thermal model, and tripping the overload protection relay in response to a predetermined running overload condition represented by the tracked stator winding temperature. In one embodiment of the invention, the stator winding temperature is tracked by use of an online hybrid thermal model that uses the resistance of the rotor as an indicator of rotor temperature and thus of the thermal operating conditions of the motor. The hybrid thermal model incorporates rotor losses and heat transfer between the rotor and the stator, and approximates the thermal characteristics of the rotor and stator.

Description

FIELD OF THE INVENTION

This invention is directed generally to electronic overload relays for mains-fed induction motors. More particularly, this invention pertains to a system for controlling electronic overload relays for mains-fed induction motors.

BACKGROUND OF THE INVENTION

The permissible temperature limit for the winding of a motor is primarily determined by the motor's insulation. Applicable standards (UL, CSA, IEC, and NEMA) distinguish different classes of insulation and corresponding temperature limits. Typical allowable temperatures inside the stator windings are given in Information Guide for General Purpose Industrial AC Small and Medium Squirrel-Cage Induction Motor Standards, NEMA Standard MG1-2003, August 2003.

If the motor is working continuously in an over-temperature condition, the aging process is accelerated. This is a chemical process that involves the deterioration of the insulation material. It is often assumed that a winding temperature that is constantly 10° C. higher than the temperature limit reduces the motor life by half. This life law shows that particular attention must be paid to adhering to the permitted operating temperature for long periods of time.

Various standards have been established to provide general guidelines in estimating stator winding temperature for motor overload protection. See, e.g., IEEE Guide for AC Motor Protection, IEEE Standard C37.96-2000, March, 2000; Guide for the Presentation of Thermal Limit Curves for Squirrel Cage Induction Machines, IEEE Standard 620-1996, June, 1996; and IEEE Standard Inverse-time Characteristic Equations for Overcurrent Relays, IEEE Standard C37.112-1996, September, 1996.

Many manufacturers provide motor protective relays based on a thermal model with a single thermal time constant, derived from the temperature rise in a uniform object.

SUMMARY OF THE INVENTION

In accordance with one embodiment of the present invention, an induction motor having a rotor and a stator is protected during running overloads by connecting the motor to an overload protection relay that can be tripped to interrupt power to the motor in the event of an overload, tracking the stator winding temperature of the motor during running overloads with a hybrid thermal model by online adjustment in the hybrid thermal model, and tripping the overload protection relay in response to a predetermined running overload condition represented by the tracked stator winding temperature. As used herein, the term “online” means that the measurements, calculations or other acts are made while the motor is in service, connected to a load and running. A running overload is typically a condition in which the stator current is in the range of from about 100% to about 200% of the rated current of the motor, with motor continuously running.

In one embodiment of the invention, the stator winding temperature is tracked by use of an online hybrid thermal model that uses the resistance of the rotor as an indicator of rotor temperature and thus of the thermal operating conditions of the motor. The hybrid thermal model incorporates rotor losses and heat transfer between the rotor and the stator, and approximates the thermal characteristics of the rotor and stator.

One embodiment of the hybrid thermal model includes adjustable thermal parameters that permit the hybrid thermal model to be tuned when it does not match the actual thermal operating condition of the motor, to adapt the hybrid thermal model to the cooling capability of the motor. The match between the hybrid thermal model and the actual thermal operating condition of the motor can be measured by a function of the difference between a rotor temperature estimated using the online hybrid thermal model and a rotor temperature estimated from the estimated rotor resistance, and the tuning is effected when the error exceeds a predetermined threshold.

Thus, the invention can provide a system for estimating stator winding temperature by an adaptive architecture, utilizing the rotor resistance as an indication of rotor temperature, and consequently the motor thermal characteristics. The hybrid thermal model is capable of reflecting the real thermal characteristic of the motor under protection. With the learned parameters for the hybrid thermal model from the rotor resistance estimation, it is capable of protecting an individual motor in an improved manner, reducing unnecessary down time due to spurious tripping.

One embodiment of the invention provides an improved on-line system for estimating the rotor resistance of an induction motor by estimating motor inductances from the stator resistance and samples of the motor terminal voltage and current; and estimating the rotor resistance from samples of the motor terminal voltage and current, the angular speed of the rotor, and the estimated inductances. The voltage and current samples are used to compute the positive sequence, fundamental components of voltage and current. A preferred algorithm computes the ratio of rotor resistance to slip, in terms of the positive sequence voltage and current components and the motor inductances, and multiplies the ratio by the slip to obtain rotor resistance. The rotor temperature is estimated from the estimated rotor resistance.

One embodiment of the invention provides an improved method of estimating the motor inductances from the stator resistance and said samples of the motor terminal voltage and current. The voltage and current samples are taken at at least two different load points, and used to compute the positive sequence, fundamental components of voltage and current. The motor inductances that can be estimated include the stator leakage inductance, the rotor leakage inductance and the mutual inductance of the induction motor, or equivalently, the stator self inductance and the stator transient inductance.

BRIEF DESCRIPTION OF THE DRAWINGS

The foregoing and other advantages of the invention will become apparent upon reading the following detailed description and upon reference to the drawings.

FIGS. 1 a and 1b are block diagrams of stator temperature tracking systems for controlling an overload protection relay for an induction motor, in accordance with two alternative embodiments of the invention,

FIGS. 2 a and 2b are equivalent circuits of a d-q axis model of a 3-phase induction motor,

FIGS. 3 a and 3b are flowcharts of online inductance estimation algorithms suitable for use in the systems of FIGS. 1a and 1b, respectively,

FIG. 4 is a steady-state equivalent circuit of an induction motor,

FIG. 5 is a phasor diagram for the equivalent circuit of FIG. 4,

FIG. 6 is a graph depicting the relationship among three parameters used in one implementation of the system illustrated in FIG. 1,

FIG. 7 is a flow chart of of an online parameter tuning algorithm for a hybrid thermal model,

FIG. 8 is a full-order hybrid thermal model of an induction motor,

FIG. 9 is a decoupled hybrid thermal model of an induction motor under a locked rotor condition,

FIG. 10 is a reduced-order hybrid thermal model of an induction machine,

FIG. 11 is a diagram of the overall architecture of an adaptive Kalman filter with noise identification and input estimation, for use in one implementation of the system of FIG. 1,

FIGS. 12 a and 12b are positive and negative sequence equivalent circuits for an induction motor,

FIGS. 13 a and 13b are graphs showing the relationship between rotor resistance and frequency/slip, and

FIG. 14 is a diagrammatic illustration of the torques produced by positive and negative sequence currents.

While the invention is susceptible to various modifications and alternative forms, specific embodiments have been shown by way of example in the drawings and will be described in detail herein. It should be understood, however, that the invention is not intended to be limited to the particular forms disclosed. Rather, the invention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the invention as defined by the appended claims.

DETAILED DESCRIPTION OF THE ILLUSTRATED EMBODIMENTS

Turning now to the drawings, FIG. 1 illustrates the overall architecture of a system for estimating the stator winding temperature in an induction motor. This estimated temperature can then be used to control the tripping of an electronic relay that protects the motor from overload conditions. The system includes an inductance estimator 10, a rotor resistance estimator 12, a rotor temperature estimator 14, and a stator temperature estimator 16. Online signals supplied to the system include signals i_s and v_s representing samples of the motor terminal current and voltage, respectively, signal R_S representing the resistance of the stator in the motor, and a signal representing the NEMA design A, B, C or D of the motor. The output of the system is a signal θ_s^HTM representing the estimated stator temperature.

The four estimators 10-16 are preferably implemented by a single microprocessor programmed to execute algorithms that calculate the desired estimates based on the input signals and/or the results produced by the estimators. Each of these algorithms will be described in detail below.

The inductance estimator 10 estimates the stator leakage inductance L_ls, the rotor leakage inductance L_lr and the mutual inductance I_m of the motor from the samples v_s and is of the motor terminal voltage and current, the stator resistance R_s, and the signal representing the NEMA design A, B, C or D. These inductance values are temperature-independent. The inductance estimation algorithm is preferably carried out within one minute from the motor start so that the stator temperature, and hence resistance, are relatively constant.

The online inductance estimation is based on the d-q axis dynamic model of a 3-phase induction motor. FIGS. 2a and 2b are equivalent circuits of the d-q axis model of a 3-phase induction motor where R_s is the stator resistance; R_r is the rotor resistance; L_ls is the stator leakage inductance; L_lr is the rotor leakage inductance; L_m is the mutual inductance, and ω is the angular frequency of the reference frame.

Applying the synchronously rotating reference frame to the equivalent circuits shown in FIGS. 2a and 2b under steady state yields: $\begin{array}{cc}\begin{array}{c}\\ \begin{array}{c}\left[\begin{array}{c}{V}_{\mathrm{ds}}\\ V_{\mathrm{qs}}\\ V_{\mathrm{dr}}\\ V_{\mathrm{qr}}\end{array}\right]=\left[\begin{array}{c}{V}_{\mathrm{ds}}\\ V_{\mathrm{qs}}\\ 0\\ 0\end{array}\right]\\ =\left[\begin{array}{cccc}{R}_s& 0& 0& 0\\ 0& R_s& 0& 0\\ 0& 0& R_r& 0\\ 0& 0& 0& R_r\end{array}\right]\left[\begin{array}{c}{I}_{\mathrm{ds}}\\ I_{\mathrm{qs}}\\ I_{\mathrm{dr}}\\ I_{\mathrm{qr}}\end{array}\right]+\\ \left[\begin{array}{cccc}0& -{\omega }_e& 0& 0\\ {\omega }_e& 0& 0& 0\\ 0& 0& 0& -s{\omega }_e\\ 0& 0& s{\omega }_e& 0\end{array}\right]\left[\begin{array}{c}{\lambda }_{\mathrm{ds}}\\ {\lambda }_{\mathrm{qs}}\\ {\lambda }_{\mathrm{dr}}\\ {\lambda }_{\mathrm{qr}}\end{array}\right]\end{array}\end{array}& \left(1\right)\end{array}$ where s is the slip under steady state operation.

