You are on page 1of 4

ACCURATE MODELLING AND SIMULATION OF A DC BRUSHLESS MOTOR DRIVE SYSTEM FOR HIGH PERFORMANCE INDUSTRIAL APPLICATIONS

R.A. Guinee
Cork Institute of Technology, Cork, IRELAND.

C, Lyden
NMRC, University College, Cork, IRELAND. inadequate for nonlinear BLMD system description with PWM inverter lag over a large dynamic control range. BLMD systems have been modelled by Leu eta1 [ 5 ] using a standard simulation package, with consequent long simulation times, and simplified with a linear PWM stage for model tractability. However BLMD modelling and identification in controller design must include the effects of inverter blanking, where this is a substantial part of the switching period TS [6],in addition to PWM transition time accuracy. This lag, which results in a considerable reduction of the available drive torque, is necessary for inverter protection and is typically 20% during active power transfer in large motor drives. Neural networks [7] have been applied for model reference adaptive control in BLMD systems but are computationally intensive during the training phase and have not included the effects of PWM inverter delay. Consequently effective physical modelling of the drive electrodynamics with inverter lag is a necessary prerequisite for motor parameter identification in optimal adaptive controller design. Furthermore accurate modelling is also an issue in feasibility studies concerning system performance prediction and evaluation, via simulation, where new embedded drive systems are proposed.

ABSTRACT
An accurate mathematical model of a three phase brushless motor drive (BLMD) system, based on physical principles, is presented for simulation in performance prediction studies of proposed embedded drive systems. This model can also be deployed for parameter identification of the adjustable speed drive (ASD) dynamics in optimal PID controller design and in adaptive system control. Model validation is provided, as an essential feature of the identification process, by step response transient simulation of motor current feedback and shaft velocity and comparison with actual drive test data. Penalty cost function simulation, based on the observed motor variables, is used to provide insight into the error surface topography and global minimizer location with the type of parameter search strategy that may be employed in the minimization process.

1. MOTIVATION
Brushless motors are widely used in ASD systems, with wide bandwidth speed and torque control loops, instead of conventional dc motors for high performance industrial applications. Typical applications [ 11 include, for example, numerically controlled machine tools and supervised actuation in aerospace and guided robotic manipulations where high torque and precision control are required. This is mainly due to the compactness and high torque-to-weight ratio of permanent magnet (PM) motors and virtually maintenance free operation by comparison with conventional DC types. Furthermore BLMD current control, for high torque applications, is accomplished through pulse width modulation (PWM) [2] of the inverter amplifier output (o/p) stage. This efficient method of power transfer provides a wide range of continuous power output in servo-amplifieroperation. Consideration of the above benefits of using BLMD systems in high performance industrial applications has generated a need for an accurate physical model description of the complete system for the purpose of parameter identification of the drive dynamics with inertial loads. This is essential for optimal drive controller design in which system identification is an implicit feature [3]. Allied to this need is a requirement for an accurate and efficient search strategy of parameter space during online system identification [3]. Two important interrelated areas where system identification plays a key role in controller design for industrial drive automation include PID auto-tuning and adaptive control. In the former case proper selection of the PID parameters is necessary to avoid significant overshoot and oscillations in precision control applications. This is dependent on an accurate model of the nonlinear electromechanical system [8] including inverter operation with turnon delay. The application of the standard black box linear model in adaptive controller design, with recursive parameter identification [4], is

2. OBJECTIVES
In this paper a model of a typical BLMD system incorporating inverter dead time [8] is presented, based on round rotor geometry with linear magnetic structures, for motor parameter identification and simulation. This tractable model has adequate complexity to describe the nonlinear electrodynamics without incurring large computational resources. High precision evaluation of the PWM transition times can be neglected without impairment in model simulation accuracy for a fixed time step of 0.5% TS with TS =200pS. A marginal improvement in simulation accuracy is obtained by application of a single iteration of the regula falsi method in a search of the PWM crossover time for all three phase current loops The model is simulated in torque control mode for a unit step input (Up). This choice of test stimulus is motivated by the fact that during normal online operation in typical ASD industrial applications input command changes are generally sudden and steplike and is sufficient for persistent excitation. The simulated motor shaft velocity and winding current feedback transient step responses are, as the chosen observed variables, compared with similar test data from a typical BLMD system for verification of model accuracy. The feedback current (FC)step response is particularly useful for BLMD model validation as it exhibits the FM characteristics of a constant amplitude swept frequency sinusoid during buildup of motor shaft speed. The simulated FC response is shown to have both frequency and phase coherence, upon parameter convergence, with target test data which attests

0-7803-5471 -0/99/$10.0001999IEEE

V-106

Authorized licensed use limited to: IEEE Xplore. Downloaded on October 24,2011 at 05:39:27 UTC from IEEE Xplore. Restrictions apply.

