Statistical poll of the drive unit issue

[marcad80]

As a new Model S owner I have been reading a lot about the drive unit issue. I have seen a couple of polls, but nothing I can really get any statistically sound analysis on (none of them have a base, or control to compare). I am a Mechanical Engineer (used to design cars) so I am interested in this data and will share the final anaylsis.

 

[breser]

This doesn't account for different parts between different cars (P85 vs 85). Doesn't account for the fact that Tesla was replacing drive units when there was no failure because it was easier to solve a problem by swapping parts. It also doesn't account for AWD vehicles that have two drive units. I'm not sure you're really going to get very useful data out of this.

For what it's worth my 85D had the rear drive unit replaced before it was even delivered to me due to a torn seal that was detected in pre-delivery testing. Which was presumably an assembly issue. No idea how you want to account for that either.

 

[omarsultan]

You generally don't use "internet poll" and "statistically sound analysis" in the same sentence for a number of reasons.

A more accurate question might be related to drive unit replacement, since, it seems there are very few actual drive unit failures (car is dead in the water), most of the replacements were done to handle noise issues that, per Elon's comments on an earnings call, mostly not related to the drive unit itself.

O

 

[marcad80]

All true, alot of data questions still about design changes through time, distribution of people that will read a drive unit failure poll, people that will shift results because they over complement or slander the car. Its difficult to get really sound data in a single question poll for a multidimensional dataset. But I have run a few polls like this in the past on the Volt forums with some success. Alot depends on how big the sample size is. Ill run some statistical significance testing and may be able to come up some level of confidence in the results.

 

[pgiralt]

I think for this to be accurate you need to factor in when the car was manufacturered as well. You can't compare a 2012 drive unit with 15k miles to a 2015 with the same mileage. There have been lots of changes in that time.

 

[supratachophobia]

We need a few more variables. The first of which is a timeline of failure. I've been trying to figure out if the whining sound and clunking sounds are surefire precursors to DU failure. Because there are replacements being done on those grounds. If we can say that, yes, those two symptoms will reliably lead to failure, then those people who had a replacement preventativly can be included in such a pole.

We also need P85/85 information as well like someone else suggested.

Additional information should be collected about when whining/clunking started for those that had actual failures. With enough data, we might be able to predict within a few hundred/thousand miles of a possible DU failure.

Furthermore, if what I've heard is true, whining/clunking will lead to DU failure so I don't believe it's a matter of if and I don't believe the issue is purely cosmetic. I've been carefully tracking mine and the whining sound just keeps getting louder and broader which means that something somewhere is in a constant state of degradation.

 

[supratachophobia]

I think for this to be accurate you need to factor in when the car was manufacturered as well. You can't compare a 2012 drive unit with 15k miles to a 2015 with the same mileage. There have been lots of changes in that time.

We should include month and year of manufacture, but not filter it out by default. If there really were obvious differences, we would already see that flushed out in the posts on here. But as it stands, people from every single year are getting replacements. I think the only place this info could illuminate is if we compare it to the whining/clunking to actual failure. In other words, for example, if there were more actual failures early on that caused the cars to be undrivable, and now, these warning symptoms appear instead first before an actual failure because something was improved.

 

[marcad80]

These statement are true, unfortunately the way these polls are set I can only isolate two variables at a time. In this case 1) failure or no failure, and 2) miles. In this pole many variables are not isolated like; version of drive unit, type of driving, version of MS, regen on standard or low, etc. With a 10 degree of freedom problem, only given a two degree tool, the only way to get to the main effect (or interaction) is to isolate two , hypothesis that the uncontrolled variables are random effect, then run a significance test to see if you captured the main effects. If it fails then isolate two more controlled with the previous test. IE.
Poll 2:

for those who had failures between xxxxx, I own a P85, and drive in standard regen
for those who had failures between xxxxx, I own a S85, and drive in standard regen
for those who had no failure and have between xxxxxx, I own a P85 and drive in low regen
......

if we had a chance to set up a 10 DOF DOE this would be alot easier but, ill take what I can get

 

[johnnyS]

About a month ago, I had a drive unit failure. My service advisor commented that the new units are much better than the old ones.

 

[omarsultan]

These statement are true, unfortunately the way these polls are set I can only isolate two variables at a time.

I appreciate what you are trying to do, but this platform is not really set up to do proper polling. Its does not provide useful insights, but its does provide a great amount of fodder for shorts and media hit pieces--the last time someone did one of these being a perfect example.

If you really want to do something better, use a platform like survey monkey and distribute the results via PM, but even that is not going to get around getting a representative and statistically valid population.

 

[AmpedRealtor]

About a month ago, I had a drive unit failure. My service advisor commented that the new units are much better than the old ones.

Yet everyone gets an older, remanufactured unit.

 

[TomServo]

About a month ago, I had a drive unit failure. My service advisor commented that the new units are much better than the old ones.

How would one determine a "new" unit from an "older" unit". Is there a VIN break point? For a potential pre-owned buyer this would be good info.

 

[cwerdna]

Besides the many questions/issues raised, what counts as "failure"? I'd imagine that many people getting theirs replaced for noise aren't counting those as "failures" but some are...

Judging by what I've seen here, I have serious doubts that anyone w/a Model S who answered "No unit failure, 75-90k miles" actually is on their on their original drive unit. I suspect they've had at least 1, maybe 2 swaps out for noise.

I also am somewhat doubtful that the person who answered "No unit failure, 45-60k miles" is on their original DU.

