Liberty, change O2 sensors all together or not?

[OceanEyes]

I started looking, I'm not terribly handy and can't see myself making a good one quickly. Are there any off the shelf products people can recommend.

Narked at 90 offers some products that may work for you.

Cell Checkers | Narked at 90

 

[Superlyte27]

For the record, Divesoft recommends replacing all 4 cells at the same time.

I never dive cells beyond their date for one reason, in the event that I die, no matter the reason, everyone of you goobers is going to say, well, he knew it had expired sensors, what did he expect? Plus, in the grand scheme, sensors are cheap.

 

[Heat Miser]

He is full of crap

Is that me or Paul or Rjack? Extreme Value Theory is not crap it is just very infrequent so its hard to measure by anyone

I just want to add one thing about the tail being left skewed in the above chart/histogram, it comes about typically from a bad batch, so if you replace 2 cells at the same time the tail underestimates the probability both cells are bad because they come from the same batch. I.E. the probability both cells are bad is not 10% * 10% its greater.

How much greater the paper doesn't try to calculate, but its does show its a risk by separately showin a similar distribution from a bad batch, so my take from the paper is swapping all your cells out of your rebreather at the same time appears to be a poor idea.

 

[beldridg]

So when one cell goes bad, you'll replace all 4? That seems very risky to replace known good cells with unknown cells




The Liberty has 2 cells per side. If one side "dies", then the voting logic is done and it has to average the two remaining cells, no different to a Shearwater voting one cell out. It won't vote a second one out because it doesn't know how. The difference with the Liberty though is you can manually vote a cell out. Say you're diving EAN32 dil and diving to 30m/100ft. The left O2 board failed and you only have the 2 cells from the right. You do a dil flush to make sure everything is OK like you are supposed to and you get 1.27 on cell 1 and 1.01 on cell 2. Any normal CCR would average those two to 1.15 and try to maintain whatever setpoint you have.
In the other breathers you have to drop the setpoint and run it manually off of cell #1 and ignore cell #2. With the Liberty you go into the menu and tell it to disable cell #2. In the cell menu it will tell you excluded if it was voted out, or disabled if you turned it off.

If you are diving trimix dil, then you can turn it to He mode and it can do some fuzzy logic with the helium sensor to back into the ppO2 and it will fire the solenoid/s based off of that.

Deleted. I didn't really just how old this thread was and had replied earlier about my experience with Shearwater and two cells and the voting logic.

Sorry for the interruption.

- brett

 

[rjack321]

Thanks @rjack321 is there any off the shelf pressure pots fit for purpose? I have thought about building my own, I'm not the most hands on, and I could see it dragging out longer than it has taken me to build my unfinished o2 analyser.

On the "Zero data", there is a histogram in the paper showing testing of 1,000 cells. Just eyeballing the chart I can see the left skew and I can tell there is a greater than 10% chance that a cell becomes current limited in less than 4 months. SImilarly eyeballing an average of 18 months it means to replicate the results would take (18 x 1000)/12 years = 1,500 years, if you did the test one cell at a time.

There's lots of data in the paper, some peer review would require quite a bit of testing, but its open to anyone to replicate the results. I'll leave that one for someone with access to 1,000 cells to replicate in 2-3 years.

View attachment 660681

This is all made up "hypothetical" data. Revo didn't buy 1000 sensors and look at their failure rates. There's no survey of 200 revo divers and the failure rates of the five cells used by each over a year plus. This whole paper has ZERO science, ZERO data, it's Paul Raymaker making up numbers to fit his preconceived notions of how cells fail - and he even says he is making up the numbers as an example in the text.

It should never have been published or disseminated in the first place. It's unfortunate that he designed the Revo to mitigate for the hypothetical failure rates of cells instead of statistical analysis of actual cell performance.

 

[Heat Miser]

This is all made up "hypothetical" data. Revo didn't buy 1000 sensors and look at their failure rates. There's no survey of 200 revo divers and the failure rates of the five cells used by each over a year plus. This whole paper has ZERO science, ZERO data, it's Paul Raymaker making up numbers to fit his preconceived notions of how cells fail - and he even says he is making up the numbers as an example in the text.

