Recognizing Brake Fade with Data Analysis
 

During our last track day, we noticed we were getting some pretty significant brake fade from our store-brand parts house pads. I decided to see if I could recognize that in the data. Knowing I was running a pretty big risk of confirmation bias, I dove in.

This was all done with an Aim MXL2, but could be done with any datalogger capable of recording brake pressure via OBD2 and deceleration at 50 Hz or so. I don't think 10 Hz would be sufficient considering how fast a braking event goes by at race pace.

Disclaimer: My Python is a little rusty, so this was done in Excel. Could definitely be streamlined with better programming skills, but I'm a mechanical engineer, not a programmer. :D

I'm using RS2Analysis, the Aim software supplied with their hardware. It makes it pretty easy to export select channels and laps to Excel, which can help prevent "data lock." I export "AccelerometerX [g]" and "TOYO_BRAKE_PREX [psi]" and only select laps at full speed. Pit in/out, heavy traffic, whatever shouldn't be included in this analysis. Anything within 120% of your best lap should be fine.

There are several different ways to do this, but the next step is to break the data up into distinct braking events. You should be able to look at a G plot on the track and see there's a set number of braking events per lap. In my case, it's 5 events per lap. You can manually go through and number each event where brake pressure is greater than zero, but it's very tedious. It's fairly easy to use some if() statements to have Excel do that for you. Here's my method:

-First column next to brake pressure is called "EVENT": If brake pressure is greater than 0, put a "Y". The OBD channel is pretty good at recognizing actual brake pedal action and not just residual pressure. It's resolution is one bar, so false positives were nonexistant in my data.
-Second column is called "BEGIN." If there's a "Y" immediately to the left of the cell, and NO "Y" immediately above THAT cell, then put a "1" in this field. What we're doing here is recognizing the beginning of each braking event.
-Third column is called "COUNT." First, check to see if the "EVENT" cell to the left has a "Y" in it. If it does, sum all "BEGIN" cells above this cell. This will number each braking event. Once this column is working properly, you can hide the "EVENT" and "BEGIN" columns. They won't be needed again.

So at this point, you've got your braking events isolated and numbered. On a new sheet, you want to use "AVERAGEIF()" to get the average deceleration and brake pressure from each braking event. I converted my brake pressure to ksi instead of psi to make the units a little bit neater, and inverted the sign of the acceleration since my goal is measuring deceleration. Then you're ready to create your deceleration gain by dividing your average deceleration by your average line pressure.

From there, it's just a matter of graphing the parameters you've created. In the plot below, I made each lap a separate color. You can clearly see there's a consistent pattern for each lap.

http://i.imgur.com/uKHWbGV.jpg

You can see after about the 5th lap, the braking force generated drops off significantly. These pads are done. Without temperature data, it's hard to get quantitative results on pad performance that could be used to compare different compounds. But trends can still be very useful tools in data analysis.

How could I use this to improve my driving?

By calculating a deceleration gain, we're removing the question of "Did I hit the pedal hard enough each lap?" from the situation. The test I used for this example has some pretty extreme fade after the 5th lap, which the driver was able to easily recognize. In less severe situations, you might notice the trend for each lap dropping slightly in the data, but it may not have been significant enough to really notice. Analyzing your sessions like this could help you identify if your brakes need attention, like a better compound, better ducting, etc.

If you felt like adding some pad temp sensors in the mix, you could pretty conclusively notice when and why your brakes are dropping off.

Why does each lap have a non-linear shape?

In short, weight transfer. We don't have wheel pots to quantify our pitch and roll, but we DO have brake line pressure. By plotting brake line pressure, we can see there's a strong negative correlation with deceleration gain that doesn't show up with other dynamic indicators like yaw rate, lateral acceleration, etc.

This tells us that's it's likely that as we get harder on the brakes, more weight shifts to the front of the car, which shifts weight off of the rear tires. The total load on all four tires is still going to be 2900 lbs or so (neglecting the marginal aero from a stock car), but tires are load sensitive and unevenly loaded tires will produce less overall force than an equally loaded pair. If we had pitch data, we could correct for this in the analysis. It's going to be interesting to see if this tendency is reduced as we modify the suspension and reduce dive under braking. If someone else wants to do this analysis on a more modified car, I'd love to see the results.

This is far from the only way to characterize braking system performance, but it's worked well for me in the past and it's a great way to take advantage of a "simple" data system without more than accelerometers and OBD connectivity. I'd love to hear everyone's thoughts on this and other methods.

Jake

You had me at Excel... Will read this in depth tonight...

-mike

I'm trying to wrap my head around the implication of G/ksi so forgive me if I'm missing something obvious.

Why does lap 2 have 6 events?
What happened lap 3 to cause such a low reading relative to 2&4 across the board?
If you've tossed out laps that are slowed up by traffic would it be better to label the laps appropriately? Lap 4, 5, 7, 9 etc.?

