Iron Loss Modelling with Sensorless Predictive Control of PMBLDC Motor Drive for Electric Vehicle Application

—Permanent magnet brushless DC motor (PM- BLDCM) is more suitable for electric vehicle (EV) with higher torque and driving performance. A suitable loss modelling of PMBLDCM with proper analysis can improve the control stability at different road surface conditions. However, the iron loss is not minimal enough to be ignored during high speed operation of PMBLDCM drives. This paper proposes a disturbance observer based sensorless drive control for accurate estimation of the rotor position. This is achieved by taking the iron loss into consideration in order to further reduce the rotor position estimation error. A modiﬁed model predictive control is also proposed to deal with the novel modelling to provide torque ripple free operation at reduced losses. The EV performances are compared for lossless and loss model of PMBLDCM through both simulation and experimental validation.


I. INTRODUCTION
Electric transportation is gaining momentum as it will overcome problems due to conventional vehicles such as global warming, resource shortages and environmental pollution. The main tools for the development of electrification of transportation system include exterior design, controller design, motor drives system, energy and power management processes, and system integration [1], [2]. The motor drive system is a key element for electric vehicle (EV) propulsion system since it is the predominant source of power in the EV [3]. Therefore, the study of high performance motor drive for EV has been a subject of active research. Permanent magnet (PM) based propulsion motors, such as PM synchronous motor (PMSM), PM brushless DC motor (PMBLDCM) are used by EV manufacturers due to their high energy density. The drive configuration of PMSM is expensive and has less Manuscript  power density compared to that of PMBLDCM especially in low power vehicles. The low cost drive arrangement makes the PMBLDCM propulsion system very attractive for use in cheaper EVs [4].
With trapezoidal back electromotive force (EMF) and quasi square-wave current, PMBLDCM can produce a constant electromagnetic torque. The high frequency torque ripple in PMBLDCM is induced due to manufacturing procedures, control strategies, and problems due to the commutation of switches [5], [6]. The torque ripple generates noise, mechanical vibration and speed fluctuations that need to be addressed. Accurate modelling of PMBLDCM considering design aspects and material properties is much more relevant to address these issues in order to enhance the performance of PMBLDCM drives for EV applications. Researchers have shown that a detailed model considering the iron loss, which consists of hysteresis and eddy current losses, plays a significant role in the performance of PMBLDCM [5], [7]- [12]. The modelling presented in [7] significantly considers lower-order harmonics, which can cause additional eddy-current losses to PM and conductive rotor sleeves when the PMBLDCM operates at a high-speed and heavy-duty region. In [8], [9], the optimal iron loss model is estimated on the basis of a finite element analysis, considering non-sinusoidal back EMFs and current. The major challenge of this approach is that a considerable amount of time must be dedicated to the finite element based simulation under different operating conditions. Temperature dependence iron loss modelling is addressed in [10], [11]. In [10], only the temperature dependence of the eddy current loss is considered, while the hysteresis loss is presumed not to be affected by the temperature. In [11], the effect of temperature on total iron loss is modelled by the implementation of an analogous temperature dependent coefficient, a combination of temperature effects on both hysteresis and eddy current losses. A non-linear phase variable model is proposed in [12] to analyse the effect of stator and rotor losses on PMBLDCM. Depending on the flux variance, the stator phase currents and the speed of the motor, the iron loss model is developed precisely in order to model fractional-slot machines. The effect of the temperature on the iron loss and the limitation of the existing model on the estimation of the iron loss without considering the temperature effect are demonstrated by [5]. Considering all aspects of modelling, simplified modelling reflecting the impact of iron loss is proposed in this manuscript, which provides a reliable estimate of the magnitude and corresponding angle contribution due to iron loss.
