Repeatablitiy of Tyre Data

[Menisk]

Just curious to start a discussion on the accuracy and repeatability of the TTC tyre data. I've also started a thread on the TTC forums which has some of the data I've been comparing (http://sae.wsu.edu/ttc/viewtopic.php?f=19&t=130) but I'd also like to start a discussion here as I think more people will see it.

Remember that TTC data can only be posted in the TTC forum, not here.

The problem I'm facing right now is the massive difference in the data for the control tyre between round 4 and round 5 of testing. I've compared the Hoosier 20.5x7-13 @ 10PSI on a 6" rim and found that the shape of the curves was essentially the same but the lateral force produced for a given normal force was remarkably different. Round 5 testing produced far more lateral force. That wouldn't concern me if that was the case for all tyres and you just had to scale data to take into account wear to the sandpaper belt or whatever it was that caused the difference. However when I compare the data for the same tyre @10PSI on a 7" rim the peak lateral force is similar, however there is a big difference in the load sensitivity and the shape of the curve is different. It's to the point I'm wondering if I'm somehow using the wrong data sets as they look like entirely different tyres. (In the TTC thread I've posted an explanation of which data sets I've used and how in the hope people can tell me if I've done something very wrong.)

So my question is, has anyone else looked at this? Does anyone have any experience with manufacturing tolerance on tyres or how the FY-SA curves change depending on the aggressiveness of the surface?

I imagine that not only the maximum force but also the curve of the tyre will change depending on the surface you're running on as the tyre will deflect differently if it's on a surface that makes less force. Is the tyre data even useful at all or is it unreliable and not representative of real world conditions?

[DMuusers]

When comparing round 4 to round 5: was the slipangle for the maximum lateral force the same or was it different? Have you also looked at the different temperatures of the tire/room? For the 7" rim: in what way is the load sensitivity and shape different? More sensitive to load? More dropoff after the peak? Also, were there significant changes in camber sensitivity?

It is also very well possible Hoosier changed something in the structure of the tire so it performs better/different on a certain rim width. Otherwise maybe the surface of the testdrum was of different material? It's very hard to say where the difference comes from without knowing the entire test setup aswell as the structure of the tire. However, in general testing on a drum is really hard to translate to a road. Even translating tire measurements from one type of road onto another one can be very hard. This is mainly caused due to the 2 ways grip is provided by the tire: adhesion and hysteresis. For a more coarse surface for example, hysteresis plays a big role in tires with a very soft tread compound and stiffness. Because it forms to the road better, more force can be created by it. For flatter surfaces adhesion can also play a big difference. Rubber sticks better to tarmac than concrete for example (even if they have the same surface coarseness).

In general, testing on drums can provide insight into how stiff or soft a tire is, so how easy it is to drive. It also provides some trends(!), for example camber sensitivity. But for exact values, in my opinion, it's of little use.

[Edward M. Kasprzak]

Doug Milliken and I are very busy this week, but we will review Menisk's question and reply next week.

I will note that all TTC data has been collected on the Calspan Tire Research Facility's flat belt testing machine. No drum testing has been performed.

[BillCobb]

I looked at the graphs you showed and have several suggestions for checking the results. The differences between the two tests is far greater than one would expect to be normal variation, even considering the likely hand built status of the two tires. This is just my opinion based on tests of dozens of tires with identical build recipes. Yes there are some yearly variations in the 'control tires' (yes there is a standard tire used in the industry to check out tire test machinery. These tires are not used on vehicles, but only to validate and verify the machinery accuracy).

First make a comparison of just the raw data points for a single load (maybe the 80% of rated load sweep). Do the same for a single slip angle and a load sweep. This will show you the range of each sweep and any transient hiccups that could confuse the fitting routines. Post it back for us.

Second, someone needs to verify that these are in fact the same tire (brand, size, construction, pressure, wheel rim and width, BUILT DATE and mounting orientation). The differences you show are highly suggestive of completely different build packages. The low slip angle comparison says (to me) tire structure is different (bead, sidewall and breakers). High slip angle area differences are usually tread and compound differences.

If CALSPAN makes calibration runs before each test group using an ASTM control tire, then let's see those results. That would indicate whether machine electronics or transducers are involved (Doug and /or Ed). I doubt very much whether any belt surface issues are open because the low slip results (slope and offsets) are so different. Think of it as a comparison of a pair of running shoes being observed at a slow walk. (A tire really acts like a millipede with a lot of little running shoes).

