I have replaced I think 7 or 8 popped diffs. Now then, to add fuel to the fire, I just inspected a diff in a tr just last week which has over 150k km on it. It is perfectly fine. I pulled it out to inspect just to be sure. I do have a feeling however it was indeed replaced by Ferrari in the past. That was indicated by more silicone then normal around the axel flange bolts and side cover housings. I have had the engine out of this car for timing belts 3 times since the current owner purchased it. This time I went the extra mile to pull the diff and have a look. I will say I'm surprised. I may or may not install an upgraded unit, that is now in the owners hands.
Guy's, I've been out of town for a while so I had to put this analysis on hold. I have also put a little more thought into the Reliability Model especially the estimation on current mileage for all TR owners. I'm using the same sample group assumptions as earlier i.e. approximately 7000 active owners with approximately 150 confirmed diff failures. For the sake of clarity, I'll be posting the results in separate posts. First, I wanted to determine a more accurate evaluation on current TR mileage out there. Luckily for us, James (Veedub00) started a survey back August 2015 where we had 79 respondents. http://www.ferrarichat.com/forum/boxers-tr-m/498445-how-many-miles-your-testarossa.html Although the survey was split into varying mileage increments, I was able to extract that data and perform the following: -Estimate its distribution -Generate some random numbers (fill) for each mileage increment Results below are the Histograms of the Original survey (Histogram of Survey 8-31-2015 (MEAN) RAW) and the Random Fill data points along with its Histogram (Histogram of Survey with Random fill) Note that the data does indeed fit a Weibull Distribution with high significance. Image Unavailable, Please Login Image Unavailable, Please Login Image Unavailable, Please Login Image Unavailable, Please Login Image Unavailable, Please Login
Next is to gather 7000 data points using Monte Caro Simulation based on the same Distribution as the original survey. Here are the Histograms and the Probability plots Again, we have a high level of confidence that the Simulated data fits the expected Weibull Distribution. Our bogey is P>0.05 or no more than 5% risk of be of being wrong. Image Unavailable, Please Login Image Unavailable, Please Login
Just to be absolutely certain that we are not introducing further error, we compare the Probability Plots and perform what are known as F and t Tests. What these tests confirm is that the data sets have the same Variance and same MEAN. In other words, the data sets (Survey data and Monte Carlo Simulated data) are statistically from the same population . Note the P values. This confirms they are the same with very little risk of being wrong. Image Unavailable, Please Login Image Unavailable, Please Login
This is where things get tricky. Again based on feedback from 2/3 aftermarket suppliers, we can estimate a total of 300 sold with ~50% used as preventative maintenance. To be safe, I assumed a Normal Distribution of mileage at diff failure between 20,000 and 60,000 miles. I also reran the Monte Carlo Simulation for cleaner results. For perspective, I have also overlaid the Histograms of all TR's vs. Failed diffs Image Unavailable, Please Login Image Unavailable, Please Login Image Unavailable, Please Login
Now we're ready to start the Reliability Analysis. I have chosen to use a Parametric Analysis with a Lognormal distribution since this had the closest fit. I began by overlaying all the data (with and without failure) and censored the vehicles that have not failed to date. Here are the results. Let's start with Survivability. This plot basically determines how many diffs will survive by mileage. As you can see, we have just under 100% at ~40,000 miles (97.4% survival rate) but the chart starts to rapidly drop. If you're lucky enough to make it to 100,000 miles, you'd be in a group of only 40% that would have also survived. By 250,000 miles, less than 1% of the diffs would have survived. Note that the bands on either side are the 95% confidence intervals. Image Unavailable, Please Login
Similar to the Survival Plot, this chart displays how many diffs would have failed by mileage. There's a point of inflection (change of slope) at approximately 100,000 miles indicating that the rate of failures starts to drop off. This is purely based on the data we have. Image Unavailable, Please Login
Finally, we have the Hazard Plot. The only take away from this plot is to say that if we were able to make it all the way to 180,000 miles, your chances of failure actually start to drop. Now that's something to look forward to. Image Unavailable, Please Login
So in summary, according to this analysis, the differential failure is real and still poses a great risk to owners out there. This of course is all based on our closest estimates of current mileage driven and the number of known failures and does not take into account driving style, environmental conditions, oil levels etc etc etc. There is always a level of risk when using statistical methods to evaluate risk but these tools have been proven to be reliable and are used in all industries, not just automotive. As usual, if you have any questions or would like to add to the discussion, please do. We're all here to learn.
sam to do this statistic it took you some time I think. so you have nothing better to do than making this? not that I don´t honor your doing, but for what it should be for use? if someone´s diff will fail because of not changing before he then will use your statistic and says that it could not be a broken diff, even he has the broken part in his hand. that the diff failure is real it does not need such a great work you have done. but thanks anyway, hope even the last one who doubt about a diff failure now will be convinced other?