A
' j

Iterative Step Size At

tL

tx

--

fH

Time

Figure 2 Regula Falsi Method

Time t (sec)

-With Delay -Without Delay


Figure 1 Transfer Function Block Diagram of a Typical BLMD System Figure 3 Phase A Base Drive BDA

to BLMD model accuracy. The observed variables chosen, along with similar test data training sets, can be used in the generation of a mean squares error (MSE) surface as the objective function to be minimized for parameter identification of motor shaft load inertia (J) changes and damping (B). The FC penalty cost surface has a smooth continuous one dimensional sinc-like profile as a function of inertia, which is the free parameter to be identified, for coarse parameter step size simulations about a known shaft value. Local minima proliferation appears to be symmetrical about the global minimizer on the sinc-like cost surface in which case the Fast Simulated Diffusion (FSD) method [9] can be used to acquire the optimal solution parameter vector X.=[J,BIT at the global extremum. The velocity objective function has a parabolic dependency in the inertia parameter variable and can be minimized by applying Powell's conjugate direction method [lo]. The accuracy with which known motor parameters are returned in various identification search schemes fiurther enhances confidence in the BLMD physical model as a simulation tool in system identification.

6 (=20pS) to reflect simulation accuracy. Increased simulation accuracy of the PWM transition time t, is obtained by a single iteration of the regula falsi method, as shown in Fig. 2, with phase j estimate given by

{ vrri ( f )-v,k ~ ( t L ) )At - {Qri (tL)-vck ( f L ) ) - { v t r i( f H ) - v c k


3.2 BLMD Mathematical Model

(fH))

(1)
+

The motor consists of a Y connected 3@ stator winding and a p pole pair PM rotor providing sinusoidally distributed magnetic flux in the intervening air gap. The assumption of linear magnetic structures and round rotor poles result in a winding self inductance Lss independent of rotor position which reduces model complexity with an accompanying reduction in simulation run-time. The total stator winding magnetic flux linkage per phase with current flow iis , for L, = 2 L,, , is given by
j s ( i j s ,t ) = L,i j s

+ hmj (8, )
2(j-1)71

(2)

3.0 MODELLING OF BLMD SYSTEM 3.1 Choice of Model Structure


The simple black box linear model is inadequate to represent the complexity and nonlinear behaviour of a three phase (3Q) commutated BLMD system, using pwm inverter control with delay, in simulation studies for system performance analysis and identification. Instead the motor drive is partitioned into its physical elements which are separately modelled by describing functions and transfer function synthesis via the Laplace Transform. These elements are then aggregated into a holistic software model of the complete system . Numerical analysis is achieved by means of subsystem model difference equations with stable solutions provided by the backward Euier integration rule using the Z Transform substitution operation
S

where

h,j (0, ) = h, sin(p0, 7 ) (3)

is the jh phase PM flux contribution which depends on rotor position 0,. The resultant phase voltage is

The developed electromagnetic motor torque is derived from the

z-1

with I=[i,s,izs,i3s]T, emf constant K,=ph, and rotor velocity 0 , = d e , / d t . The phase voltage expression (4) can be rewritten as
v . =rsijs + L - + v e j ;
JS
S

. The

numerical step size At is chosen as 5% of the inverter dead time

dij , dr

Vj

V- 107

Authorized licensed use limited to: IEEE Xplore. Downloaded on October 24,2011 at 05:39:27 UTC from IEEE Xplore. Restrictions apply.