Don't be surprised if you find out the tires are actually a different formulation. Such things are/maybe commonly done to the unsuspecting. This is why you should 'stick' with a tire with known, stable, high volume and major brand volumes. Designing a chassis and steering system to fit around a 'good' tire is less stressful, more productive, more satisfying and more fun if the tires you buy for synthesis, testing, validation, practice and competition are always the same. All it takes is one goofball out of 4 on your car to cause a headache. I always say that if your target is moving, select a belt fed weapon instead of a single round.

That's my opinion and I'm unanimous with it.

[BillCobb]

Here's an example from production car tire testing. No specific reason I ran this size, just grabbed a random wad of data. There are 7 tires on this plot. You might expect a shift in the data because of the previous test series' effect on the belt surface, but that ain't the case. Good, fast, cheap. You get to pick only two...

[Menisk]

Thanks for the replies. I'll shoot Hoosier an email and ask them if there have been changes to the tyre between the two rounds of testing. In regards to the TTC licence is it okay to send some screenshots of the curves along with my email to Hoosier so long as it's only their tyres on the plots? Not sure if this violates the licence agreement as one would assume that Hoosier would know what their tyres should be doing to begin with.

[DougMilliken]

... I'll shoot Hoosier an email and ask them if there have been changes to the tyre between the two rounds of testing. In regards to the TTC licence is it okay to send some screenshots of the curves along with my email to Hoosier so long as it's only their tyres on the plots? ...

Hi Menisk,

As Dr. Kasprzak noted above (and I posted over on the TTC secure forum), we are looking into your results but we have to fit this in with our regular work. Hopefully we will post something this week.

It seems likely that differences between the tests (which were about two years apart) generally explain your findings. Our thought was to look at the TTC test records first, make sure we know everything possible about the two tests, and then contact Hoosier if it seems necessary. The engineers at Hoosier are probably busy too (since it's the middle of the race season in N. America)...

License question: The tire companies that donate tires to TTC are "industry members" (all except Avon, TTC had to buy Avon tires). Thus they have access to all the data.

-- Doug Milliken for the TTC

[Menisk]

I'll leave it with you guys for the moment then and see where things go. Thanks for being so helpful. :)

[Edward M. Kasprzak]

Doug Milliken and I have posted a response to your question on the TTC forum.

[Z]

Doug Milliken and I have posted a response to your question on the TTC forum.

Awww..... :(

I was hoping to find out just how (IN)accurate this tyre-data can be.

Any chance of an "executive summary" of the findings being posted here? Maybe just some curves painted with a 4" brush? :)

Z

[DougMilliken]

Any chance of an "executive summary" of the findings being posted here? Z
1. Lab tire tests were run on the same tire size, about two years apart. While the tires were the same "code", it is likely that they were from different batches, since this is a popular size (we have not checked with Hoosier about this). Looking back at all the test data (including various temperatures) indicates that the test conditions were not exact repeats. With the temperature sensitivity of race tire compounds, we would expect the results to be somewhat different.

2. In a case like this (repeat test) it is probably best to compare the actual raw data, across the whole data set.

-- Doug Milliken for the TTC

[Edward M. Kasprzak]

Adding to Doug's #2: While there is some disagreement between the data from round 4 and round 5 data for this tire, the differences in the data are not as extreme as seen in the models provided by the original post. Some of the raw data peaks do not match (we discussed reasons why in the detailed response) , but not to the extent seen in the models and the raw data on-center slopes agree very well.

While comparing models can simplify the comparison it is another step away from what's really happening and can be another source of error.

I would like to see the original models plotted over the raw data (on the TTC forum, not here). I did not find raw data which matched the model results. The modeler may be making a mistake in the modeling process, or I might have misunderstood the original post. Either way I'd be happy to continue this discussion.

[BillCobb]

The plot thickens.

[DMuusers]

This is quite interesting. Any word from Hoosier yet?

[Z]

Thanks to Doug and Edward for the above updates.

I take it that the "raw data" is a series of spray-painted curves, drawn with the spray-gun quite far from the paper. On the other hand, the "models" use some data reduction magic to turn the spray-painted dots into extremely sharp and accurate lines?

Except that in this instance some of the magic might have misfired, and not all of the super-sharp lines are in the middle of the spray-painted dots???

Ahh, yes. That's the trouble with magic... :)

Z

[Z]

Further to above, does the standard sort of magic used for these things (Pacejka?) give any indication of the standard deviation of the input data?

That is, does it give any numbers for how accurate, or inaccurate, the final curves might be (ie. a +/- "tolerance" value)?

