Surface Roughness Measurement

[Anonymous]

So, here’s the scenario. We have some complex watercooled heatsinks machined from aluminium. On part of the heatsink we mount an IGBT module containing three half bridges. We need to get rid of about 2kW of heat from the IGBTs in an area a bit smaller than a paperback book.

 

I machined eight of the heatsinks in the garage last year, and we know they work, because we’ve had them running at full power without overheating. We have just had another two heatsinks machined to my drawings by a professional machine shop. They’ve done a pretty good job, apart from screwing up on some threaded holes (pun intended) where they’ve tapped M5 instead of M4. Didn’t read the drawing, tut, tut. However, on the IGBT area they have achieved a rather rougher surface than I did.

 

The manufacturers of the IGBT module call for a surface finish better than 10µm Rz. Not surprising really as Rz is an ISO standard measure of surface roughness, and the manufacturer is German. However I can only measure Ra at home. On the heatsink I knocked up in the garage I get Ra values between 0.5 and 0.9µm, whereas the professionally machined heatsink measures between 0.6 and 1.5µm. It definitely feels rougher too, in comparison to a Rubert plate and using one’s fingernail.

 

The question is, is there a generally accepted conversion between Rz and Ra values? I know that in theory there cannot be an exact conversion, as Rz and Ra measure two different aspects of surface roughness. But is there a pragmatic conversion factor? I’ve seen values for conversion between the two varying from 2 to 20 on the internet!

 

I need to get a good feel for how the surface roughness they have achieved compares with what we need, before asking them how come I got a better finish beavering away in my garage!

 

Regards,

 

Andrew

[Anonymous]

[Steve GarnettParticipant @stevegarnett62550]

I don’t know how much it will help, but there’s quite a good cross-reference chart here. I think it’s good, because it gives you a reasonable idea of what you might be able to achieve using what grades of surface treatment, and that’s generally as good as I need it. It would appear that the differences in conversion factors, even though they are essentially inaccurate anyway, probably stems from the several different definitions of Ra out there!

Edited By Steve Garnett on 02/11/2010 23:17:04

[Martin WParticipant @martinw]

Andrew

 

I found this article but I am not sure how helpful it will be to you Rough. It really highlights the problem of comparing two systems that measure roughness in different ways. Still it may prove interesting.

 

Cheers

 

Martin

[Sam StonesParticipant @samstones42903]

Hi Andrew,

I think Martin W has provided your answer. Although I haven’t checked it out, a website which might provide additional information would be that of Taylor – Hobson.

It was on their Talysurf machines that I learned a little more about RMS and CLA (in mu inches for you Imperial lovers). A CLA value of 1 mu” or less is quite a high polish.

Or am I bringing stuff up which is fifty years out of date?

What I would like to know relates to the idea that (within reason) the rougher the surface the better it will be in radiating (and adsorbing) heat, or in your case, is it the surface contact between your heat-sink and the IGBT’s that is important?

Incidentally, until I looked it up on’thinternet I had no idea that an IGBT was an insulated gate bipolar transistor.

However, as we all know, the property known as emissivity is measured on a scale of 0 – 1 (naught to one). It places a black body in space at 1, and conversely, places highly polished surfaces at an extremely low value of emissivity. An example of this property can be experienced by noting the lack of radiant heat from a highly polished, chrome plated kettle containing boiling water. But don’t burn yourself when you try it.

Where does this fit wrt to your heat-sinks Andrew?

Isn’t there a silicone grease intended for improving the thermal transmission?

Regards,

Sam

[Ian S CParticipant @iansc]

Andrew you may have answered a question I have. I was in a engineering supply shop on Friday, and among some second hand tools was a little leather pouch with a plate inside made up of a number of squares of varying roughness, did’nt spend much time looking at it, is this something to do with this subject? Ian S C

[KWILParticipant @kwil]

Andrew, What is the surface on the IGBT like?

[Martin WParticipant @martinw]

Andrew

 

 

Some more information on Ra, Ry, Rz can be found at Ra Ry Rz but I expect that you are aware of the way the various measurements of surface roughness are estimated.

