Statistical poll of the drive unit issue

[Ben W]

My drive unit just failed last night, at 29,800 miles. Clunked to a stop with strange whirring noises, completely dead, couldn't drive. P85 Sig #P00061. Sure hope they'll replace it with a new unit instead of refurbished! (I'm secretly hoping they find a reason to replace my A-battery with a newer one capable of 120kW supercharging...)

 

[LargeHamCollider]

My drive unit just failed last night, at 29,800 miles. Clunked to a stop with strange whirring noises, completely dead, couldn't drive. P85D Sig #P00061. Sure hope they'll replace it with a new unit instead of refurbished! (I'm secretly hoping they find a reason to replace my A-battery with a newer one capable of 120kW supercharging...)

I assume you mean P85 and not P85D, and that that was a Freudian slip.

 

[ChadS]

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

I have 48k on mine. No failures, no noises, nothing. I can't think of any reason you would doubt that - can you explain?

 

[marcad80]

I have 48k on mine. No failures, no noises, nothing. I can't think of any reason you would doubt that - can you explain?

Based on the data we have collected so far you are just over 3 standard deviation above mean. That places your drive unit at about the 99.8 percentile. Or better said your model S is in the top 2/10 of a percent. For every 1000 Model S 'vs with Drive Units as old as yours 2 would make it to where you are. Not impossible by any means, and not even out of the normal distribution with the sample size we have. That be said, your MS hasn't failed yet and I still adding miles......

so information from you would be very helpful in this study.
has Tesla ever evaluated or test driven the car?

 

[Ben W]

I assume you mean P85 and not P85D, and that that was a Freudian slip.

Indeed (corrected.) A man can dream!

 

[ChadS]

2 out of 1000 will make it that far? I don't follow. The (not-statistically-valid) data above shows 6 out of 84 have already made it that far, with more than half the cars still with no failures but just not enough miles yet. You can't apply mileage of failed units and assume those that haven't failed will at that point. That is pre-supposing that all will fail.

Tesla drives my car when I bring it in for service, four or five times so far.

 

[Ben W]

Based on the data we have collected so far you are just over 3 standard deviation above mean. That places your drive unit at about the 99.8 percentile. Or better said your model S is in the top 2/10 of a percent. For every 1000 Model S 'vs with Drive Units as old as yours 2 would make it to where you are.

Hmm... It's unlikely that the failures follow a normal distribution. Rather, there were probably "bad batches" of drive units that tend to fail around 20-30k miles (as mine did), and "good batches" that can be expected to last 100k-200k miles. There are also numerous cases where the drive unit was proactively replaced, even though the problem was in a different component. So a drive unit still functioning after 45k miles is not really unexpected at all. (Of the seven poll responders with 45-60k miles, just two reported drive unit failures.)

The actual failure distribution is probably more like a "bathtub curve", which applies to many types of components (e.g. hard drives). You'll see a lot of failures early in life due to defects, followed by a long stretch with very few failures, and then increasing failures as even the "good units" wear out. It's unlikely that any Model S's have yet reached their design lifetime in terms of mileage, so we don't know yet how long "good units" might typically last.

It reminds me of an apocryphal story I heard about Henry Ford... Many years after the introduction of the Model T, Ford went on a tour of the scrapyard to see for himself how the cars tended to wear out. Some had broken axles, others broken transmissions, others rusted frames... But none of them had broken kingpins. Every kingpin was in excellent shape. So Ford went back to his factory, and sent out the following memo:

"Use cheaper kingpins!"

 

[cwerdna]

I have 48k on mine. No failures, no noises, nothing. I can't think of any reason you would doubt that - can you explain?

I have plenty. There are many cases here (in addition to automotive press like Edmunds and Motor Trend) who have had their drive units replaced. And, there are numerous ones who have have gone thru several drive unit replacements, including Edmunds.