Z

[Jay Lawrence]

Z,

Assuming that people want to design their vehicles at least somewhat around the characteristics of the tyre (i.e. they'd like some information prior to buying them), are you suggesting a better way to determine these characteristics?

[DougMilliken]

... I take it that the "raw data" is a series of spray-painted curves, drawn with the spray-gun quite far from the paper...
De-identified examples of raw data from TTC Round 1 can be seen in http://www.millikenresearch.com/TTC_SAE_paper.pdf This is linked from the TTC public page, http://www.millikenresearch.com/fsaettc.html . Even with our student discount, Calspan TIRF is a damned expensive way to spray paint!

We have learned a few things about these little tires since Round 1 and I think it's fair to say that TTC has improved our spray painting with each Round. We normally leave the interpretation of the data to the students--our role is to provide raw test data, with the volunteers and budget available.


... On the other hand, the "models" use some data reduction magic to turn the spray-painted dots into extremely sharp and accurate lines? ...
Sharp? Yes.

Accurate? Maybe. What does "accuracy" mean when the test sample (tire) is wearing out during the test and is tested on a different surface than used on the road? There isn't a simple answer. But a great deal of work has been done on this problem.

Tolerance (your other post)? Someone else might comment on this. We don't use any of the commercial Pacejka/MF (Magic Formula) modeling codes since we have our own tire modeling technology.

Going off topic, I can't resist tossing in a bit of history--when Hans Pacejka was a student in the early 1960's, he worked/interned at CAL/Calspan and studied their early tire research. Then he went to Delft U where the MF approach was developed. Pacejka's right hand can be seen in the picture on page 423 of RCVD... The quote from Len Segel that starts on that page might be of interest to anyone involved in tire and vehicle modeling.

[DMuusers]

So if I read correctly the raw data of the control tire is very similar for both rounds and the fitted/filtered data isn't? So that means it's a problem with fitting data rather than differences from one test to the other?
Oh, and 'little tires' is very subjective these days ;)

[DougMilliken]

So if I read correctly the raw data of the control tire is very similar for both rounds and the fitted/filtered data isn't? So that means it's a problem with fitting data rather than differences from one test to the other?
Oh, and 'little tires' is very subjective these days ;)
It was not actually a "control tire" in the sense of getting special treatment. As noted above, it was the same "code", but no special efforts were made beyond that. It was added to Round 5 because it was a popular size.
These tires are "little" compared to the full sized race tires that we normally test, sorry for not supplying a frame of reference.

[BillCobb]

I will be open and honest in my belief that the Pacejka formulation is a convenient and thorough technology that clearly works for some tires. In fact, these tires are often designed and structurally configured to produce test data that conforms Pacejkally to the 3 regions of tire operation: Straight ahead ("linear range") driving on 2 or 3 lane roads at legal speeds and low bandwidth drivers. Then the midrange region with significant load transfer, softening cornering stiffness and significant second partial derivative changes to forces and moments due to slip, camber and vertical loads. Lastly is the region of operation seldom experienced by drivers still in control (or who believe they are in control) of the vehicle.

This is not a two way street. There are many tires with structure and tread materials which clearly do NOT fit the Pacejka paradigm either by design or by manufacture or by use. Work-arounds for this 'problem' include adjustment of the fitting algorithm's weighting function to downplay the area or poor representation. (You give up low g-level fidelity to get the best racing groove clarity). Or you give up the max-lat fidelity in order to meet a vehicle manufacturer's cornering coefficient specification. And of course there is the symmetry problem.

The 'best' data modeling in my experience comes from continuous splines that will clearly track all three regions and asymmetric characteristics. The weakness in them has to do with extrapolation and use conditioning factors (i.e. there is not practical way to adjust the splines for a change in air pressure or wheel rim width for example). The splines can be slow because of computational overhead and require large storage space in a tire library. You need to test every tire you use. No way to create clones of other tires because there is no structural basis for the coefficients produced. But if you have a good machine and a wisely chosen test procedure. You get data good enough to get a car successful right off the hauler, as they say. It's expensive to. You have to (or should) test every tire you use at each wheel position. Some of us have had the luxury of checking a box to get all this done, including the spare and also some non-sticker tires because the ones right out of the molds need to be cooked off in order to get stable results.