The flux and current have the following relationship: $\begin{array}{cc}\left[\begin{array}{c}{\lambda }_{\mathrm{ds}}\\ {\lambda }_{\mathrm{qs}}\\ {\lambda }_{\mathrm{dr}}\\ {\lambda }_{\mathrm{qr}}\end{array}\right]=\left[\begin{array}{cccc}{L}_{\mathrm{ls}}+L_m& 0& L_m& 0\\ 0& L_{\mathrm{ls}}+L_m& 0& L_m\\ L_m& 0& L_{\mathrm{lr}}+L_m& 0\\ 0& L_m& 0& L_{\mathrm{lr}}+L_m\end{array}\right]\left[\begin{array}{c}{I}_{\mathrm{ds}}\\ I_{\mathrm{qs}}\\ I_{\mathrm{dr}}\\ I_{\mathrm{qr}}\end{array}\right]& \left(2\right)\end{array}$ Combining Equations (1) and (2) yields: $\begin{array}{cc}\begin{array}{c}\left[\begin{array}{c}{V}_{\mathrm{ds}}\\ V_{\mathrm{qs}}\\ 0\\ 0\end{array}\right]=\{\left[\begin{array}{cccc}{R}_s& 0& 0& 0\\ 0& R_s& 0& 0\\ 0& 0& R_r& 0\\ 0& 0& 0& R_r\end{array}\right]+\left[\begin{array}{cccc}0& -{\omega }_e& 0& 0\\ {\omega }_e& 0& 0& 0\\ 0& 0& 0& -s{\omega }_e\\ 0& 0& s{\omega }_e& 0\end{array}\right]\\ \left[\begin{array}{cccc}{L}_{\mathrm{ls}}+L_m& 0& L_m& 0\\ 0& L_{\mathrm{ls}}+L_m& 0& L_m\\ L_m& 0& L_{\mathrm{lr}}+L_m& 0\\ 0& L_m& 0& L_{\mathrm{lr}}+L_m\end{array}\right]\}\left[\begin{array}{c}{I}_{\mathrm{ds}}\\ I_{\mathrm{qs}}\\ I_{\mathrm{dr}}\\ I_{\mathrm{qr}}\end{array}\right]\end{array}& \left(3\right)\end{array}$ Proper simplification of Equation (3) yields: $\begin{array}{cc}\left[\begin{array}{c}{V}_{\mathrm{qs}}\\ V_{\mathrm{ds}}\\ 0\\ 0\end{array}\right]=\left[\begin{array}{cccc}{R}_s& {\omega }_e{L}_s& 0& {\omega }_e{L}_m\\ -{\omega }_e{L}_s& R_s& -{\omega }_e{L}_m& 0\\ 0& s{\omega }_e{L}_m& R_r& s{\omega }_e{L}_r\\ -s{\omega }_e{L}_m& 0& -s{\omega }_e{L}_r& R_r\end{array}\right]\left[\begin{array}{c}{I}_{\mathrm{qs}}\\ I_{\mathrm{ds}}\\ I_{\mathrm{qr}}\\ I_{\mathrm{dr}}\end{array}\right]& \left(4\right)\end{array}$ where R_s and R_r are the stator and rotor resistances respectively; L_s, L_r, and L_m are the stator, rotor and mutual inductances, respectively; ω_e is the angular frequency (rad/s) of the synchronous reference frame; s is the slip; and sω_e is the slip frequency.

Aligning the d-axis with the stator current space vector {right arrow over (i)}_s(t) yields I_ds=I_s and I_qs=0. Then Equation (4) can be further simplified to: $\begin{array}{cc}\left[\begin{array}{c}{V}_{\mathrm{qs}}\\ V_{\mathrm{ds}}\\ 0\\ 0\end{array}\right]=\left[\begin{array}{ccc}{\omega }_e{L}_s& 0& {\omega }_e{L}_m\\ R_s& -{\omega }_e{L}_m& 0\\ s{\omega }_e{L}_m& R_r& s{\omega }_e{L}_r\\ 0& -s{\omega }_e{L}_r& R_r\end{array}\right]\left[\begin{array}{c}{I}_s\\ I_{\mathrm{qr}}\\ I_{\mathrm{dr}}\end{array}\right]& \left(5\right)\end{array}$

Defining $\begin{array}{cc}{V}_s\equiv \sqrt{V_{\mathrm{qs}}^2+V_{\mathrm{ds}}^2},\sigma =1-\frac{L_m^2}{L_s{L}_r},& \end{array}$ and using Equation (5), gives the following relationship: V_s²+I_s²R_s²−2R_sV_dsI_s=(1+σ)L_sω_eI_sV_qa−σL_s²ω_e²I_s²  (6)

A compact form of Equation (6) is given by: y=θ^Tu=u^Tθ  (7) where y=V_s²+I_s²R_s²−2R_sV_dsI_s, θ=[(1+σ)L_s σL_s²]^T and u=[ω_eI_sV_qs−ω_e²I_s²]^T.

The values of y and u depend on the operating point of the motor. For a given motor, different load levels lead to different values of u.

Because the parameter vector θ is composed solely of the motor inductance values, parametric estimation of the vector θ can be carried out by measuring the voltage and current at different load levels.

With recorded voltage and current data at n operating points, the following vectors can be formed: $y=\left[\begin{array}{c}{y}_1\\ y_2\\ ⋮\\ y_n\end{array}\right],U=\left[\begin{array}{c}{u}_1^T\\ u_2^T\\ ⋮\\ u_n^T\end{array}\right]$

Then Equation (8) can be expressed as: $\begin{array}{cc}\left[\begin{array}{c}{y}_1\\ y_2\\ ⋮\\ y_n\end{array}\right]=\left[\begin{array}{c}{u}_1^T\theta \\ u_2^T\theta \\ ⋮\\ u_n^T\theta \end{array}\right]\Rightarrow \left[\begin{array}{c}{y}_1\\ y_2\\ ⋮\\ y_n\end{array}\right]=\left[\begin{array}{c}{u}_1^T\\ u_2^T\\ ⋮\\ u_n^T\end{array}\right]\theta \Rightarrow y=U\theta & \left(8\right)\end{array}$

If n>2 then matrix U is not square. To determine the corresponding parameter vector θ, the Moore-Penrose inverse of U, which is designated by U^†, can be taken. For a non-square matrix, the Moore-Penrose inverse is not only a good analogy to the conventional inverse of the square matrix, but also a minimum norm least square solution with superb stability to the simultaneous linear Equation (8) under small signal perturbation.

Having obtained the Moore-Penrose inverse of U, the solution to Equation (8) can then be written as: θ=U†y  (9)

Once the parameter vector θ=[θ₁ θ₂]^T=[(1+θ)L_s σL_s²]^T is determined, the value of L_s and σL_s can be attained by solving the following second-order polynomial equation: x²−θ₁x+θ₂=0  (10)

Assuming the roots of Equation (10) are x₁ and x₂, with x₁>x₂, gives: $\begin{array}{cc}\{\begin{array}{c}{x}_1=L_s=L_{\mathrm{ls}}+L_m\\ x_2=\sigma L_s=L_s-\frac{L_m^2}{L_r}=L_{\mathrm{ls}}+L_m-\frac{L_m^2}{L_{\mathrm{lr}}+L_m}\end{array}& \left(11\right)\end{array}$

From IEEE Standard Test Procedure for Polyphase Induction Motors and Generators, IEEE Standard 112-1996, September 1996, the ratio between the leakage inductances is obtained for motors with different NEMA designs, $\frac{L_{\mathrm{lr}}}{L_{\mathrm{ls}}}=\gamma .$ The mutual inductance L_m is then the positive root of the following quadratic equation: L_m²+(γ−1)(x₁−x₂)L_m−γx₁(x₁−x₂)=0  (12) where x₁ and x₂ are the roots of equation (10).

Once the mutual inductance L_m is found, the stator leakage inductance can be acquired by: L_ls=L_s−L_m=x₁−L_m  (13) And the rotor leakage inductance is: L_lr=γL_ls  (14)

FIG. 3 is a flow chart for the inductance estimation algorithm. Sensors are needed to collect voltage and current information v_a, v_b, v_c and i_a, i_b, i_c, which is conventional for many motor applications.

The inductance estimation algorithm requires the cold state (ambient temperature) stator resistance R₅ as its input, as illustrated in FIG. 3. This stator resistance is usually provided by the motor manufacturer. If it is not, a simple direct current injection test can determine the precise value of the cold state stator resistance. In some cases, a multimeter can establish the resistance value with reasonable accuracy.

The other input needed is the machine design: NEMA A, B, C or D, which is normally printed on the motor's nameplate. This is used to determine the ratio γ between the stator leakage inductance and the rotor leakage inductance.

The online inductance estimation algorithm is based on the equations for steady state motor operation, and thus is not run during the transient state. As can be seen in FIG. 3, the algorithm executes Equations (8), (9), (10), (12), (13) and (14) sequentially.

Multiple runs at different load levels are used to construct the matrices y and U. They should be of appropriate dimensions to permit estimation of the motor parameter vector θ. Most mains-fed motors have a working cycle of starting—accelerating—running—stopped. The online inductance estimation algorithm enables the overload relay to learn the motor parameters during the first several such working cycles. After the successful recognition of those machine parameters, the electrical model of the induction machine is then properly tuned for the rotor resistance estimation algorithm. At this moment, the inductance estimation algorithm has accomplished its design objective and can therefore be safely bypassed.

