Real-Time Implementation of Extended Kalman Filter Observer With Improved Speed Estimation for Sensorless Control

This work presents an investigation on Improved Extended Kalman Filter (IEKF) performance for induction motor drive without a speed sensor. The performance of a direct sensorless vector-controlled system through simulation and experimental work is tested. The proposed observer focuses on estimating rotor flux and mechanical speed of rotor from the stationary axis components. Extended Kalman Filters’ estimation performance depends on the system matrix’s proper value (<inline-formula> <tex-math notation="LaTeX">$Q$ </tex-math></inline-formula>) and measurement error matrix (<inline-formula> <tex-math notation="LaTeX">$R$ </tex-math></inline-formula>). These matrices are assumed to be persistent and are calculated by the trial-and-error method. But, the operating environment affects these matrix values. They must be updated based on the prevailing operating conditions to get high speed and accurate estimates. The values of Q and R in the Improved EKF (IEKF) algorithm are obtained using the genetic algorithm. Besides, IEKF is incorporated to reduce in computational burden for real-time applications.


I. INTRODUCTION
For precise control of Induction Motors (IMs), values of speed, torque, rotor & stator fluxes have to be accurately estimated. In the past decade, model-based methods of IM using notations of space vector were developed. The information needed for the model-based approach is obtained by measuring the voltage and current at machine terminals. But, these model-based methods were observed to be very sensitive to machine parameters. Model-based methods need past information on the motor's electrical parameters and characteristics. Different IM parameters are measured at standstill and speed, and the current controller is tuned accordingly; the machine changes parameters during normal operation. The parameters that change during normal The associate editor coordinating the review of this manuscript and approving it for publication was Zhuang Xu .
operation are due to thermal conditions. There are changes in rotor and stator resistances; the inductance parameters depend on the flux level's value. Also, the values of leakage co-efficient change as flux and load vary. Over the last few years, there is an increased interest in applying the speed sensorless Field Oriented Control (FOC) method to highperformance IM applications. It bases this on computing the speed using machine via's parameters, the instantaneous value of current and voltage in stator [1], [2]. Also, few Adaptive Observers, namely Luenberger observer/Extended Kalman Filter (EKF) [3], [4] provide accurate speed estimates under de-tuned operating conditions. Comprehensive reviews on the sensorless drive via model-based computation methods failed to assure the machine's stability [5]. The drive systems are susceptible to inaccuracy and variations in motor equivalent circuit parameters. This may induce errors that degrade the speed holding characteristics. It intensifies the problem, due to unbalanced motor winding or any DC offset present in current controllers. Unwanted DC offset can be eradicated by auto-calibration of DC link current. [6], [7].
Various methods have been proposed over the last few decades for speed estimation and control of IM. Conventional speed estimation techniques include the MRAS [8], derivatives of rotor flux [9], [10], voltages in stator [11], a modified version of stator model, full order observer [11], [12], unscented extended Kalman filter [13], reduced-order non-linear observer [14], [15], Extended Kalman filter observer assisted with fuzzy optimization [16], sliding mode observer [17], [18]. Few other speed estimators that don't depend on measured values of voltage and current are Artificial intelligent (AI) based neural networks [19], [20], harmonic rotor slotting [21], [22], high-frequency signal injection [21]- [23]. These investigations [9]- [23] involve closed-loop strategies that use speed estimation techniques, considering the IM model to be linear. However, IM is a non-linear machine and requires accurate parameter estimation techniques and a robust speed control method. The advancements in technology, fast computing processors have led to robust speed estimation schemes. These works let to the development of the Bi-Input Extended Kalman Filter estimation method. But these methods have been shown to use more time to determine the noise covariance matrix's proper values [24]- [29]. Neuro-fuzzy techniques were applied for performance enhancement of the drive [28]- [30]. The work reported involves AI techniques using a state observer, which includes an adaptive intelligent technique using human knowledge and fuzzy-based systems. In [30], [31], new machine drive techniques are explained through other estimation methods to measure very low-speed. A shaft encoder is used to calculate both speed and load torque disturbances. The values of temperature and uncertainties in frequency are also required for accurate flux and velocity computation in sensorless control [32]- [35]. Lyapunov's theorem has proved the stability of the adaptive schemes used for speed estimation, and results obtained through simulation are analyzed [36]. The estimation of stator resistance using EKF by realizing Field Programming Gate Array (FPGA) and IM operation at very low and near zero speed areas [37]. They introduce the implementation of EKF with a fading term in the covariance matrix and hence tune the gain matrix [38]. Investigations on optimizing the performance of EKF through speed and torque fitness functions for direct torque-controlled machines are provided in [39]. For improving the robustness of the speed estimator and control system against parameter uncertainties, and improved EKF scheme is developed that rests on the correct choice of 'System and Measurement covariance matrices'.
