IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 63, …ncbs.knu.ac.kr/Publications/PDF-Files/VeluvoluKC16-TIE.pdf · 2017. 3. 13. · A permanent-magnet synchronous motor (PMSM)

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 63, NO. 12, DECEMBER 2016 7671

A Robust Observer-Based Sensor Fault-TolerantControl for PMSM in Electric Vehicles

Suneel Kumar Kommuri, Student Member, IEEE, Michael Defoort,Hamid Reza Karimi, Senior Member, IEEE, and Kalyana Chakravarthy Veluvolu, Senior Member, IEEE

Abstract—This paper investigates the problem of auto-matic speed tracking control of an electric vehicle (EV)that is powered by a permanent-magnet synchronous motor(PMSM). A reconfiguration scheme, based on higher ordersliding mode (HOSM) observer, is proposed in the eventof sensor faults/failures to maintain a good control perfor-mance. The corresponding controlled motor output torquedrives EVs to track the desired vehicle reference speed forproviding uninterrupted vehicle safe operation. The effec-tiveness of the overall sensor fault-tolerant speed trackingcontrol is highlighted when an EV is subjected to distur-bances like aerodynamic load force and road roughnessusing high-fidelity software package CarSim. Experimentswith a 26-W, three-phase PMSM are presented to demon-strate the validity of the proposed fault-detection scheme.

Index Terms—Electric vehicles (EVs), fault-tolerant con-trol (FTC), higher order sliding mode (HOSM), permanent-magnet synchronous motor (PMSM), road roughness,speed tracking control.

I. INTRODUCTION

THE high percentage of vehicle accidents due to driver’s er-ror, distractions, and drowsiness motivates the researchersin introducing the various driver assistance systems to increasethe highway safety [1]. A few of these systems such as predic-tive cruise control [2] and lane departure warning systems [3]have already been accepted as standard in many vehicle pro-duction industries. Furthermore, the dependence of the vehiclepropulsion control system on actuator/sensor components is be-coming more complex. Generally, these complex systems are

Manuscript received November 5, 2015; revised March 5, 2016 andApril 25, 2016; accepted May 8, 2016. Date of publication July 13, 2016;date of current version November 8, 2016. This work was supportedby the Basic Science Research Program through the National ResearchFoundation of Korea funded by the Ministry of Education, Science andTechnology under Grant NRF-2014R1A1A2A10056145. (Correspondingauthor: Kalyana Chakravarthy Veluvolu.)

S. K. Kommuri is with the School of Electronics Engineering, College ofIT Engineering, Kyungpook National University, Daegu 702-701, SouthKorea (e-mail: [email protected]).

M. Defoort is with the Laboratoire d’Automatique et de MécaniqueIndustrielles et Humaines, CNRS UMR 8201, University ofValenciennes, 59314 Valenciennes, France (e-mail: [email protected]).

H. R. Karimi is with the Department of Mechanical Engineering,Politecnico di Milano, 20156 Milan, Italy (e-mail: [email protected]).

K. C. Veluvolu is with the School of Electronics Engineering, College ofIT Engineering, Kyungpook National University, Daegu 702-701, SouthKorea, and also with the School of Mechanical and Aerospace Engi-neering, Nanyang Technological University, Singapore 639798 (e-mail:[email protected]).

Color versions of one or more of the figures in this paper are availableonline at http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TIE.2016.2590993

subjected to some catastrophic faults such as unknown actua-tor/sensor faults [4] and thereby increasing the concerns overfault tolerance [5], [6]. Therefore, the development of effectivecontrol systems under the aforementioned faults and challeng-ing operating conditions is still a potential research topic fromboth academic and industrial perspectives. Indeed, it is impor-tant for the overall electric vehicle (EV) system to be robustagainst such system faults/failures to enhance vehicle handlingcharacteristics like stability and comfort, which further leads tothe outright driverless cruise control.

Over the past few decades, various fault-diagnosis and fault-tolerant control (FTC) strategies (generally classified as activeand passive FTCs) have been proposed [7]–[11]. In [8], FTC isachieved based on the adaptive residual generator for identifyingthe abnormal change of system parameters and an iterative tun-ing scheme for the optimization of system performance. Amongvarious approaches for sensor FTC published in the literature,observer-based approaches are effective and reliable [12]–[17].In [12], speed sensor FTC based on a voting algorithm for EVsis proposed. The main drawback of this approach is the re-quirement of two observers and a complex voting algorithmfor the sensor fault detection. A simple adaptive observer-basedscheme is designed in [13] for the fault-tolerant cruise control ofEVs with induction motors. However, the analysis of FTC-basedcruise control performance on variable vehicle speed referencesand vehicular disturbances is not considered. A logic-based de-cision mechanism using a model reference adaptive system es-timator for the sensor FTC in an induction motor is proposed in[15]. Recently, in [16], a discrete-time estimator for simultane-ous estimation of states and actuator/sensor faults is proposed.Further, with this information, a discrete-time FTC approachfor vehicle lateral dynamics is addressed. More recently, slid-ing mode observer (SMO) based FTC without rate sensors isproposed in [17] for the attitude stabilization of satellites.

A permanent-magnet synchronous motor (PMSM) becomesan attractive candidate for electric/hybrid vehicles, includingtrains, buses, and cars, because of its high efficiency and powerdensity [18]. A high-performance PMSM drive relies on field-oriented or vector control and requires a precise rotor position.The main issues related to the sensorless speed estimation of adrive are the rotor position accuracy and the speed convergencerate. SMOs in the recent literature are popular for the robustunknown input estimation (see, for instance, [19]–[23]). Morerecently, a few schemes based on SMOs have been proposedfor the speed estimation of a PMSM [24], [25]. A high-speedSMO is proposed in [24] to avoid the time delay occurringdue to a low-pass filter, by employing a sigmoid function as

0278-0046 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

7672 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 63, NO. 12, DECEMBER 2016

the switching function. However, the boundary layer and SMOgains selection are dependent on the motor speed. Furthermore,a modified SMO is developed in [25] to estimate the speed fromback electromotive force (EMF) signals, which are estimatedwith a conventional SMO. However, this approach contains atwo-stage process for estimating the position and speed signals.

Recently, higher order sliding mode (HOSM) observers basedon a supertwisting algorithm (STA) [26]–[28] have been pro-posed to provide finite-time convergence and robustness withrespect to bounded disturbances. The structurally bounded un-certainties/unknown inputs can be estimated from the slidingmode terms without requiring external low-pass filters. Fur-ther, the well-known chattering phenomenon, which consists oflarge oscillations in the neighborhood of the sliding manifold,can also be reduced. However, the STA fails to converge in theend for a linearly growing perturbation. Therefore, a modifiedSTA is introduced in [29] to counter this problem. In this design,the SMO gain is adjusted to resist against exciting perturbationterms, which straightaway handle the linear perturbation. Re-cently, a nonhomogeneous continuous STA is proposed in [30]for systems of relative degree more than one, which assuresfinite-time convergence to the origin for all system states.

