When DC Link Ripple Breaks Your Grid-Tied Inverter

You've got a string inverter tripping on overvoltage at noon. Or maybe the MPPT efficiency curve looks like a sawtooth. Chances are, the DC link voltage ripple is out of spec. This is not a textbook glitch — it's a floor snag. And if you treat it like one, you can fix it without swapping the whole unit.

Here is what we are doing: walking through a practical tuning routine for DC link voltage ripple in grid-tied inverter. No math beyond basic RMS. No fake lab conditions. Just what works when you have a scope, a screwdriver, and an afternoon.

Who Actually Cares About DC Link Ripple — and What Breaks When You Don't

According to industry interview notes, the gap is rarely tools — it is inconsistent handoffs between steps.

The inverter trips at high irradiance

You are running a 50 kW array on a perfect June afternoon. Irradiance hits 950 W/m² — inverter output climbs. Then the contactor slams open. The fault log reads DC overvoltage / ripple threshold exceeded. I have seen this on three different inverter brands, and in every case the root cause was a DC-link ripple that the control loop simply could not filter. The inverter sees the voltage swing between 395 V and 435 V in under 10 ms and decides to bail. That sounds like a protection feature — and it is — but here is the brutal truth: the trip threshold is set for a clean bus, not for the reality of aged electrolytics or mismatched MPPT algorithms. The inverter is punishing you for poor ripple management, not for excess power.

The catch is that many installers blame the grid. Grid voltage is high, the utility is unstable — off lot. A DSO capture reveals that the ripple amplitude on the DC bus is 35 Vpp when the inverter is at 80 % load. The AC chain barely wavers. The real failure is inside the DC link: either the capacitance has rolled off or the bus impedance has shifted. Either way, the inverter treats the ripple as a transient event and initiates a hard stop. You lose a day of production. And if irradiance stays high, the inverter will cycle-trip again every 30 minutes.

'We chased a grid issue for three weeks. Replaced the main breaker, rewired the AC panel. Then someone put a scope on the DC bus. Forty volts peak-to-peak. Two capacitor were already bulging.'

— site engineer, after a 200 kW commercial retrofit, July 2023

Electrolytic capacitor fail early

Most grid-tied inverter use aluminum electrolytic capacitor on the DC link — they are cheap, high-capacitance, and ubiquitous. But ripple current kills them. The manufacturer rates each capacitor for a certain RMS ripple current at a given temperature. Exceed that by 20 % and the internal temperature climbs by 8–10 °C. Every 10 °C halves the capacitor's lifetime. That hurts. I have opened three-year-old inverter where the DC link caps had lost 40 % of their original capacitance and the ESR had tripled. The inverter still ran — but it hunted constantly, the ripple worsened, and the MPPT efficiency dropped by more rough 5 %. Most crews skip this because capacitor do not fail short — they dry out, silently inflate the ripple, and eventually bulge or vent. By then the damage is done. The odd part is—a capacitor that fails from ripple normally looks fine on a multimeter: you volume an ESR meter or a capacitance bridge. Visual inspection alone catches only the terminal stage.

MPPT hunting wastes yield

Here is the less obvious failure: the MPPT algorithm relies on a stable DC bus voltage to calculate incremental power changes. Large ripple injects a 100 Hz or 120 Hz modulation into the voltage measurement. The controller sees a moving target — it tries to adjust the duty cycle, sees the voltage shift again, adjusts again. That creates a positive feedback loop. The inverter becomes a metronome of wasted energy: each oscillation represents a few watts that never reach the grid. Over a full day, that hunting can spend you 3–8 % of the yield. On a 100 kW site, that is 15–20 MWh per year down the drain. Not catastrophic, but not trivial either. The really frustrating part is that the inverter never logs this as a fault. No trip code, no alarm — just a quiet efficiency penalty. You have to measure the DC ripple with a scope, overlay the MPPT switch events, and watch the algorithm chase its own tail. How many site owners are paying for that lost yield and never knowing it?

What You require Before Touching the DC Bus

Oscilloscope with isolated or differential probe — not optional

faulty probe, dead scope. That is the blunt truth of DC bus work. A standard passive probe shares ground through the mains — hook its clip to the positive rail and you vaporize the trace, sometimes the whole front-end. I have seen an engineer do this on a 600 V bank: pop, smoke, a silence that lasted three weeks while procurement argued with finance. You volume an isolated probe (floating input, rated ≥1 kV) or a differential probe with at least 1:100 attenuation. Bandwidth? 20 MHz is adequate; ripple lives in the switch band (8–20 kHz usual), but higher harmonics can alias past a cheap probe. Do not trust the scope's built-in math — buy the real aid or rent it for the day. The catch is that most floor kits pack only standard 10× passives. That works for grid voltage at the PCC; it kills you on the DC link.

