22-Elec-B8 Power Electronics and Drives · December 2017
Nivaar worked solution (AI-drafted; not reviewed by a licensed engineer)
Paper format. National Exams, December 2017 — 16-Elec-B8, Power Electronics and Drives. Open book, three hours, non-communicating calculator permitted. The paper is in two parts: Part 1 is twenty short-answer items (a) to (t) worth 2.5 points each, and Part 2 is five calculation problems worth 15 points each. The rubric says “attempt all parts” and that the maximum total score of 125 points includes a bonus of 25 points, so a candidate scoring 100 of the 125 available marks has a full paper. Every item and every sub-part is worked below.
Reference texts.
Question text not reproduced: the examination questions are © Engineers and Geoscientists BC. Open the official past paper (linked at the top of this page) to read the question, then follow the worked solution below.
Check — the paper prints one item three times. Items (b), (f) and (g) are the same question, printed three times, and item (a) is the smoothing-reactor half of it asked on its own. That is an error on the printed paper, not a subtlety to be decoded: four of the twenty items, ten of the fifty points, address one topic. Each is answered below rather than cross-referenced, but the emphasis is deliberately shifted — (b) gives the balanced two-component answer, (f) concentrates on the clamping capacitor and (g) on the reactor — so that the set is worth reading as a study resource. A candidate meeting this in the hall should answer (b) fully, then write “as (b)” for (f) and (g) and spend the time saved on Part 2.
A series smoothing reactor is an inductor placed in the current path of a switching converter for the sole purpose of limiting the rate of change of current. In an inverter it is used for three distinct reasons. First, it protects the switching devices during turn-on. A thyristor or IGBT turns on over a finite time, and while the conducting region is still spreading across the die the full load current must be carried by a small area. A series reactor holds the anode current rate of rise below the device rating (a typical thyristor limit is of the order of 100 to 200 A/microsecond), so the device does not fail by local overheating.
Second, it limits the fault and commutation currents. During forced commutation the firing of the incoming device momentarily short-circuits the outgoing one through the commutating capacitor, and without series inductance that discharge would be limited only by stray resistance. The reactor also converts a shoot-through fault from an instantaneous device failure into a current ramp slow enough for the protection to act.
Third, on the d.c. link of a current-source inverter the series reactor is the element that makes the source behave as a current source at all: a large link inductance holds the d.c. current nearly constant over a switching period, so the ripple that the inverter imposes on the supply is small and the input current drawn from the rectifier is smooth. The same reactor filters the harmonic voltage that the switching produces, keeping it out of the supply.
These are the two halves of the snubbing arrangement that surrounds each switch in an inverter, and they act on the two different stresses that switching produces.
The clamping capacitor is a low-inductance capacitor connected across the device, or across the d.c. bus immediately at the device terminals, usually through a diode with a discharge resistor. Its function is to absorb the energy trapped in the circuit stray inductance when the current in the device is interrupted. When a switch carrying current turns off, the stray inductance of the bus tries to maintain that current and produces a voltage spike of magnitude L times di/dt across the opening device. The clamping capacitor accepts the current for the few microseconds required, so the device voltage rises slowly and is clamped near the d.c. bus level instead of overshooting into avalanche. In doing so it also holds the reapplied dv/dt below the device rating, which for a thyristor matters twice over: an excessive reapplied dv/dt injects displacement current into the gate-cathode junction and turns the device on with no gate signal at all.
The smoothing reactor works on the complementary quantity. Placed in series with the device, it limits di/dt at turn-on, protecting the spreading conduction region as described in (a), and on the d.c. link it smooths the current so that the source sees a nearly ripple-free draw. Together the two components confine the switching locus to the safe operating area: the reactor keeps current low while voltage is still high at turn-on, and the capacitor keeps voltage low while current is still high at turn-off.
Power-electronic converters. Line-commutated rectifiers, a.c. voltage controllers, and inverter front-ends draw current in blocks rather than sinusoids. A six-pulse bridge draws a quasi-square current containing harmonics of order 6k plus or minus 1, so the 5th, 7th, 11th and 13th are present with amplitudes falling roughly as 1/n. Because these loads are now the largest single class of non-linear load on a distribution feeder, they dominate the harmonic budget of most modern systems.