Based on the recent developments and the modified STA,it is inspired to propose a sensor FTC scheme based on anHOSM observer for the cruise control of EVs. As preliminary,sensor FTC in [31] needs to be revised and adapted to moregeneral sensor fault scenarios. The contributions of this paperare summarized as follows:

1) A novel fault-tolerant cruise control architecture usinga robust HOSM observer for a PMSM-powered EV inthe presence of speed sensor faults is proposed. Unlikethe existing observer-based FTC works [13], [32], [33],the proposed HOSM observer ensures finite-time stabil-ity of the observation error for the reconfiguration-basedcontrol scheme. Furthermore, the sensor fault detectionbased on an HOSM observer is robust to bounded modeluncertainties and parameter variations, which vary ac-cording to the operating conditions [8]. Indeed, the faultcan be effectively detected using an appropriate threshold,since the estimation error is bounded, without the use ofcomplex observer-based detection techniques [12], [34].

2) In [35] and [12], the proposed approaches consider onlyEV motor speed reference to evaluate its dynamic perfor-mance under speed sensor faults. However, in this paper,the robustness of the proposed sensor FTC is further il-lustrated in the presence of vehicular disturbances likeaerodynamic load force and road roughness. The consid-ered vehicular disturbances are the main factors affectingthe vehicle speed motion. Various simulations for faultyspeed sensor cases (during constant and ramp speeds) to-gether with vehicular disturbances are provided using thehigh-fidelity CarSim software package.

3) Experimental results are presented to demonstrate theHOSM observer-based sensor fault-detection approach.Unlike the existing works [12], [13], [35], experimentson a real PMSM with a mechanical load under paramet-ric uncertainties are conducted by considering a similar

power-to-torque ratio, which is obtained in the CarSimenvironment. This is considered to justify the effective-ness of the proposed approach in a real EV scenario (whena PMSM is subject to external dynamics).

The rest of this paper is organized as follows. The modelingof a PMSM is discussed in Section II. Section III presents thedesign of the HOSM observer based on the modified STA forspeed estimation. Section IV discusses the speed sensor faultmodels and FTC architecture for EVs. Section V presents theCarSim and experimental results. The main discussions on theimplementation are presented in Section VI and Section VIIconcludes this paper.

II. VEHICLE DYNAMICS AND PMSM MODELING

In this section, the vehicle dynamics followed by awell-known PMSM model are presented.

A. Vehicle Dynamics Modeling

The load torque TL of an EV is given by [12], [31]

TL = (Frr + Fwd + Fgr + Far)rt + � (1)where TL ∈ R, Frr is the rolling resistance force, Fwd is theaerodynamic drag force, Fgr is the grading resistance force, Faris the acceleration resistance force, rt is the dynamic radius oftires, and � is the uncertainties or unmodeled dynamics.

Furthermore, the vehicle speed vs is proportional to the motorspeed ωe , which can be described in terms of tire radius rt ,differential transmission ratio nd , and the gear box ratio ni inthe gear number i as follows [36]:

vs =rt

ndniωe. (2)

Given a desired vehicle speed vs ∈ R, one needs to steer thedrive speed ωe to its corresponding motor reference speed. Astandard feedback control requires the knowledge of the currentspeed for this purpose. As the motor speed is a function of loadtorque, the forces that are acting on the EV directly affect thedriving motor speed. Therefore, it is important to provide therequired driving motor torque to track the EV’s actual speed toits desired reference speed.

B. Dynamic Modeling of PMSM

In the stationary reference frame (α − β), the dynamic modelof the PMSM is represented as follows [25]:

dijdt

=−RL

ij −1L

bj +1L

uj (3)

where j ∈ {α, β}. The stator currents are ij ∈ R, uj are thevoltages and bj are the back EMFs given as{

bα = −KE ωe sin θebβ = KE ωe cos θe

(4)

where the speed ωe dynamics are given as{dωedt =

PJ φm (− sin θeiα + cos θeiβ ) −

fvJ ωe −

TLJ

Te = 32 φm P (− sin θeiα + cos θeiβ )(5)

KOMMURI et al.: ROBUST OBSERVER-BASED SENSOR FAULT-TOLERANT CONTROL FOR PMSM IN ELECTRIC VEHICLES 7673

where R is the stator resistance, L is the synchronous induc-tance, P is the number of pole pairs, J is the moment of inertia,KE is the back EMF constant, φm is the rotor flux, fv is theviscous friction, TL is the load torque, and θe and ωe are theposition and speed of the motor, respectively.

In the above model, ij and uj are the measured quantities andbj is considered as the unknown quantity. It is well known thatthe back EMFs in (3) are functions of speed and rotor position,shown in (4). In the following section, a robust observer isdesigned for the estimation of these unknown back EMFs. Therotor position and speed are then calculated from the estimatedback EMFs.

This paper focuses on sensor faults that occur during driveoperation to ensure the uninterrupted EV operation. The mainobjective of this work is to provide the required motor torqueeven in the presence of speed sensor faults in order to auto-matically regulate the EV speed to the desired vehicle referencespeed. We aim to achieve this objective with the use of an HOSMobserver and a selection mechanism that chooses between themeasured and estimated speeds depending on the detection of asensor fault.

III. DESIGN OF HOSM OBSERVER

This section designs the HOSM observer for the estimationof back EMFs from the measured voltages uj and currents ij .The estimated back EMFs b̂j are further used to compute therotor position and speed of the drive. The actual speed of thedrive is controlled with a PI regulator to further produce thedesired current irefq in the q-axis. In the constant air gap fluxmode of operation, irefd = 0 (see the vector control architecturefor a PMSM in [25]). The HOSM observer considers the backEMFs bj of the PMSM model in (3) as unknown inputs andreconstructs them in both the α- and β-axes.

The sliding surface in the design of an HOSM observer isselected as follows:

sj (t) = îj (t) − ij (t) (6)

where îj and ij are estimated and actual currents, respectively.Let us define the SMO for PMSM currents as follows:

dîj (t)dt

=−RL

îj (t) +1L

uj (t) +1L

ϕj (t). (7)

The corrective terms proposed in the classical STA [26] are

ϕj (t) = −k1 |sj (t)|12 sign(sj (t)) − k2

∫ t0

sign(sj (τ))dτ (8)

where ki are positive gains. Since the signum function in thecorrection term of (8) is bounded, the trajectories of the algo-rithm are very slow when they are far from the origin. Therefore,in order to improve the convergence time, a slight modificationby adding linear terms is introduced. The modified STA thenbecomes [29]

ϕj (t) = −G1φ1(sj (t)) − G2∫ t

0φ2(sj (τ))dτ (9)

where

φ1(sj (t)) = sj (t) + G3 |sj (t)|12 sign(sj (t))

φ2(sj (t)) = sj (t) +G242

sign(sj (t)) (10)

+32G4 |sj (t)|

12 sign(sj (t)) (11)

where G1 , G2 , G3 , and G4 are properly chosen positive con-stants. The linear term G1φ1(sj (t)) is mainly used to improvethe convergence rate when the trajectories of the algorithm arefar from the origin. If the linear term is large, the convergencerate will increase. The gains G2 and G3 are used to obtainfinite-time stability and reject the effect of uncertainties.