Let me also add that this whole thread also needs to have mention of a vehicle test program which is NOT related to track use or lap time generation. Instead, measuring control engineering parameters (gains and response times and natural frequencies and damping factors) for at a minimum of two conditions: Tires in a base position and also with fronts and rears switched. If the results are nearly the same, you can be pretty certain that you have a stable, consistent and manageable vehicle development program. If they are not, you probably have a 'flyer' at one position and so it is a waste of time to continue any further adjustments to the car (or truck, or golf cart or go-cart). This works well ever when the tires are different constructions on left and right sides.

I've looked at some other Hoosier non-TTC race and performance tire tests and am not surprised with what I've been reading here. They are not like globally available MacDonalds cheeseburgers all made from the same process book with identical ingredients, color, smell and taste. Since we usually tested three tires on of a given car for a model correlation comparison ( a front, a rear and a virgin extra), its nice to see a stable and consistent thin line for each tire (as I've shown in my previous rhetoric). But it isn't always the machinery, or the belt prep or the operator or the mounting and pressurization sequence. Sometimes its the Chef.

[Menisk]

Massive thanks to those that have put a heap of time into helping me come to an understanding here. I'm starting to get the feeling that the tyre data is something that needs to be used slightly differently to how I imagined. I went into this assuming that I'd be able to line up my models and do an analysis where I could say that for a given situation tyre A does what I want better than tyre B. I'm starting to get the feeling that the data is only good for giving me an idea of how tyre A works and an idea of how tyre B works, but that they can't be compared exactly. I assume that while on the belt tyre A might produce more lateral force than tyre B, I could end up finding the exact opposite on the surfaces we run depending on whether it relies more on mechanical or chemical adhesion, the ambient temperatures we run in and our ability to keep the tyre at it's operating temperature with our car.

In reality I think the only way we can properly compare tyres for our purposes is to head out to track with a set of each and put them head to head with a heap of notes and setup changes. I can see the tyre data being useful in the sense we can already have a ballpark setup for each tyre when we do this.

[Edward M. Kasprzak]

Two thougths:

All laboratory tire data should be correlated with on-car performance. Occasionally the two line-up very well and you're ready to go. Usually there is a process of understanding the differences between lab and on-car performance beforethe wealth of information in the lab data can be accessed. Adjustments in tire data handling and tire modeling are often required. Direct comparison of lab data to lab data is possible if the tests are specifically designed to minimize the influence of uncontrolled variables. For example, comparing two Round 5 TTC data sets would be reasonable, and comparing two Round 4 TTC data sets would be reasonable. But as we've seen in this thread, comparing Round 4 against Round 5 may not give good results as the tests are different enough to make a difference.

All tire models have different strong points and shortcomings. Pacejka models are not the only solution. As Doug mentioned earlier in this thread, MRA has (non-Pacejka) tire modeling technology, and there are other approaches. If you're using a model that does the things you care about well while doing a sufficient job on the things you don't care about too much then you'll probably be very happy with your tire model. Otherwise you won't be. One size does not fit all.

[Z]

Thanks to Doug for the link to the Edward's SAE 2006-01-3606 "FSAE TTC..." paper (back on page 2).

From Figure 4 of that paper it seems that the curves are in fact done with a very fine-point "airbrush". Especially so, given that each distinct curve in that figure supposedly represents 5 different camber angles! (Indicating that those particular tyres have very little camber sensitivity?)

However, the paper also has the comment, on page 2, that "... on road testing ... is not repeatable."

That gets to the gist of my feelings about this whole tyre testing issue, which is best summarised by Prof. Walter Lewin's much repeated comment from his series of MIT video lectures (use Search...)

"Any measurement is MEANINGLESS (!!!!!), without knowledge of its UNCERTAINTY!!!"

(Prof. Lewy also likes to shout a lot :), and apologies to him if I didn't quite catch his emphasis correctly above.)
~o0o~

So, getting back to tyre-curves, my feeling is that if I was in the students' position, then;
1. I would FIRST want to know just how accurate (or not?) the curves are going to be.
2. Only SECOND would I want the actual curves.

This seems to be the exact opposite to the (typical) students' approach of wanting curves with Matlab-like 5 significant digit precision, but never showing any concern for just how uncertain those curves might be. (Well, hardly ever. OP and Bill and Edward and others above have qualitatively pointed it out, but still no quantitative "tolerance bands", or even estimates of such...).

So, again, "Mutter, moaaan, ... failure of the Education System, ... but at least Lewy is trying to make the point...". So once again,

"Any measurement is MEANINGLESS (!!!!!), without knowledge of its UNCERTAINTY!!!"

