Practical Transformer Winding (2010) | Hacker News Reader

Practical Transformer Winding (2010)

ludens.cl

125 pointsbearbin36 comments

Yesss! This is what I was hoping to find last time I clicked on an HN headline "transformers from scratch", which turned out to be about machine learning: https://news.ycombinator.com/item?id=29280909

This is great. It gets into the flux density and iron loss calculations that affect the physical size of a transformer core, choosing the wire, and then exquisite detail of actually winding and finishing the thing.

I need to spend a bunch more time reading this site.

I find this trend of naming software ideas after common words really quite annoying (transformers, lenses, optics etc.)...

From what I've read, that's why physicists started giving things whimsical names like Quark, Charm, Strangeness, etc. The problem with using common words is that they always suggest an analogy, the analogy is never perfect, and often causes more confusion than understanding.

(Also, Murray Gell-Mann was a whimsical character to begin with).

Or naming companies after physical units (e.g. Tesla).

Or naming physical units after people (e.g. Tesla).

FWIW, Tesla the company was named after Tesla the person, not the measurement.

Yeah, the comment you replied is almost troll-like

Why? Many professionals use the unit Tesla on a daily basis. It's like someone claiming the word Pixel for their company or product which I'm sure already happened.

https://store.google.com/category/phones

Or using the word "bandwidth" to refer to "capacity" or "data throughput" rather than "the width of the band that the signal occupies". Arrrrgh. Yes the two things are the same for certain naïve modulations, but the whole point of using a better modulation is that you get more throughput for the same bandwidth. Understanding this requires first unlearning the abuse of the term, and that's hard work for a lot of engineers who really need to understand this.

Shannon's channel theorem states that the information capacity of a channel is directly related to the bandwidth and the signal-to-noise ratio, even if you use the best possible modulation scheme:

https://en.m.wikipedia.org/wiki/Shannon%E2%80%93Hartley_theo...

He’s talking about throughput, not the information capacity of the channel.

Can you explain the difference? They seem to me like virtually the same concept, but you seem to see an important distinction.

Channel capacity is an ideal theoretical concept establishing a maximum data transmission rate over a channel with strong assumptions about the noise in that channel. Throughput is what you actually get in the real world, and it depends on many things, including a modulation scheme, retransmission protocols, specific noise models, etc.

Fair enough. But in review, they were indeed talking about the capacity:

> Or using the word "bandwidth" to refer to "capacity"[...]

I agree, but at the same time "Sequence-to-sequence model utilizing self-attention" doesn't roll off the tongue as well and would otherwise make the thing more difficult to talk about.

Touché, I opened this hoping to find a new method to train Transformers and got the opposite

The Transformer Winding community knows that you won't let us down.

A transformer and coil design guide from the same author: https://ludens.cl/Electron/Magnet.html

Ludens is absolutely legendary. This is part of his micro hydroelectric dam[1], which he uses to power his house, which he of course built himself.

Far different, but if you enjoyed this you may also enjoy another classic (which does not have explanation), Claude Paillard making a vacuum tube from glass and metal: https://www.youtube.com/watch?v=EzyXMEpq4qw

[1]: https://ludens.cl/paradise/turbine/turbine.html

Incredible. I had a dream when I was a child of a house powered by small-scale hydroelectric, but learning the reality of water rights made me abandon it. It is truly inspiring to see someone live this dream.

> Transformer steel is not all born alike. Manufacturers will provide data sheets about their products (often on their web sites), where you can see what they offer. There are usually many grades, with vastly different loss characteristics. At a given flux density and frequency, a good material might have ten times less loss than a cheap material! So it pays to look, investigate, and decide intelligently what to buy. Thinner sheets normally have lower loss, and the rest of the secret lies in the exact alloy.

Okay this is what I come to HN for, thanks everybody!

I ran a high school science lab in Bolivia in 1968. It was cheaper to get a local guy to hand-wind forty centre-tap transformers, than to order them from Germany or the U.S.

There are interesting people on thebackshed.net who have been building whole house inverters, initially repurposing some Chinese inverter chip but more recently an arduino for the brains. Unfortunately a lot of the info is spread across many different build threads but a big part of it is about rewinding toroidal transformers.

That's so retro. Useful if repairing boat-anchor equipment from the 1950s.

Here's a more modern video on winding toroids by hand.[1]

[1] https://youtu.be/sDIWNHOoNh8

Nearly all usecases for a power transformer are now better served by a switched mode power supply.

They tend to be much smaller, lighter, more efficient, cooler and cheaper. They also have features like adjustable output voltage, current limits, and overheat protection. Some can input and output AC or DC, at a configurable frequency and sometimes waveform. Some offer the same galvanic isolation that a transformer offers too.