The time derivative of sj can be obtained from (3) and (7) as

ṡj (t) =−RL

sj (t) +1L

bj (t) +1L

ϕj (t). (12)

The boundedness of back EMFs can be easily established from(4) and (5). Since ωe , TL , bj , and ij are persistent in a compactspace, there exists a constant ρj ∈ R such that the perturbationterm is bounded as

|ḃj | ≤ ρj . (13)

Theorem 1: Under condition (13), the origin of system (12)is a finite-time stable equilibrium point. After a finite time T (s0),a smooth estimation of bj is given by G2

∫ t0 φ2(sj (τ))dτ .

Proof: Let us select a Hurwitz matrix A0 as

A0 =[−(G1L +

RL ) 1

−G2L 0

]

where G1 > 0 and G2 > 0. System (12) with (9) can be equiv-alently represented by⎧⎪⎪⎪⎨⎪⎪⎪⎩

ṡ1 = s2 − (G1L +RL )

(s1 + G4 |s1 |

12 sign(s1)

)ṡ2 = −G2L

(s1 +

G242 sign(s1) +

3G42 |s1 |

12 sign(s1)

)+ ḃjL

(14)

with s1 = sj , s2 = 1L [bj − G2∫ t

0 φ2(s1(τ))dτ ], and G4 =G 1 G 3

LG 1L +

RL

= G3 G1G1 +R .

Since ‖ φ2(sj (t)) ‖ ≥ G24

2 , one gets ‖ ḃj ‖ ≤ ‖ φ2(sj (t)) ‖ if

G4 ≥√

2ρj . (15)

From the selection of gain G4 = G3 G1G1 +R in conditions (14) and(15), the gain G3 should satisfy the following inequality:

G3 ≥√

2ρj (G1 + R)G1

. (16)

The augmented state vector can be considered as

ξ =[ξ1 ξ2

]T = [ s1 + G4 |s1 | 12 sign(s1) s2 ]T . The Lyapunovfunction V (ξ) = ξT Pξ is selected to perform the stability anal-

ysis with P = PT =[

λ + 4�2 −2�−2� 1

], λ > 0 and � > 0. The

time derivative of the candidate Lyapunov function along so-lutions of the system can be found using differential inclusion


theory as follows:

V̇ =(1 + G42 |s1 |

−12

)ξT

(AT0 P + PA0

)ξ + 2ξT P

[0ḃj

].

It can be shown that

V̇ ≤(

1 +G42|s1 |

−12

) (ξT

(AT0 P + PA0

)ξ + 2P

[0ξ1

])

≤ −(

1 +G42|s1 |

−12

)ξT Qξ

≤ −(

1 +G42|s1 |

−12

)λmin(Q)‖ξ‖2

with

Q =[

Q1 Q2Q2 Q3

](17)

⎧⎪⎨⎪⎩

Q1 = 2(

G1 +RL

) (λ + 4�2

)− 4�

(G2L − 1

)Q2 = −2�

(G1 +R

L

)+

(G2L − 1

)− (λ + 4�2)

Q3 = 4�.

(18)

One can choose G2 = L(λ + 4�2 + 2�(G1 +RL )) to guarantee thepositive definiteness of the matrix Q similar to [26] and [29].The matrix Q is positive definite if

G1 > −R +2�Lλ

(λ + 4�2 − 1). (19)

One can conclude that

V̇ ≤ −γ1V12 (s) − γ2V (s) (20)

with γ1 =λm in (Q)

λ12m a x (P )

G242 and γ2 =

λm in (Q)λm a x (P )

.

The closed-loop system (14) is finite-time stable. Therefore,it follows that every trajectory of system (14) starting at initialstate sj (t0) ∈ R with initial time instance t0 reaches the origin(sj = 0) in a finite time smaller than [29]

T (sj (t0)) =2γ2

ln(γ2

γ1V

12 (sj (t0)) + 1

). (21)

Since s1 and s2 converge to zero in finite time, one gets0 = 1L [bj − G2

∫ t0 φ2(s1(τ))dτ ]. Hence, the estimation of back

EMFs is achieved as follows:

b̂j = G2∫ t

0φ2(sj (τ))dτ. (22)

Using the estimated back EMFs, from model (4), the rotorposition and speed signals are computed, respectively, as

θ̂e = −tan−1(

b̂α

b̂β

)(23)

ω̂e =1

KE

√b̂2α + b̂2β . (24)

Remark 1: Since V (ξ) is continuous but not locally Lips-chitz, the classical version of the Lyapunov theorem cannotbe used. However, it is possible to show that V (ξ) is an abso-lutely continuous function along the system trajectories, and it is

therefore monotonically decreasing if and only if V̇ is negativedefinite almost everywhere. Moreover, as the closed-loop sys-tem in (14) is discontinuous, its solutions are interpreted as theones of the differential inclusion ṡj ∈ f(sj ) obtained when theset-valued modification of the sign function assigns the interval[−1, 1] to sj = 0 [37].

IV. SENSOR FAULT MODELING AND FTC DESIGN

In this section, mechanical sensor fault conditions and thecorresponding FTC mechanism, including the fault-detectiontechnique, are presented.

A. Speed Sensor Fault Modeling

For the vector control of a PMSM-powered EV, a rotor po-sition sensor is employed to provide the required informationto be used by the control system. The speed sensor fault/failureleads to unsatisfactory or dangerous behavior if no remedial ac-tions are made. In general, mechanical sensors can exhibit thefollowing fault models [34]:

1) intermittent sensor connection;2) complete sensor outage;3) DC bias in sensor measurement;4) sensor gain drop.Generally, these faults can be described as additive

perturbations [38]:

y(t) = y0(t) + �y (t) (25)

where y is the speed measurement, y0 is the nominal one, and�y (t) represents the fault.

In some recent research works [12], [34], it is reported thatthe complete speed sensor outage is one of the most severe faultsthat may lead to instability. Therefore, the complete speed sensoroutage in the closed-loop speed control of an EV is consideredin this paper.

B. Sensor FTC Mechanism

The overall block diagram of the proposed sensor fault-tolerant cruise control architecture is depicted in Fig. 1. Themajor steps involved in the FTC design are as follows:

1) The designed robust HOSM observer (treated as a virtualsensor) estimates the vehicle motor speed in finite time.

2) A fault-detection method based on the estimated and themeasured speeds detects and isolates the eventual sensor fault.

3) A switching strategy (reconfiguration) chooses be-tween the measured and estimated speeds depending on thedetection result.