HT Busbar U d

BASE DRIVE J

Applied Statoi Voltage v , ~

,&

ozcjz

---_
N P 9

4-

Ord

2:

60

80

100

120

140 1601 Time (mS)

A Modulating Signal Vm(t) B: Triangular Carrier Vtri(t) C: Symmetrical Double Edge PWM o/p Vs(t)

Three Phase Inverter Operation


Figure 5 Phase a Inverter Circuit

Medlum Shaft Inertia J = 11.9 kg Cm A Simulated Current Feedback (smooth) B Experimental Feedback Current (Jagged)

Figure 4 PWM Comparator Input Voltages

Figure 6 Step Response Current Feedback

is the back-emf with values of torque (K,) and EMF (K,) constants equal if proper units are used. The developed torque results in shaft velocity a ,with load described by
dw re-r, = J , *+

inverter power transistor typically 20pS, is given by

TA-. The

RC delay magnitude

6,

6 = RC l n (;?v, m ) = 1.4RC if V , = 0

(1 1)

B,w,.

(8)

BLMD Torque control is achieved by varying the amplitude and frequency of the stator winding phase voltages. These adjustments are effected by PWM of the 3@inverter output with d via a comparator driven by a triangular a dc supply u switching carrier v,,i(t) with period T , and a current control input, v,,(t) of phase a for example, used to modulate the switch duty cycle as in Fig. 4. These modulated pulse sequences v,(t) are then used to control the ON and OFF periods of the inverter power transistors which require the use of a blanking period in the PWM process to avoid dc busbar short circuit to ground as shown in Fig. 3. This delay 6, typically 20 pS, is conservatively chosen for slow switching transistors in medium power applications in the low kilowatt range and implemented using an RC network in the base drive circuitry. An inverter lag of this magnitude, for switching frequencies in the audio range (SkHz), will result in a reduction of the applied stator phase voltages Consequently blanking has and thus in the developed torque be included in the BLMD model for accurate simulation studies. The duty cycle of the PWM modulator o/p control voltage vsa(t) is determined from simulation, during one switching period T, via the regula falsi method, by the relative comparison of the current control voltage v,,(t) with the dither reference v,ri(f). The switch control pulse sequence is given by

When phase j power transistors TI+ and TJ-are "OFF" during the blanking period winding current conduction is maintained through antiparallel protection diodes shown in Fig. 5. The relationship between the states of the dc to ac converter phase j switch transistor pair, denoted by , ( k ) with k E (0,1,2) , and

the base drive voltage v y and Fy can be represented by

sJ(l)

{-

vIj( t )= V,

+Tj,

V U ( # ) = ~+T,-

is " on" is"off"


"

(12)

v y (t) = 0 =+T,, is

( t )= 0

off' TJ- is " off'

The inverter o/p voltage vjg applied between the winding phase j input connection and ground, from (12), is given by
vjg

re.

={o

U , { S J ( l ) or s J ( 2 ) & ip <0) { S J ( O ) or S J ( 2 ) & iP zO)

(13)

where current flow into stator winding is assumed positive. The stator winding neutral potential is vsg =

5 E;=,vjg

with

resultant phase voltages v . = yg ,. - v sg V j e ( l = a , 2 = b , 3rc)

(9)

with modulated pulse duration

(14) The electrodynamic operation described in (2) to (14), together with the subunit description in the s domain using the Laplace transform, form rhe basis of a mathematical model of a BLMD for complete simulation of the drive system shown in Fig. 1.

4.0 BLMD TESTING AND SIMULATION

Iv,,

( t )5 -Ad

When inverter lag is introduced switching is postponed until the capacitor voltage of the complementary RC delay circuit exceeds the base drive threshold level v,h as shown in Fig. 3 for phase a

The output torque versus speed characteristic for a typicd BLMD system [8] with rated continuous power output of 1.5kW is almost constant over a 4000 rpm speed range. The dynamic transfer characteristic of motor o/p torque versus i/p torque demand ( T d ) voltage is linear in the i/p range (0,&10) volt. A torque demand step i/p was chosen to gauge the accuracy of the

v-108

Authorized licensed use limited to: IEEE Xplore. Downloaded on October 24,2011 at 05:39:27 UTC from IEEE Xplore. Restrictions apply.