It seems like the odds of a Model S w/a drive unit that hasn't been replaced for any reason (failure (stops working completely) or noise) by say the 30K mile mark is rare as it doesn't seem like they last much beyond 10K to 15K miles before developing noise (again). The repeated replacements, including recent ones, not just early ones suggests that the problem isn't under control.

I believe I've posted here what I essentially posted at Toyota Rav4 EV Forum , can you hear it in yours? re: my observations of cars and brands w/known good reliability vs. ones w/horrible reliability.

Fast forward about 1.5 years it seems like a TMC member/Tesla defender who posted at Toyota Rav4 EV Forum , can you hear it in yours? and Toyota Rav4 EV Forum , can you hear it in yours? has changed his tune and even attacked me several times (Toyota Rav4 EV Forum View topic - Check EV System warning message and Toyota Rav4 EV Forum , can you hear it in yours?).

I suspect that many people here, esp. those w/multiple DU replacements and other maladies on their S would totally disagree w/the assertions made at Toyota Rav4 EV Forum , can you hear it in yours?.

I really wonder what sort of long-term durability testing Tesla does. Rather than repeat my posting, I direct you to the last few paragraphs of Drive Unit failure symptoms and thresholds for replacement - Page 5. Or, is it a manufacturing (possibly Tesla's or that of their supplier's) consistently problem? In other words: Tesla designed a drive unit that will work perfectly w/miniscule chance of failure or developing excessive noise over say 200K miles yet there are issues w/actually manufacturing such a unit consistently.

 

[Ben W]

It seems like the odds of a Model S w/a drive unit that hasn't been replaced for any reason (failure (stops working completely) or noise) by say the 30K mile mark is rare as it doesn't seem like they last much beyond 10K to 15K miles before developing noise (again). The repeated replacements, including recent ones, not just early ones suggests that the problem isn't under control.

True, but there seems to be quite a bit of variability in the quality of the drive units being made. Clearly the drive unit design was intended to last the life of the car (~150k miles), so it's most likely a systematic but intermittent flaw in the manufacturing or assembling process. Some drive units are turning out better than others. It's not surprising that some of the "good ones" are lasting 45k miles and beyond.

My own drive unit just failed a bit shy of 30k miles, without any unusual noise preceding it. For a couple weeks prior to failure, I noticed slightly more clunkiness when transitioning between regen and accel, but wasn't sure if it was just my imagination. I'll find out from the SC in a day or two what the official diagnosis is.

 

[qwk]

I have plenty. There are many cases here (in addition to automotive press like Edmunds and Motor Trend) who have had their drive units replaced. And, there are numerous ones who have have gone thru several drive unit replacements, including Edmunds.

It seems like the odds of a Model S w/a drive unit that hasn't been replaced for any reason (failure (stops working completely) or noise) by say the 30K mile mark is rare as it doesn't seem like they last much beyond 10K to 15K miles before developing noise (again). The repeated replacements, including recent ones, not just early ones suggests that the problem isn't under control.

I believe I've posted here what I essentially posted at Toyota Rav4 EV Forum , can you hear it in yours? re: my observations of cars and brands w/known good reliability vs. ones w/horrible reliability.

Fast forward about 1.5 years it seems like a TMC member/Tesla defender who posted at Toyota Rav4 EV Forum , can you hear it in yours? and Toyota Rav4 EV Forum , can you hear it in yours? has changed his tune and even attacked me several times (Toyota Rav4 EV Forum View topic - Check EV System warning message and Toyota Rav4 EV Forum , can you hear it in yours?).

I suspect that many people here, esp. those w/multiple DU replacements and other maladies on their S would totally disagree w/the assertions made at Toyota Rav4 EV Forum , can you hear it in yours?.

I really wonder what sort of long-term durability testing Tesla does. Rather than repeat my posting, I direct you to the last few paragraphs of Drive Unit failure symptoms and thresholds for replacement - Page 5. Or, is it a manufacturing (possibly Tesla's or that of their supplier's) consistently problem? In other words: Tesla designed a drive unit that will work perfectly w/miniscule chance of failure or developing excessive noise over say 200K miles yet there are issues w/actually manufacturing such a unit consistently.