Saturating magnetic circuits. Transformers, reactors and rotating machines operated near or above the knee of their magnetisation curve draw a magnetising current that is strongly peaked rather than sinusoidal. The dominant component is the third harmonic and its odd multiples. Over-voltage on a lightly loaded feeder pushes transformers further into saturation and raises this contribution sharply, which is why harmonic distortion often worsens at light load.
Arcing and discharge loads. Arc furnaces, arc welders, and gas-discharge lighting have a non-linear and time-varying voltage-current characteristic. The arc extinguishes and restrikes each half cycle, producing both odd harmonics and, because the arc length varies randomly, a broad interharmonic and flicker spectrum that is much harder to filter than the discrete orders produced by converters.
A free-wheeling diode, also called a commutating or bypass diode, is connected across an inductive load in reverse to the applied voltage. Its function is to provide a closed path for the load current when the controlled switch turns off.
An inductor opposes any change in its current, so interrupting the current in an inductive load produces a voltage of magnitude L times di/dt of whatever size is needed to maintain it — in practice a destructive spike across the switch. With the diode fitted, the instant the load terminal voltage tries to go negative the diode becomes forward biased and takes over the current, which then decays exponentially through the diode and the load resistance with time constant L/R. Three consequences follow. The switch sees only the supply voltage, not an inductive spike, so a lower-voltage and cheaper device may be used. The load voltage is clamped near zero during the off interval instead of swinging negative, so the mean output voltage of a controlled rectifier rises and its output waveform improves. And in a rectifier the diode prevents the supply from being asked to absorb the inductive energy, which raises the input displacement factor. The energy stored in the inductance is not destroyed but dissipated in the load resistance and the diode — a genuine, if modest, loss that must be counted.
The turn-off time tq (circuit-commutated turn-off time) is the minimum interval for which the anode must be held reverse-biased, from the instant the anode current reaches zero to the instant forward voltage may be reapplied without the device turning on again. It is set by how quickly the stored charge in the four layers recombines, and five factors change it.
Junction temperature. This is the dominant factor. Carrier lifetime increases with temperature, so recombination slows and tq lengthens; a rise from 25 to 125 degrees Celsius can roughly double it. Data-sheet values are quoted at maximum rated junction temperature for this reason.
Magnitude of the anode current before commutation. A larger forward current leaves a larger excess-carrier charge in the base regions, so more charge must be removed and tq increases.
Rate of decay of the forward current, di/dt at commutation. A rapid decay sweeps carriers out efficiently and produces a larger reverse recovery current, which shortens the recovery; a slow ramp to zero leaves more charge behind and lengthens it.
Magnitude of the reverse voltage applied during turn-off. A larger reverse bias drives a larger reverse recovery current and extracts charge faster, shortening tq. Too small a reverse bias, or none, may leave the device unable to regain its blocking state at all.
Rate of reapplication of forward voltage, dv/dt. A steep reapplied forward voltage injects displacement current through the depletion capacitance and can retrigger a device that has not fully recovered, so the effective tq a circuit must allow grows with the reapplied dv/dt. Gate conditions are a fifth influence worth noting: a reverse or zero gate bias during the recovery interval shortens tq, whereas leaving the gate open or slightly positive lengthens it.
This item repeats (b) verbatim. The clamping capacitor deserves the closer look. It is a low-inductance capacitor, connected as directly across the switching device as the layout permits, normally in series with a diode and shunted by a discharge resistor so that it charges quickly on the voltage spike and discharges slowly between events. Its purpose is to hold the device voltage at or just above the d.c. bus level while the current in the stray inductance decays. It must therefore be sized on energy, not on capacitance alone: the stray-inductance energy one-half L I squared has to be absorbed for a voltage rise the device can tolerate, which fixes C directly. Its placement matters as much as its value, because any inductance between the capacitor and the die is inductance the capacitor cannot clamp. In a multilevel or a series-connected string the clamping capacitors also perform voltage sharing, forcing devices with unequal recovery characteristics to divide the bus voltage in proportion to their capacitors rather than in proportion to their leakage. The smoothing reactor, treated in (a) and (g), is the di/dt-limiting complement to it.