The motor inductances may also be estimated from phasor analysis of the steady-state equivalent circuit illustrated in FIG. 4. To eliminate the stator resistance from the circuit equations, as illustrated in the modified block diagrams in FIGS. 1b and 3b. Expressions relating V_s, I_s,r_l, L_s,σ may be derived as follows:

In FIG. 4, the stator self inductance is L_s=L_ls+L_m; the rotor self inductance is $L_r=L_{\mathrm{lr}}+L_m;\sigma =1-\frac{L_m^2}{L_s{L}_r};$ s is the slip; ω_e is the angular speed, in rad/s, of the synchronously rotating reference frame; V_qds^er and I_qds^err are the stator voltage and current space vectors under the synchronously rotating reference frame, respectively; I_qs^er and I_ds^er are the stator current space vectors along the q-axis and d-axis, respectively.

From the complex vector analysis, the voltage and current space vectors, V_qds^er and I_qds^er, are regarded as {tilde over (V)}_s and Ĩ_s rotating in space. Hence phasors can also be used in FIG. 4 to represent the steady state motor operation.

Let Ĩ_r and Ĩ_m denote the phasors corresponding to −I_qs^er and −jI_ds^er, respectively. By aligning the x-axis with Ĩ_s, the abscissa and ordinate of a plane Cartesian coordinate are completely determined in the phasor diagram depicted in FIG. 5. Let V_sx and V_sy denote the projection of {tilde over (V)}_s onto the x-axis and y-axis, and similarly I_rx and I_ry for Ĩ_r, then: $\begin{array}{cc}\left[\begin{array}{c}{V}_{\mathrm{sy}}\\ V_{\mathrm{sx}}\\ 0\\ 0\end{array}\right]=\left[\begin{array}{ccc}{\omega }_e{L}_s& 0& {\omega }_e\left(1-\sigma \right)L_s\\ R_s& -{\omega }_e\left(1-\sigma \right)L_s& 0\\ {\omega }_e\left(1-\sigma \right)L_s& r_2& {\omega }_e\left(1-\sigma \right)L_s\\ 0& -{\omega }_e\left(1-\sigma \right)L_s& r_2\end{array}\right]\left[\begin{array}{c}{I}_s\\ I_{\mathrm{ry}}\\ I_{\mathrm{rx}}\end{array}\right]& \left(15\right)\end{array}$

Eliminating I_rx and I_ry from the first two rows, eliminating r₂ from the last row, and substituting those relationships into the third row yields, after proper rearrangement: $\begin{array}{cc}\left[\begin{array}{cc}{\omega }_e{I}_s{V}_{\mathrm{sy}}& -{\omega }_e^2{I}_s^2\end{array}\right]\left[\begin{array}{c}\left(1+\sigma \right)L_s\\ \sigma L_s^2\end{array}\right]=V_s^2+I_s^2{R}_s^2-2R_s{I}_s{V}_{\mathrm{sx}}& \left(16\right)\end{array}$

By taking the voltage and current measurements at two different load levels, the induction machine parameter vector [(1+σ)L_s σL_s²]^T is solved for motor inductances, L_s and σL_s as described above.

Since the motor inductance estimation algorithm is essentially based on the fundamental frequency positive sequence machine model, other components in the voltage and current samples should be eliminated. The digital positive waveform synthesis is designed for such purpose. By extracting positive sequence instantaneous values v_a1, v_b1, v_c1 or i_a1, i_b1, i_c1 from the original sampled data v_a, v_b, v_c or i_a, i_b, i_c, filtered data can be acquired for the estimation algorithm.

The most important aspect of digital positive and negative sequence waveform synthesis is the digital phasor extraction. Its purpose is to extract from the instantaneous voltage and current data the corresponding phasors under sinusoidal steady state condition. Its theory of operation is detailed below.

Starting from the basic concept, a periodic impulse train is described by: $\begin{array}{cc}\tilde{x}\left[n\right]=\sum _{i=0}^{N-1}\delta \left[n-i\right].& \left(17\right)\end{array}$

The discrete Fourier series of this impulse train is: $\begin{array}{cc}\tilde{X}\left[k\right]=\sum _{n=0}^{N-1}{W}_N^{\mathrm{kn}}.& \left(18\right)\end{array}$ where $W_N=e^{-j\frac{2\pi }{N}}.$

When k≠0, the discrete Fourier series in Equation (18) is simplified to: $\begin{array}{cc}\tilde{X}\left[k\right]=\sum _{n=0}^{N-1}{W}_N^{\mathrm{kn}}=\sum _{n=0}^{N-1}{e}^{-jk\frac{2\pi n}{N}}=\frac{1-e^{-jk\frac{2\pi N}{N}}}{1-e^{-j\frac{2\pi }{N}}}=0& \left(19\right)\end{array}$

The real part of Equation (19) is: $\begin{array}{cc}\mathrm{Re}\left\{\tilde{X}\left[k\right]\right\}=\mathrm{Re}\left\{\sum _{n=0}^{N-1}{e}^{-jk\frac{2\pi n}{N}}\right\}=\sum _{n=0}^{N-1}\cos \left(k\frac{2\pi n}{N}\right)=0& \left(20\right)\end{array}$

Similarly, the imaginary part of Equation (19) is: $\begin{array}{cc}\mathrm{Im}\left\{\tilde{X}\left[k\right]\right\}=\mathrm{Im}\left\{\sum _{n=0}^{N-1}{e}^{-jk\frac{2\pi n}{N}}\right\}=\sum _{n=0}^{N-1}\sin \left(k\frac{2\pi n}{N}\right)=0& \left(21\right)\end{array}$

If the frequency of the power supply is f_e, then the angular frequency of the power supply is ω_e=2πf_e, and the period is $T_e=\frac{1}{f_e}.$ In each cycle N samples of the voltage and current signals are taken.

If t_n is the corresponding time for the n^th sample, $t_n=\frac{n}{N}·T_{e.}$ So from Equation (20) and Equation (21), it is known that: $\begin{array}{cc}\sum _{n=0}^{N-1}\left[\sin \left(p{\omega }_e{t}_n\right)\sin \left(q{\omega }_e{t}_n\right)\right]=\sum _{n=0}^{N-1}\left\{\frac{1}{2}\cos \left[\left(p-q\right){\omega }_e{t}_n\right]-\frac{1}{2}\cos \left[\left(p+q\right){\omega }_e{t}_n\right]\right\}=\frac{1}{2}\sum _{n=0}^{N-1}\cos \left[\left(p-q\right)\frac{2\pi n}{N}\right]-\frac{1}{2}\sum _{n=0}^{N-1}\cos \left[\left(p+q\right)\frac{2\pi n}{N}\right]& \left(22\right)\\ \sum _{n=0}^{N-1}\left[\sin \left(p{\omega }_e{t}_n\right)\cos \left(q{\omega }_e{t}_n\right)\right]=\sum _{n=0}^{N-1}\left\{\frac{1}{2}\sin \left[\left(p+q\right){\omega }_e{t}_n\right]+\frac{1}{2}\sin \left[\left(p-q\right){\omega }_e{t}_n\right]\right\}=\frac{1}{2}\sum _{n=0}^{N-1}\sin \left[\left(p+q\right)\frac{2\pi n}{N}\right]+\frac{1}{2}\sum _{n=0}^{N-1}\sin \left[\left(p-q\right)\frac{2\pi n}{N}\right]& \left(23\right)\\ \sum _{n=0}^{N-1}\left[\cos \left(p{\omega }_e{t}_n\right)\sin \left(q{\omega }_e{t}_n\right)\right]=\sum _{n=0}^{N-1}\left\{\frac{1}{2}\sin \left[\left(p+q\right){\omega }_e{t}_n\right]-\frac{1}{2}\sin \left[\left(p-q\right){\omega }_e{t}_n\right]\right\}=\frac{1}{2}\sum _{n=0}^{N-1}\sin \left[\left(p+q\right)\frac{2\pi n}{N}\right]-\frac{1}{2}\sum _{n=0}^{N-1}\sin \left[\left(p-q\right)\frac{2\pi n}{N}\right]& \left(24\right)\\ \sum _{n=0}^{N-1}\left[\cos \left(p{\omega }_e{t}_n\right)\cos \left(q{\omega }_e{t}_n\right)\right]=\sum _{n=0}^{N-1}\left\{\frac{1}{2}\cos \left[\left(p+q\right){\omega }_e{t}_n\right]+\frac{1}{2}\cos \left[\left(p-q\right){\omega }_e{t}_n\right]\right\}=\frac{1}{2}\sum _{n=0}^{N-1}\cos \left[\left(p+q\right)\frac{2\pi n}{N}\right]+\frac{1}{2}\sum _{n=0}^{N-1}\cos \left[\left(p-q\right)\frac{2\pi n}{N}\right]& \left(25\right)\end{array}$ where p and q are positive integers. From Equations (20)-(21), Equations (22)-(25) are all zero, unless p±q=0.

Typical voltage or current waveforms with fundamental frequency f, can be described by: $\begin{array}{cc}f\left(t\right)=\frac{1}{2}{a}_0+\sum _{h=1}^{\infty }\left[a_h\cos \left(h{\omega }_{e}t\right)+b_h\sin \left(h{\omega }_{e}t\right)\right]& \left(26\right)\end{array}$ where a₁ and b₁ are constant coefficients for fundamental frequency components, and a_h and b_h (h>2) are coefficients for harmonic content in the waveform.