I have said this before, and I'll say it again. You don't even own a Model S, yet you come to all of these conclusions? Noise isn't a failure, just a nuisance. One sees a lot more being in the game vs. sitting on the bleachers.....

 

[ecarfan]

There are many cases here (in addition to automotive press like Edmunds and Motor Trend) who have had their drive units replaced. And, there are numerous ones who have have gone thru several drive unit replacements, including Edmunds.
It seems like the odds of a Model S w/a drive unit that hasn't been replaced for any reason (failure (stops working completely) or noise) by say the 30K mile mark is rare as it doesn't seem like they last much beyond 10K to 15K miles before developing noise...

Your assertion is absurd.
You do not have access to any statistically valid data on which to base such a statement. You are extrapolating -- wildly -- from posts in online forums by owners complaining about noise they think may be coming from the drivetrain in their car. Are they hearing a noise? They certainly are. Do they have an accurate sense of what that noise means and it's true source? No they do not. They are owners, not Tesla service personnel.
Online forum posts about vehicle issues cannot be used to accurately determine what percentage of owners are having a problem with a specific part in a vehicle. Owners not having problems are not posting to say that "today I didn't hear a noise in my drivetrain and I'll post again tomorrow if I don't hear a noise".
Your reasoning is specious and your conclusions are unsupportable.
If it was indeed "rare" for a Model S to go for 30K miles without having a serious drivetrain issue, or even a noise that required a service visit, the company would long ago have collapsed under the weight of overwhelmingly negative news reports and falling sales. In case you haven't noticed, Model S cars are now selling at the rate of well over 50,000 per year and climbing.

 

[AmpedRealtor]

I really wonder what sort of long-term durability testing Tesla does.

Tesla's long-term durability testing is handled by its customers, clearly. If the drive units were properly stress tested prior to launch of the vehicle, we wouldn't be having all of these issues. It's quite obvious that Tesla skips steps, pushing the ramifications downstream to the owners. Do we honestly believe that Tesla's focus on world-class service wasn't motivated by the overwhelming likelihood that our cars would be in service a lot?

At the time I ordered my car in 2013, every service center in California was overflowing and backed up. I was concerned by that, thinking that was a good indicator of the car's initial quality. Folks here and elsewhere told me no, it's because a lot of cars were coming due for annual service, etc., which was not really true. Many owners don't even opt for the annual service. Clearly there were many, many problems with the cars that caused the service centers to be inundated.

Folks, this is how Tesla beta tests. And it gets people to pay $100k for the privilege.

 

[supratachophobia]

While not wholly indicative due to inflated numbers during the winter months, what is the average watts per mile for everyone? Based on severity of winter temps, I'm sure we could come up with an adjustment to make a national comparison (think of it like a golf handicap for model s). You'd have to get your original number from a 12 month period though.

Then we could corelate watts per mile to drive unit failure to find a pattern if it's there.

- - - Updated - - -

Folks, this is how Tesla beta tests. And it gets people to pay $100k for the privilege.

Yes, currently the beta test has around 80k subjects. However, I assumed a lot of this going into the purchase. I never assumed I'd be visiting the service center so often mind you, but I did think I'd be part of the experiment in one way or another.

 

[Ben W]

While not wholly indicative due to inflated numbers during the winter months, what is the average watts per mile for everyone?

P85, Los Angeles, 353Wh/mi, drive unit failure at 30k miles.

There are many confounding variables; probably the number and frequency of 0-60 runs is the best correlation, but the car doesn't keep statistics on those. At least, not user-accessible

 

[supratachophobia]

P85, Los Angeles, 353Wh/mi, drive unit failure at 30k miles.

There are many confounding variables; probably the number and frequency of 0-60 runs is the best correlation, but the car doesn't keep statistics on those. At least, not user-accessible

Moving forward collecting this data, are we assuming the whining/clunking is akin to drive unit failure or are we seperating them out? Better yet, those that had actual failures, did you detect the whining and/or clunk how far prior to actual failure?