This item also repeats (b) verbatim, so the reactor is treated more closely here. The smoothing reactor is a series inductance whose value is chosen from the di/dt rating of the device and the worst-case voltage that can be applied across it: L must be at least V divided by the rated di/dt. Because that requirement is met at quite small inductances, a few tens of microhenries typically suffices for a turn-on reactor, and it is often built as an air-cored coil or as a saturable reactor. A saturable reactor is preferred where the inductance is wanted only at low current: it presents a high inductance while the current is small, protecting the device during the critical spreading interval, and then saturates so that it contributes almost nothing to the conduction-interval voltage drop. A second, much larger, reactor appears on the d.c. link of a current-source inverter, where its job is not device protection but the maintenance of a nearly constant link current so that the inverter genuinely sees a current source. Both applications share the same principle — an inductor converts an abrupt current demand into a ramp — and both are the current-side counterpart of the voltage-side clamping capacitor.
The circuit is one thyristor in series between a single-phase a.c. supply and a resistance. During the positive half cycle the thyristor is forward biased but remains in the blocking state, so no current flows and the whole supply voltage appears across the device. At the chosen delay (firing) angle alpha, measured from the supply zero crossing, a gate pulse is applied. The thyristor latches, its voltage collapses to about one volt, and the load sees the supply voltage from alpha onwards. Because the load is purely resistive the current is in phase with the voltage, so both reach zero together at omega t = pi. The device current falls below the holding current and the thyristor turns off by natural (line) commutation; the gate has no part in turning it off. Through the negative half cycle the device is reverse biased and blocks, so the load current is zero. Conduction therefore occupies the window from alpha to pi each cycle, giving a conduction angle of 180 degrees minus alpha.
The mean and rms output voltages follow directly:
$$\begin{aligned} V_{dc}&=\frac{1}{2\pi}\int_{\alpha}^{\pi}V_m\sin\theta\,\mathrm{d}\theta =\frac{V_m}{2\pi}\bigl(1+\cos\alpha\bigr)\\ V_{rms}&=V_m\sqrt{\frac{1}{4\pi}\Bigl[(\pi-\alpha)+\frac{\sin 2\alpha}{2}\Bigr]} \end{aligned}$$so the output is fully controllable from the uncontrolled value at alpha = 0 down to zero at alpha = 180 degrees. The penalties are a large ripple, a d.c. component in the supply current that can saturate a feeding transformer, and a poor input power factor at large alpha.
A chopper is a d.c.-to-d.c. converter: a switch, now normally an IGBT or a MOSFET, is opened and closed rapidly between a fixed d.c. source and the load, and a free-wheeling diode carries the inductive load current while the switch is open. Over one period T the switch is closed for Ton and open for Toff, and because the load cannot follow the switching the load sees the mean value
$$V_o=\frac{T_{on}}{T}V_i=\delta V_i,\qquad \delta=\frac{T_{on}}{T}$$where the duty ratio delta is the single control variable. Varying the on-time changes the operating mode as well as the mean output. At small delta the free-wheel interval is long compared with the load time constant L/R, the current decays to zero before the switch closes again, and the converter is in discontinuous conduction: the output voltage then depends on the load as well as on delta, and the transfer characteristic becomes non-linear. As delta is raised, a critical on-time is reached at which the minimum current just touches zero; above it the current is continuous, the ripple is bounded by the exponential rise and decay between Imax and Imin, and the mean output is the clean linear delta times Vi. Pushing delta towards unity gives the largest output but leaves so short a free-wheel window that the ripple is squeezed into a small fraction of the period. There are two ways to vary delta: time-ratio control at fixed frequency, which keeps the ripple spectrum at known orders and is what modern drives use, and frequency modulation at fixed on-time, which spreads the harmonics over a range of frequencies and makes filtering harder.