Z

(PS. Here is link to Prof. Lewin's MIT Lectures. See at 4:30 "Any measurement is MEANINGLESS...".
Lewy Lecture 1. (http://ocw.mit.edu/courses/physics/8-01-physics-i-classical-mechanics-fall-1999/video-lectures/lecture-1/))

[Flight909]

Thanks to Doug for the link to the Edward's SAE 2006-01-3606 "FSAE TTC..." paper (back on page 2).

From Figure 4 of that paper it seems that the curves are in fact done with a very fine-point "airbrush". Especially so, given that each distinct curve in that figure supposedly represents 5 different camber angles! (Indicating that those particular tyres have very little camber sensitivity?)

How do you arrived at this conclusion? Looking at the graph I see almost 10% difference in lateral forces for the different camber angles (for sure the markings on the graphs are quite difficult to see). In you're experience, how much does force vary with camber?

[Z]

How do you arrived at this conclusion? Looking at the graph I see almost 10% difference in lateral forces for the different camber angles...

Flight909,

Following are some quotes from Edward's paper, and my comments.
~o0o~

"Typical lateral force versus slip angle raw data is shown in Figure 4. This plot contains 5 loads and 5 inclination angles worth of data, which can be sub-divided into 25 individual curves."

"Slip angle sweeps to ±12 deg. at all combinations of loads and inclination angles.
* Inclination angles: 0, 1, 2, 3, 4 deg.
* Loads: 350, 250, 150, 50, 450 lb. (1557, 1112, 667, 222, 2002 N)."

My guess is that the five reasonably distinct curves seen in Figure 4 are for the five different Fz LOADS. Each of these distinct curves would then contain the five different inclination angles (~cambers). I base this on Fz loading being a first-order (ie. big) influence on lateral force, while camber angle is usually only a second-order influence (ie. somewhat smaller).
~o0o~

BUT also (!),

"Second, there is temperature variation in the data, as seen in Figure 5. As any tire is steered the surface temperature increases—the slip angle to sweep from zero to a large slip angle often produces somewhat lower lateral force than the return sweep to zero because of the higher temperature of the return sweep."

And,

"As normal load increases the tire gets shorter, and as the tire is steered the tire also gets shorter. This is most noticeable at very high loads. A careful look at Figure 7 shows that there is some leftright asymmetry. This is due to a phase lag in the test machine’s ability to maintain normal load. These small, low spring rate FSAE tires proved to be a challenge for the test machine. The effect of the normal load variation about the target is also present in the data of Figure 4." (my emphasis)

So at any given nominal normal load (Fz) there is some "thickening" of the airbrushed curve by the above temperature effects, and also by the test machine's difficulty in maintaining a constant Fz.

So that suggests that any lateral force (Fy) variation due to camber is very small indeed, since this variation might be hidden inside the above "thickenings" of the curves. And most interesting, IMO, is that this applies over the range of cambers (Edit: INCLINATIONS!) given above of 0 to 4 degrees. Quite a lot! (Edit: So from -4 to +4 degrees of camber! See PS.)
~o0o~

I would guess the Figure 4 results are from a highish-profile, bias-ply tyre, with quite a rounded cross-section? I would also guess that a low-profile (<~45%), radial-ply tyre, as commonly used on sporty production cars, would be much more camber sensitive. Perhaps losing a third (or more?) of its lateral grip over a camber range of -4 to +4 degrees???

Corrections to above welcome from anyone who knows more about these things.

Z

(PS. "Camber" and "inclination" angles are the same numerically, but may have opposite signs (+/_).)

[exFSAE]

I would guess the Figure 4 results are from a highish-profile, bias-ply tyre, with quite a rounded cross-section? I would also guess that a low-profile (<~45%), radial-ply tyre, as commonly used on sporty production cars, would be much more camber sensitive. Perhaps losing a third (or more?) of its lateral grip over a camber range of -4 to +4 degrees???

Rule #1 with tires: Do not make unbased speculation / assumptions / guesses. Even after years of working with gigabytes of data, you can be wrong more often than not.

[DMuusers]

Rule #1 with tires: Do not make unbased speculation / assumptions / guesses. Even after years of working with gigabytes of data, you can be wrong more often than not.

I agree. In my experience with designing tires, camber sensitivity is much more dependent on internal structure than shape or dimensions (though they do contribute). From tire data you can only say how stiff or soft a tire is, what its lateral stability is, camber sensitivity, etc. in trend like behaviour (not actual numbers). Linking those parameters to the actual size or structure type or whatever physical properties is in my opinion impossible (without seeing the tire).