Power Corner: How to Test Solid-State and Line-Frequency Transformers for AI Data Centers

Description

In this Power Corner interview, Patrick Fuchs, Global Application Segment Manager for Power Electronics Test at Rohde and Schwarz (R&S), discusses key line-frequency transformer (LFT) and solid-state transformer (SST) tests and the benefits of using a power analyzer to perform them. Fuchs introduces the purposes of the precision power analyzer to “measure the relevant voltage and current signals with the lowest time error…to eventually determine and output power readings directly for efficiency and loss analysis.” He notes that power analyzers need a “precisely synchronized base” for accurate measurements.  

When asked about the difference between the power analyzer and oscilloscope, Fuchs says these instruments complement each other. Power analyzers measure continuously and very accurately, but with a lower bandwidth and sampling rate than an oscilloscope. Meanwhile, oscilloscopes are fast and wide-bandwidth, capturing a signal window with lower accuracy. A power analyzer shows users how much power is lost and where (e.g., voltage, current, power, efficiency, losses, power factor, impedance, harmonics, K-factor, etc.), while an oscilloscope generally explains why with detailed waveforms (e.g., analyzing switching events, oscillation spikes, transients, control behavior, and gate-drive performance). 

LFTs

Fuchs zooms into precise power analyzer use cases within the data center, namely SSTs and LFTs. He begins with LFTs, where tests often follow the IEC 60076 framework, which applies to three-phase and single-phase power transformers. The two core LFT tests are the no-load test and the short-circuit (load) test (Figure 1). 

These transformer tests involve “hundreds up to thousands of amps,” while “a power analyzer can measure up to 32 A or 50 A directly,” so Fuchs says users “go with external transducers” that keep “PPM-class linearity and very good stability” and scale the signal down to avoid temperature drift. He says the same approach applies to SST efficiency testing.

These two tests let engineers determine the parameters of the simplified equivalent circuit (Figure 2). The no-load test measures iron (core) loss. The short-circuit test measures copper loss and impedance. Fuchs explains that at no load, the power factor is very low, so a tiny phase error causes a large power error, requiring “excellent synchronization” between voltage and current in the power analyzer.

Fuchs explains that three-phase transformer testing requires voltage, current, active power (R values), reactive power (X values), and apparent power, measured separately for each phase and then for the whole transformer. He also describes additional measurements, including frequency (AC cycles per second), power factor, and impedance (combination of X and R). 

Fuchs describes add-on measurements including: 

• K-factor: Calculated from harmonic measurements, allowing users to quantify the additional thermal stress that harmonics can cause.
• Phase symmetry: Taken from the amplitude and phase relationship of three voltage and current systems to confirm the three phases behave as expected.
• Efficiency + phase displacement together: With the primary and secondary three-phase systems both wired to the power analyzer, the analyzer’s channel grouping gives the efficiency and the phase shift between secondary and primary in a single measurement to identify incorrect wiring, unbalance, and asymmetry. 

SSTs

For SST testing, Fuchs explains that the testing scope changes because there is no longer a single input and output like LFTs. He continues, noting that testing begins at the board level with oscilloscopes to capture switching performance, then moves up to system-level testing with power analyzers. Fuchs states that because SSTs switch so fast, the test equipment must be fast and accurate, with high-bandwidth inputs and megasample- to gigasample-per-second (MSPS/GSPS) sampling rates that capture SSTs’ high-frequency harmonic content (Figure 3). 

Beyond input-to-output measurements, Fuchs notes that what happens in between matters as well: an analyzer should offer start-delta conversion to “directly transform the delta-wye voltage measurement into star, to then get the actual, correct phase readings and the collective sum readings.”

Efficiency, for example, is measured by stepping from idle to full load in 10% increments, then going past 100% (at ~ 20 ms intervals) to test dynamic load steps. Fuchs notes that it’s also important to measure power flow in both directions, since SSTs are bidirectional, to ensure the two directions behave symmetrically. 