4) Using either the measured speed (if no fault has beendetected) or the estimated speed (in the presence of fault), a PIcontroller achieves the speed tracking control objective.

Using results from the previous section, the fault-detectionscheme, which uses the measured and HOSM estimated speedsusing a threshold, is presented here. One should note that, due tosensor noise and system parameter uncertainties, there will be abounded error between the measured and the estimated speeds.The threshold selection should also depend on the noise and


Fig. 1. Proposed sensor FTC architecture.

estimation error to avoid false alarms. For the fault detection,the speed index is chosen as

Rotor speed index : ωe,index = ω̂e − ωe. (26)

The index value in the absence of a fault will be a boundederror that will be less than the threshold. For industrial drives,the minimum operating speed must be determined to selectthe threshold value ωe(th) . In practice, the speed threshold isset smaller than the minimum speed of the drive to identifythe sensor fault. For instance, if the rotor speed index ωe,indexexceeds the preset threshold value ωe(th) , a speed sensor faultis detected.

Based on the detection result, the reconfiguration strategyfeedbacks either the estimated speed or measured speed to thefeedback controller. The control law that forces the speed totrack the motor speed reference [33] is designed as follows:

ΥPI = P1z1 + P2∫ t

0z1(τ)dτ (27)

with positive gains P1 , P2 > 0 and

z1 =

{ωrefe − ω̂e , if the fault is detectedωrefe − ωe, otherwise.

Remark 2: Since the proposed HOSM observer provides anestimate of a speed sensor in finite time, the stable aforemen-tioned controller design achieves the speed tracking control ob-jective either in faulty or in healthy sensor operations.

Remark 3: The controller output ΥPI may be saturated if thereference speed is given a large step change or if a large externaltorque is loaded. The controller gains P1 and P2 are properlytuned to minimize the speed tracking error z1 .

For the state-driving cruise control problem, complete sensoroutage or abrupt sensor faults are commonly detected in theliterature using a fixed threshold [32], [39]. However, a pos-sible large fault-detection time may occur due to the use of afixed threshold. Here, the threshold is carefully tuned based onthe minimum operating speed of the drive in order to minimize

TABLE ISPECIFICATIONS OF PMSM

Symbol Quantity Value

PW Rating 50 [kW]ω Rated speed 628 [rad/s]R Stator resistance 0.79 [Ω]L Stator inductance 0.47 [mH]KE Magnetic flux linkage 0.2709 [Vs/rad]P Number of poles 4

the fault-detection time while avoiding false alarms. Indeed, thecontroller successfully manages the control objective since thefault is detected within a short period of time due to the appro-priate selection of the threshold. Adaptive threshold schemes areconsidered in [15] and [40], where model uncertainties highlyinfluence the complex system dynamics. Recently, in [40], aLuenberger observer-based adaptive threshold is proposed sincethe generated residuals may deviate from zero even without anyfault. These time-varying deviations may cause missing or falsealarms, which are not acceptable in the wide speed operationrange of EVs. Here, it should be highlighted that the speed indexis always bounded using an HOSM observer due to its robust-ness properties and thus does not require an adaptive thresholdto reduce the complexity.

Remark 4: The stability of the system can be established inthe practical sense. As the system is assumed to be healthyduring the initial drive operation, the observer converges infinite time. Later in the event of a fault, because of the finite-time stability of the proposed observer, the decoupling betweenthe observer and controller is achieved in the practical sense.Therefore, the stability of the closed-loop system holds even inthe presence of a sensor fault.

V. SIMULATIONS AND EXPERIMENTAL RESULTS

Computer simulations are performed on a high-fidelity full-vehicle model constructed in CarSim software to provide theclosed-loop behavior of the proposed sensor FTC. Experimentalresults are conducted on a three-phase PMSM with a mechanicalload to support the simulation results.

A. CarSim Results

The CarSim simulations are performed to analyze the closed-loop responses of vehicle dynamics under vehicular distur-bances by using the developed sensor FTC approach. The em-ployed CarSim vehicle (front-wheel drive A-class, Hatchback)consists of nonlinear tire dynamic models and also takes the sus-pensions and steering effects. The vehicle under considerationhas an autonomous front suspension system. The specificationsand parameters of the PMSM used are given in Table I.

In the simulations, it can be seen that the cruise control ac-tion is activated when the controller and motor dynamics inthe closed loop reach their steady state. To demonstrate the ef-fectiveness of the proposed approach, two different referencespeed profiles are considered in which faults are applied in twosituations: one during the constant speed drive operation and


Fig. 2. FTC for a given speed profile with a sudden aeroforce loadapplied at t = 33 s. (a) Aerodynamic force applied. (b) Motor torque.(c) Actual motor speed. (d) Vehicle actual and reference speeds. Thezoomed-in portion from t = 32.5 s to t = 37.5 s is highlighted. (e) Vehiclespeed error.

the other during the ramp speed drive operation. The HOSMobserver requires the tuning of four positive gains G1 , G2 , G3 ,and G4 . The bound of the rate of change of the perturbation wasestimated as ραβ = 5.5, which is verified a posteriori throughnumerical simulations. This procedure is usually employed insliding mode applications [29]. With regard to this bound, theconstant gains can be selected as G1 = 0.9, G2 = 50, G3 = 16,and G4 = 8. It can be seen that the chosen gains satisfythe following:⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

G1 > −R + 2�Lλ (λ + 4�2 − 1)

G2 > L(λ + 4�2 + 2�

(G1 +R

L

))G3 ≥

√2ρj (G1 +R)

G1

G4 = G3 G1G1 +R .

(28)

If the gain G1 is high, it results in a large control input (possiblesaturation in practice); however, the convergence rate becomeshigher, which is used to compromise between the size of controland convergence rate. Further, the gains G2 and G3 are tunedto compromise between the robustness and size of the control.

Fig. 3. FTC for a given speed profile with an aerodynamic load att = 33 s and a speed sensor fault at t = 37 s. (a) Motor torque. (b) Speedindex and threshold. (c) Actual motor speed after reconfiguration: Thezoomed-in portion from t = 34.5 s to t = 39.5 s is highlighted. (d) Longi-tudinal vehicle actual and reference speeds. (e) Vehicle speed error.

A tradeoff between robustness, rate of convergence, and size ofthe control should be determined to tune the parameters. Thespeed threshold is chosen as ωe(th) = 50 rad/s. Furthermore, thespeed controller (PI) gains can be selected to minimize the speedtracking error as P1 = 6.0 and P2 = 15.