Classified by the supply they take, d.c. drives fall into three groups. A.C.-fed (rectifier or phase-controlled) drives take a single-phase or three-phase a.c. supply and use a line-commutated thyristor converter — half-wave, semi-converter, full converter, or dual converter — to produce a controlled d.c. armature voltage. A single converter gives one-quadrant or two-quadrant operation; a dual converter in back-to-back connection gives all four quadrants. D.C.-fed (chopper) drives take an existing d.c. supply such as a traction third rail, a battery, or a rectified and filtered bus, and use a chopper to vary the armature voltage. Ward-Leonard and motor-generator schemes are the historical third class, now almost entirely displaced by the first two, in which a motor-generator set supplies the variable armature voltage.
In a variable-speed d.c. drive there are two controlled variables, and they act in different speed ranges. The armature voltage is the control below base speed: with the field held at rated value, speed is very nearly proportional to armature voltage, and constant-torque operation is available over that whole range. The field current (and hence flux) is the control above base speed: with the armature voltage held at its rated ceiling, weakening the field raises the speed inversely with flux, and the drive operates at constant power with torque falling as 1/speed. In closed loop the armature current is controlled as well, as the inner loop of a cascade structure, because current is what sets torque and what must be limited during acceleration.
The distinction is the nature of the d.c. link. A voltage-source inverter (VSI) is fed from a stiff d.c. voltage: a large capacitor across the link holds the voltage constant, the link impedance is low, and the inverter output voltage waveform is imposed by the switching pattern while the current waveform is whatever the load draws. A current-source inverter (CSI) is fed through a large series reactor from a controlled rectifier operating in current control: the link current is held constant, the link impedance is high, and the inverter imposes the output current waveform while the voltage is whatever the load develops.
Several practical differences follow. The VSI output voltage is a rectangular or PWM waveform and its current is roughly sinusoidal for an inductive load; the CSI output current is a quasi-square block and the voltage carries the switching spikes. A VSI tolerates an output short circuit badly and needs fast overcurrent protection, whereas the CSI is inherently current limited and rides through a short circuit. A VSI needs anti-parallel feedback diodes to return reactive energy to the link; a CSI needs series-blocking diodes instead, because its devices must block reverse voltage. Regeneration in a CSI is straightforward — the link current keeps its direction and the link voltage simply reverses — while a VSI needs either a dual converter or a braking chopper. Against that, the CSI is sluggish because of the large link reactor and it cannot easily run more than one motor from one inverter, so the VSI with PWM is the general-purpose choice and the CSI survives in large single-motor, high-power, regenerative applications.
The fully controlled three-phase bridge has six thyristors in three legs. At any instant one device of the upper group and one of the lower group conduct, connecting the most positive and the most negative of the available line terminals to the load, so the output is composed of segments of the line-to-line voltages. Each device conducts for 120 degrees and the devices are fired in the sequence T1 to T6 at 60-degree intervals, giving six pulses of output per supply cycle. Measuring the firing angle alpha from the natural commutation point (the crossing of two line voltages), the mean output voltage is
$$V_a=\frac{3\sqrt{2}}{\pi}V_{LL}\cos\alpha = 1.3505\,V_{LL}\cos\alpha$$which is fully controllable from its maximum at alpha = 0, through zero at alpha = 90 degrees, to a negative value in the inverting region beyond 90 degrees. Six-pulse operation means the lowest output ripple harmonic is the sixth, which is easy to filter, and the lowest input current harmonic is the fifth.
For a separately excited d.c. motor the armature back e.m.f. is E = k phi n and the armature loop gives Va = E + IaRa. Below rated speed the field is held at its rated value and the bridge feeding the armature is used as the control: advancing alpha raises Va, raises E, and raises the speed almost in proportion, with full rated flux available so the drive delivers constant torque throughout. Above rated speed the armature bridge has run out of voltage at alpha = 0, so control passes to a second, much smaller, controlled rectifier feeding the field winding. Retarding the firing angle of that field converter weakens phi, and since n = (Va minus IaRa)/(k phi) the speed rises as the flux falls. Torque T = k phi Ia falls in the same proportion, so this is the constant-power region, and it is limited by commutation and by the maximum safe mechanical speed, typically to two or three times base speed.
Adding inductance and a back e.m.f. changes the circuit in three ways, and the change caused by each is different.