And I'm assuming the LA region is the zero point for weather adjustment. The yearly median temp is like 74 degrees.

 

[marcad80]

We have collected about twice as many samples from the last data analysis. The control (no failure) is now statistically significant, but the test group (failed) is still a little low.

But the data now shows a Mean of Not Failed = 21100 miles and Failed = 24300 miles. These are close to the previously calculated means with half the sample size. Good sign for stabilizing data.

The Standard Deviation has gone up a bit through, from 8,600 miles to 12500 miles. Meaning that some higher mileage failures are more common than previously calculated in the lower sample size.

IE. MS fails at: 24300miles you are in the top 50%, 36800 miles top 32%, 49300 miles top 5%, 61800 miles top 0.3%

The data has not changed significantly enough to change my original possible effects though. The Non failed to failed mileage is still way to skewed to be coincidence, and the failed data still appears to be close to a normal distribution.

 

[ChadS]

I appreciate the effort at using data to answer a tough question, but you are still assuming your conclusion and reading way too much in to the data. While your interpretation may happen to be correct, there are many others that could be as well.

Why is the non-failed group "skewed"? The oldest cars are only 3 years old and they are shipping them much faster than they used to; we should expect more cars with fewer miles so that could be reasonably random (though, like with any internet poll, we can't tell if it's for that reason or a sampling error).

And while I do see the rough appearance of a bell curve on the failed units, similarly there are many possible explanations other than failures center around 15-30k miles. Like maybe more miles always means greater likelihood of failure, but there are fewer cars with more miles than 15-30k. Or that the failures center around a certain manufacturing error which didn't exist at first and was later corrected. Or a sampling error again - you are only looking at 25 failures out of 90k or so cars shipped, and you have no way of knowing whether they are representative or what the real failure rate is.

 

[marcad80]

ChadS,

the reason the non failed group is skewed is the way the questions are set-up. The voter is in data set number one (non failed) unit you have a failure, then you move to data set number two and record that mileage there. If all of these Drive units / drivers / environment etc. were the same and the build rate constant, every unit would fail at 24300 miles and the data point in data set two would be 27 units at 24300miles with an average of 24300miles. Consequently the other data set, data set one (non failed units) would be a linear line of units (because of build rate) from 0-24300 miles and the average would be 12150 miles. Once you hit 24300 miles your unit fails and you move to data set two and stay there.

Now variances are all over this data in the miles you drive per day, sensitivity to the failure, etc, but in an increasing sample size the one thing that should null this noise is the averages of the data......

The average of data set number one should be approx. half of the average of data set two....... unless there is a difference in the propensity to fail between data set one and two. The average are no where near half, they are almost equal. BTW. If they are making more cars now than in the past (ramping up) this would skew the data to a lower average miles in dataset one, not higher (exactly the opposite of whats happening). In this case we would expect to see a lower than 1/2 average in data set one, and explain it by a rampup.

The reason I said that I believe this drives us to possible effect #2 and / or #3 is because the data average being lower and not higher indicates a positive change with respect to vehicles with lower miles (newer cars). The differences in averages can even give a VERY rough estimate of when the less prone to failure mode changed.

You are correct about sampling error, I am assuming that if we get a big enough samples size it will represent the total population.... ?

 

[jerry33]

I have 48k on mine. No failures, no noises, nothing. I can't think of any reason you would doubt that - can you explain?

Mine is coming up to 50K with no failures and no noises.

- - - Updated - - -

You are correct about sampling error, I am assuming that if we get a big enough samples size it will represent the total population.... ?

Yes and no. The sampling only comes from those who frequent TMC. That alone skews the data when compared to the general population. It would be representative for TMC owners.

 

[qwk]

Noise doesnt mean failure. There are many noises that Tesla drive units make, and some are purely cosmetic. How in the world can you differentiate between all of these variables? You can't.