The use of position sensors for the electric traction system has been verified for years in the case of EVs. However, in order to build a smart and autonomous on-board fault detection and diagnostic system, the EVs need to be more reliable in the context of a fail-safe operation. Such performance can be achieved by sensorless operation. Researchers around the globe therefore sought to provide an effective, sensorfree framework for PMBLDCM-fed EV applications [13]- [18]. The research approaches concentrate primarily on either the enhancement of the motor architecture or active ripple compensation of the torque. Back electromotive force (EMF) sensing-based hybrid sliding mode control for PMBLDCM in vehicle applications is proposed in [13], which is more sensitive to parameter uncertainties at lower speed. The sensorless anti-lock braking control system based on back-EMF given in [14] for estimating the rotor speed of PMBLDCM is validated for low-speed application and uses wavelet transformation which does not require the deployment of separate break sensors in every section of the vehicle. In [15], a sensorless scheme is designed to compensate for commutation error along with zero crossing detection from line-to-line back EMF. In addition, [17] has implemented a commutation error compensation technique based on the harmonic analysis of phase back EMF. However, lower amplitude harmonic content prevents its use for high-speed applications as it is less susceptible to phase delay due to low-pass filtering. [16] has implemented a sensorless operation to minimise commutation torque ripple using line-to-line flux linkage. The aim of this approach is to obtain a precise rotor orientation and to determine the motor parameters by using Kalman filter for line-to-line flux linkage. The extended Kalman filter involves a complex computation thus not appropriate for real-time implementation as the interrupt routine time for a wide-speed range motor is very limited.
Recently, a stable state feedback controller to mitigate the effects of disturbances is designed for PMSM using sensorless disturbance observer [18]. These techniques have not yet fundamentally solved the issue of the iron loss of the motor, so there is scope for further study as the effect of motor iron loss has been ignored. A disturbance observer oriented sensorless approach is proposed in this article and experimentally validated for a wide range of PMBLDCM speed variations. The proposed disturbance observer estimates the rotor position along with the rotor speed by considering iron loss effect. The torque ripple minimization of PMBLDCM based EVs has received considerable attention from researchers to ensure a smooth and convenient ride for passengers [19]. Various closed-loop control techniques, such as vector control [20], [21], direct torque control [22], Fourier series based control [23], are available in literature to reduce the torque ripple of PMBLDCM. A unique petal-wave current dependent vector control approach is presented in [20] without considering the effect of iron loss component. However, the [20] provides a descriptive study of copper loss minimisation when achieving a maximum torque per ampere. Even achieving a lower copper loss, the proportional-integral (PI) dependent vector control approach has a disadvantage of higher settling time leading to higher torque ripples at reduced bandwidth. In order to address these issues, a finite control set model predictive control (FCS-MPC) technique has recently been introduced [24]- [30]. FCS-MPC are easy to configure, provide fast transient response and optimal motor current control to maintain a large bandwidth. The FCS-MPC for induction machine (IM) is explored in [24]. This process has been implemented with space vector pulse width modulation and utilizes the Kalman filter to obtain state responses that estimates the non-measured load torque. FCS-MPC developed in [25] is proposed for PMSM drives using constant input and online optimization control regulations. The technique improves the reliability of predictionoriented scheme with an independent modulation-state, given the uncertainty of the optimization problem. The performance of FCS-MPC for PMBLDCM relies on the estimation of the optimum voltage vector, which significantly minimises the cost function [27]. Predictive deadbeat control method for PMBLDCM is proposed in [28], with an accurate estimate of the switching frequency from the selected voltage vector over the next sampling interval. FCS-MPC presented in [29] integrates a ranking analysis to reduce the uncertainty due to tuning of weighting factors. However, this approach does not recognise the effect of iron loss, nor does it include any information on temperature dependence when used for EV applications.