With inductance present the current can no longer follow the voltage. Once the thyristor fires at alpha the current builds against L, and when the supply voltage passes through zero at pi the inductor still holds current, so the device continues to conduct into the negative half cycle. Conduction ends at the extinction angle beta, where the current finally reaches zero, and beta exceeds pi. The load therefore sees a negative voltage over the interval from pi to beta, which pulls the mean output down; and because the mean voltage across an inductance is zero over a complete cycle, the mean current is still exactly the mean load voltage divided by R. The conduction angle gamma = beta minus alpha is no longer 180 degrees minus alpha but is fixed by the load angle phi = arctan(omega L/R) through the transcendental condition sin(beta minus phi) = sin(alpha minus phi) exp((alpha minus beta)/tan phi).
Adding a d.c. voltage source Ec in the load — the back e.m.f. of a motor, or a battery being charged — changes the conduction window again. The thyristor cannot be fired usefully until the supply instantaneous voltage exceeds Ec, which sets a minimum firing angle alphamin = arcsin(Ec/Vm). With resistance alone the load then fixes the extinction angle at beta = 180 degrees minus alphamin, independent of where the gate pulse was applied; with inductance also present, beta is pushed later still. Between beta and the next firing the load terminals sit at Ec, not at the supply voltage, so the mean load voltage is not simply the integral of the chopped sine. The mean current becomes
$$I_{dc}=\frac{1}{2\pi R}\Bigl[V_m(\cos\alpha-\cos\beta)-E_c(\beta-\alpha)\Bigr]$$and the power delivered to Ec is the useful (mechanical, or charging) power, the copper loss being accounted separately in R. In practice a free-wheeling diode is added so that the negative-voltage interval is suppressed, which restores much of the lost mean voltage at the price of a longer conduction path.
The consequence is over-fluxing, and everything else follows from it. Neglecting stator resistance, the stator voltage is balanced by the rate of change of flux linkage:
$$V_1\approx 4.44\,f\,N\,k_w\,\Phi \quad\Longrightarrow\quad \Phi \propto \frac{V_1}{f}$$so holding V1 at rated value while reducing f raises the air-gap flux in inverse proportion. A machine designed to run just below the knee of its magnetisation curve is driven hard into saturation. Four consequences follow. The magnetising current rises far faster than the flux does, because beyond saturation the magnetisation curve is nearly flat, so the no-load current can reach or exceed rated current and the stator overheats even with no shaft load. Core losses rise steeply: hysteresis loss goes roughly as flux density to the power 1.6 to 2 and, although the reduction in frequency partially offsets it, the net iron loss and the local heating in the saturated tooth tips increase. The magnetising current is badly distorted, peaked and rich in third-harmonic content, which pollutes the supply and produces additional stray loss. And the machine draws heavily reactive power, so the power factor collapses.
The breakdown torque does rise, since it varies as (V/f) squared, but this is not a usable benefit: the machine cannot be operated at the resulting flux without overheating. The correct practice is constant volts-per-hertz — reduce the voltage in proportion to the frequency, with a small boost at very low frequency to compensate the stator resistance drop that the approximation above neglects. That keeps the flux and hence the torque capability constant all the way down.
Pulse-width modulation controls the output of an inverter by varying the width of the output pulses rather than their height, so that the fundamental component of the switched waveform is controlled while the d.c. link voltage is left fixed. In the sinusoidal (sine-triangle) scheme a sinusoidal reference at the wanted output frequency f1 is compared continuously with a triangular carrier at a much higher frequency fc. Whenever the reference exceeds the carrier the upper device of that inverter leg is turned on; otherwise the lower device is. The resulting pulse widths are wide near the crest of the reference and narrow near its zero crossings, so the moving average of the switched waveform tracks the reference.