Wow, very impressive Sam. Thank you! One quick question, it seems that 1 in 50 cars has been affected, and statistically this number will worsen as our cars accumulate miles, but was it right to account for 7000 cars in total (the TR production number-ish) or should the production of 512 (up to 94) and all BBs be included? That would add another 4000 cars, up to 11000. Unless the 512, more powerful, have a higher rate than TRs, who had more than BBs? I'm just wondering if 7000 is the right number. Fascinating!
So you believe metal gets stronger over time. Please tell me you don't do analysis for the airline industry. Know about caos theory.
metal gets not stronger over time, but less tsress BMW for the formula 1 race engines for example used engineblocks from the junk yards for the turbo powered race cars when piquet get world champion. this all because the metal was old, often gets warm and cold again, so no stress anymore. so I think if a diff will last more than 100k miles it will never get destroyed when there is still the same driver. but may be with another driver and so other handling it could get destroyed
I must admit, I put some good time thinking about how to eliminate as much error as possible and how to set the model up appropriately but the analysis itself takes a few hours of trials, checks and validation. I have plenty to do, trust me, just see my Brake Caliber rebuild thread . As for the usefulness of this study, I say why not?? It's been a long debate and a lot of anacdotal evidence on whether the diff failure is significant or not. At least this study tilts the scales towards risk based on mileage only. Additionally, if anyone was to tell me that the failure rate is miniscule or not worth thinking about, I can refer them back to this study and argue with more than just here say. If we keep the failed population (150) constant and add more data points for non failed vehicles (11,000 censored), then the curve gets stretched out to ~300,000 miles. Below is a summary file of the results but I did my due diligence confirming the added 4,000 vehicles fit the distribution. You'll also notice that the risk is only slightly reduced on the low end. No I don't. It's based on the number of vehicles that have gone the distance so far without failure and the probability of those vehicles failing at a higher mileage. Since we don't have any confirmed failures after 60,000 miles, then whats to say they must fail at any arbitrary mileage? The analysis just tells us that the likely hood of failure does indeed peek since we have evidence of high mileage cars that have survived so far. But you bring up a good question and it goes back to the other variables we have already mentioned. This phenomena may very well be more related to other variables than just mileage. Consider infant mortality for instance. We just don't know yet. Correct. Fatigue analysis is valid when stresses are concerned. So long as the loads cycle within design limits, then the part should technically last forever. This is how springs are designed (10 million cycles = infinite life) Image Unavailable, Please Login
Joe, I'm pretty sure that is urban legend like pissing on the blocks. Regardless, as I have said the question of replacing the diff is entirely upto the individual. Statistics, legend, folklore and voodoo mean nothing if the failure effects you personally. For me I just replaced mine and it looked perfect, however it looked like someone was there before and my car has over 55k on it.
Right on https://en.wikipedia.org/wiki/Fatigue_limit These diffs are Steel but one correction I must make is that I should not assume Ferrari (or ZF) ever determined the fatigue life. It's a shame because these harmonic tests are usually accelerated (less than a week) and easy to set up. Image Unavailable, Please Login
lenz, the man who did for my competition the electronic was involved in the BMW F 1 project and he told me so. also a former mechanic from BMW F1 told me. this mechanic still has someparts form this area, also a complete engine and this block looks like the same as the F1 blocks. he lives in munic and I will ask him to send me some photos from this block. I have seen it about 10 years before but cannot remember right anymore. when I have those photos I will put it in here. lenz seems now located in the US ( Lenz Motorentechnik USA manufacturers of Formula 1 based engine management systems - BMW ) because he had trouble here in germany. you may contact him and ask himself and say a nice hello form me
This thread cracks me up. Never has one topic been so overly debated. What cracks me up more is all the people who are in denial. Ostrich's. When Newman has posted REPEATEDLY that he has seen cars with no symptoms only to find a completely separated diff upon disassembly, what do you think you're going to find? Hmmm? It takes me about 1.5 hours to tear it down to the diff. It's worth the inspection folks. Don't quote me, I won't reply. I hate this topic due to all the supidity/denial. If you don't believe me, keep it to yourself. I don't care.