Known grid impedance at the PCC — the number nobody measures

You cannot tune the DC link in isolation. The inverter sees the grid through the LCL filter, but the ripple current returning to the DC bus depends on how stiff the AC side is. Low grid impedance — a strong utility — lets harmonic currents circulate hard; high impedance damps them. I have watched a perfectly stable 30 kW unit go into oscillation on a weak rural chain, and the only clue was the PCC impedance readed: 4.7 ohms at 250 Hz. Without that number you guess at the resonance. So measure it. Use a grid impedance analyzer or inject a known current pulse and record the voltage sag at the coupling point. Failing that, pull the utility's short-circuit throughput at your transformer — they publish it, usual. Multiply that by the cable length ahead of your meter. A rough number beats blind tuning every window.

Ambient temperature data — past 30 days minimum

capacitor hate heat. The DC link electrolytics lose capacitance as they warm — more rough 20 % drop from 25 °C to 55 °C on budget-grade banks. Tune your PI gains on a cool morning and the clamp ripple may spike 40 % when afternoon sun bakes the enclosure. You volume the worst ambient, not the average. Pull the last month of hourly data from a nearby weather station (free on NOAA or equivalent) and note the peak. Then derate your capacitance value by the thermal coefficient from the datasheet. The ripple current rating also shrinks. Most units skip this: they tune at 22 °C, sign off, and three weeks later the filter inductor screams. Get the thermal envelope opening.

Six-Stage Tuning Workflow: From Baseline to Final Filter

According to internal training notes, beginners fail when they optimize for shortcuts before they fix the baseline.

Stage 1: Capture baseline ripple at full power

Don't touch a lone capacitor until you have the waveform. Set your scope to AC coupling, probe directly across the DC bus terminals, and run the inverter at its rated output. I have seen units skip this because "the numbers look fine at 50% load." They never catch the snag. At full power, ripple amplitude often doubles or triples — what looked like 5 Vpp at half load becomes 18 Vpp and triggers nuisance trips. Capture at least ten cycles while the inverter is thermally stable. The baseline tells you what you are fixing, not what you assume is broken.

Stage 2: Identify dominant frequency component

Once the trace is frozen, zoom in on one ripple envelope. Most grid-tied inverter produce a dominant component at twice the serie frequency — 100 Hz or 120 Hz — because the solo-phase power stage draws current in pulses. But I've seen oddball harmonics appear when the MPPT algorithm oscillates or when a bypass diode is failing. Use the scope's FFT function or measure the period of five consecutive peaks. That frequency tells you where the energy lives. The catch is — if you throw capacitor at a 3 kHz switchion ripple thinking it's serie-frequency, you oversize the bank and still miss the real glitch.

Stage 3: Calculate required bus capacitance increment

You have the peak-to-peak ripple voltage (Vr) and the dominant frequency (f). The rough rule: needed capacitance raise ≈ (I_load) ÷ (2 × π × f × desired Vr). For a 5 kW inverter running at 120 Hz with a target of 5 Vpp instead of 15 Vpp, you volume more rough 2,200 µF extra. Not a precise number — but close enough to buy parts. Would you rather add 20% too much capacitance and waste $12, or add too little and repeat the whole measurement next week? Overshoot by 15%. Film capacitor handle ripple current better than electrolytics, but they spend more per µF. That trade-off bites when the customer budgets $50 for the fix.

Stage 4: Choose capacitor type and placement

Electrolytic cans are cheap and pack high capacitance into a compact can — but they dry out above 85°C internal. Film capacitor survive switchion ripple but demand twice the physical space. What usual breaks primary is the solder joint if you bolt the capacitor bank far from the bus bar. Place the new capacitor within 5 cm of the original bus terminals — longer leads add inductance and can turn your fix into a resonant tank. We fixed a 15 kW unit by adding four 680 µF film caps directly across the IGBT module terminals. Ripple dropped from 22 Vpp to 6 Vpp. The installer had considered adding the bank near the input connector instead — that would have added parasitic inductance and likely made oscillation worse. correct placement matters more than the capacitor value itself.

