Beruflich Dokumente
Kultur Dokumente
2, MARCH/APRIL 2009
Implementation of a Four-Pole
Dead-Time-Compensated Neutral-Point-Clamped
Three-Phase Inverter With Low
Common-Mode Voltage Output
Robert M. Cuzner, Senior Member, IEEE, Ashish R. Bendre, Member, IEEE,
Peter J. Faill, and Boris Semenov, Member, IEEE
Abstract—This paper describes the implementation of a four- In addition, successful integration of these systems requires
pole neutral-point-clamped (NPC) three-phase inverter that pro- further insight into how the power electronics and associated
duces virtually no common-mode voltage. The low common-mode loads and sources interact on a common bus. Furthermore,
voltage output is achieved by constraining the switch states of the
NPC inverter to only those states that produce zero common-mode innovative circuit topologies and control techniques are needed
voltage in dead-time compensation, which enhances the capability to make these systems compatible without incurring the cost
of the circuit to produce low common-mode voltage by compen- and weight of additional components, such as harmonic filters
sating common-mode voltages produced by reverse diode commu- and electromagnetic interference (EMI) filters.
tations. The low common-mode voltage performance is achieved A neutral-point-clamped (NPC) inverter [1] properly con-
at the expense of reduced voltage utilization and loss of dc-link
voltage balance control. In order to overcome this limitation, figured, optimized, and controlled can achieve very good
a fourth pole and associated control are added to balance the performance from the standpoint of compatibility with a three-
upper and lower dc-link voltages. This paper describes the control phase ac bus. Previous work has demonstrated how the addi-
implementations and simulation results compared to measured tional choices in switching states of the NPC inverter provide
results used in tuning and commissioning of the system. A 450-V many opportunities for optimizing a system for a particular
78-kW system is implemented in hardware, and experimental
results are provided, showing its differential mode transient and desired performance, whether low differential mode harmonic
steady-state harmonic performance as well as its common-mode performance, minimum dc-link ripple current, or low common-
output voltage. mode output voltage [2]. The focus of this paper is a viable
Index Terms—Electromagnetic compatibility, grounding, power NPC inverter implementation that produces very low common-
systems, pulse width modulated inverters. mode voltage output. This paper reports the implementation of
a fully commissioned 450-V 78-kW dc-to-ac inverter, which
I. I NTRODUCTION can be applied as the power source to an ac grid having
multiple loads (deriving its power from some other dc source,
Fig. 1. Neutral-point inverter with output filter and fourth pole added.
Fig. 6. Simulated voltage to ground produced by an NPC inverter with (left) NTV modulation versus (right) ZCM modulation during a ground fault on the dc
bus, with capacitance to ground at the load.
TABLE I
MAPPING FROM TWO-LEVEL INVERTER SWITCH STATES TO
NPC INVERTER ZCM INVERTER SWITCH STATES
TABLE III
DEAD-TIME COMPENSATION FOR T1/T2
A key requirement for any NPC implementation is having utilizing the redundant inner switch states in Fig. 4 [2]. For
the ability to balance the voltage across the upper and lower ZCM modulation, only the outer switch states are utilized, so
dc-link buses in Fig. 1. This is typically accomplished by there are no redundant switch states available.
CUZNER et al.: IMPLEMENTATION OF A FOUR-POLE NPC INVERTER WITH LOW COMMON-MODE VOLTAGE OUTPUT 821
TABLE IV
SIZE AND WEIGHT COMPARISONS OF FOURTH-POLE COMPONENTS
WITH EMI FILTER COMPONENTS
Fig. 11. Simulated fourth-pole voltage and current (I_4th ) for a 500-kW
design.