If the brakes are thoroughly faded after lap 5 why does lap 8 show the highest stopping acceleration on turns 4 & 5 with the exception of lap 2&3? Was that a result of babying the brakes? Wouldn't that indicate that there is still some time left on the table assuming the brakes don't need to be managed?

Assuming laps 5-8 are consecutive that would have been an excellent dataset to show with an accompanying video.

I enjoyed it, thanks.
:burnrubber:

if you're using Aim's Race Studio software then all you had to do was select the "math channel" graph options and plot your selected lap's (or all laps, if you wish to colour code) longitude g's vs brake pressure, you would end up with a user-friendly scatter-plot graph, (after some unit spread massaging)


at which point you just click on the points of interest (those with high brake pressure but low longitude values) while having the map open and it would tell you where this event is happening.

I am very interested in your data as well.

I have been researching brake systems as I am rebuilding my system on my racecar and adjusting rear bias.

One thing I always try to look at from data on a Traqmate system is:
Do you get the same braking force (in G's) as you do in cornering?

We have found that we consistently do NOT.
If you see more lateral Gs (during cornering) than straight forward (during braking) this means the car is capable of more braking.
We consistently see 1.2 Gs in cornering forces and less than .8 Gs in braking.
This means we should be capable of braking harder.

I have also considered adding thermocouples to one pad on each caliper.
If you lined up that thermal data with what you have above, you would know if your fade is due to pads overheating or something else.

just to be clear, you mean to say


If you see more lateral Gs during cornering than longitudal Gs during braking

which I have also wondered (braking g's always less than turning g's) but i had always thought there would be a logical explanation rooted in physics


they teach us the friction circle but I assumed it was never meant to be 100% circular. (ie, if a tire can brake 1g it should turn 1g)

The driver hit the brakes in a location he usually doesn't. We didn't have the track to ourselves, and I believe it was due to traffic.

I'm assuming you mean 4 relative to 3 and 5. So this may not make sense if I'm wrong. It looks like the driver was hardest on the brakes that lap, looking purely at brake pressure, which would keep the gain down (weight transfer and all that). The notes say the brakes fell off after laps 4 and 5, so that matches that too. I agree it does look abnormally low, but without more sensors on the car I don't want to speculate.

Certainly could. The actual numbers were 2, 3, 4, 5, 6, 7, 9, and 10. I don't a good reason for that other that clarity for a theoretical situation.

You mean lap "7"? I'm honestly not sure. You'll have to forgive my somewhat vague answers here, I was out of the country for this test, so I'm piecing together this analysis from driver notes and interviewing coworkers. :D

Sure

My SWAG says that you should be able to get MORE during straight braking then you would ever get in max cornering.
The geometry of the tires is more consistent and has a larger contact patch during stright braking than in hard cornering.

Interesting idea, I'll look into that a little more. My initial reaction is you lose the ability to get an average value for each braking event, since you're looking at every measurement taken over the course of the lap. By taking the time to massage it in excel, you can get five nice, neat values per lap that are a little easier to deal with.

I see the same thing in our data. We're flirting with 1G decelerating, but seeing 1.2-1.3G lateral on a regular basis. I agree, it's pretty counterintuitive. The only thing I can think of, is the anti-roll bars allow the inside wheels to contribute more to the cornering than the rear wheels can contribute to braking. This is another location wheel pots could really help figure out what's up. If anyone has some insight on this, I'd love to hear it. I'll do some reading.

Yep! That would make this much easier to understand.

Kudos! Interesting stuff. I've thought about doing some extra analysis along the same lines as this in Python, but...I just haven't. Lots of potential here though so it's good to see someone going for it.

One thought is flagging corners where ABS is activated as it may have an influence on those averaged values. (not sure if it was for your data, just throwing it out there).

With a firmer suspension, you will reduce pitch/dive but the total transferred load should be roughly the same. The load will transfer more quickly so you may find it easier to activate ABS (depending on the tire). It would be interesting to see what happens.

- Andrew

well yes, if you want to quantify the amount of g's lost vs pressure lap after lap you'd need to sit down and do the spreadsheet


however, if need to quickly confirm with the driver if and where he's fading, a quick math channel (which could be preset) will give you a graph that will confirm at which point on the track exactly the brakes begin to fade out.

I only have an Aim Solo DL, but it would be possible to do this on the fly without the driver even getting out of the car, just need a decent laptop and quick fingers.

.......something I wish to test out one day, haha. (on site data logging services).

Yeah, you're right. I was thinking from a perspective of lowering the CG height since we're still at 4x4 status.

I'll look into a good way to recognize ABS activation, I don't have a channel for that currently.

That would be very interesting to see. It would be nice to correlate that to weight transfer/pitch to hopefully dial out some of the variability I'm seeing across a given lap.