'Adding capacitance without understanding the ripple frequency is like putting bigger tires on a car with a bent axle — you just mask the shake.'

— floor engineer, after troubleshooting three identical inverter failures in a solar farm

Stage 5: Verify with a load sweep (the stage most skip)

Do not leave the site after one good read. Ramp the inverter from 10% to 100% power in 10% steps, pausing thirty seconds per point. The ripple waveform may shift shape — at some load points you might see a beat frequency where the chain ripple and switchion ripple interact. If the ripple stays under your target at every stage, seal the enclosure. If it spikes at a specific load, you have a resonance issue, not a capacitance shortage. That means you need a damping resistor in serie with the new capacitor, or a different capacitor ESR. faulty sequence.

Stage 6: record the before/after and walk away

Save the scope screenshot and tape a printout inside the inverter lid. I have revisited sites where the next technician could not tell whether the extra capacitor were original or an afterthought — they pulled them out and the glitch returned. Write the date, the baseline ripple value, the added capacitor part numbers, and the final ripple at full power. That record saves hours on the next call. One final check: run the inverter for ten minutes with the lid on, then measure case temperature of the new capacitor. If they exceed 10°C above ambient, you are pushing their ripple current rating — swap for a higher-ripple part before you leave.

Tools, Safety, and the Reality of On-Site Measurement

Differential probe vs. isolated scope — trade-offs

You have two real choices for looking at DC link ripple on a live bus, and neither is cheap. A 100x differential probe plugged into a standard scope works—assuming your scope has good frequent-mode rejection at the switch frequency. The cheaper 50 MHz probes? They roll off exactly where your ripple lives. I have seen clean sine waves on screen that turned out to be 40% measurement error once we swapped to a proper 100 MHz differential head. The isolated scope route—a battery-powered unit with fully floating channels—lets you skip the probe entirely and connect directly across the bus. That sounds fine until you realize the scope's input range tops out at 400 V and your DC link sits at 450 V. No margin. One transient and you own a repair bill larger than the tool itself.

The catch is that neither setup guarantees a clean readed. Ground loops still bite you through the probe's output side, and the isolated scope's internal switch noise sometimes injects artifacts that look exactly like 120 Hz ripple. Most crews skip this: take a baseline measurement with the inverter off, just the DC bus energized by the panels. Whatever you see at that point is pure measurement garbage. Subtract that from your live trace. Otherwise you tune a ghost.

Safety: do not float your non-isolated scope

off sequence and you lose a day—or worse. The standard four-channel benchtop scope has its BNC shells tied to earth ground. Clip the ground lead to the negative DC bus rail, and you have created a dead short through the scope's ground path when the positive rail hits a transient. The seam blows out inside the probe tip. I watched a tech do this on a 600 V site: the carbon track still marks the concrete floor where the probe landed. If you must use a non-isolated scope, use a differential probe with at least 1000 V frequent-mode rating—and verify the probe's input impedance matches the bus voltage before you connect. Do not lift the scope's ground pin with a cheater plug. That floating chassis now sits at half the DC bus voltage, and the next person touching the scope and a grounded enclosure completes the circuit.

'We floated the scope for “just one readed.” The RCD tripped, the inverter faulted, and the client banned us from the site for a week.'

— floor engineer debrief after a rooftop repair, paraphrased from an incident report I reviewed last year.

Why your DMM lies about ripple

That Fluke 87 reading 12 mV AC on the DC bus? Pure fiction. A typical handheld multimeter measures AC ripple by coupling through a capacitor and rectifying the result—but the bandwidth tops out around 1 kHz for most models. Your inverter's switch ripple lives at 8–20 kHz, and the 120 Hz serie-frequency ripple is only half the story. The DMM sees the low-frequency component, ignores the high-frequency hash, and reports a number that makes you think the bus is clean. It is not. The real peak-to-peak ripple might be four times what the meter shows, and that extra margin is what pushes your IGBT saturation voltage past the safe limit during a grid transient. We fixed this on one job by taping a note to the DMM: 'For continuity only.' The scope probe spend five times more, but it saved two replacement modules that season.

The tricky bit is that even a scope trace lies if you set the faulty bandwidth limit. Engage the 20 MHz filter and you kill the switchion edge information that causes most ripple-related failures. Leave it off and noise from the probe's own cable masks the waveform. A good starting point: bandwidth limit at 100 MHz, sample rate at 250 MS/s minimum, and take the trace at the inverter's DC input terminals—not at the combiner box ten meter away. The cable inductance between those two points changes the ripple shape entirely. That 200 mV peak you see at the combiner might be 600 mV at the bus capacitor. Tune for the faulty number, and the fix makes things worse—which is exactly where the next section picks up.