Based on Equations (22) through (25), the following Equations (27) through (28) can be employed to determine the constant coefficients a₁ and b₁ in the voltage or current waveform f(t): $\begin{array}{cc}{a}_1=\frac{2}{N}\sum _{n=0}^{N-1}\left[f\left(t_n\right)·\cos \left({\omega }_e{t}_n\right)\right]& \left(27\right)\\ b_1=\frac{2}{N}\sum _{n=0}^{N-1}\left[f\left(t_n\right)·\sin \left({\omega }_e{t}_n\right)\right]& \left(28\right)\end{array}$

Once the coefficients, a₁ and b₁, in the fundamental component of the instantaneous voltage, v(t)=a₁ cos(ω_et)+b₁ sin(ω_et), are known from Equations (27) and (28), the corresponding voltage phasor, $\tilde{V}=\sqrt{\frac{a_1^2+b_1^2}{2}·e^{-\mathrm{j\phi }}}$ can be calculated. $\begin{array}{cc}\phi =\{\begin{array}{cc}\arctan \left(\frac{b_1}{a_1}\right)& a_1>0\\ \pm \frac{\pi }{2}& a_1=0\\ \arctan \left(\frac{b_1}{a_1}\right)+\pi & a_1<0\end{array}& \left(29\right)\end{array}$

FIG. 6 can be used as an aid in determining the angle φ from Equation (29). When the point (a₁, b₁) is in Quadrants I and IV, $\phi =\arctan \left(\frac{b_1}{a_1}\right);$ when point (a₁, b₁) is in Quadrants I and III, $\phi =\arctan \left(\frac{b_1}{a_1}\right)+\pi ;$ when point (a₁, b₁) is on the vertical axis, $\phi =\frac{\pi }{2}$ for b₁>0 and $\phi =-\frac{\pi }{2}$ for b₁<0.

After extracting the phasors from the instantaneous values, the second step is to use a symmetric components method to decompose them into positive and negative sequence components for future analysis.

Denoting a=e^j120 yields: $\begin{array}{cc}\left[\begin{array}{c}\widetilde{V_1}\\ \widetilde{V_2}\\ \widetilde{V_0}\end{array}\right]=\frac{1}{3}\left[\begin{array}{ccc}1& a& a^2\\ 1& a^2& a\\ 1& 1& 1\end{array}\right]\left[\begin{array}{c}\widetilde{V_a}\\ \widetilde{V_b}\\ \widetilde{V_c}\end{array}\right]& \left(30\right)\end{array}$

By decomposing the voltage and current information into positive and negative sequence components, further analysis can be done on the machine's sequence model.

Once the positive sequence phasors are computed, the corresponding instantaneous values for each individual phase can be obtained by the following mapping: $\begin{array}{cc}{\tilde{V}}_1=V_1{e}^{\mathrm{j\varphi }}\Rightarrow \{\begin{array}{ccc}{\tilde{V}}_a={\tilde{V}}_1& \iff & v_a=\sqrt{2}{V}_1\cos \left({\omega }_{e}t+\varphi \right)\\ {\tilde{V}}_b={\tilde{V}}_1{e}^{-\mathrm{j120°}}& \iff & v_b=\sqrt{2}{V}_1\cos \left({\omega }_{e}t+\varphi -120°\right)\\ {\tilde{V}}_c={\tilde{V}}_1{e}^{\mathrm{j120°}}& \iff & v_c=\sqrt{2}{V}_1\cos \left({\omega }_{e}t+\varphi +120°\right)\end{array}& \left(31\right)\end{array}$

The rotor resistance estimator 12 estimates the rotor resistance based on the estimated inductances of the motor, the samples v_s and i_s of the motor terminal voltage and current, and the angular speed ω_rud of the rotor. The rotor resistance value is temperature-dependent.

The digital positive sequence waveform synthesizer (DPSWS) block is also used in the rotor resistance estimation algorithm to filter out the negative sequence components and harmonics to insure a reliable estimate of the rotor resistance.

Rotor speed can be obtained, for example, from a saliency harmonic speed detection algorithm based on K. D. Hurst and T. G. Habetler, “A Comparison of Spectrum Estimation Techniques for Sensorless Speed Detection in Induction Machines,” IEEE Transactions on Industry Applications, Vol. 33, No. 4 July/August 1997, pp. 898-905.

Using a synchronously rotating reference frame, the operation of a symmetrical induction machine can be described by the following set of equations: $\begin{array}{cc}\left[\begin{array}{c}{v}_{\mathrm{qs}}\\ v_{\mathrm{ds}}\\ v_{\mathrm{qr}}\\ v_{\mathrm{dr}}\end{array}\right]=\left[\begin{array}{cccc}{R}_s+{\mathrm{pL}}_s& {\omega }_e{L}_s& {\mathrm{pL}}_m& {\omega }_e{L}_m\\ -{\omega }_e{L}_s& R_s+{\mathrm{pL}}_s& -{\omega }_e{L}_m& {\mathrm{pL}}_m\\ {\mathrm{pL}}_m& s{\omega }_e{L}_m& R_r+{\mathrm{pL}}_r& s{\omega }_e{L}_r\\ -s{\omega }_e{L}_m& {\mathrm{pL}}_m& -s{\omega }_e{L}_r& R_r+{\mathrm{pL}}_r\end{array}\right]\left[\begin{array}{c}\begin{array}{c}\begin{array}{c}{i}_{\mathrm{qs}}\\ i_{\mathrm{ds}}\end{array}\\ i_{\mathrm{qr}}\end{array}\\ i_{\mathrm{dr}}\end{array}\right]& \left(32\right)\end{array}$ where R_s and R_r are the stator and rotor resistance respectively; L_s, L_r and L_m are the stator, rotor and mutual inductance respectively; ω_e is the angular frequency (rad/s) of the synchronous reference frame; s is the slip; and p is the differential operator $\frac{d}{dt}.$

Aligning the d-axis with the stator current vector {overscore (i)}_s(t) and assuming steady state operation so that p=0, Equation (32) can be simplified to: $\begin{array}{cc}\left[\begin{array}{c}\begin{array}{c}{V}_{\mathrm{qs}}\\ 0\end{array}\\ 0\end{array}\right]=\left[\begin{array}{ccc}{\omega }_e{L}_s& 0& {\omega }_e{L}_m\\ s{\omega }_e{L}_m& R_r& s{\omega }_e{L}_r\\ 0& -s{\omega }_e{L}_r& R_r\end{array}\right]\left[\begin{array}{c}\begin{array}{c}{I}_s\\ I_{\mathrm{qr}}\end{array}\\ I_{\mathrm{dr}}\end{array}\right]& \left(33\right)\end{array}$

Solving Equation (33) for R_r gives: $\begin{array}{cc}{R}_r=s{\omega }_e\sqrt{L_r\left(\frac{{\omega }_e{L}_m^2}{{\omega }_e{L}_s-\frac{V_{\mathrm{qs}}}{I_s}}-L_r\right)}& \left(34\right)\end{array}$

Equation (34) is the primary equation used for rotor resistance estimation. Since it requires only inductances L_s, L_r and L_m; voltage, V_qs; current, I_s and slip frequency, sω_e′ the outcome gives an accurate estimate of rotor resistance. This rotor resistance estimate is independent of the stator resistance and hence the stator winding temperature.

The rotor resistance may also be estimated from phasor analysis of the steady-state equivalent circuit illustrated in FIG. 4.

First, the last two rows in Equation (15) are solved for I_rx and I_ry: $\begin{array}{cc}\left[\begin{array}{c}{I}_{\mathrm{rx}}\\ I_{\mathrm{ry}}\end{array}\right]=\frac{-I_s}{r_2^2+{{\omega }_e^2\left(1-\sigma \right)}^2{L}_s^2}\left[\begin{array}{c}{{\omega }_e^2\left(1-\sigma \right)}^2{L}_s^2\\ {\omega }_e\left(1-\sigma \right)L_s{r}_2\end{array}\right]& \left(35\right)\end{array}$

Substituting I_rx in the first row of Equation (15), solving for r₂ and using the notational definition r₂=R_r/s(L_m/L_r)², gives: $\begin{array}{cc}{R_r\left(\frac{L_m}{L_r}\right)}^2=s{\omega }_e\left(1-\sigma \right)L_s\sqrt{\frac{\frac{V_{\mathrm{sy}}}{I_s}-{\omega }_e\sigma L_s}{{\omega }_e{L}_s-\frac{V_{\mathrm{sy}}}{I_s}}}& \left(36\right)\end{array}$

The rotor temperature estimator 14 estimates the rotor temperature or θ_r^EST from the estimated rotor resistance and the well known linear relationship between the rotor temperature and rotor resistance.

The stator temperature estimator 16 estimates the stator temperature of an induction machine from its rotor angular speed ω_rud, the stator current i_s and the estimated rotor temperature θ_r^EST, using a full-order hybrid thermal model (HTM) illustrated in FIG. 7. The HTM approximates the rotor and stator thermal characteristics. The model parameters are loosely associated with aspects of the machine design.

The quantities θ_s and θ_r are temperature rises [° C.] above ambient on the stator and rotor sides, respectively. The power input P_s [ ] is associated directly with the I²R loss generated in the stator winding. In addition, under constant supply voltage, the core loss generated in the stator teeth and the back iron contributes a fixed portion to the rise of θ_s. The power input P_r is associated mainly with the I²R loss in the rotor bars and end rings. These losses are calculated from the induction machine equivalent circuit.

The thermal resistance R₁ [° C./W] represents the heat dissipation capability of the stator to the ambient through the combined effects of heat conduction and convection; the thermal resistance R₂ [° C./W] is associated with the heat dissipation capability of the rotor to its surroundings; the thermal resistance R₃ [° C./W] is associated with the heat transfer from the rotor to the stator across the air gap. Since the radiation only accounts for a small amount of energy dissipated for most induction machines, its effect can be safely ignored without introducing significant errors in the stator winding temperature estimation.