The core of a switched mode supply usually is a transformer, or at least an inductor, but the key difference is that it operates at far higher frequencies than classic uses of transformers, which allows them to be far far smaller, and therefore cheaper for the same power output.

Are you saying large Transformers 1 mva, 10 mva, 100 mva, 500 mva are obsolete and should be replaced by switched mode power supplies? What mva is the threshold?

For new installation, yes, for any size a switched mode supply will usually be better in all dimensions. For the largest sizes, you can't buy them off the shelf, and there design costs may dominate. But after the thing is designed, in component costs, switched mode will win. Everything scales linearly with kVA, so there is no economics crossover point for the fundamental materials.

Eventually it will be worth switching out old transformers - they contain a massive amount of valuable copper and quite valuable steel, and their lower efficiency means every year they remain in service they are wasting $$$'s of electricity.

Transformers in cities can often be replaced with much smaller switched mode units underground, allowing the building housing the old transformer to be rebuilt as luxury flats to make the project much more profitable too!

Large power transformers have efficiencies in the 98-99.75% range.

I don't doubt switch mode could be smaller and cheaper up to some size, but I am struggling to see transformers larger than about 5 mva being replaced with power electronics.

Solar farms etc have inverters in modules I believe 500 kva each - and of course the power electronics are necessary there, there is no substitute.

I have a 20 MVA transformer that is nearly at end of life and would be open to cheaper replacements.

> but I am struggling to see transformers larger than about 5 mva being replaced with power electronics.

We already see these at the substations at each end of a HVDC link, China is operating well over a dozen of them with capacities of up to 12 GW (https://en.wikipedia.org/wiki/Ultra-high-voltage_electricity...).

For 5-20 MVA you're talking about 67 kV substations or similar, where a transformer costs in the low 6 figures. HVDC converter stations in the same range would cost somewhere around 8 figures, although that's mostly a guess- you typically need maintenance and supervision in a way that you don't with transformers. 1%+ downtime is pretty common, which absolutely sucks if you aren't a full grid and can't pull extra generation.

Yeah, but that price tag is mostly due to the fact that HVDC isn't a widespread technology yet. Once factors of scale come into play, the situation will look different and the prices come down.

Additionally, the price of copper is already at an/near the all-time high and it's not going to get cheaper, and the land on which huge transformers sit is shooting up in value... so in the end, market forces may push towards solid-state technology anyway.

Those properties are all positive, but there is one glaring negative: rf radiation. I know some ham radio station builders that will not allow any switched mode power supply on their property, including wall warts and smartphone chargers.

That's a problem, especially with low-end power supplies.

I designed a switching power supply a few years ago for a specialist application - driving 1930s Teletype machines which need 120VDC 60mA.[1] A switching power supply is a spike generator. Here's the schematic.[2]

This is reasonably RF-quiet. The transformer of the switching power supply is a toroid in a metal can, so you don't get too much RF from the transformer itself. But that's not where it usually comes from. The important thing is to keep spikes out of both the input and output wiring. That's dealt with by using LTSpice to simulate the circuit, and adding small capacitors and inductors until the spikes have disappeared in both voltage and current. At the output end, note the snubber C7, R1, D10, and D11, to soak up any spikes from the inductive load, and L2, to soak up output side current spikes. There's L1 and C12, to soak up kickback from the switched input side of the power supply. Plus C10 at the power input, which is from a USB port. They're all tiny surface mount components. The inductors are ferrite beads in surface mount form.

So there are eight extra components, just to prevent unwanted RF generation. The LTSpice simulation shows that they're all needed. The simulation was used to choose the values.

This is why good switching power supplies have more parts than bad ones. You see that in teardown videos.

[1] https://github.com/John-Nagle/ttyloopdriver

[2] https://github.com/John-Nagle/ttyloopdriver/blob/master/boar...

> This is reasonably RF-quiet

I wonder if you have actual measurements?

>to prevent unwanted RF generation

Perhaps actually "to reduce unwanted RF generation"

Winding your own transformer was one of those things which amazed me as a teenager: you do a simple math by calculating winding ratio and the voltage is changed at the secondary by magic...

Christ, I still have a welder that I made myself. It uses a round ferromagnetic core I found in a scrapyard.

I forgot how many windings it has, I think around 300 on the input side and 40 on the output.

Wiring it was a major pain in the ass, as the core is round, a closed circle.

And it draws too much current when idle, which means I have not used enough wire on the input side or the core is too small, fuck knows, I was 14 when I made it.

It does weld well, especially on DC through 4 huge diodes I also found in some scrap.

Sorry for my English, I'm drunk and tired.

Reply on news.ycombinator.com