 

If this isn’t bad enough you might like to read this article Rz Flavours re various flavours of Rz. See the text in the blue column

 

Cheers

 

Martin

Edited By Martin W on 03/11/2010 10:12:49

[Chris TriceParticipant @christrice43267]

J&L Industrial show a Surface Roughness Standards kit with samples in their 2009 catalogue (and presumably online) – item number SPU-30695J if that helps.

[Anonymous]

Thanks for all the information; plenty for me to read. Don’t know why I didn’t think of looking on the Taylor Hobson website; after all they made my surface roughness gauge, doooh!

 

Right, now to try and answer a few points.

 

Ian: I’ve posted a picture of my Rubert set; did the one you see look like this? Rubert, and others, also made sets for specific operations. I have one with six squares on it aimed at shapers.

 

Sam: Sorry, I tend to forget that not everyone knows the esoteric abbreviations. In simple terms IGBTs look like a bipolar device at the output, but the input looks like a MOSFET, so it’s easier, but not simple, to drive. I won’t bore everyone with the details unless people are interested.

 

Chris: I did get one (well used) standard when I bought my roughness gauge, but some new ones would be good!

 

I’ve added a photo showing the bottom of the IGBT module. For reference it’s about the size of a paperback book. Each half bridge has the IGBT and diode dies mounted directly on an alumina substrate, which is what the gold rectangles are. Why you might ask are there ICs outlined on the bottom? We need to use a thermal paste specified by the manufacturer between the device and the heatsink. They recommend a thickness of 35µm and the use of screenprinting. Screenprinting, through a mesh, seems a little crude, so we had the idea of using a stainless steel paste mask, as used in the automated assembly of PCBs. The picture shows a test done at our assemblers using a gash paste mask, before we went to the expense of designing and having made a custom mask that covered only the alumina substrates.

 

In essence the IGBT-metal-coolant interface is a damn nuisance, but we have to have the metal in the way for other reasons. We just need to minimise the thermal resistance.

 

Right, back to the surface roughness measurement. The Ra measurement is essentially an arithmetic mean, easy to do with analogue electronics. The Rz measurement seems to be an average of the peak values only. This would be a bit tricker with analogue, but easy with digital measurement. The use of Rz would seem to be better intuitively, as it gives a measure of the total depth of the roughness, rather than just an average. So, in electrical terms it takes into account the crest factor of the surface. So, as far as I can see there is no simple way to relate Ra and Rz?

 

It would seem that we cannot really say whether the professional machine shop has or hasn’t meet the specification, unless they can measure Rz directly. However, we know that the heatsinks I made work at full power and high ambient temperature (100°C). Personally I think they ought to be able to match, or better, what the peasant writing this can achieve in his garage.

 

Should you be curious about the use of this heatsink look at http://www.provector.co.uk and look under ‘News’ and ‘610 Series’. It will tell you a bit more about the unit, who the customer was, and where it might be used.

 

Regards,

 

Andrew

 

[Anonymous]

An addendum:

 

For the idly curious I’ve added a photo of my surface roughness measurement unit. Given the relatively small differences in Ra values between the two heatsinks I find it amazing that a quick fingernail test can easily differentiate between them.

 

I’ve just had a proper look at the standard I got with the measurement unit. It says Ra is 2.97µm and Rz is 11,38µm. So it looks like the machine shop did meet the specification, but I’m still going to gently raise it as a issue. We’re about to give them a whole load more work, so they need to do what I want, or at least tell me why I can’t have it!

 

Regards,

 

Andrew

[Martin WParticipant @martinw]

Andrew

 

 

I hope your ref to:    ‘However, we know that the heatsinks I made work at full power and high ambient temperature (100°C).‘     is a typo otherwise unless you have a high pressure system you should be producing steam according to your first post re water cooling  !!.

 

 

Cheers

 

 

Martin

 

PS I know I’m not helping .

[Anonymous]

Hi Martin,

 

First, thanks very much for the links to the Ra versus Rz notes. They have been very useful. It would seem that our machine shop may or may not have met the specification, according to how we interpret the data.

 

On the other point, no it’s not a typo! The whole system consists of a diesel engine driving a three phase generator into our AC-DC inverter, which then produces a high voltage DC bus. The ambient temperature in the engine bay, where our inverter is located, can reach 100°C. We’re also about six inches from the exhaust pipe; nice!