The thermal capacitances C₁ and C₂ [J/° C.] are defined to be the energy needed to elevate the temperature by one degree Celsius for the stator and rotor, respectively. The capacitance C₁ represents the total thermal capacity of the stator windings, iron core and frame, while C₂ represents the combined thermal capacity of the rotor conductors, rotor core and shaft.

A motor's stator and rotor thermal capacitances C₁ and C₂ are fixed once the motor is manufactured. The motor's thermal resistances, however, are largely determined by its working environment. When a motor experiences an impaired cooling condition, R₁, R₂ and R₃ vary accordingly to reflect the changes in the stator and rotor thermal characteristics.

The thermal characteristics of a motor manifest themselves in the rotor temperature, which can be observed from the rotor temperature via the embodiment of the rotor resistance estimator. Based on the estimated rotor temperature, the parameters of the HTM can therefore be adapted in an online fashion to reflect the motor's true cooling capability.

Compared with the conventional thermal model using a single thermal capacitor and a single thermal resistor, the HTM incorporates the rotor losses and the heat transfer between the rotor and the stator. When properly turned to a specific motor's cooling capacity, it is capable of tracking the stator winding temperature during running overloads, and therefore provides adequate protection against overheating. Furthermore, the HTM is of adequate complexity compared to more complex thermal networks. This makes it suitable for online parameter tuning.

For most TEFC machines up to 100 hp, C₁ is usually larger than C₂ due to the design of the motor. R₁ is typically smaller than R₂ because of the superior heat dissipation capacity of the stator. R₃ is usually of the same order of magnitude as R₁, and it correlates the rotor temperature to the stator temperature. These relationships are justified by the detailed analysis of the machine design.

1) Relationship between C₁ and C₂: For most small-size induction machines with the TEFC design, the frame is an integral part of the machine. The frame acts as a heat sink with a large thermal capacitance. Therefore, even thought the bulk parts of both the stator and rotor are laminated silicon steel of the same axial length, the thermal capacitance C₁ is often larger than C₂ due to the inclusion of the frame. In addition, the stator winding overhang provides some extra thermal capacity to C₁. Normally, C₁ is 2 to 3 times larger than C₂.

2) Relationship between R₁ and R₂: The frame of a small-size TEFC motor prevents the free exchange of air between the inside and outside of the case. Consequently, the heat transfer by means of convection from the rotor to the external ambient is restricted at all conditions. Only a limited amount of heat created by the rotor is transferred by means of conduction from the rotor bars and the end rings through the rotor iron to the shaft, and then flows axially along the shaft, through the bearings to the frame and finally reaches the ambient. In contrast, the stator dissipates heat effectively through the combined effects of conduction and convection. The axial ventilation through the air gap carries part of the heat away from the stator slot winding and the stator teeth, with the balance of the heat transferred radially through the stator iron and frame to the external ambient.

As evidence of the good thermal conductivities of the stator iron and frame, the temperature of the stator slot winding is usually 5-10° C. lower than that of the stator end winding. Some TEFC motors are even equipped with pin-type cooling fins on the surface of the frame, which makes it easier for the heat to be evacuated to the ambient. R₂ is usually 4-30 times larger than R₁ at rated conditions.

3) Relationship between R₁ and R₃: Most induction machines with TEFC design in the low power range are characterized by a small air gap (typically around 0.25-0.75 mm) to increase efficiency. Furthermore, only a limited amount of air is exchanged between the inside and outside of the motor due to the enclosure. Therefore, among the rotor cage losses that are dissipated through the air gap, a significant portion is transferred to the stator by means of laminar heat flow, while the rest is passed to the endcap air inside the motor. These rotor cage losses, combined with the stator losses, travel through the stator frame and end shield and finally reach the ambient. Hence, the rotor and stator temperatures are highly correlated due to this heat flow pattern.

It has been estimated that 65% of the overall rotor losses are dissipated through the air gap at rated condition. This indicates that R₃ is much smaller than R₂, and is usually of the same order of magnitude as R₁. Otherwise, the rotor losses would be dissipated mainly along the shaft instead of through the air gap.

Assuming P_s and P_r are the inputs and θ_r is the output, the state space equations that describe the hybrid thermal model in FIG. 7 are: $\begin{array}{cc}\left[\begin{array}{c}{\stackrel{.}{\theta }}_s\\ {\stackrel{.}{\theta }}_r\end{array}\right]=\left[\begin{array}{cc}-\frac{1}{C_1}\left(\frac{1}{R_1}+\frac{1}{R_3}\right)& \frac{1}{R_3{C}_1}\\ \frac{1}{R_3{C}_2}& -\frac{1}{C_2}\left(\frac{1}{R_2}+\frac{1}{R_3}\right)\end{array}\right]\left[\begin{array}{c}{\theta }_s\\ {\theta }_r\end{array}\right]+\left[\begin{array}{cc}\frac{1}{C_1}& 0\\ 0& \frac{1}{C_2}\end{array}\right]\left[\begin{array}{c}{P}_s\\ P_r\end{array}\right]& \left(37\right)\\ y=\left[\begin{array}{cc}0& 1\end{array}\right]\left[\begin{array}{c}{\theta }_s\\ {\theta }_r\end{array}\right]& \left(38\right)\end{array}$ Since the magnitude of the core loss is independent of the motor load level under constant supply voltage, and the core loss is far less than the stator I²R loss for most modern induction machines, it is therefore neglected in P_s. By taking into account only the I²R losses in the stator and rotor, P_s and P_r are: $\begin{array}{cc}{P}_s=3I_s^2{R}_s& \left(39\right)\\ P_r=3I_s^2\frac{s^2{\omega }_e^2{L}_m^2}{R_r^2+s^2{\omega }_e^2{L}_r^2}{R}_r& \left(40\right)\end{array}$ where I_s is the stator rms current; R_s and R_r are the stator and rotor resistances, respectively; L_m and L_r are the mutual and rotor inductances, respectively; s is the per unit slip; and ω_e is the synchronous speed.

K is the ratio of P_r to P_s: $\begin{array}{cc}K=\frac{P_r}{P_s}=\frac{s^2{\omega }_e^2{L}_m^2}{R_r^2+s^2{\omega }_e^2{L}_r^2}·\frac{R_r}{R_s}& \left(41\right)\end{array}$ Since the temperature increases gradually inside the motor at running overload conditions, K remains constant if the rotor speed does not change.

By substituting P_s with KP_r in Equation (37) and taking the Laplace transform, the transfer function between the input P_s, and the output θ_r in the s-domain, where s now denotes the Laplace operator, is: $\begin{array}{cc}Y\left(s\right)=\frac{b_{1}s+b_0}{a_2{s}^2+a_{1}s+a_o}{P}_s\left(s\right)& \left(42\right)\end{array}$ where

• • a₀=R₁+R₂+R₃; a₁=C₁R₁R₃+C₁R₁R₂+C₂R₂R₃+C₂R₁R₂; a₂=C₁C₂R₁R₂R₃;
  • b₀=R₁R₂+R₁R₂K+R₂R₃K; b₁=C₁R₁R₂R₃K.

When the input to the hybrid thermal model is a step signal with a magnitude c, the output, in the time domain, is: y(t)=α·u(t)+β₁e^λ₁^t+β₂e^λ₂^t  (43) where $\alpha =\frac{b_{0}c}{a_0};$ u(t) is a unit step; ${\lambda }_1=\frac{-a_1+\sqrt{a_1^2-4a_2{a}_0}}{2a_2};{\lambda }_2\frac{-a_1-\sqrt{a_1^2-4a_2{a}_0}}{2a_2};{\beta }_1=\frac{\left(b_1{\lambda }_1+b_0\right)c}{a_2\left({\lambda }_1-{\lambda }_2\right){\lambda }_1};\mathrm{and}{\beta }_2=\frac{\left(b_1{\lambda }_2+b_0\right)c}{a_2\left({\lambda }_2-{\lambda }_1\right){\lambda }_2}.$

In Equation (43), a represents the magnitude of the steady-state rotor temperature; β₁e^λ₁^t and β₂e^λ₂^t correspond to the rotor thermal transient.

Information Guide for General Purpose Industrial AC Small and Medium Squirrel-Cage Induction Motor Standards, NEMA Standard MG1-2003, August 2003, specifies the major physical dimensions of induction machines up to 100 hp, with squirrel-cage rotors. Analysis of these design specifications gives a typical range for the thermal parameters R₁, R₂, R₃, C₁ and C₂ in the hybrid thermal model. Therefore, for a typical small-size, mains-fed induction machine, β₁e^λ₁^t corresponds to a slow exponential change with large magnitude, while β₂e^λ₁^t corresponds to a fast but rather small thermal transient. Since the latter term vanishes quickly, β₁e^λ₁^t is identified as the dominant component in the rotor thermal transient. Similar conclusions can also be drawn for the stator thermal transient. Consequently, Equation (43) is simplified to the following: $\begin{array}{cc}y\left(t\right)=\alpha ·u\left(t\right)+{\beta }_1{e}^{{\lambda }_{1}t}=\alpha ·u\left(t\right)+{\beta }_1{e}^{\frac{t}{{\tau }_{\mathrm{th}}}}& \left(44\right)\end{array}$ where τ_th is the thermal time constant obtained from the rotor temperature, estimated from the rotor temperature estimation algorithm.

As a consequence of the strong correlation between the rotor and stator temperatures at running overload conditions, the stator winding thermal transient has the same thermal time constant as the rotor.