Transcript

AS: Welcome, everyone, to Power Corner. I’m your host, Aalyia Shaukat, editor-in-chief of Power Electronics News. Today we’re speaking with Patrick Fuchs, global application segment manager of power electronics test at Rohde & Schwarz. Today we’re talking about testing data center power electronics—namely, where to use a power analyzer. We’ll mostly be talking about transformers, right?

PF: Yeah.

AS: Patrick, thank you for joining me, first of all. Thank you so much.

PF: Thanks for having me. Yeah, it’s great.

AS: I appreciate it. Let’s start by framing the conversation. Let’s talk about the surge in power demand from AI data centers and how it’s affecting the power electronics industry.

PF: Sure, yeah. Let’s dive directly into that a little bit. What AI is changing is not only the total amount of electricity a data center consumes, but also the power density and the dynamics of that load, right? 

Giving exact numbers is tricky, as it changes nearly weekly. We see it in the news every time—more and more data centers—so it also depends on geopolitical situations. But there’s the IEA, the International Energy Agency in Paris, and they estimate data centers consume roughly around 480-ish terawatt-hours (TWh) globally—sorry, last year. And they project around twice that already by 2030, so it’s crazy.

That already directly indicates that AI-focused data centers are growing considerably faster than the overall data center market and the energy we can basically deliver. At the same time, we are moving from racks that historically operated in the tens of kilowatts toward now hundreds of kilowatts, and the industry is already designing for even megawatt-class AI racks, right?

So this changes almost every part of the power chain, as I call it, especially in technologies. Later we will perhaps talk about solid-state transformers, wide-bandgap materials, leveling up power density and efficiency. From grid transformers, UPS, through power conversion and distribution, down to the server power supply, point-of-load converter, and eventually the server, basically.

AS: Let’s talk about test and measurement. From the Rohde & Schwarz perspective, what are the main test and measurement applications you’re seeing for AI data centers? I’m sure there are loads, but from your perspective.

PF: Yeah, loads of applications, because—I mean, Rohde & Schwarz stands for almost everything, right? We cover digital design testing, power electronics testing—my part—EMC, cybersecurity, facilities security, so when we talk about body scanners. But keeping it in the power electronics testing corridor, we focus on all major relevant power conversion systems.

So we look at the complete power chain—the data center energy infrastructure, commonly also called the grid-to-chip power chain. That starts on the grid and facility side. It includes conventional transformers, but now also increasingly solid-state transformers, UPS systems, and power distribution. 

On the load side, we have server power supplies and power shelves, battery backup units, isolating DC/DC converters at the server PCB level, and point-of-load converters close to the processor. 

From a test and measurement perspective, we go up to a testing level of mainly: power conversion, quality, and efficiency. That can be complemented, but rather from a digital design testing scope again, when we talk about power integrity and signal integrity. But I focus on power electronics testing.

So, for power electronics testing, the measurements concerned change along the chain, this infrastructure. But several questions always remain the same: where are my losses? That’s the most important question here. How efficient is each conversion stage? What happens under rapidly changing load because of the server prompts? What does the current drawn from the grid look like? And what happens to power quality and harmonics?

So concretely, we go into test scenarios like efficiency and power loss analysis. We cover power quality and harmonic measurement, load-dependent dynamics, and transient switching behavior analysis. So, from system level to board level, or, from the actual design process view, vice versa.

AS: Thank you so much. That was a very robust answer. I mean, it’s a very big question, so I do appreciate it. What benchtop instruments do you typically use for power efficiency measurements? 

I know we’ve got the multimeters, the power meter, the oscilloscope. And I know there are loads at Rohde & Schwarz—I’m quite familiar with the VNAs and the spectrum analyzers myself—but maybe you could talk a little more about some of the benchtop instruments.

PF: Yeah. If you ask me, I would definitely use a power analyzer and an oscilloscope, right? But—okay, you just said multimeter, power meter. The optimal product fit, or the core instrument for highly accurate efficiency measurement, is our precision power analyzer, because it measures the relevant voltage and current signals with the lowest time error, simultaneously, to eventually determine the power readings directly for efficiency and loss analysis.