1) Case 1: In the first simulation, it is considered that thevehicle is moving on a flat and dry road. Fig. 2 depicts theperformance of the developed approach when the vehicle speedreference is set from 90 to 120 km/h in the presence of a suddenaerodynamic load force (three times the normal aerodynamicforce) applied at t = 33 s onwards. Fig. 2(a) shows the appliedaerodynamic force and Fig. 2(b) shows the corresponding motortorque Te , in which a significant effect of aerodynamic force canbe clearly observed. The actual motor speed ωe in the presenceof a sudden aerodynamic force is shown in Fig. 2(c). Further-more, the vehicle’s actual and reference speeds, vs and vrefs ,respectively, are depicted in Fig. 2(d). It can be clearly seen thatthe vehicle speed accurately tracks the reference speed even inthe presence of a sudden aerodynamic load. The vehicle speederror is shown in Fig. 2(e).


In the second simulation, Fig. 3 depicts the performance of thedeveloped approach under an aerodynamic load and an abruptmotor speed sensor fault. An aerodynamic load is applied att = 33 s and a speed sensor fault is applied at t = 37 s. As soonas the motor speed sensor fault is properly diagnosed using theHOSM observer speed, the motor speed is controlled by a speedcontroller to provide a torque input shown in Fig. 3(a). Further-more, Fig. 3(b) shows the speed index and the correspondingthreshold to detect the faulty measurement speed. The actual(reconfigured) motor speed under a sudden faulty sensor speedis depicted in Fig. 3(c). The corresponding vehicle speed and itsreference are shown in Fig. 3(d), despite the motor speed sensorfault. The vehicle speed error is shown in Fig. 3(e). Therefore,it can be observed that the vehicle speed tracks the referenceaccurately even in the presence of a motor speed measurementfault, which shows the effectiveness of the proposed speed sen-sor fault-tolerant cruise control for the considered EV scenario.

In the third simulation, the robustness of the proposed methodis verified in the presence of road roughness, an aerodynamicload, and a speed sensor fault, as depicted in Fig. 4. A Class Droad profile is generated according to the International Organi-zation for Standardization, in which standards of road rough-nesses are classified into different road classes. The generatedClass D road profile data are provided as an input for the vehiclesimulation. Fig. 4(a) shows the motor torque of an EV under theroad roughness profile. The load torque contains noise due tothe roughness in the road profile. Fig. 4(b) represents the presetspeed threshold and the corresponding speed index, which de-tects the faulty measurement speed. The actual (reconfigured)speed of the motor in the presence of a speed sensor fault is de-picted in Fig. 4(c). The estimated speed is engaged (or switched)in the sensorless controller when a speed sensor fault occurs.The transition between the sensor and the observer’s estimationwith the help of a proper threshold is smooth, which resemblesthe validity of the developed sensor FTC method. However, dueto the existence of noisy measurements, the actual speed of themotor also contains noise. Moreover, the corresponding vehiclelongitudinal speed in the presence of roughness is depicted inFig. 4(d), which accurately tracks the vehicle reference speed.The corresponding vehicle speed error between the referenceand actual, as shown in Fig. 4(e), is low, which proves the sta-bility of the overall closed-loop sensor FTC. It can be pointedthat the proposed FTC scheme provides a good rotor speed in thepresence of noise as the EV operates smoothly even in the pres-ence of vehicle vibration characteristics due to the roughness inthe road profile. The considered road profile for the simulationis depicted in Fig. 4(f).

2) Case 2: To further prove the effectiveness of the pro-posed sensor FTC, a ramp-change vehicle speed profile vary-ing between 90 and 120 km/h is considered. The proposed ap-proach is tested in the presence of an aerodynamic load (threetimes the normal aerodynamic force) and road roughness, asshown in Fig. 5. Fig. 5(a) shows the generated motor torque andFig. 5(b) shows the corresponding speed index and its thresh-old. The actual motor speed (after reconfiguration) is depictedin Fig. 5(c). Fig. 5(d) represents the vehicle reference and actualspeeds achieved with the developed FTC architecture. The actual

Fig. 4. FTC under road roughness for a given speed profile with anaerodynamic load at t = 33 s and a speed sensor fault at t = 37 s.(a) Motor torque: The zoomed-in portion highlights the effect of the roadroughness profile. (b) Speed index and its threshold. (c) Actual motorspeed after reconfiguration. (d) Longitudinal vehicle actual and refer-ence speeds: The zoomed-in portion from t = 34.5 s to t = 39.5 s ishighlighted. (e) Vehicle speed error. (f) Road profile.

vehicle speed follows the reference (desired) speed in the pres-ence of a sudden motor speed sensor fault. The zoomed-inportion of the speed reconfiguration is highlighted and shownin the same plot. The corresponding vehicle speed error andconsidered road roughness are depicted in Fig. 5(e) and (f).

B. Experimental Validation

To show the effectiveness of the proposed HOSM observerwith fault detection of the motor speed sensor in a real-timevehicle control scenario, an experimental validation with thefollowing motor specifications is presented. A three-phasePMSM of TBL-i model TS4632N2050E510 powered by asmart power module (switching frequency is 15 kHz) is set up


Fig. 5. FTC under road roughness for a given speed profile withan aerodynamic load at t = 25 s and speed sensor fault at t = 29 s.(a) Motor torque. (b) Speed index and its threshold. (c) Actual motorspeed after reconfiguration: The zoomed-in portion highlights the effectof road roughness. (d) Longitudinal vehicle actual and reference speeds.(e) Vehicle speed error. (f) Road roughness profile.

in the laboratory, as shown in Fig. 6. A 32-bit floating pointTMS320F28335 DSP is utilized as the controller and the spacevector pulse width modulation is employed as it is the mostreliable modulation technique. An incremental encoder of 2000pulses/rotation is used to measure the actual motor speed. Thespecifications and parameters are provided in Table II.

In order to validate the results obtained in the CarSim soft-ware, the mechanical inertia load (JL = 0.0746 kg · cm2) isaccordingly added on the real PMSM rotor shaft to maintaina similar power-to-torque ratio to that achieved on the simu-lated car. The mechanical load is fixed to the rotor shaft of thePMSM drive. For a speed profile of 135–185 rad/s, the realPMSM with an applied inertia load produces a motor torque of0.017– 0.027 N·m, which is approximately similar to the CarSim

Fig. 6. Overall experimental setup.

TABLE IISPECIFICATIONS OF PMSM

Symbol Quantity Value

PW Rating 26 [W]T Torque 0.062 [N·m]ω Rated speed 418 [rad/s]R Stator resistance 2.0 [Ω]KE Back EMF constant 0.156 [Vs/rad]L Stator inductance 0.51 [mH]P Number of poles 8

vehicle motor power-to-torque ratio in the presence of vehicu-lar load disturbances (see Fig. 5). This means that the PMSMis driving with the same power-to-torque ratio under loadingand parameter variation conditions as that of the vehicle PMSMemployed in CarSim. In the experiments under a similar power-to-torque ratio, the real PMSM torque is driving the same ratioof load as that employed in the CarSim vehicle powered by aPMSM. Although the real PMSM is not subjected to all loadtorque dynamics of a real EV, it is intended to show the valida-tion of the proposed approach with a considered mechanical loadunder parametric uncertainties. Therefore, by performing the ex-periments on a real PMSM with respect to the power-to-torqueratio obtained in CarSim, it can be justified that the EV givessimilar performance under real conditions (actual load torquedynamics) in the presence of considered speed sensor faults.In other words, the scaled-down motor (26-W PMSM) perfor-mance shows the effectiveness and reliability of the proposedapproach with a similar power-to-torque ratio to that obtainedin the CarSim vehicle (50-kW PMSM). The HOSM observergains are selected as follows: G1 = 0.5, G2 = 40, G3 = 25, andG4 = 4.5. The speed threshold value is set to ωe(th) = 15.