Since a typical small-size, mains-fed induction machine's internal losses are dissipated chiefly through the stator to the ambient, and the thermal resistance R₁ is directly connected to the stator side in the full-order HTM in FIG. 7, R₁ is the most significant factor in determining the steady-state rotor temperature and its thermal time constant in a thermal transient. Due to the strong correlation between the rotor and stator temperatures, R₁ is also the most important factor in determining the steady-state stator temperature and its thermal time constant in a thermal transient. The stator steady-state temperature is relatively insensitive to the changes in R₂ and R₃.

To ensure comprehensive motor overload protection under the running overload condition, an online parameter tuning algorithm is used to protect the motor against running overloads with stator current between 100% and 200% of the motor's rated current.

FIG. 8 shows the flowchart of the online parameter tuning algorithm. R₂ and C₂ are first computed offline from the motor nameplate data, including the full load current, the service factor and the trip class. Then R₁, R₃ and C₁ are calculated online based on the data from heat runs at various load levels.

To check whether the tuned hybrid thermal model matches the motor's true thermal behavior, the sum of the squared error (SSE) is used as an index. It is defined as the error between the rotor temperature predicted by the hybrid thermal model and the rotor temperature estimated from the rotor resistance: $\begin{array}{cc}\mathrm{SSE}=\sum _{n=1}^N{\left[{\theta }_r^{\mathrm{HTM}}\left(n\right)-{\theta }_r^{\mathrm{EST}}\left(n\right)\right]}^2& \left(45\right)\end{array}$ where N is the number of samples used in computing SSE.

Once the SSE becomes smaller than a predetermined threshold value ε, the tuning is done. Otherwise, a mismatch exists between the hybrid thermal model and the motor's true thermal operating condition, and more iterations are needed with additional data from heat runs to tune the parameters until a solution is found.

Since R₁ is more closely related to the motor cooling capability than R₂ and R₃ due to the heat flow inside an induction machine, R₁ is the focus of the online parameter tuning algorithm. The tuning of R₁ alone can yield a sufficiently accurate estimate of the stator winding temperature.

The thermal parameters R₂ and C₂ are determined from the motor specifications at the locked rotor condition. When the rotor is locked, there is no ventilation in the air gap. Consequently, it can be assumed that no heat is transferred across the air gap by means of laminar heat flow during this short interval. Then this locked rotor condition is modeled by letting R₃→∞ and decoupling the full-order HTM into two separate parts, the stator part and the rotor part, as shown in FIG. 9.

During locked rotor conditions, the thermal capacity of an induction machine is usually rotor limited. At an ambient temperature of 40° C., a typical maximum allowable rotor temperature of 300° C. is assumed.

Associated with the rotor thermal capacity at locked rotor conditions, the trip class is defined as the maximum time (in seconds) for an overload trip to occur when a cold motor is subjected to 6 times its rated current. It is related to R₂ and C₂ by: $\begin{array}{cc}\mathrm{TC}=R_2{C}_2\ln \left(\frac{6^2}{6^2-{\mathrm{SF}}^2}\right)& \left(46\right)\end{array}$ where TC is the trip class given by the motor manufacturer; and SF is the service factor.

Alternatively, a manufacturer may specify a Stall Time and the Locked Rotor Current. One may substitute Stall Time for Trip Class in Equation (46) and the ratio of Locked Rotor Current to Full Load Current for the constant 6, to obtain the rotor thermal parameters.

The rise of the rotor temperature during a locked rotor condition is: $\begin{array}{cc}{\theta }_r\left(t\right){{}_{t=\mathrm{TC}}-{\theta }_r\left(t\right)}_{t=0}=P_r^{\prime }{R}_2\left(1-e^{-\frac{t}{R_2{C}_2}}\right){❘}_{t=\mathrm{TC}}& \left(47\right)\end{array}$ where P_r is the rotor I²R loss at the locked rotor condition; θ_r(t)|_t=0 is the initial rotor temperature; and θ_r(t)|_t=TC is the rotor temperature after TC elapsed.

Substituting θ_r(t)|_t=TC and θ_r(t)|_t=0 in Equation (47) with 300° C. and 40° C., respectively; and combining Equations (47) and (46) to obtain R₂ and C₂: $\begin{array}{cc}{R}_2=\frac{9360}{P_r^{\prime }·{\mathrm{SF}}^2}.& \left(48\right)\\ C_2=\frac{P_r^{\prime }·{\mathrm{SF}}^2·\mathrm{TC}}{9360·\ln \left(\frac{6^2}{6^2-{\mathrm{SF}}^2}\right)}& \left(49\right)\end{array}$

The thermal parameters R₃, R₁ and C₁ are calculated online based on the data from heat runs at various load levels. First, R₃ is determined from the motor specifications at the rated condition: $\begin{array}{cc}{R}_3=\frac{{\theta }_r\left(t\right){{}_{t=\infty }-{\theta }_s\left(t\right)}_{t=\infty }}{P_r-\frac{{\theta }_r\left(t\right){❘}_{t=\infty }}{R_2}}& \left(50\right)\end{array}$ where θ_s(t)|_t=∞ and θ_r(t)|_t=∞ are the steady state stator and rotor temperature rises above the ambient at the rated condition, respectively; P_r is the rotor I²R loss at the rated condition; and R₂ is computed previously by Equation (48). θ_s(t)|_t=∞ is either specified in the motor manufacturer's data sheet or obtained by measuring the stator resistance at the rated condition through the direct current injection method.

Then the coefficients in the dominant component of the rotor thermal transient, β₁ and λ₁, are identified. The identification process involves regression analysis, i.e., fitting an exponential function in the same form as Equation (44) to the rotor thermal transient, which is derived from the estimated rotor resistance.

Finally, R₁ and C₁ are solved from β₁ and λ₁ by numerical root finding techniques. Given R₂, R₃ and C₂, R₁ and C₁ are related to β₁ and λ₁: $\begin{array}{cc}\{\begin{array}{c}{\beta }_1=f_1\left(R_1,C_1\right)\\ {\lambda }_1=f_2\left(R_1,C_1\right)\end{array}& \left(51\right)\end{array}$ where the nonlinear functions f₁ and f₂ are formulated from Equation (44). Newton's method is chosen as the numerical root finding method due to its superior convergence speed.

After successful completion of the tuning, SSE is constantly monitored to detect changes in the motor's cooling capability, such as those caused by a broken cooling fan or a clogged motor casing. If such a change is detected, R₁ is tuned again since only this parameter is highly correlated with the motor's cooling capability. In this case, λ₁ is first updated through regression analysis. Then it is expressed as a function of R₁: λ₁=f₂(R₁)  (52) where f₂ has the same form as f₂ in Equation (51), but C₁ is now regarded as a known constant. Finally a new R₁ is solved from Equation (52) by the Newton's method to reflect the change in the motor's cooling capability.

Through the online parameter tuning, the hybrid thermal model ensures a correct prediction of the stator winding temperature, even under an impaired cooling condition, thereby providing a comprehensive overload protection to the motor.

Most induction machines of TEFC (totally enclosed fan-cooled) design in the low power range (<30 hp) are characterized by a small air gap (typically between 0.25 mm and 3 mm) to increase the efficiency. In addition, the machines are shielded by frames to prevent the free exchange of air between the inside and outside of the case. Therefore, among the rotor cage losses that are evacuated through the air gap, a significant portion is transferred to the stator slots, while the rest is passed to the endcap air inside the motor. These rotor cage losses, combined with the stator losses, travel through the motor frames and finally reach the ambient. Hence the rotor conductor temperature and the stator winding temperature are correlated due to the machine design.

Based on the analysis of the full-order hybrid thermal model, a reduced-order hybrid thermal model may be used, as illustrated in FIG. 10. This simplified thermal model is capable of predicting the stator winding temperature rise from only current and rotor speed information. The model parameters are loosely associated with aspects of the machine design.

The quantities θ_s and θ_r are temperature rises [° C.] above ambient on the stator and rotor, respectively. The power input P_s [Watt] is associated chiefly with the I²R loss generated in the stator winding. The power input P_r is associated mainly with the I²R loss in the rotor bars and end rings. They are calculated from the induction machine electrical model. The thermal resistance R₁ [° C./W] represents the heat dissipation capability of the stator to the surrounding air through the combined effects of heat conduction and convection, while R₃ is associated with the heat transfer from the rotor to the stator through the air gap, mainly by means of conduction and convection. The thermal capacitance C₁ [J/° C.] is determined by the overall thermal capacities of the stator winding, iron core and motor frame. The voltage V in FIG. 10 represents the rotor temperature estimated from the rotor resistance. Compared with the full-order HTM, R₁ in the reduced-order HTM is the same as R₁ in FIG. 7, R₃ in the reduced-order HTM is the same as R₃ in FIG. 9, and C₁ in the reduced-order HTM is the same as C₁ in FIG. 7.

According to I. Boldea and S. A. Nasar, The Induction Machine Handbook. Boca Raton: CRC Press, 2002, p. 406, at rated speed, approximately 65% of the rotor cage losses are dissipated through the air gap to the stator and ambient. This figure may be different at other operating points. Therefore, ΔP_r is incorporated in the reduced-order HTM to indicate the variation of power losses across the air gap at different load levels. This reduced-order HTM not only correlates the stator winding temperature with the rotor conductor temperature, but also unifies the thermal model-based and the parameter-based temperature estimators. In addition, the reduced-order HTM is of reasonable complexity for real time implementation. With proper tuning, the HTM provides a more flexible approach than most state-of-the-art relays to the problem of estimating the stator winding temperature under running overload conditions. The increased flexibility and versatility are due to the real time tuning of the HTM thermal parameters to reflect the true thermal characteristics of the motor.

The operation of the reduced-order HTM can be divided into two stages:

Stage I involves the online parameter estimation of the thermal parameters in the hybrid thermal model; while Stage II involves the optimal stator winding temperature estimation based on an adaptive Kalman filtering approach.