10 kHz. In order to meet the conducted EMI requirements,
it is necessary to include a second-order common-mode filter
In order to address the issue of dc-link balancing, the (common-mode inductor with capacitors tied either to chassis
topology proposed in Fig. 1 has an additional fourth pole, or some return point within the inverter). The main driver to the
with its output tied back to the dc-link midpoint through an size of the EMI filter is the common-mode current produced by
inductor [17]. The fourth pole may be implemented with either the inverter. Since the common-mode current of the proposed
a two-level inverter pole or an NPC inverter pole as shown, topology is essentially zero and minimal EMI is produced
depending on the need to control the voltage stress on the in the switching frequency range (assuming that dead-time
devices. The fourth-pole control is accomplished by measuring compensation is implemented), this EMI filter is not needed.
the difference between the upper and lower dc-link voltages On the other hand, it is recognized that some EMI filtering will
and then commanding either upper devices or lower devices be required for higher frequencies due to switch commutation;
depending on whether charge must be pulled out of upper however, these higher frequencies would not be handled well by
or lower dc link capacitors in order to maintain the balance. the EMI filter for the conventional two-level or NPC inverter,
One way to accomplish this is through a delta modulator, as so it may be assumed that a second-stage EMI filter would be
shown in Fig. 12, which commands either upper devices or required.
lower devices to be on depending on the voltage error at each Given these assumptions, a determination is made of the
switching frequency period. Fig. 11 shows the simulated fourth- achievable size and weight reduction by comparing the size and
pole voltage (referenced to the neutral) and the fourth-pole weight of the added fourth-pole inductor, fourth inverter leg,
current. The key to neutral balancing is to use the current I_4th and associated controls to the size and weight of an EMI filter.
in the fourth-pole inductor of Fig. 1 to drive the error between These comparisons were made based on actual designs and
the upper and lower voltages to zero. The controlled current procured hardware for a 250-kW system that is presently being
has a third harmonic of the fundamental component along with built. Size and weight of components at other kilowatt levels
a switching frequency component, as shown in Fig. 11. were scaled from this design. Table IV shows a comparison
When the fourth-pole inductance and switching frequency of between the size and weight of the added components for the
the additional inverter leg are optimized, the fourth-pole current proposed topology versus EMI filter components that would
rating will be approximately 50% of the output current rating. otherwise be required. It is clear that this topology is more
From a size and weight standpoint, the optimal point is found beneficial from a size standpoint at power ranges above 100 kW.
by increasing the fourth-pole inductance to the point where the The reason for this is that a reasonable assumption that switch-
fourth-pole current controlled to a minimum value and the LI ing frequency can be doubled for the 100- and 75-kW appli-
product of the fourth-pole inductor has reached a global mini- cations is made, which allows for a smaller EMI filter. The
mum over the range of inductance values, achievable switching anomaly of low improvement in weight at the 100-kW rating
frequencies and control gains. is due to the assumption that the capacitance of the EMI filter
and the control hardware of the proposed topology remain fixed
for all power levels. The result shows that at some point, the
VI. S IZE AND W EIGHT I MPACTS weight, in particular, of the required additional components is
An important consideration is whether it will be more ad- balanced out by the weight of the EMI filter components. The
vantageous to utilize common-mode filtering either with a results of Table IV indicate that the proposed topology is more
two-level inverter or with an NPC inverter that utilizes all advantageous for higher power inverters and is probably not the
of its states rather than to utilize the ZCM modulation along best option for lower power inverters.
with the additional hardware and control as proposed in this
paper. This approach is best applied where the common-mode
VII. C ONTROL I MPLEMENTATION
conducted emission limits extend to a range that includes low-
order harmonics of the switching frequency, such as the CE102 The control block diagram for the system is shown in Fig. 12.
requirements of Mil-Std-461E where the low-end frequency is The control is implemented on a hardware platform which
822 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 45, NO. 2, MARCH/APRIL 2009
utilizes an AD 21160 DSP for computational algorithms and The voltage and current feedbacks used in the controller are
a Xilinx XC2V3000 FPGA for logical operations and gate sampled and averaged eight times over one 100-μs interval.