The estimation performance of Standard EKF is observed to deteriorate at low speed, transients, and under variations in equivalent circuit parameters. A real-coded integer Genetic Algorithm with improved precision using a new concept of the chromosome was reported in [40]. It optimizes the EKF with different values of fitness functions for speed-sensorless control of IM. The work has reported real-time experiments to state the best fitness function for accurate motion control [41]. Differential Evolution Algorithm is implemented to optimize the covariance matrices of reduced-order Kalman Filter in offline mode and realize the sensorless control of induction motors [42]. Also, the use of differential evolution algorithm for optimizing EKF is studied in [43]. This method, however, cannot address the sensitivity of the observer to stator resistance variations. Performance of GA-optimized EKF of constant V/Hz controller considering variation in parameters is proposed and implemented [44]. As speed sensorless control applications to induction motors are predominantly increasing, especially in EV applications, IM's dynamic performance for several speed profiles and different load torque disturbance conditions is to be achieved. The proposed research work intends to identify IM parameters considering the associated non-linearities and achieve improved speed estimation and control over an extensive speed zone. The effect of stator resistance and sensitivity of observer to variations in resistance of stator [43] is investigated.
In this research work, an improved Extended Kalman filter for control of induction motor in sensorless mode is performed. The values of the system (Q) and measurement (R) matrices are updated in Improved EKF (IEKF) algorithm using the genetic algorithm. The performance of sensorless vector-controlled drive systems is tested through simulation and experimental work. Earlier research investigations have focused on the estimation of rotor speed under different constraints. However, there is a need to address the vector control scheme for achieving good dynamic performance. Hence, in this research, the Bus Clamped Space Vector Modulation scheme is implemented, which has improved the IM drive system's complete performance. Re-configurability of Digital Signal Processing (DSP) devices gives researchers the chance to test and validate a novel controller; which does not damage the actual system through co-simulation, also known as Rapid Control Prototyping (RCP). The present work uses the Code Composer Studio (CCS) environment [45] to establish and validate an adaptive Improved Extended Kalman Filter (IEKF) speed estimator based on a Real-Coded-Genetic Algorithm (RC-GA-EKF). The research work models the system to improve the direct rotor field-oriented or the vectorcontrolled speed sensorless induction motor.
The paper is organized as: Section II presents the parameter estimation using an extended Kalman filter algorithm with a modified machine model. Section III describes the basic idea of Tuning Q & R Matrices Using Real-Coded-Genetic Algorithm (RC-GA). Section IV presents the hardware configuration of the process of EKF using RC-GA. The experimental and simulation results with suitable discussion are shown in section V. Finally, conclusions in section VI.
EKF is considered a recursive estimator with optimum stochastic characteristics. A full order state observer is used in the state-space model to estimate the components viz., the flux of rotor, rotor speed, and stator current. This approach significantly reduces the execution time and simplifies the tuning of the covariance matrix. The fundamental steps to speed estimation using EKF include discretizing the machine model. For determining state and noise covariance matrices, and implementing and tuning the discretized EKF algorithm using real coded Genetic Algorithm to get proper rotor speed.

A. DISCRETISED STATE SPACE MODEL OF INDUCTION MOTOR
There are 4 state variables: stator currents in d & q axes (i ds , i qs ) and the rotor fluxes ( dr , qr ) in the dynamic model of 3-phase IM. Including the rotor speed ω r as an additional variable a modified model of IM results [40]. The discretized extended model is as got [39]. where, for an IM, the values of R r , R s , L r , L s , L m, and ω r are constant. Therefore, matrices of state vector input and output are taken as constant.

