AI Cells Again

diverreb

Call me a beginners checker player here.....

I'm having a very hard time wrapping my head around this concept.

Accurate cell readings are the desired end result of the "system."

Cells are that only way to get that measurement, that I'm aware of.

You do your cell checks & MV checks pre dive. You check linearity, if you have that capability. If everything is good, that means that at that moment the cells are working.

Is your point that the 'system' has to fully analyze the cells functionality & accuracy continually during the dive before you feel a unit is safe to dive?

I'm not aware of any 'system' that will do that today. Voting logic seems to be the closest thing and all that's doing is dropping a cell from the average if it's reading varies a certain percent from the other two.

With the cells being the key and only thing providing information to get the desired results, if they are notoriously faulty, then if will be very difficult to get accurate results, no matter what the 'system' is doing. Yes?

Richie

jmurba

I don't even try ....
there are 2 was of seeing a forum
the first is to explain a supposed new concept from A to Z in simple words.
the second is not to ...
you can't understand what hss not been properly explained neither can I
regards

__________________
when will I be able to think about something else then spending hours underwater, when will I be normal ?

http://web.mac.com/jmurba/Site_2/home.html
http://web.mac.com/jmurba/Site_3/home.html

Dave Sutton

No.

Accurate Setpoint Maintenance is the desired end result of the system.


That can be done using a few statistical analysis and artificial intelligence tricks with a wide range of cell data-quality levels. Think assignment of higher FOM (Figure of Merit) to cells producing expected outputs V/S assigning lower FOM to ones producing unexpected outputs and biasing the output calculations accordingly. You can predict with fair accuracy what a cell SHOULD be reading at any given time and compare what data the cells are ACTUALLY offering to derive a very accurate estimation of loop contents. It's like the way an inertial navigation system works: It knows where it SHOULD be, but can also detect that it is NOT where it should be and it then figures out the difference and displays it.


Grasp that and we can continue. Alex will step up the plate, I'm sure, as he's already done it. I could propose schemes but they would be hypothetical solutions, so I'll defer to Alex. Go back and read the earlier part of the posts and look especially at the output graph for the test he posted. One good cell, 3 bad ones, and a system that holds setpoint. That is the desired end result of the system. The way you achieve that takes a higher level of data processing than is currently used, but is fairly easy to do.

Rich: Think FAR outside of the current paradigm of PP02 calculation. In 10 years we'll look back at our very primitive systems of today and wonder why more of us were not killed. I'm not even sure that voting logic as you understand it is better than the averaging system used in the analog Mark-15 pods. In any event, compared to what is possible, simple 'voting logic' is nearly as primitive as averaging logic.

Best,

Dave

.

__________________
"Silent Diving with No Bubbles and No Politics".... www.nobubblediving.com

diverreb

Dave,

Having three bad cells and one good one and having a system that figures that out and controls the set-point accurately is Way beyond my pay grade.

My point was relating to existing systems.... and there the end result is accurate set-point maintenance, and currently the only way of doing that well is to have accurate cell readings.

Thinking outside the box also may bring about a much more reliable cell, where accurate readings and reliable performance will be the norm.

Richie

Tienuts

The sad truth is, a more reliable cell can be made now - Mark from AII will be the first to say this. It's just we would have to pay $400 for it.

Lancer4545

Dave, Seeing as you have jumped in on the subject.....Please let me point out the Topic Title for you so you can refresh your memory...... AI Cells Again.

The subject is about the UN-Reliability of some batches of AI cells. We're not talking about the reliability of the electronics package, or the batteries, or even the diver trying to decypher the readings from his handset which has just gone T!ts up because his cells have failed. There is no need to preach to us the unreliability of these cells.....WE ALREADY KNOW.

Great so you agree with me then, I'm glad we have that bit sorted!!

Who's talking about cores....This topic is ALL about cells.........That is unless someone is trying to dazzle us with brilliance but that's not what we are discussing...is it!!