However, it is normally not the only instrument on the bench, or rather, let’s say, in the rack, right? As you already said. An oscilloscope is extremely valuable and complements the high-resolution signal level. So when I need to understand switching events and resulting oscillation spikes caused by parasitics in the commutation path, transients, control behavior, gate drive performance—it’s endless, right? Whatever counts into signal analysis. 

And a multimeter or power meter is used here and there in use, but not if we are particularly talking about precise efficiency measurement. I know power meters or energy meters are used for overall monitoring to get the PUE value, the power usage effectiveness value. 

But I would say the power analyzer is the right tool. It tells me accurately how much power is being used. It verifies how much is coming out and where the losses are, and the oscilloscope helps explain why those losses or dynamic behaviors occur, and where I—now looking from an R&D engineer’s perspective—have open potential to optimize the efficiency.

AS: You’ve discussed a little bit of the differences between the oscilloscope and the power analyzer. What are the main components you’re looking at in its front end? Can you talk about that a little bit—what’s inside? Obviously, to the degree that you can talk about it.

PF: You mean inside a power analyzer? 

AS: Right. 

PF: Also, the question of what the difference is to an oscilloscope—that’s such a famous question. 

I always say: a power analyzer is destined to continuously run measurements on a cyclic basis with very high accuracy—low bandwidth, lower sampling rate compared with the oscilloscope. Oscilloscopes definitely have a high sampling rate and high bandwidth, but work on an acquisition window, with low accuracy.

Okay, front end, or what’s inside a power analyzer. In the end, we need to measure signals: voltage, current, and ultimately power. So we have the current and power measurement inputs, or, let’s say, a combined power channel. I say combined because we always need to make sure the current and voltage are perfectly in time, so there is no considerable time deviation between them—otherwise that would directly affect the power accuracy. 

This is followed by analog-to-digital converters to get them into digital values, and then the overall processing. Of course, this is the secret that makes the power analyzers different from each OEM. Finally, what is done is integration on a very precisely synchronized base, so that in the end we can get the derived readings.

But if you ask what needs to be optimized in the front end—when we look at what is happening directly at the interface, at the power channel, at the measuring inputs—there are three things I would say to pay particular attention to: the voltage path, the current path, and, as already said, the timing relationship between them. A lot of things.

AS: Thank you so much for illuminating the front end a little bit. Let’s talk about a particular test—we’ll talk about transformers first. Data center operators are increasingly anticipating a shift away from conventional LFTs, or line frequency transformers, and transformer rectifier units, or TRUs, towards solid-state transformers. I was wondering if you could talk about that a little more—the differences between these technologies and the thinking behind this shift.

PF: Yeah. I’m a big fan of power electronics, so I very much welcome that shift. However, I would be slightly careful with the premise here. Solid-state transformers—let’s call them SSTs—are definitely emerging. 

Nearly multiple times a day I can see it in the news or on LinkedIn: so many SSTs here and there, and it’s just starting. So it’s emerging. But the first actual scheduled rollouts, as far as I know, are deployments, especially for data center hyperscalers.

AS: Later this year?

PF: Yeah, probably around 6 to 12 months from now, if I’m not totally wrong. But what I would like to say is that they are not yet simply replacing conventional transformers throughout our data centers, but adding the proper alternative. I reckon it’s a question of empirical data on reliability and durability that we have for dozens of years for LFTs, but not for SSTs in such a kind of application, in direct comparison. So it definitely needs to prove itself over time. 

And you also asked about the technology difference here, right? Shall I explain a little bit?

AS: Yeah, yeah.

PF: Great—because this is really interesting. The typical, traditional, conventional transformer, the line frequency transformer, is fundamentally a magnetic device. So there’s not much going on. It changes the AC voltage level at a frequency, for instance, of 50 or 60 Hz, the mains frequency. 

It luckily provides galvanic isolation and can do that pretty efficiently and reliably. I mean, if we talk about the high-efficiency transformers that are usually used for data centers, that comes with high-level core materials: high-grade silicon steel, amorphous metals. 