B. EXTENDED KALMAN FILTER ALGORITHM
With the modified model of induction motor, it is noticed that the state-space model is not linear. This is because it has self-product terms of the state variables. The method's essence stems from the linearization of the non-linear model, so use the system's first derivatives and written the filter structure according to this. The EKF is a recursive procedure that helps produce estimations of unknown variables using a series of measurements observed over time more precisely [40]. The EKF uses a non-linear model and observations and assumes error propagation to be linear. The other non-linear parameters are linearized. In the proposed method, we have attempted to develop the estimator to give improved performance using d-q axes stator current and flux values. The rotor speed is estimated via the reference speed added as a state variable and further tuning the EKF algorithm using a real coded Genetic Algorithm. The estimates are got by EKF as stated below: • State Vector Prediction This vector is predicted with a sampling interval of (n + 1). The value is obtained from input vector U(n), state vector, matrices A, and B.
The values of matricesX (n), U (n), A, B can be got from (1), • Estimation of the Prediction Covariance matrix The covariance matrix has the order of 5 × 5. A gradient matrix f can represent the terms in (3) as, where, X = X * (n + 1) • Computing Kalman filter gain value The order of gain matrix of the filter is taken as 2 × 5, and it is calculated as: where parameter CX is another gradient matrix represented by h as follows.
where X = X * (n + 1) • Estimating state vector Correct estimation of state estimator is done at the time interval (n + 1) and is done as given, • Error covariance matrix This matrix is attained as, • Updating parameters is done as follows: In (2)-(9),^refers to the estimated value, and * refers to the prediction parameter. Updated the algorithm with the values arrived through the (2)-(9) and develops covariance matrix that is modified upon iterations, and accurate estimates of the states are obtained by a trial-and-error method. Getting a faster response through correct tuning of Q and R matrices. An improvisation in the observer's performance is done in the present work using a real coded genetic algorithm that tunes matrices Q and R and gets accurate and quicker speed estimates.

III. TUNING OF Q AND R MATRICES USING REAL-CODED GENETIC ALGORITHM
In this study, the EKF algorithm is taken as an improved current model-based observer where it can provide very lowspeed and high-speed zone operation. But, the values of filters affect the performance of estimated state vectors. Hence, in this work, a Real Coded Genetic Algorithm (RC-GA) optimizes the EKF observer and, in turn, helps optimize the controller's multi-dimensional performance.The primary aim is the minimization of error which is the mean squared value of the speed estimate E 1 (x) which is given by, where n is the sampling time, for improved speed estimate through EKF, three of its initializes matrices. It will update the covariance matrix P at each sample period of EKF iterations (10). Hence, only Q and R are optimized. Thus, RC-GA is employed, and the corresponding algorithm flow diagram is as in Fig. 1. The RC-GA for EKF (RC-GA-EKF) is implemented using MATLAB. For tuning the matrices Q and R of IEKF for vector control, parameters of RC-GA are set as follows: • The initial population size : 100 • The maximum number of generations : 20 • The probability of crossover : 0.8 • The mutation probability : 0.1 • The initial range of real coded strings : [.1; 5] • The performance measure : MSE obtained from the actual speed of the motor and estimated speed. The selection of RC-GA parameters used to identify the vector P and hence tune Q and R of IEKF is as given above. The steps followed for implementing the RC-GA algorithm is given: 1) The five diagonal elements which are the decision variables representing the population index are (q 11 , q 22 , q 33 , q 44 , and q 55 ) of Q matrix, and the elements (r 11 , r 22 ) are a diagonal representation of the R matrix. These are taken as a long string and coded further than shown below: 2) The non-domination rule is applied and the initial population of variables is randomly generated.
3) Every variable is converted into diagonal terms of the two matrices Q & R of EKF observer at each time of iteration. Then the diagonal components are sent to the RC-GA-EKF speed observer, and hence the fitness function is obtained. This method gives the fitness evaluation criterion and thus provides solutions for fitness functions.
4) The operators of crossover and mutation are used to get the off-springs. The present population is sorted along with the current off-spring population based on the rule of nondomination. This gives the best individual P op (population size), which is selected for further operations. 5) Steps 3 & 4 are repeated till the maximum iterations are reached. MATLAB coding is applied to optimize RC-GA EKF observer thus, enhancing the performance.