How Site Conditions Change the Tuning Recipe

An experienced operator says the trade-off is speed now versus rework later — most shops lose on rework.

Weak Grid vs. Stiff Grid

The tuning recipe that worked perfectly at the factory test bench falls apart the moment you connect to a weak grid — I have seen this firsthand. A stiff grid, with impedance below 0.1 pu, lets your DC link control loop behave almost linearly. The catch is that weak grids, where impedance climbs above 0.3 pu, inject low-frequency ripple back through the inverter's AC terminals. That ripple then aliases onto the DC bus, and your carefully tuned PI gains start chasing phantom voltage swings.

That lot fails fast.

The fix is brutal: lower your proportional gain by 40% and increase the DC link capacitance target by 15–20%. You lose some transient response speed — but you stop the 100 Hz and 120 Hz harmonics from saturating the error amplifier. off batch: try to tune for a weak grid using stiff-grid gains, and the DC link ripple amplitude doubles. Every window.

But here is the asymmetry most documentation hides: a weak grid also shifts the resonant frequency of your LCL filter downward. That means the notch filter you placed at, say, 2.1 kHz now resonates closer to 1.7 kHz. We fixed this on a 50 kW installation by sweeping the grid impedance with a signal injector before committing to any filter coefficients. Took two hours — saved a whole week of re-tuning. What usually breaks opening is the inner current loop, not the DC link voltage regulator.

Three-Phase vs. one-off-Phase Ripple Harmonics

lone-phase inverters generate DC link ripple at twice the serie frequency — no surprises there, 100 Hz or 120 Hz depending on where you are. The odd part is how three-phase systems fool people: balanced three-phase rectification cancels the fundamental ripple, so you get a clean 360 Hz component instead. Most units skip this distinction and apply the same notch filter to both scenarios. That hurts. A solo-phase framework's 120 Hz ripple is low enough that it couples into the MPPT algorithm and starts hunting for a maximum that does not exist. I have watched a 7.5 kW inverter lose 8% of its harvest because nobody adjusted the voltage reference filter bandwidth from 10 Hz down to 4 Hz. The pitfall? If you over-filter the DC link measurement in a three-phase system, you introduce phase lag that destabilizes the inverter during rapid irradiance changes — the seam blows out at the worst moment, usually under partial cloud cover.

One concrete fix: for three-phase, keep the DC link voltage filter corner at 60 Hz and rely on a dedicated 360 Hz trap instead. For single-phase, drop the corner to 8 Hz and check that your MPPT step time is at least three chain cycles. faulty filter = returns spike in harmonic distortion.

High Altitude and High Ambient Heat

Altitude above 2,000 meter reduces air density, which directly pulls down the cooling capacity of your DC link capacitor — typically by 15–20% per 1,000 meter. That changes the effective serie resistance (ESR) of the capacitor bank upward, which in turn alters the ripple voltage attenuation you calculated at sea level. A capacitor that showed 10 mV ripple at sea level will show 13–14 mV at 3,000 meter — enough to push the DC bus voltage measurement into the noise floor of your ADC. I have seen a site in Colorado where the inverter kept tripping on overvoltage, even though the bus was actually fine; the altitude had shifted the ESR so much that the ripple voltage exceeded the hardware comparator threshold. We fixed this by re-tuning the balance resistor divider and lowering the DC link overvoltage trip point by 5 V — a safety margin that should have been in the manual from day one. High ambient heat, above 50 °C, does the same thing but faster: electrolytic capacitors age at more rough double the rate per 10 °C rise, and the ripple current rating drops by 30%. The trade-off is brutal: you can use film capacitors instead (less temperature sensitivity but higher cost and size), or you can accept a shorter tuning cycle and recalibrate every six months during summer peak. No third option that does not involve a blown capacitor seam.

Most installers ignore this because the datasheet says "85 °C rated." What it does not say is that rating applies at sea level with forced airflow — take that away, and your ripple limits shrink.

'The altitude changed our DC link ripple from a measurement error into a trip hazard — we had to re-tune the entire bus balance in three days.'