Also, since you said with an additional rectifier unit—it’s a secondary winding, typically at the output side, and we can also rectify into DC.

But a solid-state transformer, in contrast, is fundamentally different, because here we basically speak about power electronics—a power converter, so to speak. It combines power semiconductors, high-frequency conversion, and a smaller high-frequency isolation transformer. So I think that’s where this word “transformer” still comes from. 

That allows us not only to transform voltages, but potentially to perform not only AC-AC, but also AC-DC conversion directly, with different voltage levels. We can regulate the output actively; we can have power factor correction (PFC) at the input side, or three-phase with an active front end. We can manage bidirectional power flow and interface directly to a data center 800-V HVDC bus. And I mean, that’s particularly interesting for future AI data centers.

I don’t think it will remove the traditional transformer, but what it removes, or can remove, are some in-between parts of the power chain—conversion stages from medium-voltage AC to the DC distribution system. And this is great, because in doing so, we are not only skipping potential loss stages—as a result, we can increase the overall efficiency. And this is top-notch.

AS: Let’s zoom into the line frequency transformers in more detail—testing them. As you were saying, they’re still very relevant in the short and medium term. They’re still very critical. What parameters are you generally looking for when you test LFTs, and how do you measure them? Can you talk about some of those tests? I know, it’s a big question again.

PF: Sure, yeah. So typically we see there is this type of acceptance test, which specifically follows IEC 60076 for transformer testing. For a traditional transformer, we have two fundamental measurements, which are the no-load test and the short-circuit, or load, test. They are split because what we eventually want to do is determine the simplified equivalent circuit parameters.

In the no-load test, we energize, let’s say, one winding of the transformer side, while the other output side is usually open, or not loaded. The current is relatively small, but at the applied rated voltage. And what we’re primarily measuring is the iron, or core, loss. 

In contrast to that, for the short-circuit test, one side is shorted—as it’s called, short-circuit, so it’s shorted—and we apply enough voltage to just produce the rated current. So now we are primarily interested in winding, or copper, losses and the resulting measured impedance as well.

Now, if you think of a three-phase transformer, we want voltage, current, all the powers—active, reactive, apparent power. We want to have all these values phase by phase and in total, together with frequency, power factor, and the impedance. Efficiency, of course—not to neglect that. 

And then, depending on the application itself, we can also add, as part of these types of acceptance tests, harmonics for the K-factor determination, phase displacement, vector group verification, phase symmetry. Oh, yeah. So that sums up basically all the measurements with the power analyzer, I would say.

AS: I was wondering, for the power factor in a no-load test, how much can it vary based on the core?

PF: Yeah, that’s a good question. Not only because of the core, but—so basically, yes. Indeed, it can vary significantly with the transformer design, as you said: the core lamination, the material—as I said before, silicon-based steel or amorphous material—but also the flux density and the operating voltage. 

And the important point is that the no-load power factor during no-load tests can become very low, because the magnetizing current is largely reactive, while the real iron loss is comparatively small. Basically, because we do not have any considerable resistive load applied, right?

And there we already get to suitable test equipment. That’s exactly the tricky, difficult measurement condition. We may have a substantial voltage and current, easy to measure, but only a relatively small amount of real, active power at high phase shift. So this is a critical measurement, definitely.

AS: So why is a precision power analyzer important for that test?

PF: I think I have mentioned that before, for the front-end question—I emphasized that this time correlation, or the smallest time deviation between voltage and current, is very important, which comes into play here. Because at low power factor, phase accuracy becomes critical. Imagine a worst case: voltage and current are almost 90 degrees apart, theoretically. 

A tiny phase error in the measurement system can then create a relatively large error in the calculated real, active power, especially if the measuring tolerance exceeds the remaining phase margin. I mean, this is horrible.