A modified MPC controller for PMBLDC motor drive considering standard automotive temperature range is proposed in this paper. The improved dynamic model of the PMBLDCM by taking iron loss into consideration is used. The PMBLDCM model, effect of iron loss on the torque and stator current is given in section II. An improved optimum current vector-based FCS-MPC by considering the effect of iron loss is proposed in section III. A three-phase three-level (TPTL) neutral point clamped (NPC) voltage source inverter (VSI) is used to drive the PMBLDCM. The TPTL-VSI gives greater flexibility in the determination of switching vectors, lower voltage stress and improved total harmonic distortion (THD) compared to that of two level-VSI [31]. The low THD helps in reducing the torque ripple. The proposed MPC controller not only achieves reduced torque ripple and minimal copper losses; but also includes a technique to minimise the operating switching frequency of the drive. In addition, DClink capacitor voltage balancing by considering additional objectives in its main objective function is also achieved. This

II. PMBLDCM MODELLING CONSIDERING IRON LOSS EFFECTS
A schematic representation of PMBLDCM driven rear motor driven EV is shown in Fig. 1. It comprises of a PMBLDCM, an TPTL VSI stack and a battery pack as dc power source. The equivalent circuit model of the PMBLDCM in dq-plane is shown in Fig. 2(a) by taking into account the iron losses.

A. Calculation of R i and Iron Loss Modelling of PMBLDCM
The equivalent resistance representing iron loss (R i ) for a given PMBLDCM is determined experimentally as follows. At steady state and no load condition, when the motor is running at constant speed of ω m , the iron loss (P i = P in − 3R s I 2 s,rms ) is computed from Fig. 2(a). The input power (P in ) along with root mean square (RMS) value of stator phase voltages and currents (V s,rms and I s,rms ) can be measured using a power analyzer. From Fig. 2(a), the iron loss at steady state, can be written as in (1).
where R s is stator phase resistance, L s is stator inductance, e s,dq is back EMF and ω r is the electrical speed of the rotor. The back emf is given by e s,dq = (ω r λ s,dq ) where λ s,dq (= λ s,d + jλ s,q ) represents the back EMF per ω r . It can be noted from the established relation in (1) that R i can be evaluated from the slope of the P i vs. (e s,dq ) 2 curve i.e. P i vs. ω 2 r curve. The R i is calculated at various motor speed ω m varying from 40 rad/sec to 360 rad/sec. These R i values are plotted against ω m in Fig. 2(b) and a linear regression polynomial is obtained as in (2) using curve fitting algorithms. The curve extrapolated until ω m = 0 rad/sec, thus an offset resistance 12 Ω is reported. The R i is estimated from the ω m as in (2) [32].
In steady state, the magnetising component of the stator current (i m,dq ) is obtained from Fig. 2(a) as in (3). The relation in (4) introduces a variable C which represents the variable due to iron loss with magnitude k c = 1 + (ω r L s /R i ) 2 and From Fig. 2(a) and relation (3), the dynamics of PMBLDCM can be represented as in (5) where the stator current (i s,dq = i s,d + ji s,q ) is the state of the equation and v s,dq is the controlled stator voltage. τ s (= L s /R s ) represents the electrical time constant of PMBLDCM.
The electromagnetic torque (T em ) developed can be deduced by considering the iron loss as in (6).
The notation denotes the real part of the complex variable andλ s,dq represents the complex conjugate of λ s,dq . The additional electromagnetic torque component due to iron-loss (T i ) varies with sectors (k = S 1 , . . . , S 6 ) as in (7).
It should be noted that, when iron loss effect is ignored, in (3), the i m,dq = i s,dq . So the first term in right hand side of (6) , the λ s,dq i s,dq represents the electromagnetic torque produced by motor when the iron-loss effect is ignored, denoted by T es (= λ s,dq i s,dq = λ s,d i s,d + λ s,q i s,q ).