synthesis. Control blocks that are implemented by the FPGA This greatly improves the differential mode output harmonic
are indicated by light-gray blocks. All other blocks shown are performance of the inverter. The “Reg Select” block in Fig. 12
implemented by the DSP. The software was programmed so that shows how the two types of regulators are selected. If the
two kinds of control could be investigated. “voltage-regulator-only” mode is selected, then the outputs
The first type of control is a cascaded outer loop voltage from which “m” and “γ” are constructed come directly from
regulator with an inner loop current regulator. The components the “PI Reg” blocks of the outer loop voltage controller.
of the inner loop current regulator are shown to be highlighted If the “cascaded voltage regulator with inner loop current
by the light-gray box in Fig. 12. regulator” is selected, then the outputs of the inner loop current
Both of these regulators are implemented in the synchro- regulator form “m” and “γ.” The block diagram also shows
nous frame, i.e., the three-phase voltages and currents are how the ZCM modulator, dead-time compensator, and fourth-
transformed from stationary “abc” frames to “dq” frames pole controller are implemented.
that are synchronous with the instantaneous electrical angle
derived from the commanded frequency. The q-axis voltage
VIII. E XPERIMENTAL R ESULTS
corresponds to the peak output phase voltage of the dc-to-
ac inverter. The q-axis and d-axis current commands to the The experimental setup is shown in Fig. 13. The hardware
inner loop current regulator correspond to the real and reactive includes two parallel multiuse H-bridge power electronic mod-
currents, respectively, that must be delivered to the load in ules which implement the four-pole three-phase NPC inverter
order to satisfy the outer loop voltage regulator. The inner loop of Fig. 1. The input to this system was a 12-pulse transformer
current regulator also includes load decoupling, which has the rectifier. All experimental results are at a 78-kW output power
measured load currents in the dq reference frame multiplied by condition, limited only by the available power source. Fig. 14
a gain Ko (close to unity) and added to the current regulator shows the input dc line-to-ground voltage from the dc source,
references. These are required to improve the voltage transient the input dc line-to-line voltage, the output voltage measured
response of the cascaded regulator. from one output phase to the neutral point (i.e., midpoint of
CUZNER et al.: IMPLEMENTATION OF A FOUR-POLE NPC INVERTER WITH LOW COMMON-MODE VOLTAGE OUTPUT 823
Fig. 15. Voltage from line to neutral (VLN ) NPC inverter with ZCM modula-
tion, no dead-time compensation, 200 V/Div.
Fig. 14. (Top trace) Input voltage to ground of source Vs_g , (second trace)
line to neutral voltage VLN , (third trace) input voltage Vdc , and (bottom trace)
fourth-pole current I_4th . Note: Volt/Div and A/Div are shown at the bottom. Fig. 16. VLN (zoomed in) without dead-time compensation.
Fig. 18. Line-to-line voltage (VLL ) spectrum, with outer loop voltage Fig. 19. Line-to-line voltage (VLL ) spectrum, with outer loop voltage regula-
regulator/inner loop current regulator. tor only.
Peter J. Faill received the B.S. degree in physics and Boris Semenov (M’90) received the B.Sc. and M.S.
mechanical engineering from Columbia University, degrees in electrical engineering from St. Petersburg
New York, NY, in 1987, and the M.S. degree in State Technical University, St. Petersburg, Russia,
theoretical physics from Tufts University, Medford, in 1982.
MA, in 1992. He has over 20 years of industrial power converter,
He has over 20 years of digital controls design and motor drive design, and development experience. In
development experience, both in semiconductor tool the past, he was with Ametek, Otis Elevator Com-
manufacturing and in PM motor/generator controls. pany, and International Rectifier. He is currently with
He is currently a Switching Topologist with DRS DRS Power Systems, Hudson, MA, working on the
Power Systems, Hudson, MA, where he develops development and design of high-power conversion
and implements a wide range of device switching equipment. His primary areas of interest are in power
schemes for PM machines and power conversion. electronics, high-performance ac drives, intelligent power management, and
power systems.