The performance of the proposed HOSM observer for rampand sudden change speed profiles varying between 135 and 185rad/s is shown in Figs. 7 and 8. The actual and estimated cur-rents for a ramp-change speed profile are shown in Fig. 7(a) and(b), which are identical in phase and magnitude. Despite thenoisy currents, the back EMFs shown in Fig. 7(c) are relatively


Fig. 7. Estimation using HOSM observer for ramp-change speedprofile: (a) Actual currents. (b) Estimated currents. (c) Estimated backEMFs. (d) Estimated speed.

Fig. 8. Estimation using an HOSM observer for a step-change speedprofile. (a) Estimated back EMFs. (b) Estimated speed.

smooth. The corresponding speed estimate is algebraically com-puted from the back EMFs depicted in Fig. 7(d). Due to theexistence of noise in current measurements, the estimated speedalso contains noise. Several experiments have been conductedto verify the robustness of the proposed observer by varying±10% of the motor parameters (R and L) and similar perfor-mances have been observed for a step change in the speed.Fig. 8(a) and (b) shows the estimated back EMFs, which areobtained using the HOSM from (22), and the corresponding es-timated speed. For comparison, the performance with the recenthigh-speed SMO [24] with a sigmoid function under a ramp-change speed profile is depicted in Fig. 9. Fig. 9(a) shows the

Fig. 9. Estimation using a high-speed SMO with a low-passfilter for a ramp-change speed profile. (a) Estimated back EMFs.(b) Estimated speed.

Fig. 10. Fault detection of a speed sensor for a ramp speed profile.(a) Reference and actual speeds. (b) Speed index and its threshold.

estimated back EMFs, which are obtained with the use of a low-pass filter, and Fig. 9(b) depicts the estimated speed, which stillcontains the chattering phenomenon. The main drawbacks ofthis observer is the use of a low-pass filter to smoothen the backEMFs and less steady-state accuracy due to the observer gain forvariable speeds.

Fig. 10 shows the fault-detection performance when the mo-tor is driving with a load of JL = 0.0746 kg · cm2 . To evalu-ate the performance of the fault-detection algorithm, a suddenspeed sensor fault is applied at t = 0.5 s, as shown in Fig. 10(a).Then, the corresponding speed index is more than the selectedthreshold value ωe(th) , as depicted in Fig. 10(b). Therefore, aspeed sensor fault is detected when the speed index ωe,index ex-ceeds its threshold value. When the sensor fault is detected,the estimated speed can be provided as a feedback to achievethe continuous operation of the drive. Further, the proposedHOSM-based fault-detection approach is compared with therecently proposed high-speed SMO-based fault detection tech-nique in Fig. 11. Fig. 11(a) shows the reference speed and mea-sured speed, where a fault is applied at t = 0.5 s. Then, thecorresponding speed index exceeds the threshold, as depicted in


Fig. 11. Fault detection of a speed sensor using a high-speed SMO fora ramp speed profile. (a) Reference and actual speeds. (b) Speed indexand its threshold.

Fig. 11(b). Due to the steady-state error obtained with the ob-server, the threshold must be set higher than the previous valueto avoid the false alarms. Major drawbacks/observations withthis approach are as follows: Though the high-speed SMO uses asigmoid function, it still requires a low-pass filter (design tradedoff) to smoothen the back EMFs, whereas our HOSM observerprovides smooth estimation of back EMFs without requiring anexternal low-pass filter. Furthermore, the threshold must be sethigh due to the steady-state speed error. Hence, the efficiencyof the fault-detection scheme is decreased.

VI. DISCUSSION

The proposed results of this paper provide the automatic ve-hicle speed control to track its reference speed in the presence ofmotor speed measurement faults and external vehicular distur-bances, which reduces the driver fatigue and improves comfortfor long drives across highways and sparsely populated roads.A threshold-based fault-detection technique using the robustHOSM observer (giving reduced chattering and finite-time con-vergence) is employed for the reconfiguration to provide thefault-tolerant motor torque for an EV.

The proposed HOSM observer gains (G1 , G2 , G3 , and G4)selection is important during the implementation. For a desiredspeed range, the observer gains have to satisfy the conditions in(28). The observer gains must be appropriately chosen depend-ing on the motor speed operation. However, the observer gainsthat are tuned for a particular speed of operation work well forthe wide speed range.

It is well known that the algorithms relying on thresholdvalues are sensitive. Therefore, the threshold value is selectedso as to avoid the false alarms caused by loading conditions,machine parameters, load transients, etc. In this paper, two dif-ferent threshold values are used in simulations and experimentsaccording to the motor ratings and loading characteristics.

The effectiveness of the developed sensor FTC approach withsudden or abrupt speed sensor faults by using a threshold valueis presented. One should note that the incipient (and/or inter-mittent) sensor faults can also be detected using the proposed

approach within a short time, until the speed index crosses thethreshold value. However, this detection time can be minimizedwith a proper selection of the preset threshold.

VII. CONCLUSION

This paper presented an HOSM-observer-based sensor FTCapproach for EVs using a variable load torque of the PMSMdrive. This HOSM-based realization provides the fault-tolerantmotor torque for an EV to automatically regulate the vehiclespeed to a desired vehicle reference whenever a fault occurs inthe speed sensor. Various vehicle reference speeds with motorspeed sensor faults were chosen to control the actual vehiclespeed in the presence of vehicular disturbances using the pro-posed FTC approach. CarSim results on a flat and dry road witha road roughness profile and in the presence of aerodynamicloads confirmed the effectiveness of the proposed HOSM-basedsensor FTC architecture. Experimental validation was reportedto justify the working of the proposed HOSM observer andfault-detection approach in a real-time EV scenario.

REFERENCES

[1] R. A. Auckland, W. J. Manning, O. M. J. Carsten, and A. H. Jamson, “Ad-vanced drive assistance systems: Objective and subjective performanceevaluation,” Veh. Syst. Dyn., vol. 46, no. 1, pp. 883–897, Jan. 2008.

[2] N. Murgovski and J. Sjoberg, “Predictive cruise control with autonomousovertaking” in Proc. 54th IEEE Conf. Decis. Control, Osaka, Japan,Dec. 2015, pp. 644–649.

[3] H. Dahmani, M. Chadli, and A. R. Hajjaji, “Road curvature estimationfor vehicle lane departure detection using a robust Takagi–Sugeno fuzzyobserver,” Veh. Syst. Dyn., vol. 51, no. 5, pp. 581–599, May 2013.