B. Optimal Current Vector formulation
Cascaded control structure as shown in Fig. 3 is used for the speed control of PMBLDC. The estimated speed feedback (ω m ) is compared with the speed reference (ω * m ). The optimal dynamic speed response is achieved by using a proportionalintegral (PI) controller. The output of the speed controller is the torque reference T * em which does not account for the copper loss and iron loss [20]. Therefore, an acceptable modification must be introduced, as described in the following. The optimal  (8).
where P * e (= 2EI * ) is the electrical power of PMBLDCM ignoring the losses, E (= λ n ω m ) is the back EMF magnitude, and λ n is the nominal value of back EMF constant. In certain, further analysis is made in αβ-plane to obtain a proper visual representation and the previous derived equations can be used without any modification. The variables in αβplane can be obtained by multiplying e jθr to the variables in dq-plane. The discussion on the estimated rotor electrical position of PMBLDCM (θ r ) is presented in section III-A. Considering T * em as the command signal, there is possibility of infinite current vector (i 1 m,αβ , i 2 m,αβ , . . . , i * m,αβ , . . . , i ∞ m,αβ ) as shown in Fig. 4(a) with a projection length of T * em / |λ s,αβ |. However, only one optimal current vector (i * m,αβ ) is available which is in phase with λ s,αβ . The i * m,αβ vector provides the minimum copper loss with constant torque and can be derived from (8) and Table I, as given in (9). The αβ-components of λ s,αβ and the i * m,αβ in six different sectors over one cycle of the supply current is defined in Table I, where the variable µ = 2θ r /π. The trajectory in each sector in (10) represents a circle with centroid (0, I * / √ 3), shown in Fig. 4(b).
where u λ is unit vector in the direction of λ s,αβ . The obtained i * m,αβ is evaluated by avoiding the iron loss that results to (i) inaccuracies in rotor position estimation (ii) reduces generation of torque than the actual as per (6) and (iii) ripple due to error in estimated rotor position. A solution must therefore be introduced to overcome this issue by incorporating additional torque due to iron loss i.e. T i as in (7). Again as depicted in Fig. 4(a), there will be infinite current vectors with a projection length of C −1 (T * em + T i ) / |λ s,αβ |. However, only one optimal current vector i * s,αβ , which is in phase advance of θ c with λ s,αβ , can provide minimal copper loss considering the iron loss effect. The i * s,αβ also substantially decreases the torque ripple and can be calculated from (6) as per (11).
III. SENSORLESS PREDICTIVE CONTROL OF PMBLDCM INCLUDING IRON LOSS A sensorless architecture is conducted using a disturbance observer to estimate unknown disturbances due to loss in functioning units [18]. The operational schematic diagram for the proposed disturbance observer based sensorless operation of PMBLDCM is presented in Fig. 5. In addition, the control structure of the proposed MPC is shown in Fig. 7. The overall control system utilizes the speed controller, current controller, rotor speed estimation, rotor flux estimation, prediction model, TPTL switching logic table and the cost function minimization blocks. Proposed disturbance observer is utilised to estimate the rotor position (θ r ) and rotor speed (ω r ).

A. Estimation of Rotor Position, motor Speed and iron-loss resistance
Flux position is detected to extract the exact position of the rotor since no hall sensors are used.θ r andω r estimation is performed in dq-plane to avoid complexities of the disturbance observer. The expression given in (5) can be rearranged as (12) where the disturbances (δ s,dq =δ s,d + jδ s,q ) is separated from the original modelling without considering the iron losses.
disturbanceδ s,dq due to iron loss (12) Substituting C in (12), theδ s,dq can be related to the speed of the motor (ω r ) and R i as in (13).
Variations in rotor position (∆θ r ) can be computed with the obtainedδ s,dq , that can therefore be used to measure ω r . In Fig. 5, theω r is estimated from the relative disturbance error using an PI controller followed by a low-pass filter (ω r / (s + ω r )). Theω r T s is added with estimated rotor positionθ r to compensate for the delay due to different pulse width modulation techniques. The actual estimated motor speed (ω m ) is described byω m =ω r /p, where p represents the number of pole pairs.