 

[marcad80]

Not sure if these owners post on this forum, but I would sure like to here more information on the person that made it 75-90 without a failure, or even the people that made it 45-60 without, and 30-45 with failure. Not that there is anywhere near enough data yet, but these are the ones that are high in the standard deviations (if not outliers) in the data that are going to help us isolate the next round of variables....

 

[ZBB]

Not sure if these owners post on this forum, but I would sure like to here more information on the person that made it 75-90 without a failure, or even the people that made it 45-60 without, and 30-45 with failure. Not that there is anywhere near enough data yet, but these are the ones that are high in the standard deviations (if not outliers) in the data that are going to help us isolate the next round of variables....

I think there are too many variables for that to be valuable… For example, the person with 75-90k miles likely has an older Model S. Current models have updated designs and Tesla likely has changed their process on building and/or installing them -- Tesla has iterated and improved. Likely not comparable to current production or cars that have had drive units replaced at different points in time.

I'm not going to respond to the poll, but I did have a drive unit replaced at ~12.5k miles and ~9 months after delivery. The issue was detected at around 10k miles almost 2 months earlier -- it was the whine issue that seemed fairly common in the early cars (my VIN is in the low 11,000s, May 2013 production) -- and I only heard it after the service center manager pointed it out to me on a test drive to validate an alignment issue. He advised to wait since he'd already replaced some drive units twice without the issue going away, but a "3rd gen" drive unit had just gone into production that was supposed to resolve the issue. So I had it installed during the next scheduled service (I'm using the mileage schedule…).

That was almost 15 months ago and we've had no drive unit-related issues during the >25k miles since. The newer DU appears to be more efficient -- I averaged 297 Wh/mi on the first one, and have averaged 286 on the replacement...

 

[qwk]

Like stated above, the poll is useless, and will only cause confusion. Having said that, my car is an early build, has 37k miles on the original DU. It did develop a slight drone early on at higher speeds, but it doesn't bother me. In my experience, a silent DU doesn't stay silent for long. I would bet that the majority of DU replacements are for noise issues.

 

[jerry33]

At 49.5k miles I am still waiting for the noise to appear.

 

[marcad80]

I know this logic of this isn't just popping out, so ill try to keep explaining as questions come up.

as far as the "new" drive unit fix. This is a variable I will isolate in round two in the under 15k and 15-30k groups

ie
I have not had a failure, under 15, drive unit built xxxxxx (or vehicle build)
I have had a failure under 15, drive unit built .....
I have not had a ....... 15-30, drive unit ......
I know the logic trail is hard to follow, but the data will start asking its own next questions.


btw, its to bad you are not voting, your data would be key in the control group for proving the new fix worked. By not voting you are making the old unit look better....

also I will ask if you did not vote in round one, do not vote in later rounds. All rounds must be from the same data set, or I will loose a ton of analysis ability.



that means subtracting data from later rounds out of round one to isolate entirely new variables with out running another poll. After I'm able to pull new drive unit data out of round one I will be able to tell the rest of us on the old unit an average mileage when they are failing..... Etc.

 

[marcad80]

I’m sure some people will have a lot of interesting thingsto say about this preliminary analysis, so I have included some logic as well. Of course I would like to have a lot moredata, but…..

(calculated by an average of averages i.e. 15-30K = 22500miles)
The average mileage of MS that have not had a “failure” yet à20526 miles
The average mileage of MS at failure à 21562 miles

This is a key datapoint. On average (all other variables being the same) you would expect theaverage mileage of people who haven’t had a failure to be exactly half of the averagefailure miles. i.e. 10263 miles

And, since Tesla has been ramping up there production, theoverall average of mileage of cars without failures should be even lower thanthe expected 10263 average….. But thisis not reflected in the data set. I evenremoved the possible outlyers of (1) 75-90, and (3) 45-60, and the data isstill skewed…

Possible effects.

1) The people that have recorded previous failures aremore sensitive than the more recent buyers… (not really likely)
2) Tesla is now less sensitive to replacing the failure innewer cars than older cars (maybe)
3) There is a design or manufacturing cause that has been changed(maybe)
4) Data is skewed, people with newer cars are not voting,etc… (maybe)
5) People that have yet to have a failure will eventuallyhave them at very high miles (not likely, the limited data on failures isforming a normal distribution centered on 21.5K with a std. dev. of 8.6K)
6) Model, driving habits, climate, (not likely, would haveexpected to see a non-normal distribution in the failed miles as these shouldhave linear dist. across person to person, place to place, model to model. I can explain this one more if necessary)
7) Driving regen mode (Stand vs Low) or other single pointbinary failure cause (maybe but…., based on the tight dist of data in thefailure miles, I would believe Tesla would have easily identified these longago)

Of course not enough data yet, but based on what we have, inorder to effect the distribution shift like this with the number of MSreporting, the “reduced likelihood of failure” MS’s would have to be into the 15K and overdataset. If the average mileage per yearis 15K miles, that would put a shift in the likelihood of failure at around 1.5years ago. Unfortunately if the designchange (#2) was recent (6months or so) this would have not created thiseffect. Whatever created this effect hadto be in the 15K miles and over population. I really hope the change in dist is because of a design change a while ago and not sensitivityby Tesla to the issue but… to answer this would take another poll.

Hopefully we’ll get more data to so I can test somesignificance.

 

[AmpedRealtor]

Not sure if this helps, but here is the serial number of my last DU installed in February. Service confirmed that it was BRAND NEW from the factory, not remanufactured:

ASY,P-TRAIN,SPORT,MECH,COMPLETE
(1002633-01-M)

I'd be curious to know if anything can be gleaned from the above?