So especially for transformer no-load testing, where we have this kind of condition, it’s not enough to have good voltage accuracy and good current accuracy independently. You also need excellent synchronization between them. Phase accuracy between those two measurement inputs, or paths, is mandatory. We talk about nanoseconds of time adjustment that we at least provide for our inputs. And this is exactly the use case for which we at Rohde & Schwarz highlight our LMG671 power analyzer, with the S-channel. So that is the right tool for it.

AS: I noticed in the block diagrams there were the Danisense current transducers. I know these are very accurate, low-PPM, high-voltage current transducers. How are they necessary for these tests? Could you talk about that a little more?

PF: Yeah, sure, I will. The Danisense sensors are great, and it’s required for them to be great. With “high voltage,” I assume you mean the voltage withstand capability—primarily measuring the current. 

The reason for using a high-precision external current transducer is that transformer testing can involve very high currents. We are speaking about hundreds up to thousands of amps, but we still want PPM-class linearity and very good stability. 

Usually, a power analyzer can measure up to 32 A or 50 A directly, but there’s no chance [we can measure the current directly] for these power transformers. So we go with external transducers. As a result, we have a measurement chain consisting of both devices, which we still want to maintain with the highest accuracy.

 That is ensured by using a much more accurate external sensor compared to the power analyzer channel itself, which is given with the ones from Danisense.

So not only do we need to measure high currents, but we need to measure high currents externally, while keeping high precision, with these flux-gate transducers. They provide galvanic isolation without necessarily inserting a large shunt into the high-current path. 

And what is really good with that is that they have a turn ratio, so the signal itself will then be scaled down to the secondary side in the hundreds-of-milliamps range. That is good for the power analyzer, because then we can use lower measuring ranges for the current that do not introduce any considerable temperature drift.

AS: I see. Could you talk a little more about, I guess, the phase symmetry test? How do you measure that? You might have already.

PF: For phase symmetry, for transformers, right? Let’s go ahead with three-phase transformers, because there it makes sense. There, we do not only care about the total power or efficiency losses. We also want to know whether the three phases behave as expected: if we have a good, as by the standard mentioned before, specified, beautiful three-phase system, with a 120-degree phase shift between the phases. 

The phase symmetry analysis compares the amplitudes and phase relationships of these three voltage and current systems, particularly the fundamental in scope here. The fundamentals come from harmonic computation as a derived measurement, since we already have the PA wired to the transformer. We have the three-phase groups, which means we do the FFT, and as a result we get the amplitudes and phases—in our scope here, the fundamentals.

They can also be expressed not only in numbers, so numerically, but also as a vector diagram. And then, together with such a thing as phase displacement measurement—which basically determines the phase shift between the secondary and primary system that enables us to verify the transformer vector group and identify incorrect wiring, any unbalance, or asymmetry.

AS: Maybe we can shift to the K-factor, because you did mention the K-factor measurement earlier. Could you talk about that a little more as well?

PF: It’s not always typically very important, but now it is for data centers, where everyone—and when I say everyone, I mean the government—is saying, “Hey, they need to run very efficiently.” And we know the loads in a data center—everything is switching power electronics, so there are a lot of harmonics, and this is where we now make the shift to the K-factor getting more and more important.

The K-factor is essentially a way of, let’s say, quantifying the additional thermal stress that current harmonics, as mentioned, can cause in the transformer. So it basically describes—or, let’s rather say, to be more correct, rates—the withstand or handling capability of harmonic load currents without exceeding the transformer’s specified thermal limits. 

This is particularly relevant, as I said, in data centers, but not only there: also in commercial buildings, hospitals, telecommunication stations, and industrial production sites. All those facilities that usually come with many nonlinear loads. 

So, all the switch-mode power supplies of the supplied equipment, and any other power electronics load in the chain, or in general, that can draw distorted current. Like data centers again: the server PSUs, we have these UPS systems, even for the cooling, we have the frequency converters. We have also dynamic load behaviors, and so on.

What’s important to mention about the K-factor here is that it shouldn’t be compared with THD [total harmonic distortion]. Sometimes it will be mistaken. But for the K-factor rating, we measure individual current harmonics, and higher-order harmonics receive progressively greater weighting because they can contribute disproportionately to heating. 

