PRELIMINARY M-DCT shift times - results and data

[lucid]

Originally Posted by swamp2

As well look at the size of the oscillations at the beginning of the D3 data and just after that shift. A couple of points show data with negative acceleration. That surely is not real, no way.

Although I agree that it is unlikely for those dips to be real, I think it is too early to say what is and what is not real at this point. Who knows, maybe the issue I brought up with clutch slippage/engagement is plausible since things are happening so fast in the D3 shift, and maybe those oscillations that occur after the interval you have marked as "the shift" are artifacts of the engagement process as the clutch loses and regains traction. Unlikely but not impossible.

[swamp2]

Originally Posted by lucid

Although I agree that it is unlikely for those dips to be real, I think it is too early to say what is and what is not real at this point. Who knows, maybe the issue I brought up with clutch slippage/engagement is plausible since things are happening so fast in the D3 shift, and maybe those oscillations that occur after the interval you have marked as "the shift" are artifacts of the engagement process as the clutch loses and regains traction. Unlikely but not impossible.

Some of the oscillations AFTER the shift may indeed be real. High frequency mechanical osciallations from a spike type input are the norm and possible here. As well the slipping you described may be actually measured. However, the deceleration (as visible just at the beginning of the D3 shift data trace - and to be 100% clear at the beginning of the displayed data, which is well before any sign of any clutching) cleary is not real. Since that represent a period in reality of relatively constant, smooth acceleration you can draw similar conclusions about similar noise levels in all of the rest of the data. The exception seems to be a lack of noise during high jerk (i.e. slope of acceleration). Particularly good for us but I have no idea why the data gets cleaner!

[swamp2]

Originally Posted by footie

Based on the evidence that swamp has uncovered the whole lag after flicking the paddle is purely interest and has no real bearing on the performance of the car. Acceleration seems to be unbroken with DCT as all the manufactures have claimed and with surge there is a slight increase during that 200ms transition from one gear to the next, the only benefit with discovering to amount of lag at different levels is to learn why it's there at all.

Absolutely agree. But this data was not needed to realize that. We all knew simply from driving that all acceleration continued truly as if nothing had happened in the period between the paddle actuation and the beginning of clutching! What is novel is the the surge seems to occur across the entire shift, not after it.

Originally Posted by footie

We all know that SMGIII might quote a shift in 80ms but I doubt anyone believes that to be the case, I have read here that some reckon it's closer to 180ms which even using the none surge modes mean the DCT improves on this by up to 5~6 times. One can only imagine what difference it will bring to the likes of the M5/6, RS6 and other truly bonkers cars.

I will be interested to see how quick the shift can actually be after additional testing, is it truly possible that DSG and DCT actual shift in the often quotes 8ms. That's mind blowing.

We have seen accelerometer data from an M5 (or M6) showing the ~150 ms number. The suspicion is about half and half clutching vs. moving the synchros. BMW chose to report half the picture. Again my strong hunch is we may see 20 ms or so on D modes from 6->7th but 8 ms, VERY doubtful. This is going to be very difficult to test as well. It will require a WOT shift from 6->7 due to signal to noise, that is fast!

[lucid]

Originally Posted by swamp2

The exception seems to be a lack of noise during high jerk (i.e. slope of acceleration). Particularly good for us but I have no idea why the data gets cleaner!

Well, we don't know that the data get cleaner at that point. How do you know that is pure signal? That's why I asked about the frequency of the noise during that interval in another post, which seems to be a variable across the data range you plotted and is a function of how the sensor is mounted among other things. Anyway, maybe we are all reading too much into this preliminary, and somewhat noisy, data. We'll see how things look after better mounting.

[swamp2]

Originally Posted by lucid

Well, we don't know that the data get cleaner at that point. How do you know that is pure signal? That's why I asked about the frequency of the noise during that interval in another post, which seems to be a variable across the data range you plotted and is a function of how the sensor is mounted among other things. Anyway, maybe we are all reading too much into this preliminary, and somewhat noisy, data. We'll see how things look after better mounting.

I am only saying this based on pure empericism. I saw the clean data in shift after shift (D modes only!) as the derivative of the value became very large or very small. I do agree fully that this data is VERY noisy and significant improvement needs to happen here. Hopefully Kens mount will help. The next step would be replacing the accel followed by that nice clean power supply.

[southlight]

Maybe I'm completely off but I'm not sure about the 35ms shift in D3. What says you that the shift is completed at the point you were drawing the "end of shift" line, swamp? My understanding is that the shift can't be completed as long as the acceleration is higher than it was before the shift and that's some time longer. In case I'm plain wrong here say it in a nice way please.




Best regards, south

[ArtPE]

Originally Posted by swamp2

The accel is quoted as:

• 0.5 mV/g nominal output (sensitivity)
• bandwidth 0-1.5 kHz, 3% down on output at 500Hz, no specs given for higher frequencies
• 0.3% non linearity
• an odd accuracy spec., 0.3 mg/sqrt(Hz). ???.

