Equipment Failure Rates - real data?

[gr8jab]

Hi all,

Everyone here has opinions on pony bottles and spare-air and doubles and sidemount. I'm not asking to rehash any of that stuff. I've read them all and am trying to form my own opinions. As an engineer, I try to stay away from 'feelings' or 'assumptions' or 'guesses' and go with facts (data).

For this analysis, I'd like to ignore human error (such as poor planning or bad decisions).

I'm wondering if there is a source of data concerning equipment failures. I'm specifically interested in mechanical failure that would cause a loss in breathable air. I can't name all the causes, but the obvious ones are things like o-ring failures, hose ruptures, primary or secondary regulator failures, and free-flow. Is there a source of data, such as Mean Time Between Failure (MTBF), or maybe a historical incident rates? I know some here have analyzed scuba accident data (fatality info). Maybe there is other interesting data captured and stored somewhere.

I'm avoiding anecdotal stories or single-event data points, which don't tell me the probabilities of equipment failure. "This one time, at band camp..." may be entertaining but is also irrelevant.

Thanks!

 

[tbone1004]

data is not there because there is no central collection body. Even if there was, it would be nearly impossible to extrapolate because use cases are very different.
Does the 4 hour dive that I do with a bailout bottle count as a "dive" even though it only has maybe 4 breaths total per dive because it's a bailout bottle? If I use a standard DIR configuration set of doubles where the inflator hose comes off of the right post, how do you count use for the left post? Technically it's pressurized and in the water, but there is no cycling of that regulator during the dive with the exception of the few breaths you take to check functionality.

 

[johndiver999]

Also, many of the failures could ultimately be considered to be human error such as failure to inspect, failure to follow service guidelines, incorrect service, wrong parts used etc.

Anecdotal evidence and observation of failures is not irrelevant. It may not have the statistical significance you are seeking, but years of observations of thousands of divers, (as might occur on a live aboard operation or daily charter) should yield some useful information.

 

[gr8jab]

Too bad there is no source of field data. Even if there was, the set would likely be small, more susceptible to things Tbone points out. I analyze field data for a living and you would be surprised what useful information can be gleaned from it. If the data set is large enough, ambiguities as you mentioned become corner cases or are balanced by other data points from the opposite tail of the distribution.

MTBF is generally established through manufacturer testing, not field data. It (or similar data) is readily available in many industries. Though I can understand it is less widely published for consumer products. Maybe I could find something in military/government RFQs and bids.

Unfortunately, I trust anecdotal and casual observational data the least. The human brain is wonderfully bad at storing, recalling, and relaying such information. Even with good intentions and experienced people, observational bias and selective recall error is huge. Maybe throw in some logs, ledgers, or billing data and you have something useful.

Oh well...

 

[KenGordon]

Read a few years of Annual Diving Incident Report

If the incident involved equipment failure it will be mentioned. Often equipment failure is really a failure of technique or maintenance. I think that actual equipment failure is extremely rare.

I am not sure it matters though, there are enough actual free flows to make up for a lack bits metal of falling off or badly made widgets.

 

[ams511]

The military or scientific diving community may have some data but for the recreational diving the data is not there. Also most failure modes is diving result in a freeflow not a lost of air.

 

[Caveeagle]

data is not there because there is no central collection body. Even if there was, it would be nearly impossible to extrapolate because use cases are very different.
Does the 4 hour dive that I do with a bailout bottle count as a "dive" even though it only has maybe 4 breaths total per dive because it's a bailout bottle? If I use a standard DIR configuration set of doubles where the inflator hose comes off of the right post, how do you count use for the left post? Technically it's pressurized and in the water, but there is no cycling of that regulator during the dive with the exception of the few breaths you take to check functionality.

Very true. I tend to cringe a bit when I hear the “this regulator is bad, brcause ...this happened to me” type of stories. Typically these are just single DATA points, with no context missing all sorts of factors that likely contribute to the result.

Small samples of DATA with no consistency in measuring system, make it real hard to prove out solid findings.