[4] M. Pacas, “Sensorless drives in industrial applications,” IEEE Ind. Elec-tron. Mag., vol. 5, no. 2, pp. 16–23, Jun. 2011.

[5] S. Aouaouda, M. Chadli, and H. R. Karimi, “Robust static output feedbackcontroller design against sensor failure for vehicle dynamics,” IET ControlTheory Appl., vol. 8, no. 9, pp. 728–737, Jun. 2014.

[6] J. Hu, D. Yin, and Y. Hori, “Fault-tolerant traction control of electricvehicles,” Control Eng. Pract., vol. 19, no. 2, pp. 204–213, Feb. 2011.

[7] X. Li, S. Yin, and H. Gao, “Passivity-preserving model reduction withfinite frequency H∞ approximation performance,” Automatica, vol. 50,no. 9, pp. 2294–2303, Sep. 2014.

[8] S. Yin, H. Luo, and S. X. Ding, “Real-time implementation of fault-tolerant control systems with performance optimization,” IEEE Trans.Ind. Electron., vol. 61, no. 5, pp. 2402–2411, May 2014.

[9] Z. Feng and G. Hu, “Distributed fault-tolerant control for a large-scalepower generator network” in Proc. Amer. Control Conf., Chicago, IL,USA, Jul. 2015, pp. 5521–5526.

[10] J. Qiu, S. X. Ding, H. Gao, and S. Yin, “Fuzzy-model-based reliable staticoutput feedback H∞ control of nonlinear hyperbolic PDE systems,” IEEETrans. Fuzzy Syst., vol. 24, no. 2, pp. 388–400, Apr. 2016.

[11] S. Yin, G. Wang, and H. Gao, “Data-driven process monitoring based onmodified orthogonal projections to latent structures,” IEEE Trans. ControlSyst. Technol., vol. 24, no. 4, pp. 1480–1487, Jul. 2016.

[12] B. Tabbache, M. E. Benbouzid, A. Kheloui, and J. Bourgeot, “Virtual sen-sor based maximum likelihood voting approach for fault tolerant controlof electric vehicle powertrains,” IEEE Trans. Veh. Technol., vol. 62, no. 3,pp. 1075–1083, Mar. 2013.

[13] R. Marino, S. Scalzi, P. Tomei, and C. M. Verrelli, “Fault-tolerant cruisecontrol of electric vehicles with induction motors,” Control Eng. Pract.,vol. 21, no. 6, pp. 860–869, Jun. 2013.

[14] K. C. Veluvolu, M. Defoort, and Y. C. Soh, “High gain observer withsliding mode for nonlinear state estimation and fault recosntruction,” J.Franklin Inst., vol. 351, no. 4, pp. 1995–2014, Apr. 2014.

[15] C. Chakraborty and V. Verma, “Speed and current sensor fault detectionand isolation technique for induction motor drive using axes transforma-tion,” IEEE Trans. Ind. Electron., vol. 62, no. 3, pp. 1943–1954, Mar. 2015.

[16] Z. Gao, “Fault estimation and fault-tolerant control for discrete-timedynamic systems,” IEEE Trans. Ind. Electron., vol. 62, no. 6, pp. 3874–3884, Jun. 2015.


[17] B. Xiao, M. Huo, X. Yang, and Y. Zhang, “Fault-tolerant attitude stabiliza-tion for satellites without rate sensor,” IEEE Trans. Ind. Electron., vol. 62,no. 11, pp. 7191–7202, Nov. 2015.

[18] K. T. Chau, C. C. Chan, and C. Liu, “Overview of permanent magnetbrushless drives for electric and hybrid electric vehicles,” IEEE Trans.Ind. Electron., vol. 55, no. 6, pp. 2246–2257, Jun. 2008.

[19] K. C. Veluvolu, Y. C. Soh, W. Cao, and Z. Y. Liu, “Observer with multiplesliding modes for a class of nonlinear uncertain systems.” in Proc. Amer.Control Conf., Portland, OR, Jun. 2005, pp. 2445–2450.

[20] K. C. Veluvolu and Y. C. Soh, “Multiple sliding mode observers andunknown input estimation for lipschitz nonlinear systems,” Int. J. RobustNonlinear Control, vol. 21, no. 11, pp. 1322–1340, Jul. 2011.

[21] T. Floquet, C. Edwards, and S. K. Spurgeon, “On sliding mode observersfor systems with unknown inputs,” Int. J. Adapt. Control Signal Process.,vol. 21, no. 8–9, pp. 638–656, Oct./Nov. 2007.

[22] K. C. Veluvolu and Y. C. Soh, “Discrete-time multiple sliding mode stateand unknown input estimations for nonlinear systems,” IEEE Trans. Ind.Electron., vol. 56, no. 9, pp. 3443–3452, Sep. 2009.

[23] K. C. Veluvolu and D. Lee, “Sliding mode high-gain observers for aclass of uncertain nonlinear systems,” Appl. Math. Lett., vol. 24, no. 3,pp. 329–334, Mar. 2011.

[24] H. Kim, J. Son, and J. Lee, “High-speed sliding-mode observer for thesensorless speed control of a PMSM,” IEEE Trans. Ind. Electron., vol. 58,no. 9, pp. 4069–4077, Sep. 2011.

[25] Z. Qiao, T. Shi, Y. Wang, Y. Yan, C. Xia, and X. He, “New slidingmode observer for position sensorless control of permanent magnet syn-chronous motor,” IEEE Trans. Ind. Electron., vol. 60, no. 2, pp. 710–719,Feb. 2013.

[26] J. A. Moreno and M. Osorio, “Strict Lyapunovfunctions for the super-twisting algorithm,” IEEE Trans. Automat. Control, vol. 57, no. 4,pp. 1035–1040, Apr. 2012.

[27] J. J. Rath, K. C. Veluvolu, M. Defoort, and Y. C. Soh, “Higher-ordersliding mode observer for estimation of tyre friction in ground vehicles,”IET Control Theory Appl., vol. 8, no. 6, pp. 399–408, Apr. 2014.

[28] L. Zhao, J. Huang, H. Liu, B. Li, and W. Kong, “Second-order sliding modeobserver with online parameter identification for sensorless inductionmotor drives,” IEEE Trans. Ind. Electron., vol. 61, no. 10, pp. 5280–5289,Oct. 2014.

[29] T. Gonzalez, J. Moreno, and L. Fridman, “Variable gain super-twistingsliding mode control,” IEEE Trans. Automat. Control, vol. 57, no. 8,pp. 2100–2105, Aug. 2012.

[30] M. Basin, R. R. Pablo, S. Ding, and S. Dominic, “A nonhomogeneoussuper-twisting algorithm for systems of relative degree more than one,” J.Franklin Inst., vol. 352, no. 4, pp. 1364–1377, Apr. 2015.