The closed-loop controller comprises the training set: the voltages and currents of the stator (v d , v q , i sd , and i sq ). This training set makes the EKF observer's input and takes the obtained rotor speed as the target set for getting the RC-GA fitness evaluation. Around 40000 sets of vector generations, training profiles are generated. These profiles have been seen to provide wide speed range operation from low to high speed both during the transient condition and steady-state conditions.
A comparative analysis is made between the trial-and-error method in EKF and RC-GA using simulation and experimental research. Also, this article presents the hardware configuration for implementing the technique above in section IV.

IV. HARDWARE CONFIGURATION FOR REAL-TIME IMPLEMENTATION OF EKF ON SENSORLESS VECTOR CONTROLLED IM
The TMS320F28335 connects the improved EKF observer and process implementation on target DSP [45] for the hardware test. First, the observer is expressed in mathematical form through the CCS toolbox and then it is optimized via the processor and later is transferred to the hardware platform This step occurs in two parts: • First, the EKF observer is loaded in the target DSP and the IM model is described in Simulink.
• Second, the IM model comprises the sensorless fieldoriented control strategy.
Building the hardware requires a few initial processes.
• The model is built, arranged, simplified, and tested in a Simulink environment. This prevents timing errors. Implementing the observer in this work has 3 subblocks: 1) current estimator, 2) rotor flux estimatorrotor flux is estimated from voltage model of IM and 3) parameter estimation as in Fig. 2.
• Fig. 2 represents a framework of the planned estimator on the RCP platform or the so-called hardware co-simulation platform.
• Using the CCS toolbox in the target DSP of Simulink is the next step. Through this the hardware platform for the estimator is set up. An environment of convenience and precision is created to select the data size and allow the same signal to traverse between the hardware blocks built. The inner information of variables and precise computation are two parameters that must be considered achieving the required data size. Time analysis is performed and the instant at which the improved EKF algorithm gets synchronized with the hardware at various stages of the design process is noted.
• In the previous step before synchronization, the EKF algorithm is verified along with the design. In this stage, simulation along with time analysis and co-simulation with hardware is performed. This process validates the performance of the design parameters on the target DSP board. TMS320F28335 is used for the co-simulation function. A code generator present will synthesize to give a DSP program in C file from the hardware. A new JTAG is used to replace the previously built design (Fig. 2). The block obtained after this is now put in the design. Further, implementation of hardware through the board connected to the PC is performed. In a hardware implementation, the JTAG has a value at one of the input posts and this JTAG now sends the data to the hardware. The hardware provides the estimates and these values are read back to simulation through USB. The output port further functions converting the data type to simulink readable format and further feeds this data to the vector control module (Fig. 2).
To exploit the advantages of the proposed research work, both simulation analysis and experimental analysis are done for diverse working environments of the drive and the estimator, and discussion of results work in section V.

V. RESULTS AND DISCUSSIONS
This section demonstrates the dynamic response of the speed sensorless control scheme of 2 HP IM using RC-GA-EKF.

A. SIMULATION RESULTS
Presented results of the speed response for various load changes. Recorded typical measurements of speed behavior to show the efficacy of the estimation scheme developed in this work. The results show the quality of the speed sensorless control, and relative speed errors. Simulated the sensorless control of IM using GA based EKF on Simulink platform of MATLAB software and study the performance of the observer under different operating conditions. Realization of the model is done using MATLAB language file. Inserts a subsystem into the system that contains the RC-GA-EKF, which is further made into the block as S-function. The RC-GA-EKF in the drive system is implemented for  Pulse Width Modulated inverter driven IM drive without the need of an additional speed sensor. Fig. 3 depicts the speed response of IM at No-load, over a speed range of [0 100 139 0] rad/s. Observed from Fig. 3 that the estimated speed follows the set speed at every step change in speed (the legends in red and blue colors are aligning one over the other). The speed response got in the simulation is steady with minor errors (in the band of [−0.001 to +0.001] as in Fig. 3. It is witnessed that an improved EKF based estimation mechanism overcomes drift and distortion issues that were noticed in MRAS [46], [47] based speed adaptation scheme. An accurate speed estimate at above 100 rad/s (i.e., near rated speed of 139 rad/s) is achieved, as observed in Fig. 3. The performance of the improved EKF at low speeds is entirely satisfactory and the speed. The improved EKF state observer gives better speed estimates even at a low-speed range of [0 30 50] rad/s as shown in Fig. 4, compared to the similar speed range estimates got using the MRAS model [40]. However, the simulation results corresponding to a load torque condition of 10 Nm are presented in this part of the work. The results for different low and high speed values are shown in Fig. 5 and Fig. 6. Fig. 5, shows that the speed estimation obtained is matching with the set speed with an applied load of 10 Nm at an instant of 1 second in a low-speed range which  is between 30 and 50 rad/s. The simulation results for a speed range of [0 10 50 100 139] rad/s for 10 Nm load are shown in Fig. 6. It is observed that performing improved EKF is satisfactory over a wide range of speeds. Thus, improved EKF provides a better speed estimate.
