Search results

Interestingly, the only 2020 aircraft to be inspected had 3300 hours, all commercial, and the right front bolt hole failed.

Let's look at what the data tell us. Here's failure rate with CIs by year (in 5-year buckets): Note: this does not account for the fact that some year buckets may tend to have more hours: Picking out a trend in the ECI failure rates would basically be speculation given the large error bars.

Interestingly, Piper originally wanted a 5000h TIS requirement in the old SB, and the FAA created FSH for the AD. Piper has since proposed 2:1; who knows what the FAA will propose next. I definitely plan to create an extensive write-up and send it to the FAA/Piper. If anyone is interested I...

I've made substantial changes to the original post based on your (and others') feedback. Let me know if this helps clarify things!

2002 It doesn't. Unfortunately most PA28/32s do not record cycles. FYI I have updated the original post above for brevity and to include the most important results from follow-on analysis.

Given people's questions about what is a good indicator of ECI failure, I performed a new analysis to determine how commercial and non-commercial hours compare. To do this, I did the same MTBF analysis as above, but also optimize over a commercial hour weight (i.e., how many non-commercial...

This is absolutely correct. To compute statistical power (i.e., how confident we can be in the negative result) we'd need to hypothesize some other distribution. Things get very tricky very quickly.

Great points here. Unfortunately the available data can only get us so far. The random (memoryless) failure distribution is consistent with failures resulting from incidents/occurences like these. Perhaps a post-wing-damage or hard-landing AD (similar to AD 2004-10-14 for prop strikes) makes...

Since statistical tests are a random process, it's difficult to say for sure whether there is "some evidence". All we can say is that, if there were no difference, we'd randomly see evidence at least this compelling 1 in 3 times. Such statements are patently disatisfying, but such is life with...

Failure rates by model. Tough to make much out of this. Here's the data aggregated by family: Keep in mind this doesn't correct for the fact that different models may have different average FSH.

It's important not to read too much into noise in the plot, keeping in mind that those spikes are just a few ECI failures. At the risk of polluting this analysis with even more stats, we can actually determine whether two given distributions are significantly different using another statistical...

Piper and the FAA unfortunately reduced the quality of the data collection by limiting inspections to 5000+ FSH. The failure rate below this is consistent with this memoryless process, but there are too few samples to say anything conclusively. Expanding the AD to all airframes would...

By "random" here I mean something very specific. The data suggest that ECI failure is a memoryless random process. You can think of this like every flight hour is a roll of the dice, and there's some probability that will cause a failure (e.g., there's some probability of a hard landing causing...

These plots are CDFs. Commonly used in statistics, but unfortunately difficult to interpret if you're not used to looking at them. You can normalize these so all the series add to one, but then the failure line would be basically impossible to see (hence why they don't). You can think of these...

I completely agree (see my note in Conclusion 4), with the exception that generally the goal should not be “no accidents”, but rather a number where costs are balanced with VSL. If the goal were no accidents at all costs the solution would be to scrap every PA28 today! Piper no doubt has good...

My claim is somewhat more nuanced than this. If failure is a random process, rather than fatgique, then a life limit is not the correct way to prevent failures since failures can also occur at low times. Note: throughout here I'm considering "failure" to be the point at which a spar is damaged...

How about this. Here's a KDE plot (basically a histogram but continuous) of the same data. You can see the failed spars do skew higher time, but the blue and red skew by the same amount.

Thanks for the feedback, Jim. This data is all hot-off-the-presses so I'm definitely looking for thoughts on how best to present the results. The line plots are cumulative distribution plots: https://analyse-it.com/docs/user-guide/distribution/continuous/cdf-plot You can think of them as...

Note: I am updating this post incorporating feedback from below. In 2021, the FAA released AD 2020-26-16, requiring inspection of most PA28/32 wing spars. This AD also required reporting results of the inspections to the FAA and Piper. This data has not been made publicly available, until now...

I don't see how someone who truly understands how brittle GPS is would tolerate a panel without backup nav. Personally, I want a monitored ground-nav approach at either my destination or alternate. Civilian GPS can be (with difficulty) spoofed, and trivially blocked. Further, the "redundancy"...