B. Model Predictive Control with Iron Loss Consideration
In traditional MPC, the cost function associated with the current control does not provide accurate control of PM-BLDCM without considering the iron loss. A modified MPC is therefore proposed in this paper to resolve the effect of iron loss in current controls, as shown in Fig. 7. For this process, the switching frequency control is also achieved along with the DC-link capacitor voltage balancing of the NPC TPTL-VSI based EV drive. The TPTL-VSI uses three different states ('P' {1100}, 'O' {0110} and 'N' {0011}) to represent the four IGBT switches of each leg. The binary number '1' shows that the switch is 'ON' while the binary number '0' shows switch status as 'OFF'. The TPTL-VSI has 27 voltage vectors (VVs) corresponding to available switching states. In order to reduce/nullify common mode voltage (CMV), only 19 VVs are used for the proposed MPC, as shown in Fig. 6. These VVs are grouped as 1) large vectors V lk , 2) medium vectors V mk , 3) small vectors V sk and 4) zero vector (O). The entire αβ-plane is converted to 6 different sectors (k = 1 . . . 6) at π/3 intervals and every sector is subdivided into two sub-sectors (a, b) at π/6 intervals. From the estimated rotor position information, the sector and sub-sector is identified and the voltage space vector v  s,αβ /T s and the back EMF can be estimated as (14).
As back EMF varying slowly with higher time constant as compared to sampling instant T s , one step delay/advance will not affect the overall performances of the controller, i.e. e The objective function of the proposed MPC considering iron loss component is expressed with three objective i.e. current control, switching frequency control and DC-link voltage balancing control. The proposed method considers sampling frequency in such a way that the switching frequency (f sw ) of generated pulse is within the maximum limit as per the Nyquist criteria. By considering sampling time of T s = 50 µs, maximum f sw must be less than 10kHz. Further reduction in f sw of overall drive performance can be achieved by introducing additional objective to the main objective function. The cost minimization function expressed in (16) is to enable the torque ripple-free operation of PMBLDCM drives with reduced copper losses.

DC-link Voltage Balancing
is the stator voltage of PMBLDCM at k th switching instant. The DC-link voltages v dc1 and v dc2 are measured through two voltage sensors and can be feedback to the MPC controller directly through appropriate DC filter. The weighting factors (W i , , W f , W v ) corresponding to the current control and switching frequency control are set to 0.33 to eliminate the shortcomings of online/offline tuning. This approximation can be validated by applying two-stage prediction method with ranking analysis, as mentioned in [29].
• The first step of the proposed method is the selection of the voltage vectors groups from (16). The speed and torque robustness of the PMBLDCM drive can be greatly affected by voltage vector selection. Large vectors can be selected for high torque changeover at higher speed if g ≥ 0.7. The medium vectors can be selected with 0.3 ≤ g < 0.7, if DC-link capacitor voltage balancing is necessary. The small vectors are specifically important at low speed requirement of EV drive and it can be selected on 0 ≤ g < 0.3. • The second step of the proposed method is to apply the similar ranking analysis as in [29], where three ranks (R 1 , R 2 , R 3 ) are assigned to the objectives g i , g f , g v . The average value of these ranks i.e. R = (R 1 + R 2 + R 3 )/3 is evaluated for each of the voltage vectors in the selected voltage groups. Finally, the optimum selection of voltage vector can be obtained by minimizing the average rank 'R'.

IV. RESULTS AND DISCUSSIONS
The validation of the proposed MPC controller in addition to the disturbance observer is verified by simulation and experimental findings. All the internal parameters of the PMBLDCM are extracted through various open-loop tests. The parameters obtained are shown in Table II. These PMBLDCM parameters are used in the simulation. The proposed MPC-based closed-loop drive performance is verified through simulation in MATLAB/Simulink environment.