To find out the robustness of improved EKF state observer, performed simulation analysis for the effect of 'Current noise' and 'Machine parameter (R s ) changes'. Simulink model of improved EKF observer subjected to noise in stator current and a 10% increase in the value of stator resistance. The noise signal (randomly generated) was injected to show the stator currents of IM in Fig. 7 [40]. The characteristic of the noise is ''zero mean, white and Gaussian''. The estimator's response for a set speed of 30 rad/s is recorded, when a load torque of 10 Nm is applied for 10% variation in the stator resistance of the IM. The nominal value of stator resistance R s of IM is 4.57 . 10% decrease in the stator resistance applies to performing improved EKF state observer. The result got for a set speed value of 30 rad/s and 139 rad/s Figs. 8 and 9, respectively. Observed that speed estimate tracks the set speed with minimal deviations (± 0.3 rad/s). As shown in Figs. 8 and 9, for low-speed and high-speed range, the computed value of speed is not affected drastically by noise signal injected. From simulation VOLUME 9, 2021  analysis performed, find it that there are almost no differences between the set and the estimated values of speed. Thus, performing the improved EKF model is stable even under the presence of injected noise, variations in stator resistance R s, and load torque variations. Therefore, the Genetic Algorithm used to tune the state matrices of the EKF state observer, provides a better speed estimate at a widespread speed range, justifying the efficacy of an improved EKF observer. Simulation results presented correspond to a load of 10 Nm and stator resistance variation of 10% of its nominal value. Further, the performing improved EKF is analyzed for load torque variations of 10%, 50%, and 100% of the nominal value (10 Nm) at a common near the rated speed of 139 rad/s. Table 1 show the Mean squared error of speed estimate values got using the 'Trial-and-error' method and improved EKF state estimator. Observed in Table 1 that the Mean squared error of speed estimate recorded for different load torque variations is substantially low compared to the Mean squared error of speed estimate got through the 'Trial-and-error' process. Therefore, the improved EKF speed estimator is proved to perform better than the 'Trial-and-error' scheme of speed estimation. Also, performing improved EKF for stator resistance R s variation of 10%, 20%, and 50% are analyzed. The nominal value of stator resistance R s of IM is 4.57 . For a set speed of 139 rad/s (near rated speed) of IM, the mean squared error of speed estimate got upon 10%, 20%, and 50%    Table 2. It is observed from the values stated in Table 2 that the mean squared error of speed estimate got through the 'Trial-and-error' method is more sensitive to stator resistance variation than those obtained by improved EKF state observer.

B. SIMULATION ANALYSIS OF RC-GA-EKF STATE OBSERVER APPLIED TO SENSORLESS VECTOR CONTROL OF IM DRIVE
The simulation analysis of sensorless Field Oriented Control (FOC) of the IM using RC-GA-EKF state observer is performed. Sample time of 10e-6s is selected. Fig. 10 shows the generic diagram of a vector control scheme in sensorless mode using an improved EKF state estimator. For a good performance of speed sensorless system accurate rotor speed and flux estimation is necessary. This is because any error in flux estimation can affect the estimated speed and vice versa. Hence, in the presented work, improved EKF is employed only for estimating speed. The flux estimator is used extensively for tracking the position of the rotor-flux linkage. In this subsection, discuss sensorless FOC of the IM  subsequent results. The four PI controllers (flux controller, speed controller, and two current controllers) in the control system are tuned individually by adjusting the proportional constant K p and the Integral constant K I .