A recent poster here, called out a certain brand/type of gear where he and another diver have seen x # of failures within the last year or so. I happen to know that both these divers dive 20+ types per month, do deep, long cave penetrations and us multiple gear configurations. In addition to several other factors that might contribute to the result. As far as “anecdotal” evidence goes, there are few people’s opinions I would trust more. However it is still an unreliable way of collecting DATA.

 

[gr8jab]

This is in response to an inquiry on my related thread:

The plural of anecdote isn't data. What exactly are you looking to accomplish with your "data" request?

To answer your questions:
None.
A lot.
A lot.

Part 1:

I'm not a huge fan of collecting data this way (the other thread), but I don't consider 1st person data as bad as 2nd person.

Questionable observational data: "I've been a DM at XYZ resort for 10 years, and I've seen one person with a burst hose."

Less questionable data: "I've logged 1066 dives and I had one regulator fall apart."

Part 2:

I want to quantify the chances of a life endangering equipment failure. I can run a reliability model life distribution. Assuming I maintain my equipment properly, and I don't make any mistakes, I can use a beta of 1, and calculate a probability of non-wearout failures (defective parts, damage, etc).

I've been asking myself "why do I carry this pony bottle?" As a human, my answer is "because it makes me feel safer." As an engineer, I don't have any data describing the system. If I did, I could modify my human emotional response to more accurately fit the truth.

 

[gr8jab]

Also transferred from the other thread:

G...

Everything you jump into the water with is prone to failure...including the diver...

Service...maintenance...inspection...training...using the right gear for the right purpose...adherence to check lists...regular doctor visits/checks...staying fit and healthy...

Numbers have nothing to do with ''failures''...and sometimes...the ''God's gift to scuba diving'' are the worst offenders...when deviation from norm becomes common practice...

W...

Actually, given enough data, you absolutely can calculate the probabilities of any and all of what you describe. Each of these things you list (health, procedures, maintenance frequency, training, etc) are known as 'factors'. I do it all the time in our manufacturing environment where my factors include equipment failure and human data (such as recency of training, shift schedule, and sick/vacation days).

However, that not withstanding, I'm currently only interested in air-supply related equipment failure. I acknowledge there are a multitude of other factors, many worthy of investigation and understanding, and likely first-order (highly influential). I know my overall chances of a catastrophic emergency due to a screw up are much higher than an equipment failure. But that's not what I'm after right now.

Cheers!

 

[tbone1004]

@gr8jab FWIW my day job is as a textile engineer, but I'm also global quality manager for our company....

what we have learned from accident analysis is that there have been exceptionally few incidences of death due to equipment failure on open circuit scuba *maybe one a decade.....* and those failures are pretty much always due to some sort of error in either maintenance or failure to comply with manufacturer recommended service intervals.
The rest of them are either heart attacks, or Darwin Awards.

In terms of trying to quantify a probability of failure, the probability of failure is highest immediately after service, usually due to faulty parts from the manufacturer or bad service in the first place. Those failures are usually HP creep which causes the second stage to freeflow and you'll likely have that happen before you get in the water. It may happen in the water, but if you're buddy diving, that's what your buddy is for, and if not, that's what your redundant gas supply is for.

At work we can track machine hours, failures, and failure modes in order to actually generate statistically valid data. For scuba you would need to know dive profile, specific equipment used *the data for an Apeks sealed diaphragm first stage is not relevant to a Scubapro MK2 because even though they perform the same function, their design is completely different. Same with a Poseidon Xstream first stage vs. an Apeks XTX, vs a Scubapro R095 which are all completely different designs*, time and use since most recent service, and specific failure mode. It is functionally impossible to compile that data to get any sort of realistic data.
We can SWAG it fairly quickly.
PADI claims they issue on average 900k certifications per year for the last 20 years. If we say 3 dives per certification on average, that's 2.7m dives per year, just in training dives, just from PADI. I haven't dug through DANs incident report, which isn't going to include all deaths because some aren't reported, and certainly won't include the majority of any equipment failure incidents because people don't report it, but even if there was 1 fatality due to equipment failure per year *which we know is not the case, it may be one every 3-5 years if I had to guess*, that's 1/2.7m which is IMO irrelevant.