A. Simulation Results
The PMBLDCM drive is simulated to evaluate the proposed MPC considering iron losses and the corresponding results are compared with lossless model presented in [20]. Fig.  8(a) reflects the simulation performance while considering iron loss by considering the proposed MPC controller. Fig.  8(b) is a simulation validation without taking the iron losses into account with PI controller based vector control approach. The efficiency of the proposed MPC is assessed with load changes at different speed ranges to determine the robustness of PMBLDCM, by considering iron loss or without iron loss. At t = 0.07 s, the no load PMBLDCM drive speed is at steady-state withω m = 500 rpm. A sudden load change with   a load torque of T L = 5 N m is applied at t = 0.2 s. It is observed that the proposed disturbance observer based speed estimation provides accurate tracking in speed during sudden load change, as shown in Fig. 9. A step change in motor speed from 500 rpm to 1500 rpm is commanded with same load of T L = 5 N m at t = 0.3 s and the motor achieves its stable operating speed in duration of 0.17 s. In order to observe the robustness of the proposed observer, PMBLDCM is suddenly unloaded at t = 0.6 s. The proposed disturbance observer has good dynamic behaviour asω m tracks the ω * m with almost zero steady state error and meets the desired transient response. At t = 0.7 s, a negative step change in ω * m from 1500 rpm to zero speed is commanded with T L = 0 N m i.e. at no load. The simulation results of motor torque (T em ), active power (P e ), reactive power (Q e ), and dq-axis stator currents (i s,d and i s,q ) are also given in Fig. 8. The comparative results of the transient speed and torque response is given in Fig. 9.
The active power requirement corresponds to the motor torque profile and the zero reactive power ensure minimized power losses. The expanded view of simulation result for stator current both in abc-plane and αβ-plane is presented in Fig. 10 while applying sensorless MPC under steady-state at load of 5 N m. It is observed that the αβ-plane stator current generates petal-wave lissajous pattern atω m = 500 rpm andω m = 1500 rpm, as shown in Fig. 11(a) and 11(b). The simulation results of optimised stator phase back EMF forω m = 500 rpm and ω m = 1500 rpm are shown in Fig. 11(a) and 11(b). These trajectories validates the theoretical analysis given in section

B. Experimental Verification and Results
The Fig. 12 shows an experimental hardware prototype. The overall drive system comprises of a TPTL NPC VSI, a PMBLDCM and a digital control unit. The PMBLDCM is coupled to a DC machine, that functions as load. The SKM100GB063D insulated-gate bipolar transistor (IGBT) and SKKD100 fast recovery diode based TPTL NPC VSI with 20 kHz switching frequency is used. A DC-link voltage of 240 V and 1000 Ah battery pack is used for powering VSI. Two current sensors (LEM LA55p) and two voltage sensor (LEM LV25p) are utilised to measure two-phase stator current and voltage across DC-link capacitors, respectively. The measured  [20] voltage and current signals are given to the digital controller dSPACE1102 in which the control algorithms are executed. The motor torque, motor speed, active and reactive power are estimated in real time using this digital controller. These signals are converted to analog signals using inbuilt digitalto-analog (DAC) converter and analyzed using digital storage oscilloscope. To demonstrate the effectiveness of the proposed modelling, the drive is evaluated experimentally for both (a) proposed sensorless MPC taking into account the iron loss and (b) vector control with hall position sensors, ignoring iron loss as per [20]. The corresponding experimental results are given in Fig. 13. In both cases identical conditions are maintained. The experimental results of T es ,ω m , i s,dq , P e and Q e , under different operating conditions of both cases is given in Fig.  13. The PMBLDCM is accelerated from stand still to 500 rpm at t 1 with a step speed command under no load. The motor speed reaches its steady state in 70 ms in case of proposed MPC controller and proves faster than the conventional vector control approach. At t 2 , load torque of T L = 5 N m is applied andω m remains constant without any fluctuation and proves the faster dynamic of disturbance observer. At t 3 , a step speed command of 500 rpm to 1500 rpm is given with the load torque of 5 N m and steady-state speed of 1500 rpm after 170 ms. It can be observed that sudden load removal at t 4 does not affect motor speed output. At t 5 , the proposed controller performs a regenerative action by detecting a sudden brake operation and finally, the PMBLDCM achieves zero speed at t 6 . The active power demand (P e ) follows the T es . The reactive power Q e remains constant at zero verifying that the phase current and phase back EMF are in phase. The plots of the stator current and the current drawn from the battery pack (i bat ) are shown in Fig. 14 which shows less ripple for the proposed algorithm. The time scale expanded results of these plots at steady state forω m = 500 rpm andω m = 1500 rpm with load torque of 5 N m is presented for the proposed model with loss and the model without iron loss [20] in Fig. 14(a) and Fig.14(b), respectively.