And the rating factor. Maybe to give an idea about that: we have, for instance, a K-1, so that’s what it starts with. That means we have more or less a linear load, so no significant harmonics. And that can even range up to K-50, which describes extreme, harsh harmonic environments. 

And for data centers, typically, we should optimally be using transformers that are rated at around K-20. Let’s say we neglect the cost; we can, of course, go even higher. But the higher the K-factor, the more the transformer costs. But let’s say we have a lot of nonlinear load, a lot of power electronics, so we should go for something that can handle up to 100% nonlinear loads—or starting from 60 to 100%. So usually there we have these K-20. Sometimes, if cost matters, K-13-rated transformers. And yeah, again, this is where power analyzers come into play.

AS: Please carry on. Sorry.

PF: Since I’m thinking about, okay, a lot of story about this K-factor, but how to measure it, and then why with the power analyzer. 

The K-factor measurement itself is typically not done during the factory tests mentioned before, but during actual operation, or even in simulation, if we have a dedicated harmonic load-test setup using a programmable, regenerative power source acting as a load. Because we need to generate these specific harmonics, right? 

The power analyzer, as the core measurement instrument, then performs aliasing-free harmonic analysis on the power measurement, usually at the transformer’s load side. And then it calculates the K-factor as a derived measurement. But this is not a standard parameter a typical power analyzer has, so we need to find a way to weight these harmonics and calculate the K-factor as per the formula specified in the applicable standard. The good thing is that our power analyzer, the LMG671, comes standard with an integrated script, or formula editor, let’s call it that way, which is kind of the optimal tool for this calculation.

And particularly for the harmonics computation precision—to get it even more precise—we need to apply an anti-aliasing filter (AAF) for the harmonic configuration. It’s a digital filter, sometimes, and we know whatever we have programmed there, so we can even compensate for any considerable effect that we have by applying the AAF.

AS: And could these functions be used in other aspects of the transformer testing?

PF: Yeah. So whenever there is a parameter that is not standard or available as a derived measurement, we need to find a way to calculate that with the script editor. This is the way to go, and this is why we have it. It’s not only “A is equal to B multiplied by C.” No, but you can also write scripts with it.

AS: Okay, gotcha. Well, I was going to shift to solid-state transformers, but I want to make sure we have the transformers tied up in a nice little bow. Is there anything that I missed that you wanted to mention? Any key aspects of testing that you really wanted to drive home?

PF: We found a way to get, with a single measurement, the efficiency as well, by wiring the primary and secondary three-phase systems to the power analyzer and directly measuring the phase displacement. This is not easy, and there should be a way to facilitate that. And with our power analyzer, we found a way, with our pretty comprehensive grouping concepts for these three-phase systems, so that we can directly determine what the phase displacement is. Very important.

AS: What design topologies are you commonly seeing for data center SSTs, and how does that impact the instrumentation needed to test them?

PF: First of all, nice—SSTs. Good, making this shift. Still keeping it with transformers, but, as I said before, I’m into power electronics, and this introduces a lot of power electronics. 

In my view, I don’t think that there is one universal SST topology, right? But conceptually, we’re usually dealing with several power conversion stages out there: single-stage, two-stage, or three-stage concepts. And typically we address data centers with a three-stage solid-state transformer concept. 

That means, on the grid side—the grid, like the mains—which is then connected to make the transformation from the AC mains supply to all the other things inside the data center. So on the grid side, there’s typically an active AC/DC rectification stage, three-phase, then an active front end, for instance.

Then we have a high-frequency isolated DC/DC stage, often based on a dual active bridge (DAB) or a related topology, which lets us shrink the transformer. Finally, the third stage is the required DC-to-AC conversion, or DC/DC conversion—all of it, nowadays, with the demand to operate bidirectionally. This is good.