In Stage I, a reliable tracking of the stator temperature requires that three thermal parameters, C₁, R₁ and R₃ be determined. Online parameter estimation can be performed with information from estimated rotor temperature to yield fairly accurate values for C₁, R₁ and R₃.

Sophisticated signal processing techniques are required to tune the thermal parameters C₁, R₁ and R₃ of the reduced-order HTM from its transfer function: $\begin{array}{cc}H\left(s\right)=\frac{{\theta }_r\left(s\right)}{P_s\left(s\right)}=\frac{R_1\left(1+0.65K\right)}{{\mathrm{sR}}_1{C}_1+1}+0.65{\mathrm{KR}}_3& \left(53\right)\end{array}$ where s is the Laplace operator in the s-domain, K is the ratio of rotor I²R losses to stator I²R losses, defined by Equation (41). Adaptive recursive filtering techniques based on an infinite impulse response (IR) filter can be used to estimate the thermal parameters in the reduced-order HTM.

In Stage II, since the reduced-order HTM reasonably correlates the stator winding temperature with the rotor conductor temperature without explicit knowledge of the machine's physical dimensions and construction materials, an adaptive Kalman filter is constructed to track the stator winding temperature based on the reduced-order HTM. A bias term is incorporated in the Kalman filter equations to allow for the variations in power losses at different motor operating points. By using the rotor conductor temperature estimated from the rotor resistance, the variances in the Kalman filter are identified via a correlation method. Once the variances and the bias are determined, the Kalman filter becomes an optimal online stator winding temperature estimator.

In the continuous-time domain, the hybrid thermal model is described by the following equations: $\begin{array}{cc}{C}_1\frac{d{\theta }_s\left(t\right)}{dt}+\frac{{\theta }_s\left(t\right)}{R_1}=P_s\left(t\right)+0.65P_r\left(t\right)+\Delta P_r\left(t\right)& \left(54\right)\\ {\theta }_r\left(t\right)={\theta }_s\left(t\right)+\left[0.65P_r\left(t\right)+\Delta P_r\left(t\right)\right]·R_3& \left(55\right)\end{array}$

When Equations (54) and (55) are transformed into the discrete-time domain with a sampling interval T and written in the discrete Kalman filter form, they become: $\begin{array}{cc}\mathrm{Process}\mathrm{equation}:x\left(n\right)=\mathrm{Ax}\left(n-1\right)+B\left[u\left(n-1\right)+\Delta u\left(n-1\right)\right]+w\left(n\right)& \left(56a\right)\\ \mathrm{Measurement}\mathrm{equation}:y\left(n\right)=\mathrm{Cx}\left(n\right)+D\left[u\left(n\right)+\Delta u\left(n\right)\right]+v\left(n\right)\mathrm{where}x\left(n\right)\mathrm{is}{\theta }_s\left(n\right);A\mathrm{is}{e}^{-\frac{T}{R_1{C}_1}};B\mathrm{is}{R}_1\left(1-e^{-\frac{T}{R_1{C}_1}}\right)·\left[\begin{array}{cc}1& 1\end{array}\right];u\left(n\right){\mathrm{is}\left[P_s\left(n\right)0.65P_r\left(n\right)\right]}^T;\Delta u\left(n\right){\mathrm{is}\left[\begin{array}{cc}1& \Delta P_r\left(n\right)\end{array}\right]}^T;y\left(n\right)\mathrm{is}{\theta }_r\left(n\right);C\mathrm{is}1;D\mathrm{is}{R}_3·\left[\begin{array}{cc}0& 1\end{array}\right].& \left(56b\right)\end{array}$

In Equation (56a), w(n) corresponds to the modeling error of the HTM. In Equation (56b), v(n) designates the measurement noise. They represent the intrinsic uncertainties associated with the HTM and decide the optimal combination of the output from the thermal model-based temperature estimator and the output from the parameter-based temperature estimator. Let {circumflex over (x)}(n|n) denote the best estimate of x(n) at time n given the measurements y(i) for i=1,2, . . . ,n, and {circumflex over (x)}(n|n−1) denote the best estimate of x(n) at time n given the measurements up to n−1. P(n|n) is the error covariance matrix computed from x(n) and {circumflex over (x)}(n|n), and similarly P(n|n−1) computed from x(n) and {circumflex over (x)}(n|n−1). The recursive estimate of x(n), which is the stator winding temperature rise, is: Prediction: {circumflex over (x)}(n|n−1)=A{circumflex over (x)}(n−1|n−1)+B[u(n−1)+Δu(n−1)]  (57a) Innovation: P(n|n−1)=AP(n−1|n−1)A^T+Q_w(n) K(n)=P(n|n−1)C^T[CP(n|n−1)C^T+Q_v(n)]⁻¹  (57b) P(n|n)=[I−K(n)C]P(n|n−1) Correction: {circumflex over (x)}(n|n)={circumflex over (x)}(n|n−1)+K(n){y(n)−C{circumflex over (x)}(n|n−1)−D[u(n)+Δu(n)]}  (57c)

By computing the appropriate Kalman gain K(n), {circumflex over (x)}(n|n) is obtained as an estimate of the stator winding temperature with minimum mean-square error. Rotor conductor temperature from the parameter-based temperature estimator 14 is used as y(n) in each step to provide correction to the estimate of stator winding temperature from the thermal model. Therefore, built on the hybrid thermal model, the adaptive Kalman filter provides an optimal estimate of the stator winding temperature.

The HTM and the subsequent formulation of the Kalman tilter present an adaptive structure in estimating the stator winding temperature without any temperature sensors. The adaptive structure takes into account the bias and the variances through Δu(n), w(n) and v(n) in Equations (56a) and (56b). However, as indicated in Equations (57a) through (57c), Q_w(n), Q_v(n), and Δu(n) must be known before the Kalman filter starts to produce an optimal estimate of the stator winding temperature. This problem is solved in two steps: (1) the hybrid thermal model described in Equations (56a) and (56b) is split into two subsystems: a stochastic subsystem driven purely by noise terms w(n) and v(n), and a deterministic subsystem driven purely by input terms u(n) and Δu(n); and (2) the noise identification technique is applied to the stochastic subsystem to obtain Q_w(n) and Q_v(n), and the input estimation method to the deterministic subsystem to obtain Δu(n). FIG. 11 shows the overall architecture of the algorithm. It illustrates the system parameters, the signal flow and the location of noises and bias terms for the adaptive Kalman filtering system. In FIG. 12, the system parameters A, B, C and D are defined as in Equations (56a) and (56b); Z⁻¹ represents a unit delay in the discrete-time domain; I is an identity matrix. The signal flow is from the left to the right, with u(n) (the power losses) as the input and the y(n) (the rotor temperature) as the output. The noise terms are w(n+1) and v(n), the bias term is Δu(n).

There is an intrinsic modeling error associated with the hybrid thermal model. When formulating the Kalman filter, this modeling error is quantified by an load-independent equivalent noise w(n). Similarly, v(n) is used to indicate the presence of measurement noise in the rotor temperature, which is calculated from the parameter-based temperature estimator. v(n) consists of a load-independent part v₁(n) corresponding to the uncertainty from the measurement devices, and a load-dependent part v₂(n) originating from the slip-dependent rotor resistance estimation algorithm. The noise terms v₁(n) and v₂ (n) are assumed to be uncorrelated due to their load dependencies.

The autocorrelation matrices Q_w(n) and Q_v1(n) capture the statistical properties of w(n) and v₁(n), respectively. They can be identified at rated load, where Δu(n)=0, by observing the property of the innovation process, y(n), defined as: γ(n)=y(n)−C{circumflex over (x)}(n|n−1)−Du(n)  (58)

The autocorrelation function of γ(n) can be used to identify Q_w(n) and Q_v1(n) when the innovation process is not a white Gaussian noise (WGN). Also, the autocorrelation matrix of the load-dependent measurement noise Q_v2(n) is determined from the properties of the rotor resistance estimation algorithm. Because v₁(n) and v₂(n) are uncorrelated, Q_v(n)=Q_v1(n)+Q_v2(n). After the successful identification of Q_w(n) and Q_v(n), the Kalman filter is able to give an optimal estimate of the stator winding temperature at rated load.

When the motor is operating at load levels different from the rated condition, the power losses that traverse the air gap from the rotor to the stator may no longer be fixed at 65% of the total rotor cage losses. This change, denoted by Au, is detected by regression analysis. In the deterministic subsystem, when the rotor temperature differences at two instants, Δθ_r(t₁) and Δθ_r(t₁+Δt), are acquired after the load changes from u to u+Δu, then ΔP_r is found by solving the following equation: $\begin{array}{cc}{R}_1{C}_1\ln \frac{\Delta P_r\left(R_1+R_3\right)-\Delta {\theta }_r\left(t_1\right)}{\Delta P_r\left(R_1+R_3\right)-\Delta {\theta }_r\left(t_1+\Delta t\right)}=\Delta t& \left(59\right)\end{array}$ The rotor temperature difference Δθ_r (t) is defined to be the difference between the real rotor temperature, derived from the rotor resistance, and the rotor temperature predicted from the reduced-order HTM without considering the presence of Δu.

Once Q_w(n), Q_v(n) and Δu have been determined, the adaptive Kalman filter is initiated and produces an optimal estimate of the stator winding temperature at different load levels.

The additional heating created by an unbalanced power supply is taken into account by adapting the model inputs to reflect the additional heating due to the presence of negative sequence current.

The HTM adaptation to an unbalanced supply involves the calculation of additional losses in P_s and P_r introduced by the negative sequence current. Voltage unbalance is a common phenomenon in power distribution systems. A small unbalance in voltage may lead to significant unbalance in current. This can be explained by the induction motor negative sequence equivalent circuit shown in FIG. 12b.