The plots of e s,αβ , i s,αβ with proposed model with iron loss and without iron loss at T L = 5 N m are given in Fig.  15(a) and Fig. 15(b) for 500 rpm and 1500 rpm, respectively. It can be found that the phase current waveforms appear like petal wave current with low ripple from the proposed method which is in line with the theoretical observations. For the conventional method [20], where iron losses are neglected, the current trajectory shows high ripple. Current THD and torque THD are validated by both simulations and experiments as shown in Fig. 16(a) and 16(b), respectively during steady state operation of the PMBLDCM with T L = 5 N m. The  Table III shows the % improvement in the ripple in T es , P e , Q e , i abc s and i bat . These comparative results demonstrate the significant reduction in ripple of various parameters by the proposed method with iron loss when compared with that of [20] where iron losses are neglected, thus validating the analytical observations. The average switching frequency of an inverter is measured as count of switching pulses over one second time period. The average switching frequencies of the TPTL VSI when operated under the proposed MPC scheme are plotted with respect to speed variation in Fig. 16(c)(top plot), where the switching frequency verses rotor speed curves are plotted for different load torque. It is clear from Fig. 16(c)(top plot) that the average switching frequency increases with the increase of speed irrespective of load torque. The variation of average switching frequency with the variation of speed are plotted for the vector control scheme of [20] and the proposed MPC scheme both for a load torque of 5 N m in Fig.  16(c)(lower plot). It is observed from Fig. 16(c)(lower plot) that the TPTL VSI is operated with lower average switching frequency when operated under the proposed MPC scheme as compared to the vector control scheme of [20].
The Fig. 16(d) displays the timing information of the proposed MPC technique in real time implementation, and compared with that of the MPC method [29] and vector control method [20]. The microprocessor specifications (timing, PWM generation, etc.) are imposed by proposed MPC, MPC in [29] and vector control method [20]. However, the proposed MPC requires 11 µs from a computational point of view, MPC in [29] needs approximately 33 µs while vector control [20] consumes just approximately 8 µs of the sampling time. The proposed MPC represents a clear advantage against MPC approach in [29], especially in low-powered drives, where switching frequencies (up to 20 kHz) are utilized. However, EV propulsion drives appear to be relatively medium powered, thus the sampling time of 50 µs doesn't create difficulties in practical operation.

V. CONCLUSION
An appropriate loss modelling of PMBLDCM with detailed analysis has been developed to achieve the control stability of EV for wide speed range at different road surface conditions. A sensorless scheme using a disturbance observer is carried out for high precision drive control taking iron loss component into account. A modified predictive control algorithm is validated for torque ripple free operation of EV at reduced losses. In addition, the impact of iron loss on the optimal current vector has been investigated and the modified optimal current reference vector ensures low torque ripple and reduced losses. The efficient drive behaviour employing the proposed model is demonstrated, validated with simulation and experimental findings. Consequently, the comparative experimental results that examine the effectiveness of the proposed method to that of the conventional method without considering of iron loss are presented. The potential outcomes with proposed algorithms, while taking into account the effect of iron losses, includes: 1) Modified optimal current reference vector reduces torque ripple and reduces the losses. 2) Reduced voltage stress on switches, reduced torque and current ripple, elimination of CMV and improved THD using TPTL NPC VSI. 3) Significantly improved dynamics of EV with minimal steady state error using disturbance observer to estimate motor speed. This method yielded a high performance solution with less computation burden, reduced torque ripple and lower impact of losses on PMBLDCM. The results demonstrate the adequacy of the method for a 2.5 kW PMBLDCM proposed for electric vehicle application.