 

Our system utilises the normal engine coolant for cooling the power electronics. There are two points to note. One, we’re first in line after the radiator, so the inlet coolant will be a lot cooler than 100°C, and second, the heatsink is pressurised, so it shouldn’t boil at 100°C. You wouldn’t believe how many screws there are to hold the lid down and make sure the O ring seals work. You also wouldn’t believe how small a piece of swarf is needed to cause a dribble of water past the seals.

 

Don’t worry about ‘not helping’; I love a good technical discussion.

 

Regards,

 

Andrew

[John OlsenParticipant @johnolsen79199]

Well, as has already been implied, measurement of surface roughness can be compared with the measurement of AC electrical signals. If we imagine tracing a probe along the surface, it will produce a waveform with a varying amplitude and with a range of spacial frequencies…this can be compared to an AC voltage waveform with multiple frequencies present. The difficulty comes in deciding how to measure it, and usually in doing so we make some assumptions which will not always hold true.

For AC we usually assume that the waveform is a nice sine wave, and that it will therefore have a fixed relationship between its mean value, its root mean square value, and its peak value. The older style of  moving coil meter actually relied on this, so did not necessarily read correctly for waveforms other than a sine wave. This causes problems when trying to measure say square waves, or noise waveforms, neither of which are sine waves.

Now, the surface roughness of our piece of metal is unlikely to be a nice sine wave shape. So we need to choose a suitable method of measuring it. We could use an RMS measurement, but that would actually allow through some things that we might not like. For instance a surface that was generally smooth, but with the occasional big spike could measure the same as a surface that had a smaller but more regular ripple. Hence the need to use measurements like the average of the peaks…which is a bit of a tricky concept to nail down.

For a heat sink application, when you bolt it up the two random surfaces will come together and meet at the high spots, leaving the paste to fill the gaps. Since the paste never has quite such a high conductivity as the metal, it is actually desirable for the paste to be minimised, although not to the point where there are air gaps instead of paste. Of course the higher the clamping force, the more the elasticity of the metals will cause the contact points to spread, improving the conductivity, at least to a point. 

regards

John

[Billy MillsParticipant @billymills]

Andrew

Why worry about surface irregularities of  fractions of a uM when you might have 35uM of much higher thermal resistance in the way? The surface finish is far less significant than flatness over the area of contact. Some semiconductors are best lapped against the heatsink. Many other power modules are on sheared integral heatsinks which are far from flat. 

It is always worth testing with blue to see how much contact area you actually have. You can use the regular irregularities to dither the binary  blue/no blue by getting the thinnest coat posible then looking at the blue /silver graduation.

 

 There is one common app where the power density is much higher, CPU’s . The interesting detail is that there is a mounting pressure advisory from the manufacturer which few motherboard mounts can attain long term. Some CPU’s dissipate 150 W from a 10mm x 10mm area, the ordinary blown finned al heatsink + smear of grease can give 0.1C/W, better with a silver loaded grease. Think that I would worry about the device contact over the very large area and how it will be maintained.

regards,

Alan

Edited By Alan Gray 1 on 04/11/2010 00:16:21

[Chris TriceParticipant @christrice43267]

Incidentally, as any car mechanic will tell you, the coolant temperature can exceed 100 degrees c providing the system is pressurised to stop it boiling. Cylinder head temperatures are easily in excess of 100 degrees.

[Anonymous]

I assume that the use of roughness parameters that mirror AC electrical signals is no accident, but was developed because the early measurement units used analogue electronics. Hence, parameters were devised that were simple to measure with analogue filtering and processing. The measurements are also altered according to how long a measuring stroke you use, how quickly the probe moves, and the radius on the tip of the probe.

 

All in all it’s more of an art than a science, unless you have an electron microscope to hand and can ‘see’ the surface directly.

Chris: I’m well aware that coolant temperatures, and cylinder heads, are normally above 100°C. However, this is not a conventional vehicle and we had a gaurantee from the overall systems developer that we would get coolant at an agreed temperature below 100°C.