It should never have been published or disseminated in the first place. It's unfortunate that he designed the Revo to mitigate for the hypothetical failure rates of cells instead of statistical analysis of actual cell performance.

Do they have to buy 1,000 cells to do the test? They just have to test all the cells on the rEvo's they do a 5 year service on. I'm not trying to argue. When I read the paper that is how I assumed the data would be collected. Most people I know who have put their units into a 5 year service seem to come back with a couple of new cells anyway.

how many rEvo's would they have sold? I guess its feasible if the numbers 1,000 prob not if the numbers 600.

I'll ask my service guy next time we speak what he thinks.

 

[Heat Miser]

This is all made up "hypothetical" data. Revo didn't buy 1000 sensors and look at their failure rates. There's no survey of 200 revo divers and the failure rates of the five cells used by each over a year plus. This whole paper has ZERO science, ZERO data, it's Paul Raymaker making up numbers to fit his preconceived notions of how cells fail - and he even says he is making up the numbers as an example in the text.

It should never have been published or disseminated in the first place. It's unfortunate that he designed the Revo to mitigate for the hypothetical failure rates of cells instead of statistical analysis of actual cell performance.

I read the disclaimer

"Some graphs in the presentation are only shown for didactic purposes and do not reflect real quantities of data."

OMG

 

[cathal]

This is all made up "hypothetical" data. Revo didn't buy 1000 sensors and look at their failure rates. There's no survey of 200 revo divers and the failure rates of the five cells used by each over a year plus. This whole paper has ZERO science, ZERO data, it's Paul Raymaker making up numbers to fit his preconceived notions of how cells fail - and he even says he is making up the numbers as an example in the text.

It should never have been published or disseminated in the first place. It's unfortunate that he designed the Revo to mitigate for the hypothetical failure rates of cells instead of statistical analysis of actual cell performance.

Ever hear about statistics? You don’t need n-100

I read the disclaimer

"Some graphs in the presentation are only shown for didactic purposes and do not reflect real quantities of data."

OMG

Agreed, but if you read on up to just Page 13, the key data that he extrapolates from is the failure modes from an actual batch of 1,000 cells:

"This type of graph show the number of cell that have failed before or at the moment of reaching a certain age, out of a batch of 1000 sensors. It is this graph we will later use for computer simulation, when we want to estimate the lifetime of a sensor: each time we install a new sensor in our rebreather, we let the computer pick a random number between 1 and 1000, and using the graph we get a lifetime prediction for this particular sensor." ​

​

 

[rjack321]

Ever hear about statistics? You don’t need n-100

Agreed, but if you read on up to just Page 13, the key data that he extrapolates from is the failure modes from an actual batch of 1,000 cells:

"This type of graph show the number of cell that have failed before or at the moment of reaching a certain age, out of a batch of 1000 sensors. It is this graph we will later use for computer simulation, when we want to estimate the lifetime of a sensor: each time we install a new sensor in our rebreather, we let the computer pick a random number between 1 and 1000, and using the graph we get a lifetime prediction for this particular sensor." ​

​

Except there is no N at all. We don't know the failure rates at any age beyond anecdotes. If there are actual failure data in that paper the brand, storage, condition, usage, EVERYTHING important is missing or bootstrapped from a completely mysterious dataset. The whole paper is garbage

 

[cathal]

Except there is no N at all. We don't know the failure rates at any age beyond anecdotes. If there are actual failure data in that paper the brand, storage, condition, usage, EVERYTHING important is missing or bootstrapped from a completely mysterious dataset. The whole paper is garbage

The paper says the data was got from an actual batch of 1,000 cells. As this is a study done by rEvo they would only have used their own cells as they are the only cells that they supply, so the brand dataset would not be relevant. But can you explain what exactly do you mean by 'cell condition'?