Two ratios describe the modulation. The modulation index M = Vref/Vcarrier sets the amplitude: in the linear region the peak fundamental is M Vdc/2 per leg, so output voltage is proportional to M for M up to 1, and beyond that the inverter enters over-modulation and finally square-wave operation, where the fundamental saturates at 4Vdc/(2 pi) and low-order harmonics reappear. The frequency-modulation ratio mf = fc/f1 sets where the harmonics land: the dominant harmonics cluster in sidebands about mf and its multiples, so a high carrier ratio pushes all significant distortion to frequencies the load inductance filters easily. Choosing mf odd and a multiple of three gives half-wave and three-phase symmetry, which eliminates even harmonics and triplens from the line-to-line voltage. The great advantage of PWM over the older practice of varying the d.c. link voltage is that voltage and frequency are controlled together in the inverter itself, so a simple diode rectifier can feed the link and constant volts-per-hertz operation of a motor becomes a software matter.
Motor insulation stress from reflected-wave over-voltage. A modern IGBT switches in 50 to 100 nanoseconds, which makes the cable between drive and motor a transmission line. The impedance mismatch at the motor terminals reflects the wavefront, and for cable lengths beyond a critical value the reflected and incident waves add to give terminal voltages of up to twice the d.c. bus. The steep front also distributes itself very unevenly across the stator winding, so the first turn of the first coil sees a large fraction of the whole step. The result is accelerated partial-discharge ageing of the turn insulation and premature winding failure unless inverter-duty magnet wire, an output reactor, or a dv/dt filter is used.
Bearing currents and shaft voltages. The three-phase PWM output has a non-zero common-mode (zero-sequence) component that switches at the carrier rate. This drives capacitive current through the stator-to-rotor and rotor-to-frame capacitances, and it induces a shaft voltage. When that voltage exceeds the dielectric strength of the lubricant film it discharges through the bearing, producing electric-discharge machining of the races: fluting, pitting, and bearing failure in months rather than years. Insulated bearings, shaft grounding rings, and common-mode chokes are the countermeasures.
Electromagnetic interference and switching losses. The steep edges radiate and conduct broadband interference, coupling into instrumentation, encoder feedback and communications, so screened cable, careful grounding and an EMI filter become mandatory. And because switching loss in the devices is proportional to the switching frequency, a higher carrier raises drive losses and heat-sink size directly — the price paid for the improved current waveform. A fourth effect worth naming is acoustic: a carrier in the audible band produces the characteristic magnetostrictive whine of a drive.
Silicon carbide is a wide-bandgap semiconductor, and the two properties the question asks about follow from that one fact. The bandgap of 4H-SiC is about 3.26 eV against 1.12 eV for silicon, roughly three times as large.
High temperature. A device stops working when thermally generated carriers become comparable with the doping, because the junction then loses its blocking ability and leakage runs away. The intrinsic carrier concentration depends exponentially on the ratio of the bandgap to the thermal energy, so tripling the bandgap reduces the intrinsic concentration at a given temperature by many orders of magnitude — at room temperature SiC has of the order of 10 to the minus 8 carriers per cubic centimetre against 10 to the tenth for silicon. In consequence SiC junctions still block properly at 400 to 600 degrees Celsius where silicon has long since failed, and the practical ceiling is set by packaging and metallisation rather than by the semiconductor. Its thermal conductivity, about three times that of silicon, helps further by carrying the heat out of the die.
High voltage. Avalanche breakdown begins when carriers gain enough energy between collisions to ionise the lattice, and the field needed for that scales with the bandgap. The critical field of SiC is about 2.2 to 3 MV/cm, roughly ten times silicon's 0.3 MV/cm. Because the blocking layer thickness needed for a given voltage is inversely proportional to the critical field and the specific on-resistance of a drift region scales as the cube of the breakdown field in the denominator, an SiC device of a given rating can use a drift layer about ten times thinner and about ten times more heavily doped than the silicon equivalent. That gives on-resistances of the order of a few hundred times lower for the same blocking voltage, which is why SiC MOSFETs and Schottky diodes have displaced silicon IGBTs and fast-recovery diodes at 600 V to 1700 V and made 10 kV devices practical. The saturated drift velocity, about twice silicon's, also permits faster switching, so the conduction and switching advantages compound.