 

Alan: It’s all a bit more complicated than just bolting the module down! The module is fastened with a number of high tensile bolts that have to be tightened in the correct sequence and to an exact torque. This actually pulls the parts of the module into place, forcing the internal contacts together, and forcing the alumina substrates into contact with the heatsink. The manufacturer’s quoted force per alumina substrate on the heatsink is over 3000N. So for the whole module, with three substrates, that’s over 9000N. At these sort of forces we assume that the thermal paste layer is going to be pretty thin, and is really just filling the voids. I can’t remember exactly what thickness of paste mask we went for, but I think it was 4 thou. For each substrate the paste was masked as a series of small squares, separated by gaps.

 

Regards,

 

Andrew

[Billy MillsParticipant @billymills]

The paste layer may be limited by particle size and compressability then. So any old paste is a no go. The case for lapping the heatsink seems stronger still, if you ensure that any high spots are lapped off with a non-embeding abrasive then the surface defects should be pits rather than peaks and fill in with paste.But how flat is the module?

 

Your original question on surface roughness remains interesting. How do you depict a three dimensional surface in a single number? the answer has to be an “engineering approximation ” that is very measuring method dependant- just as hardness testing methods are- but more so because of modality. SEM on E-bay? For heatsinking we want the plateau to valleys area ratio with no mountain ranges at all! 

Regards,

Alan.

 

 

[John HaineParticipant @johnhaine32865]

I was reading yesterday that the thermal density on the latest Intel PC chips is higher than a domestic hotplate….

[Anonymous]

Here’s the denouement.

 

We went to see the professional machine shop yesterday morning. They held their hand up and admitted a screw up on the wrong threads on the heatsink. That doesn’t worry me, we’ve all been there, they’ve had the honesty to own up. There were also a few other issues to do with swarf in blind threaded holes and deburring techniques.

 

With regards to the finish on the pad where the IGBT goes, there was no arguement. Just a quick fingernail test and no question that I’d got a much better finish. I got the impression he was rather embarrassed. They have no means of measuring Ra or Rz. They’re going to re-work one heatsink to see if it meets my requirements. Overall a positive and friendly meeting and we will certainly be working with them in the future.

 

Regards,

 

Andrew

[Dinosaur EngineerParticipant @dinosaurengineer]

If the surface finish spec was clearly on the drg. , why did the commercial M/C shop accept the job if they had no means of measuring/comparing it ?

[Anonymous]

D.Eng:

 

That’s a damn good question, and one to which I don’t have an answer. Only they know that. Given that a lot of their work involves waveguides, and hence talking to electronics and radio types, I assume that they thought the same of me. Presumably they weren’t expecting to be questioned about cutters, speeds, feeds and in what order they machined things, oh and why didn’t they try this type of cutter?

 

Regards,

 

Andrew

[Steve GarnettParticipant @stevegarnett62550]

Hmm. If Andrew gave them the link I posted at the start of this thread, at least they’d have some sort of idea what sort of roughness grade they were going for…

[MICHAEL WILLIAMSParticipant @michaelwilliams41215]

When sending drawings of prototype parts out to machine shops I have always found it useful to fully describe the part , its purpose and background in actual words either on the drawing itself or in a page of notes . Within this any specific requirements for finish and accuracy can be described in a more readily understood way than simply by symbology . In your instance I would put something like ‘ the machined areas indicated on drawing (reference) are are intended to be mountings for electronic modules and a high level of flatness and a very fine finish are required . We can show you samples of the type of flatness and finish required on other existing parts ‘ . The machine shop will then make the part with full understanding of what is needed and in my experience will make a superb job of it – often exceeding the actual specification requirement by a good margin . At a later time drawings for production can have rigorous specification but the same machine shop will then know in a more complete way what is actually needed .

 

 

[Anonymous]

Michael: All very sensible suggestions. Before we placed the order, we visited the machine shop in question. It’s only a few miles away, which was one of the reasons for selecting them. We took one of the prototype heatsinks that I’d machined, so that they could look at it and ask questions. We also explained it’s purpose and the finish required. It’s may be slightly ironic that the finish on other parts of the heatsink is better than that where the IGBT module goes.

 

Still at least they’re doing better than the sheet metal company we’re dealing with; ‘order, what order, we didn’t know you’d placed an order’!

 

Regards,

 

Andrew

[Author]

Posts