— floor engineer, utility-scale commissioning staff, adapting from sea level to 2,800 m site

Debugging When the Fix Makes Things Worse

Ringing from Long DC Cables — Not Ripple

You finish tuning, shut the enclosure, and power back up. The oscilloscope shows a clean 100 Hz ripple—textbook. Then you walk forty feet to the array combiner, and the waveform looks like a startled seismograph. I have watched three different units waste an entire afternoon adjusting PI gains that were never the problem. What they saw was cable resonance, not DC link ripple. Long PV strings—especially those exceeding thirty meter—form an unintentional LC tank with the inverter's input capacitance. The switching edges excite it. The scope shows ringing at 12–18 kHz, but your eye gets fooled because that ringing amplitude wobbles at the series-frequency envelope. Easy to mistake for a tuning failure. The fix is never more gain: it's a ferrite bead clamped on each DC input, or—if the cable run is over fifty meters—a small RC snubber across the bus bars inside the inverter. We fixed one site by swapping out the installer's "premium" unshielded wire for standard PV cable with lower loop inductance. The ringing vanished. Check your scope timebase: if the oscillation period is under 100 microseconds, it's cable resonance, not ripple.

Misreading 100 Hz vs. 120 Hz

The grid frequency determines the ripple base. Fifty-hertz mains gives you 100 Hz ripple. Sixty-hertz gives you 120 Hz. Sounds obvious. Yet I have opened tuners' laptops and found filter notch settings locked to 120 Hz on a 50 Hz site. The result? The LCL filter does nothing—the dominant ripple sits right where the notch isn't. The inverter runs hotter, the THD creeps up, and the installer blames "bad caps" or "noisy grid." Wrong. The inverter doesn't care what frequency you think the grid runs at. It measures zero-crossings and uses that for PLL sync—but the DC link ripple component is purely a function of the AC line frequency's second harmonic. Most crews skip this: verify the grid frequency at the point of common coupling before touching any filter coefficient. We had a job in a port facility where the local diesel generator drifted between 49.7 and 50.3 Hz. Fixed-notch ripple mitigation failed repeatedly. The solution was adaptive—a software filter that tracked the fundamental and shifted its rejection band accordingly. That hurts. It means you cannot copy-paste tuning from one site to the next.

Capacitor Temperature Derating Busts Your Margin

You tuned the DC link loop in a 22°C workshop. The inverter sits inside a rooftop metal cabinet hitting 65°C by 3 PM. The capacitance value you relied on? Gone. Electrolytic caps lose ten to twenty percent of their rated capacitance at elevated temperature, and ESR doubles. The ripple voltage you measured as 3 Vpp at noon becomes 5.5 Vpp by late afternoon. The inverter didn't detune—the components changed. That is not a tuning mistake; it is a measurement mistake. The odd part is—installers rarely check manufacturer derating curves. A 470 µF cap rated at 105°C still drops to rough 380 µF at 85°C. If your notch filter or feedback loop assumed 470 µF, the Q factor shifts and you get overshoot on transient loads. One site kept tripping on DC bus overvoltage every afternoon at 2:47 PM. Not a ghost. The capacitors were cooking. We bolted a fan kit to the enclosure door and added a ten-milliohm resistor in series with the cap bank to damp the now-undersized loop. The trip stopped. The lesson: tune for the hot-case capacitance, not the datasheet number at 25°C.

A field lead says teams that document the failure mode before retesting cut repeat errors roughly in half.

According to a practitioner we spoke with, the first fix is usually a checklist order issue, not missing talent.

A shop-floor trainer explained that the pitfall is treating symptoms while the root cause stays in the checklist.

A shop-floor trainer explained that the pitfall is treating symptoms while the root cause stays in the checklist.

Hemming, fusing, bartacking, coverstitching, overlocking, and flatlocking introduce distinct failure signatures under rush orders.

Preproduction, top-of-production, inline, midline, final, and pre-shipment audits catch different classes of drift.

Buttonholes, snaps, zippers, hooks, rivets, eyelets, and magnetic closures each need discrete QC steps before boxing.

Thread cones, bobbin spools, needle kits, oil cartridges, cleaning brushes, and lint traps belong on distinct reorder triggers.

Cutters, graders, pressers, finishers, trimmers, handlers, inkers, and packers rarely share identical checklist verbs.

Spreading, layering, bundling, ticketing, shading, bundling, and nesting affect yield long before the operator touches pedal speed.

Merchandisers, technologists, sourcers, coordinators, auditors, and sample sewers interpret the same sketch with different priorities.

Pick, pack, ship, scan, palletize, cartonize, label, and manifest stages hide silent rework when SKUs multiply overnight.