You can get the entire spec sheet in my post #10.

The data sheet for the DAQ is attached. So many "accuracy" specs available it is tough to determine how much noise may be from the DAQ. Not my area of expertise. My gut tell me very, very little.

In general a car can slow down during a shift (well D3 yes, S4, no, if we believe this data!). My point was that the car can not be decelerating before any shift initiation (as the car was nearly pegged!). This makes the mounting and noise problems very obvious.

As well, as I mentioned previously, this accel was overloaded on input based on a typo on a supplied data sheet. I am going to get it repaired and recalibrated or replaced (on their dime!) soon, just to be sure.

Lastly a truly regulated DC power supply like this one. Might improve noise as well.

Any more donors??

thnx

I think the units for noise is an RMS value (sqrt(noise V^2 ~ 'power')), hence the sqrt of (hz)
mg = g x 10^-3
the sensor looks like it was designed for vibration/shock analysis
say the g force range is 1g
if the f of oscillation is say 1hz, noise is 0.3 mg
if the f is 400hz the noise is 0.3/20 mg 0.015 mg
noise http://www.endevco.com/resources/tp_pdf/TP324.pdf

this is a good primer http://www2.usfirst.org/2005comp/Manuals/Acceler1.pdf

non-linearity is 0.3%, which is kinda like the accuracy or repeatability...

how do I donate?

[swamp2]

Originally Posted by southlight

Maybe I'm completely off but I'm not sure about the 35ms shift in D3. What says you that the shift is completed at the point you were drawing the "end of shift" line, swamp? My understanding is that the shift can't be completed as long as the acceleration is higher than it was before the shift and that's some time longer. In case I'm plain wrong here say it in a nice way please.

Attachment 179820


Best regards, south

This is a very good point. Results here have plenty enough signal to noise ratio to be carefully questioned. Particularly in this trace it looks like some significant oscillations (say 4 or so local peaks after the 30 ms) may in fact be part of the clutching operations or some part of the shift dynamics. Quite a few things point to this not being the case though.

• Look at the size and shape of the oscillations near the beginning of this data. Again these are very similar shape, size and frequency compared to the oscillations I am calling post shift. As I explained to lucid, these are occuring during near WOT acceleration, in 3rd gear well before the shift. They must be noise
• Examining other shifts in D3 there is almost always this characteristic down spike followed by the up spike and these truly have clear physical meaning/cause related to the clutches.
• If you examine much longer periods of the raw data you can see a fairly clear drop in the average (average to get rid of all these noise effects) of the acceleration comparing pre and post shift levels. You can see a bit of this in the S4 trace I originially posted. Those pre/post shift acceleration levels are consitent with simulation results as well! Previous work in this domain for MTs could use sampling at only 50 Hz or less to capture a manual shift on the order of 1/4 s. At 50 Hz you would filter the data carefully and almost all of this noise would disappear leaving you a fairly clear indication of pre/during/post shift acceleration levels. Everything is much tougher at 300-500 Hz!
• The magnitude of the first down spike and following upspike are generally not matched anywhere in else the data set except right at the shifts. This is how I picked them out from the raw data.

These are the very reasons I have provided very large +/- uncertainty values on the actual shift times. Accurate and certain values are a way off and will require quite some additional work.

Below is a fully annotated run. I should have posted this long ago. Not entirely relevant to your question but loosely. I think it will help others see the big picture. The previosuly posted shifts are just serious zooms on both axes but more so on the horitontal/time axis.

[ArtPE]

Originally Posted by southlight

Maybe I'm completely off but I'm not sure about the 35ms shift in D3. What says you that the shift is completed at the point you were drawing the "end of shift" line, swamp? My understanding is that the shift can't be completed as long as the acceleration is higher than it was before the shift and that's some time longer. In case I'm plain wrong here say it in a nice way please.


Best regards, south

I'll chime in...
I think he picked those points because you can see the g drop (clutch disengaged), then rise(engaged)

but you may be right...the ripples may be the clutch slipping due to slower engagement or the computer adjusting it and the throttle for smoothness...so the actual shift may be much long...when the large cycles end and they settle down to the pre-shift level

[lucid]

Originally Posted by swamp2

Below is a fully annotated run. I should have posted this long ago. Not entirely relevant to your question but loosely. I think it will help others see the big picture.

Yes, this helps the rest of us. Seeing what you were seeing, I can now understand what you meant. I'm sure we'll get this cleaned up over time with the help of members like Ken, and additional donors like Art.

[swamp2]

Originally Posted by ArtPE

thnx

I think the units for noise is an RMS value (sqrt(noise V^2 ~ 'power')), hence the sqrt of (hz)
mg = g x 10^-3
the sensor looks like it was designed for vibration/shock analysis
say the g force range is 1g
if the f of oscillation is say 1hz, noise is 0.3 mg
if the f is 400hz the noise is 0.3/20 mg 0.015 mg
noise http://www.endevco.com/resources/tp_pdf/TP324.pdf

this is a good primer http://www2.usfirst.org/2005comp/Manuals/Acceler1.pdf

non-linearity is 0.3%, which is kinda like the accuracy or repeatability...

how do I donate?