Due to the skin effect, the rotor resistance is a function of supply frequency. In addition, rotor resistance is also dependent on the physical dimensions of rotor bars. It is a common practice to model the relationship between rotor resistance and supply frequency with a linear model, as shown in FIG. 13a according to E. O. Schweitzer and S. E. Zocholl, “Aspects of overcurrent protection for feeders and motors,” in Industrial and Commercial Power Systems Technical Conference, May 1994, pp 245-251. A typical value of R_r at power frequency f_s is three times its dc resistance. Thus during normal operation, the positive sequence rotor resistance in FIG. 12a is R_r≈R_dc, and the negative sequence rotor resistance in FIG. 12b is R′_r≈5R_dc. FIG. 13b interprets the same physical phenomenon as FIG. 13a, but in terms of rotor resistance and slip.

The current unbalance introduces excessive rotor heating associated with the negative sequence current component. FIG. 14 explains the rotor heating due to the negative sequence current component according to B. N. Gafford, W. C. Duesterhoeft, and C. C. Mosher, “Heating of induction motors on unbalanced voltages,” AEEE Transactions Power Apparatus and Systems, vol. 78, pp. 282-288, 1959. The motor can be considered to have two tandem virtual rotors. One is driven only by the positive sequence current I₁, which is symmetrical and balanced. The other is driven only by counter-rotating negative sequence current I₂, directly related to unbalanced current; and producing torque in the reverse direction. The motor is designed to produce torque only from the positive sequence current component, not from the negative sequence current component, and thus the reverse torque produced by I₂ in the second rotor works against the main action of the motor. Because the negative work caused by I₂ stays within the rotor, it is completely transformed into heat and therefore has a far more significant effect on the rotor heating than I₁.

Based on the above analysis, the inputs to the HTM in FIG. 7 and FIG. 10, P_s and P_r, should be modified to P′_s and P′_r to accommodate the additional heating produced on the stator and rotor side by negative sequence current as follows: P′_s=P_s+P⁻_s, P′_r=P_r+P_r⁻  (60) where P_s and P_r are losses under normal balanced supply; and P_s⁻ and P_r⁻ are additional losses produced by negative sequence current. P_s⁻ and P_r⁻ are determined by: $\begin{array}{cc}\{\begin{array}{c}{P}_s^{-}={{\tilde{I}}_s^{-}}^2{R}_s\\ P_r^{-}={{\tilde{I}}_r^{-}}^2\frac{R_r^{\prime }}{2-s}\end{array}& \left(61\right)\end{array}$

Claims

1. A method of estimating the stator winding temperature of an induction motor having a rotor and a stator, said method comprising estimating the stator temperature from the motor current and the angular speed of the rotor using an online hybrid thermal model.

2. The method of claim 1 further comprising estimating motor inductances from the stator resistance and samples of the motor terminal voltage and current, estimating the rotor resistance from samples of the motor terminal voltage and current, the angular speed of the rotor, and said estimated inductances, estimating a first rotor temperature from said estimated rotor resistance, and identifying thermal capacity and thermal impedance parameters of said hybrid thermal model online based on said estimated rotor temperature.

3. The method of claim 2 further comprising estimating a second rotor temperature using said hybrid thermal model, comparing said first and second estimated rotor temperatures and producing an error signal representing the difference between said first and second estimated rotor temperatures, and adapting the parameters of said hybrid thermal model online in response to said error signal when said error signal exceeds a predetermined threshold.

4. The method of claim 2 wherein said estimated motor inductances include (a) the stator leakage inductance, the rotor leakage inductance and the mutual inductance of said motor, or (b) the stator self inductance and the stator transient inductance.

5. The method of claim 2 wherein the positive sequence, fundamental components of said samples of the motor terminal voltage and current values are used in estimating said rotor resistance.

6. The method of claim 2 wherein said hybrid thermal model uses the angular speed of said rotor, samples of the motor terminal current, and said error signal representing the difference between said first and second estimated rotor temperatures.

7. The method of claim 1 which includes using said estimated stator temperature as an indication of the overload of said motor.

8. The method of claim 1 wherein initial values of selected parameters of said hybrid thermal model are determined from parameter values available from the motor manufacturer.

9. The method of claim 8 wherein said parameter values available from the motor manufacturer include at least one parameter selected from the group consisting of full load current, service factor, trip class, stator resistance, stall time, steady state temperature rise above ambient at full load condition, and NEMA machine design class for said motor.

10. The method of claim 1 wherein said hybrid thermal model approximates thermal characteristics of said rotor and stator.

11. The method of claim 1 wherein said hybrid thermal model is a full-order or reduced-order hybrid thermal model.

12. The method of claim 1 wherein said hybrid thermal model includes thermal capacitance values representing the thermal capacity of the stator core, windings and frame of said motor, and the thermal capacity of the rotor core, conductors and shaft.

13. The method of claim 1 wherein said hybrid thermal model includes thermal resistance values representing the heat dissipation capabilities of said stator and rotor, and the heat transfer from said rotor to said stator across an air gap between said stator and rotor.

14. A method of estimating the rotor resistance of an induction motor having a rotor and a stator, said method comprising estimating motor inductances from the stator resistance and samples of the motor terminal voltage and current, and estimating the rotor resistance from samples of the motor terminal voltage and current, the angular speed of the rotor, and said estimated inductances.

15. The method of claim 14 wherein said samples of said motor terminal voltage and current used in estimating said motor inductances are obtained by measuring the motor terminal voltage and current at a plurality of different load points, and computing the positive sequence, fundamental components of the measured voltage and current values.

16. The method of claim 14 wherein said samples of said motor terminal voltage and current used in estimating said rotor resistance are obtained by computing the positive sequence, fundamental components of the measured voltage and current values.

17. The method of claim 14 wherein said estimating of rotor resistance includes obtaining the slip of the motor, computing the ratio of rotor resistance to slip, and multiplying said ratio by said slip to obtain rotor resistance.

18. The method of claim 14 further comprising estimating the rotor temperature from said estimated rotor resistance.

19. The method of claim 14 further comprising estimating the rotor resistance from samples of the motor terminal voltage and current, the angular speed of the rotor, and said estimated inductances.

20. The method of claim 14 wherein said estimating of motor inductances includes forming equations in preselected parameters for each of a plurality of said load points, combining the equations for each operating point into a matrix equation, solving said matrix for said preselected parameters, and computing motor inductances from said preselected parameters.

21. A method of protecting an induction motor during running overloads, said motor having a rotor and a stator, said method comprising connecting the motor to an overload protection relay that can be tripped to interrupt power to the motor in the event of an overload, tracking the stator winding temperature of the motor during running overloads with a hybrid thermal model by online adjustment in the said hybrid thermal model, and tripping the overload protection relay in response to a predetermined thermal condition represented by the stator winding temperature tracked with said hybrid thermal model.

22. The method of claim 21 wherein said running overload is a condition in which the stator current is in the range of from about 100% to about 200% of the rated current of said motor.

23. The method of claim 21 wherein said hybrid thermal model uses the resistance of said rotor as an indicator of rotor temperature and thus of the thermal operating condition of said motor.

24. The method of claim 21 wherein said hybrid thermal model includes thermal resistances and capacitances of said rotor and stator.

25. The method of claim 21 wherein said hybrid thermal model includes adjustable thermal parameters, and further comprising tuning said thermal parameters of said hybrid thermal model to adapt said thermal parameters to the cooling capability of said motor.

26. The method of claim 25 wherein said tuning is effected when said hybrid thermal model does not match the actual thermal operating condition of said motor.

27. The method of claim 26 wherein the match between said hybrid thermal model and the actual thermal operating condition of said motor is measured by a function of the difference between a rotor temperature estimated using said hybrid thermal model and a rotor temperature estimated from the estimated rotor resistance, and said tuning is effected when said function exceeds a predetermined threshold.

28. A system for estimating the stator winding temperature of an induction motor having a rotor and a stator, said system comprising a microprocessor programmed to estimate the stator winding temperature from the motor current and the angular speed of the rotor using an online hybrid thermal model, and a current sampler responsive to the motor terminal current and supplying said microprocessor with samples of said current.

29. The system of claim 28 wherein said microprocessor is programmed to estimate motor inductances from the stator resistance and samples of the motor terminal voltage and current, estimate the rotor resistance from samples of the motor terminal voltage and current, the angular speed of the rotor, and said estimated inductances, estimate a first rotor temperature from said estimated rotor resistance, estimate a second rotor temperature from the motor current and the angular speed of the rotor, compare said first and second estimated rotor temperatures and producing an error signal representing a function of the difference between said first and second estimated rotor temperatures, and adapt the parameters of said hybrid thermal model in response to said error signal.

30. A system for estimating the rotor temperature of an induction motor having a rotor and a stator, said system comprising a sampler responsive to the motor terminal current and voltage and producing samples of said current and voltage, a microprocessor receiving said samples and programmed to estimate motor inductances from the stator resistance and said samples of the motor terminal voltage and current, estimate the rotor resistance from said samples of the motor terminal voltage and current, the angular speed of the rotor, and said estimated inductances, and estimate the rotor temperature from said estimated rotor resistance.

31. A system for protecting an induction motor during a running overloads, said motor having a rotor and a stator, said system comprising an overload protection relay connected to said motor to interrupt power to the motor in the event of an overload, a sampler responsive to the motor terminal current and voltage and producing samples of said current and voltage, a microprocessor receiving said samples and the stator resistance, and programmed to track the stator winding temperature of the motor during running overloads with a hybrid thermal model by online adjustment in the said hybrid thermal model, and to generate a signal to trip said overload protection relay in response to a predetermined thermal condition represented by the stator winding temperature tracked with said hybrid thermal model.