[31] S. K. Kommuri, J. J. Rath, K. C. Veluvolu, and M. Defoort, “Robust fault-tolerant cruise control of electric vehicles based on second-order slidingmode observer” in Proc. IEEE Int. Conf. Control, Automat. Syst., Seoul,South Korea, Oct. 2014, pp. 698–703.

[32] A. Raisemche, M. Boukhnifer, C. Larouci, and D. Diallo, “Two activefault-tolerant control scheme of induction motor drive in EV or HEV,”IEEE Trans. Veh. Technol., vol. 63, no. 1, pp. 19–29, Jan. 2014.

[33] G. F. H. Beng, X. Zhang, and D. M. Vilathgamuva, “Sensor fault-resilient control of interior permanent-magnet synchronous motordrives,” IEEE/ASME Trans. Mechatronics, vol. 20, no. 2, pp. 855–864,Apr. 2015.

[34] A. Akrad, M. Hilairet, and D. Diallo, “Design of a fault-tolerant controllerbased on observers for a PMSM drive,” IEEE Trans. Ind. Electron., vol. 58,no. 4, pp. 1416–1427, Apr. 2011.

[35] B. Tabbache, N. Rizoug, M. E. H. Benbouzid, and A. Kheloui, “A controlreconfiguration strategy for post-sensor FTC in induction motor basedEVs,” IEEE Trans. Veh. Technol., vol. 62, no. 3, pp. 965–971, Mar. 2013.

[36] R. N. Jazar, Vehicle Dynamics: Theory and Application, 2nd ed. NewYork, NY, USA: Springer, 2014.

[37] A. F. Filippov, Differential Equations With Discontinuous RighthandSides, 1st ed. New York, NY, USA: Springer, 1988.

[38] D. U. Campos-Delgado, D. R. Espinoza-Trejo, and E. Palacios, “Fault-tolerant control in variable speed drives: A survey,” IET Control TheoryAppl., vol. 2, no. 2, pp. 121–134, Mar. 2007.

[39] G. H. B. Foo, X. Zhang, and D. M. Vilathgamuva, “A sensor fault detectionand isolation method in interior permanent magnet synchronous motordrives based on an extended Kalman filter,” IEEE Trans. Ind. Electron.,vol. 60, no. 8, pp. 3485–3495, Aug. 2013.

[40] C. Choi, K. Lee, and W. Lee, “Observer-based phase-shift fault detectionusing adaptive threshold for rotor position sensor of permanent-magnetsynchronous machine drives in electromechanical brake,” IEEE Trans.Ind. Electron., vol. 62, no. 3, pp. 1964–1974, Mar. 2015.

Suneel Kumar Kommuri (S’15) received theB.Tech. degree in electrical and electronics en-gineering from Acharya Nagarjuna University,Guntur, India, in 2010, and the M.S. degreein the field of embedded systems and controlengineering from the School of Electronics Engi-neering, Kyungpook National University, Daegu,South Korea, in 2013, where he is currentlyworking toward the Ph.D. degree.

His current research interests include sliding-mode observers/controllers, fault diagnosis and

fault-tolerant control, sensorless industrial drives, and automatic vehiclecontrol.

Michael Defoort received the Ph.D. degree inthe field of automatic control from the Ecole Cen-trale de Lille, Lille, France, in 2007.

Since 2009, he has been an Associate Pro-fessor (HDR) with the University of Valenci-ennes and Hainaut-Cambresis, Valenciennes,France. From 2007 to 2008, he was a Re-search Fellow with the Department of Sys-tem Design Engineering, Keio University, Japan.His research interests include nonlinear con-trol and observer design with applications to

power systems, multi-agent systems, and vehicles. He is as an As-sociate Editor for the Journal of The Franklin Institute. He has pub-lished more than 90 papers (34 journal papers and 1 book) in his areasof research.

Hamid Reza Karimi (M’06–SM’09) was born in1976. He received the B.Sc. (First Hons.) de-gree in power systems from Sharif University ofTechnology, Tehran, Iran, in 1998, and the M.Sc.and Ph.D. (First Hons.) degrees in control sys-tems engineering from the University of Tehran,Tehran, Iran, in 2001 and 2005, respectively.

He is currently a Professor of applied me-chanics with the Department of Mechanical En-gineering, Politecnico di Milano, Milan, Italy.His current research interests include control

systems and mechatronics.Dr. Karimi is an Editorial Board Member for several international

journals, such as the IEEE TRANSACTIONS ON INDUSTRIAL ELECTRON-ICS, IEEE TRANSACTIONS ON CIRCUIT AND SYSTEMS—I: REGULAR PAPERS,IEEE/ASME TRANSACTIONS ON MECHATRONICS, Information Sciences,IEEE ACCESS, IFAC-Mechatronics, Neurocomputing, Asian Journal ofControl, Journal of The Franklin Institute, International Journal of Con-trol, Automation, and Systems, International Journal of Fuzzy Systems,International Journal of e-Navigation and Maritime Economy, and Jour-nal of Systems and Control Engineering. He is also a Member of theIEEE Technical Committee on Systems with Uncertainty, the Committeeon Industrial Cyber-Physical Systems, the IFAC Technical Committee onMechatronic Systems, the Committee on Robust Control, and the Com-mittee on Automotive Control.

Kalyana Chakravarthy Veluvolu (S’03–M’06–SM’13) received the B.Tech. degree in electricaland electronic engineering from Acharya Na-garjuna University, Guntur, India, in 2002, andthe Ph.D. degree in electrical engineering fromNanyang Technological University, Singapore,in 2006.

During 2006–2009, he was a Research Fel-low with the Biorobotics Group, Robotics Re-search Center, Nanyang Technological Univer-sity. Since 2009, he has been with the School of

Electronics Engineering, Kyungpook National University, Daegu, SouthKorea, where he is currently an Associate Professor. He is also a Visit-ing Professor in the School of Mechanical and Aerospace Engineering,Nanyang Technological University, Singapore, for the period 2016–2017.He has been a Principal Investigator or a Coinvestigator on a number ofresearch grants funded by the National Research Foundation of Koreaand other agencies. He has authored or coauthored more than 100 jour-nal articles and conference proceedings. His current research interestsinclude nonlinear estimation and filtering, sliding-mode control, brain–computer interface, autonomous vehicles, biomedical signal processing,and surgical robotics.

/ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /CropGrayImages true /GrayImageMinResolution 150 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 300 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 1.50000 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages false /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /CropMonoImages true /MonoImageMinResolution 1200 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 600 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.50000 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile (None) /PDFXOutputConditionIdentifier () /PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped /False

/CreateJDFFile false /Description >>> setdistillerparams> setpagedevice

Documents

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 63, …ncbs.knu.ac.kr/Publications/PDF-Files/VeluvoluKC16-TIE.pdf · 2017. 3. 13. · A permanent-magnet synchronous motor (PMSM)