Got me wondering Dave what you're on about.......Aren't we discussing AI cell failures and you're blazing off on some quest about chess and checkers. I know all I need to about the reliability of cells thankyou, and the TOPIC is quite clear, if you were to take the time to read it.

Huh what are you on about???

When outdoors for any length of time, cover up, wear a wide brimmed hat and wear plenty of sunscreen. The damage can be perminant!

Lance

__________________
You can run but you can't hide!

ISC Appointed Megalodon Dealer for East Australia
http://www.closedcircuitdivers.com.au
info@closedcircuitdivers.com.au

deepwrecksc

I too would be very interested to learn more about the algorithms Alex is talking about. I know that this is technically off-thread and so I apologize to those most interested in AI cell problems for potentially pushing it farther off (I think a new thread is in order??).

Maintaining a desired ppO2 with one out of four oxygen sensors operating properly is definitely a good thing - my question is: does this work under all conceivable failure modes?

In the field of analytical chemistry, we always deal with "derivative" data - turning a physical property (in this case oxygen partial pressure) into a signal that can be observed and analyzed (electrical potential here). In the real world, most transducers that can do this type of conversion are imperfect devices (just like the galvanic O2 cell). and this is one reason why three independent measurements are the minimum in order to arrive at an answer that should be close to what is REALLY happening in the system being measured.

With highly unreliable instruments, we must (1) make many more than three measurements, (2) cross-check the data with another type of transducer, and/or (3) find algorithms that can identify and cancel the errors in the faulty transducers. When faced with this problem in the laboratory, I tend to combine 1&2 rather than face #3, since in most (well ALL really) situations it is virtually impossible to find an algorithm that so perfectly describes the behavior of a transducer that it will predict all failure modes. Remember, complex and fancy algorithms are nothing more than models. Models can and do fail unpredictably.

So what would happen if your algorithm picked the wrong sensor (of the four possibilities) to be the "correct" one? I have not seen the decision tree matrix and don't know what failure modes that you have built in to the model, but clearly this risk would have to be minimized.

It is tempting in making measurements to allow precision (having several measurements close to one another) to predict accuracy (having a result close to the true value), and this is exactly what voting logic attempts to do. Accuracy and precision can be mutually exclusive - you can (and often do) have one without the other. This is one of the first principles I teach to beginning grad students and it never ceases to blow their minds when I show them examples. Lucky for us that our oxygen sensors often DO agree on a result that we believe to be correct.

Lee

jmurba

Hi,

Some work as they can and sell rebreathers, ap valve, meg, draeger etc etc
some other spend their time posting litterature on the internet explaining "how do rebreather kill people" and how they would not kill people if their fantastic future machine was on the market.
but it is not in the market....

the present topic is "AI cells"

so I have only once tested those sensors 5 years ago and returned to teledyne cause the sensor I had recieved had nothing to compare with teledyn's ones

I think that there has been an effort on this forum and others to promote AI cells certainely because teledyne must not be easy to deal with.

I just can't technicaly knows id teledyn's are better but for the moment I have not reason not even the price to swith to AI.

reading dozens of diver posts who complain will not make me change my mind

but mabye Am I missing something....mabye my belief is wrong.

is there anyone out there having dived for years with r'22 and who switch for years to AI cell and who can say that they are at least as good or even better


regards
jean mi

__________________
when will I be able to think about something else then spending hours underwater, when will I be normal ?

http://web.mac.com/jmurba/Site_2/home.html
http://web.mac.com/jmurba/Site_3/home.html

diverreb

I have about 80 hours on my O2ptima and have never had a problem with AI cells.... The first three worked fine for nine months and the the second set is performing well.

I also know a few of my O2ptima buddies that haven't had any problems with them either....

Maybe we're just lucky.... Or maybe it has something to do with how they are placed in the 'system' that keeps them from going bad.

Richie

Dave Sutton

Are you running for Moderator?

Dave

.

__________________
"Silent Diving with No Bubbles and No Politics".... www.nobubblediving.com