This sensor is actually quite special in that is has true DC response and very consistent sensitivity up to 500 Hz. I felt this was required for this testing. It was quite tough to locate. Many accels used in vibration and modal testing do not offer ANY DC output.

The error specs do not lend themselves to estimating errors during the DC or nearly DC periods and this is what I would like to have. They do explain why the sharply rising data section have less noise. Peaks have a great deal of high frequency content and if the noise decreases with increasing frequency that likely explains this observation.

I understand the linearity . FYI: I was an ME designing, building and testing precision pressure and force sensors for some time....

Donations can be made to me through paypal. I'll pm you. Thus far we have spent about $600 and I have been given 4 $100 donations. As well, as you probably read, Ken is donating time and materials to make a real mounting system.

Ken: One other point: If there is any way at all to make this a universal mount (i.e. for all or most cars) that would be great. I seriously doubt it is possible but if you examine a bunch of seat rails you may come up with a clever solution. Otherwise it is a new mount for each bloody vehicle.

[southlight]

Fair enough, thanks guys!


Best regards, south

[swamp2]

Originally Posted by lucid

Yes, this helps the rest of us. Seeing what you were seeing, I can now understand what you meant. I'm sure we'll get this cleaned up over time with the help of members like Ken, and additional donors like Art.

Yeah, absolutely should have included this from the get go. 1. I'm (somewhat) lazy. 2. It is never so easy to understand what others may not be seeing. 3. I viewed this initial post as preliminary data. Overall S/N during a~constant is about 3:1, ugh. Perhaps 8:1 at best, given the much smaller noise during high slope portions.

[swamp2]

Originally Posted by southlight

Fair enough, thanks guys!


Best regards, south

Keep up the skepticism! I am not even fully convinced yet.

[swamp2]

ArtPE has generously donated $75 to help fund this fun little project. Thanks again Art, your financial and non financial contribution is appreciated. We will be interested to see the results of your matlab post processing of the data (if that was your plan...).

[Irb Digital]

Plz PM me, I'd also like to donate.

[Wads]

Has anyone tried to compare and contrast the actual acceleration curves between the different modes yet (I'm sure somebody will have done this?!?!)..... I pick up my M3 next Monday and plan to time it in the various modes (once run in) using a Racelogic PerformanceBox (used by many of the top car magazines)...... if anything interesting shows up i'll post up the results.

[swamp2]

Originally Posted by Wads

Has anyone tried to compare and contrast the actual acceleration curves between the different modes yet (I'm sure somebody will have done this?!?!)..... I pick up my M3 next Monday and plan to time it in the various modes (once run in) using a Racelogic PerformanceBox (used by many of the top car magazines)...... if anything interesting shows up i'll post up the results.

You could try this. These preliminary results are not at all controlled enough to compare them directly (not even identical throttle used in all cases). Better yet you should compare velocity vs. time curves. You need to numerically integrate the acceleration signal to get velocity vs. time. This would be basically just like using a Vtech or similar acclerometer based device. As you can see these signals are pretty noisy and that is going to be a problem. It would probably be much better to test for this in a more direct fashion with an arm/wheel tester more commonly used for vehicle performance testing. The effects are fairly small so you probably would have to make a lot of runs and then average the results to see the difference (no matter which method was used). Also, if you had two cars that could otherwise be seen as nearly identical you could just do an direct head to head test only changing the Drivelogic mode.

You should be fairly convinced of this effect just looking at the data I already posted. You can clearly see a longer duration and increased acceleration profile which is consistent with what you feel. You can easily calculate the extra velocity just as I did in my OP and despite the simplicity of that calculation it should be quite accurate (I basically just smoothed and averaged the curve and then numerically integrated it "by eye"). Once in gear everything will be identical. You only get the changes right during the actual shift events. You will get these tiny bursts of extra velocity comparing modes with surge vs. those without (just like it feels!).

We'll certainly be interested in your Racelogic results.

[swamp2]

Member Irb Digital has generously donated $40 to this project. Irb is considering a new M and an automated manual for the first time. A very gracious donation when Irb isn't even an owner (yet ). Cheers Irb.

[Wads]

I'm not expecting any noticable difference in the curves but happy to test in the name of science and autogeekology

[swamp2]

For those of you REALLY following all of the details. The accel was likely overloaded on its input due to a typo on a spec sheet (and my fearless experimentation). I knew something was wrong when it quickly shot up to 175+ deg F . All of the data in the OP was taken after the overload. The frequency resolution seemed unaffected but it did have a zero shift and small change in output which I accounted for in that data. I have sent it off for repair/recalibration or replacement. There will be no additional testing until I get the unit back. It will probably take 1-2 weeks.

[footie]

Sorry to hear it, hope things get sorted for you soon.