And for the emerging data center energy and its internal voltage distribution architectures, the interesting case is medium-voltage AC converted to—now, everyone talks about that—the 800-V high-voltage DC distribution bus. But when needed, the SST can also output low-voltage AC and even lower-voltage DC, like 50 V. And yeah, if you now look from the measurement perspective—

AS: Yes, testing SSTs, let’s talk about that.

PF: I mean, with the transformer and—we already have a lot of things going on in terms of measurement. But now SSTs, that changes a lot, because now I don’t have only system-level testing or validation with one input and one output. I have overall testing, starting at the board level—like converter modules and PCBs—with an oscilloscope for efficiency and switching-performance analysis, and system-level testing with a power analyzer at several power-conversion stages, multiple DC links, and switching modes. And, as I think I already mentioned, we need to support potentially bidirectional power flow measurements.

And I think, since the main focus for data centers is size and heat emission. Everything needs to be small, and in a perfect world there should not be any heat emission, which means high efficiency. This application forces the SST to achieve a higher system efficiency of at least 96% to also keep up with the traditional high-efficiency line-frequency transformers. 

And again, in order to enable that, we increase switching frequencies for the single conversion stages. But doing so introduces a lot of harmonic content. That means our test instrumentation must provide high-bandwidth inputs and, for power or signal analysis, mega-sample to giga-sample sampling rates. Always ensuring the highest measurement accuracy, especially for efficiency verification. And low-noise measurement for scoping on switching-semiconductor-related signal levels, and so on.

AS: Let’s focus on maybe one measurement—say, efficiency. You said that SSTs need to have a very high efficiency. Could you talk about that test a little more? How would you hook it up and start measuring? I know it’s a large system, but just throwing one test out there.

PF: Yeah, well, let’s shrink it to efficiency, neglecting all the thousands of measurements we can do in between, right? Efficiency, as always, we measure from the output to the input to get the difference, which gives us the efficiency and the losses. 

Also, here we use very accurate external current transducers to extend our measuring range for the current. For power and voltage ranges, we are quite fine; we can measure up to 1,000 V. Maybe not on the medium-voltage supply side, so there we use external voltage sensors, but again, we need the same accuracy. Also, here we have high-voltage dividers, for instance.

What is then critical is that we need to run the tests not only for a single operating point. So we start from idle, going to full load in 10 % steps. But we don’t stop at 100%. We still have the safety margin, so we go up to 120%, sometimes even 150%, for very dynamic load steps—especially during validation, which can happen. But then we only talk about something like 20 milliseconds, right? 

But anyhow, we need to test that. So also the power ratings, the power ranges, go crazily up.

AS: What kind of ranges are we looking at?

PF: Yeah, and I mean, since we not only have the AC in, AC out as with the traditional transformer—so not considerable harmonics—now here, since the output sides are switching, we have a lot of switching going on in between. And with wide-bandgap materials, or high switching frequencies again, that means we need equipment that can measure this high bandwidth at the output side. And also at the input side, for power factor correction and bidirectional power flow. 

So we need the right equipment to measure not only in one energy-flow direction, but also in the other. In the end, I want to validate whether my system is working symmetrically: if the positive energy direction matches, does the opposite direction match as well? Or do I have to change anything here for the semiconductor switch control?

AS: Well, I wanted to close the chapter on solid-state transformers and testing there, but not quite yet. Is there anything else you wanted to talk about a little more?

PF: Perhaps what else is very important and required is what happens in between. As I said before, we have a lot of different stages. It’s not only output and input. 

And also, now that I think about it, we have three-phase systems. So when we wire a power analyzer for the efficiency test, what we need to do—or what we sometimes do, or mostly do—is measure the voltages between the phases, but the current in the phase. So the equipment, the power analyzer, should have a feature that is usually called star-delta conversion, to directly transform the delta-wye voltage measurement into star, to then get the actual, correct phase readings and the collective sum readings. This is very important.

AS: Very good. Well, thank you so much for your time, Patrick. I really appreciate it.

The post Power Corner: How to Test Solid-State and Line-Frequency Transformers for AI Data Centers  appeared first on Power Electronics News.