The gain values of the four PI controllers through the 'Trial-and-error' process. Tuning of these controllers enables the controller error signal to zero. Initially, the inner current loops are tuned (having speed and flux controllers disabled) with constant values of their reference inputs. Later, speed and flux PI controllers are tuned. Table 3 shows the value of K p and K I used in this part of the work for the four PI controllers [40]. After tuning the current speed and flux controllers appropriately with the values stated in Table 3, simulation analysis of sensorless vector control for IM is made. The corresponding responses got concerning phase voltage (V ab ) in volts, Phase Currents (I a , I b , I c ) in amperes, rotor speed in rad/s, electromagnetic torque in Nm, is as in Fig. 11. Fig. 11 depicts vector control of IM at no load in the sensorless mode for a value of speed 20 rad/s using an improved state observer. It is realized that the estimated speed tracks the reference speed, and the speed estimate is smooth after initial oscillations die down. The initial drift and distortions are because of integrators in PI controllers used for speed control and current control. The performance of the sensorless drive using an improved EKF speed observer is satisfactory when operated under a very low-speed range, between 10 rad/s-20 rad/s. Attributes the higher amplitude of the transients in stator currents to the inertia of the motor. Observed these amplitudes to decrease drastically by the time the steady-state is reached, as shown in Fig. 11. We make the simulation runs for one load condition of 10 Nm common for two-speed ranges. The low-speed range of [30], [50] rad/s and high-speed 139 rad/s are considered. The responses got in this context are shown in Figs. 12 and 13, respectively. The results got to imply that there are few deviations among the set and estimated speed during load and speed changes in both cases. The conditions realized with such reduced deviations are deemed as a stable operation of sensorless vector control of the IM. It is seen from simulation results, improved EKF overcomes the problem of the MRAS speed estimation scheme for achieving shorter settling time and tuning of PI controller gain value [46], [47]. Therefore, in the present research work, an improved EKF state observer provides accurate speed estimation at a higher speed range. Even at low speeds, even a slight variation in stator resistance does not affect the performance of the drive system. Hence, for high and low-speed estimates in connection with sensorless control of IM, improved EKF stably responds. Table 4 shows the response of speed sensorless control and relative speed error at different speeds. The measurements for four reference speed settings. The values stated in Table 4 VOLUME 9, 2021   show the efficacy (speed control errors being close to zero) of the control scheme using improved EKF. Thus, accurate speed estimate, sensing, and control are got at an extended speed range, starting from high values of speed down to very low values of speed. Further, the comparative values of speed recordings under no-load condition got through simulation analysis of improved EKF and Model Reference Adaptive System (MRAS) speed estimator are stated in Table 5. The values specified in Table 5 show the capability of improved EKF speed estimator in estimating the rotor speed of IM over a wide speed range demanded. Whereas, the speed estimates using MRAS for high-speed demand like 120 rad/s and 139 rad/s were not at all achieved. Further, comparative analysis of speed estimates got by earlier investigators and made those got in the present research work. The results got under No-load conditions for different ratings of IM and different speed estimation techniques reported in the literature are compared and presented in Table 6. It is observed   that satisfactory speed estimation is got using improved EKF developed in the present research work. Also, the advantage of the improved EKF speed estimator lies because it does not require 'Change of algorithms' or 'Adjustments of gains' or 'Changing parameter signals' for convergence at a steady state.
The arrived RC-GA-EKF speed estimation algorithm offers a more generalized and effective solution for speed estimation of speed sensorless IM in a wider speed zone.

C. EXPERIMENTATION RESULTS
The performance of improved EKF is tested using hardware setup built as in Fig. 14 for different scenarios of variations in load torque, changes in speed command, no-load conditions, and the worst case of zero speed state [14]. Hardware-in-loop is done by inserting the observer in the controller scheme in the Simulink environment. A generic block diagram depicting the implementation of sensorless vector control using 'hardware co-simulation is shown in Fig. 15. It connects  Obtained the data traces of motor currents and speed responses of IM. Checked the suggested estimator under step changes of speed and different load conditions. Results obtained emphasize that when variations are introduced on the parameters estimated, it is observed that the improved EKF gives a good estimation for rotor speed at its rated value of 139 rad/s as shown in Fig. 16. Fig. 17 shows the ability of the improved EKF to sustain its performance at near-zero speed with 80% of load applied to the drive system. From   the experimental investigation's values of speed at near-zero speed (100 rpm) under load variations are recorded and the corresponding speed error obtained is stated in Table 10. Also, the rotor flux value recorded through the experiment is shown in Fig. 18. The value of the rotor flux remains unaffected even with load and parameter changes. The speed response recorded exhibits good dynamic response both under no-load and loaded conditions. The time response of getting from the speed results is shown in Table 7. Further, compared the time response values stated in Table 8 with the time response got for speed characteristics using the V/f control scheme.