Renewable sources do not naturally produce what the grid requires, and power electronics is the entire interface that reconciles the two. A photovoltaic array produces a d.c. voltage that varies with irradiance and cell temperature and whose power-voltage characteristic has a single maximum. A d.c.-d.c. converter continuously adjusts the operating point to sit on that maximum — maximum power point tracking — and a grid-tied inverter then converts to a.c. at grid frequency, synchronised in phase and compliant with the anti-islanding and power-quality requirements of the interconnection standard. Without that converter chain a PV array can deliver only a fraction of its available energy.
A wind turbine presents a different problem. Its aerodynamic efficiency is maximised at a fixed tip-speed ratio, so the optimum rotor speed varies with wind speed, and a fixed-speed machine directly connected to the grid gives up a substantial part of the available energy. The doubly fed induction generator uses a partial-rating back-to-back converter in the rotor circuit to decouple rotor speed from grid frequency over a range of plus or minus thirty per cent; the full-converter topology puts a back-to-back converter at full rating between generator and grid and decouples them completely. In either case the grid-side converter also controls reactive power, giving the voltage support and fault-ride-through that grid codes now demand of generation.
Beyond the source itself, power electronics is what makes storage and grid integration possible: bidirectional converters for battery and flywheel storage, HVDC links for long-distance and offshore transmission, and STATCOMs for voltage control on weak feeders carrying large amounts of variable generation. The general point is that renewable generation is inherently variable and unsynchronised, and every mechanism by which such a source is made to behave like a controllable, dispatchable, grid-compliant plant is implemented in a converter.
A smart grid is one that senses its own state and acts on it, and the acting is done by power electronics; the sensing and the communications only decide what the converters should do. Four roles stand out.
Controllable power flow. FACTS devices — the static VAR compensator, the STATCOM, the thyristor-controlled series capacitor, the unified power flow controller — let the operator set voltage, reactive power, and even the real power routed down a particular corridor, on a timescale of a cycle rather than the minutes a mechanical tap changer or switched capacitor bank needs. HVDC and back-to-back links do the same between whole systems, and their voltage-source-converter form can also start a dead network.
Integration of distributed and variable resources. Every rooftop PV inverter, battery, electric-vehicle charger and small wind machine reaches the network through a converter, and a modern grid code requires those converters to do far more than export watts: volt-VAR and volt-watt control, frequency-watt droop, ride-through of voltage dips, and increasingly grid-forming behaviour that supplies synthetic inertia to a system losing its synchronous machines.
Power quality and protection. Active filters cancel harmonic currents at their source, dynamic voltage restorers ride through sags, and solid-state circuit breakers and fault-current limiters interrupt in microseconds rather than cycles, which is what makes meshed d.c. distribution feasible at all.
Storage and flexibility. Bidirectional converters make the charge and discharge of grid-scale batteries dispatchable, and the same hardware in a vehicle charger turns a fleet of cars into a controllable resource. In short, the intelligence of a smart grid is in software, but its authority over the physical system is exercised entirely through switching converters.
Inside a house, power electronics does two things: it makes each load able to consume only the energy the task actually needs, and it makes the house able to respond to a signal from outside.
On the load side, almost every appliance that has become markedly more efficient in the last two decades owes that to a converter. Variable-speed drives on refrigerator and heat-pump compressors, and on furnace and ventilation fans, replace on-off cycling by continuous modulation, so the machine runs at the capacity the conditions call for instead of oscillating around it; a heat pump with an inverter-driven compressor also keeps working usefully at low outdoor temperature, where a fixed-speed unit would need resistive backup. LED lighting depends on a driver that supplies constant current and dims by pulse-width modulation. Induction hobs are simply resonant inverters. Every piece of consumer electronics is fed by a switched-mode power supply, whose efficiency and standby draw are converter design problems.
On the system side, the house meets the grid through converters as well: a rooftop PV inverter, a home battery with a bidirectional converter, and an electric-vehicle charger that in its bidirectional form can support the house or the network. Those converters are what allow a demand-response or time-of-use signal to translate into physical action — shifting a charging session, discharging a battery through an evening peak, riding through an outage on stored energy. A home energy management system decides; the power electronics does. It is also where the losses now are: a house may contain fifty converters, so their standby consumption and light-load efficiency are a real component of the annual energy bill.