Ifa Cost Surface

A Shaft Velocity Cost Surface

Time (mS)

- - - z

"

~Jm=3.0e-4 0 kg.m-2*20%and 30 AJ,"

A: Simulated Shaft Velocity (smooth) B: Experimental !Shaft Velocity (jagged)

Figure 9 Shaft Velocity Objective Function Figure 8 Current Feedback Cost Surface model simulation and parameter extraction process based on FC test data, possessing FM attributes, with BLMD simulation for transient response which has FM characteristics. The simulated identification purposes. Excellent fit of motor drive trial FC step response can be compared with similar test data for simulations with test data for known shaft inertial load is model validation using a frequency and phase coherence check. achieved. Cost function formulation, based on current feedback and shaft velocity target data, provide insight and visibility as to A Td step i/p of lvolt is sufficient, to guarantee linear PWM modulator operation and retardation of shaft speed increase, for the nature of the response surface topography and location of adequate resolution of swept frequency in the FC test data global minima in each case with the parameter optimization coherence check with BLMD trace simulation. The intrinsic search strategy that may be employed. motor dynamic coefficients of shaft inertia J , and viscous 6.0 REFERENCES damping B, are determined initially, in the absence of exact [ 11 EL-Sharkawi M.A. "Advanced control techniques for High apriori parameter values, by inspection from simulated MSE Performance Electric Drives". response surfacer; based on current feedback and shaft velocity Control and Dynamic Systems. Vol44, 1991, as shown in Figures 8 and 9. An example of known shaft inertial [2] Tal J. "Design and Analysis of Pulsewidth-Modulated load ( J L =9.3kg.cm2) is then used in BLMD simulation as a Amplifiers for DC Servo Systems". IEEE Trans. on check against test data for model accuracy and validation. Industrial Electronics and Control Instrumentation. The multiplexed step response current feedback Ifa and shaft Vol IECI 23, No. 1, Feb 1976. [3] Ljung L. "Issues in System Identification". velocity V, test waveforms, acquired at a 5OkHz sample rate for IEEE Control Systems. Jan 1991. loaded shaft conditions, are illustrated in Figures 6 and 7. [4] Ljung L. System identification: Theory For The User. Verification of simulation accuracy and BLMD model validation PHI, 1987. are immediately established by comparing the 1 , and V, trace [SI Leu M.C., Liu S. and Zhang H. "Modelling, Analysis and simulations with the measured data in Figures 6 and 7 in both Simulation of Brushless DC Drive System". cases. The simu1,ated current transient provides an excellent fit Winter Annual Meeting of ASME. Dec 1989. in terms of frequency and phase coherence with test data. The [6] Murai Y., Watanabe T. and Iwasaki H. Waveform Distortion accuracy of the fiit of V, trace simulation with experimental data And Correction Circuit For PWM Inverters With Switching provides additional confirmation of BLMD model integrity. , 1985. Lag Times". IEEE CH2207-9/85/0000-0436 Response surface simulation is an effective system identification [7] El-Sharkawi M.A., El-Samahy A.A. and El-Sayed M.L. tool which provides visibility into deciding the most suitable "High Performance Drive of DC Brushless Motors Using optimization strategy along with formulating the objective Neural Network". IEEE Trans. on Energy Conversion. function to be minimized for accurate parameter extraction. The Vo1.9, No.2, June 1994. cost function, based on the errors of fit between the actual [8] "Moog Brushless Technology User Manual:D3 1X-XXX BLMD sampled test data and the simulated model response, is Motors,T158-O1X Controllers, T157-001 Power Supply". minimized in terms of the parameter vector X=[J,BIT to be Moog Irl. Ltd, Feb 1989. identified using tlhe least MSE criterion. A suitable optimization [9] Guinee R.A and Lyden C. "A Novel Application of the Fast method can be applied with the BLMD model, to either the 1 , Simulated Diffusion Algorithm in Brushless Motor or V, response surface obtained for a given inertial load, in the Parameter Identification ". The Third IEEE European extraction of the the optimal vector XOpt. The Powell [lo] and Workshop on Computer-Intensive Methods in Control and FSD [9] optimization techniques for example, can be applied to Data Processing. Prague, Czech Republic. Sep7-9, 1998. [lo] Fletcher R. Practical Methods of Optimization. J. Wiley the respective V, and Ifa response surfaces to identify X,. & Sons, 1987.

Figure 7 Shaft Velocity Step Response

5.0 CONCLUSIONS

A very accurate model of a high performance BLMD system for industrial applications has been presented. Model validation is provided by frequency and phase coherence of step response FC

V-109

Authorized licensed use limited to: IEEE Xplore. Downloaded on October 24,2011 at 05:39:27 UTC from IEEE Xplore. Restrictions apply.

You might also like