We design both control schemes to operate in the RCP platform for a speed of 1400 rpm under a load current of 1.8 A.   From the responses tabulated; observed that the sensorless vector control scheme provides a better response than the V/f control scheme. Table 9 shows the speed recordings measured for a set speed of 600 rpm under No-load conditions and different load conditions. The rated current of IM is 3.3 A, loading the IM up to ∼80% of its rated value (i.e., 2.6 A). We have carried similar investigations out for different set speed values. Table 10 shows the speed recordings made for different speed ranges and load conditions. From the values stated in Table 10, we observe the IM can operate at various speed ranges up to a near rated speed of 1400 rpm with the sensorless vector control scheme. The average speed control error is well within the tolerance band of ± 2%. This shows better speed regulation. Observed that% error of speed of response decreases as the motor speed increases and is <1 at near rated speed. We have presented the comparison of experimental results and simulation results to validate the efficacy of the closed-loop control, and the same is stated in Fig. 19. We record the simulation and experimental results  for random speed demands of 600 rpm, 800 rpm, 1000 rpm, and 1400 rpm under loaded conditions. The simulation results stand validated upon the experimental results proving the efficacy of the control system designed on the RCP platform. Table 11 presents the percentage difference in speed got between simulation results and experimental results for different speed ranges under loaded conditions. Observed that the % error in speed values well within the tolerance band of ± 2%.
The results from simulation and experimentation find out that the IM drive-in closed loop operates satisfactorily for different speed and load torque zones. It also reduces the error in speed with load changes and mismatches in parameters of the IM. The transient and steady-state performance of a drive control system is improved. It is observed that results obtained from hardware show fluctuations when compared with results of the simulation environment. This situation depends on the value of the data size chosen. But the fluctuation in the estimated parameter does not have any adverse effect on the dynamic response which does not affect the speed (Fig. 16), and rotor flux (Fig. 18). Neglecting the fluctuations caused during the operation of the drive, results shown in Fig. 17, reports the ability of improved EKF speed observer to sustain its performance at very low speed. Tests validated the proposed work for parameter variations concerning a change in stator resistance and load conditions. We have tested the performance of the algorithm through simulation and experimental results for 50% variations in stator resistance and 100% load variation (through simulation) and 80% load variation through experimental investigations. Under these conditions, the proposed work has tracked speed at all variations in parameters in the limit described above and has proved less sensitive to any fluctuations in load or parameter variations.

VI. CONCLUSION
In this work, a real-time investigation of a non-linear state observer is done to improve the IM sensorless control in motion control applications by properly estimating the speed of the rotor under changing conditions. Unlike previous studies, where only one of the covariance matrices tuned, the proposed strategy can tune both Q and R matrices and optimize the components of speed and rotor flux. The real coded genetic algorithm based EKF speed observer improves the performance of the sensorless drive system and optimizes the state estimate values the behavior of the system in different zones of speed and load is observed. Results obtained are promising and show that the algorithm has the potential to provide optimal and accurate speed estimates in both low and high-speed ranges.
UMASHANKAR SUBRAMANIAM (Senior Member, IEEE) worked as an Associate Professor and the Head of VIT Vellore and a Senior Research and Development and a Senior Application Engineer in power electronics, renewable energy and electrical drives. He is currently with the Renewable Energy Laboratory, College of Engineering, Prince Sultan University, Saudi Arabia. He has over 15 years of teaching, research and industrial research and development experience. Under his guidance, 24 master's students and more than 25 bachelor's students completed the senior design project work. Also, six Ph.D. scholars completed Ph.D. thesis as a Research Associate. He is also involved in collaborative research projects with various international and national level organizations and research institutions. He has published over 250 research articles in national and international journals and conferences. He has also authored/coauthored/contributed 12 books/chapters and 12 technical articles on power electronics applications in renewable energy and allied areas. He is a member of IACSIT, IDES, and ISTE.