Will http://www.ncbi.nlm.nih.gov/pubmed/25525213 change deco procedures?

[lowviz]

That would be a purely empirical model, nothing more than fitting a random equation to your datapoints.

Yes, exactly. That would be one extreme. I was intending to find the two extremes.

But, if it were possible to do so, I'd bet that it would provide a great algorithm. The utilitarian point would be well served and for most divers, who needs to progress to a theoretical model?

... Both supersaturation models and bubble models have known physical/chemical processes as their fundament, so by nature they should be more robust as you venture outside the data points that were used for fitting the model. ...

Well, I'm not entirely sold on that concept.

The effects of supersaturation and bubble formation should be "perfectly" and obtusely coded into the purely statistical approach. You just can't tease any of it out.

It may well be that a new mathematical construct that isn't based in the physical world is a better fit to reality. Someone would give it a name and it would become the new gold standard.

 

[rossh]

I only skimmed over that long thread, here is what I found. I would be grateful if you could help me by pointing out where I got it wrong.

What did I miss?

These extra points and information were added to the discussion after the RBW thread closed. They are only available in other threads. These points below really complete the picture.

**********

The supposition and assumption is that deep stops were actually tested in the nedu test, but they were not tested. The test instead fiddled the stops between 60 and 30 ft, using navy (shallow stop) test design models.

You can see the deep stops part is the green where VPM /RGBM / RD / ZHL+GF will make stops (100-110ft). But the test profile had its first stop much later (70-60ft start) - a shallow stop model.

They in fact made a test of extended shallow stops, coupled with thermal stress - no deep stops anywhere.. They tested thermal stress on two shallow stop profiles that have deco time that is over 2x longer than needed. This is an example of a "low stress" test method, where the results are then assumed to mean something else.


So, no actual deep stops in the nedu test.


*************

The assumption to connect this shallow stop test to deep stops, lies in A\ the fallacy of the reports title, and B\ these 'plausable explanations' that Simon makes:

The most plausible explanation for the NEDU result is that protecting the fast tissues from high levels of supersaturation early in the ascent at the expense of increased supersaturation in slower tissue late in the ascent (a characteristic of most deep stop profiles including many used by technical divers) is not a successful strategy.
Simon M

Lets examine those more closely:


".. is that protecting the fast tissues from high levels of supersaturation early in the ascent... "


No. The A2 starts with a supersaturation peak of 120 kPa (1.2 ATA). Clearly A2 doesn't have a deep stop, because real deep stop profiles start with several small supersaturation values. The protection mentioned above has not occurred. The A2 has double the VPM start pressure, and A2 appears to start as a shallow stop example. (shown below)

****

... at the expense of increased supersaturation in slower tissue late in the ascent ...

No. The A2 has less than half of a typical supersaturation pressure from a normal ZHL or VPM dive through the ascent. The A2 does increase its supersaturation pressure as it ascends but that condition is unique to A2 only. This inverse condition is not found in normal shallow or deep stop models. Both ZHL and VPM will either hold a steady maximum supersaturation or decrease it as the ascent progresses. A2 a not typical of deep or shallow model. (shown below)

*****

.... (a characteristic of most deep stop profiles including many used by technical divers) ...

No. Deep stop profiles typically finish with about the same maximum supersaturation pressure as the shallow stop equivalent profiles. (shown above)

Also this can be checked with individual cell pressure levels in MultiDeco - not significant as suggested above.


*******************

So Mr. kr2y5. that's what you missed.

All 'plausible explanations' given to connect this shallow test to deep stop conditions are invalid, and the assumptions that connect the test to deep stops, do not exist.

They tested something, but it wasn't deep stops.






Note: Anyone can make these profiles and diagrams above in MultiDeco. Supersaturation tracking and graphing is part of the program. Supersaturation pressure is the very core of deco calculations and its the basis of the limits in VPM and ZHL. The GF adjustment is directly based on these values.
Also a new feature, is that you can look at individual cell pressure that make up those graphs (right click, show cell #).

 

[Storker]

But, if it were possible to do so, I'd bet that it would provide a great algorithm. The utilitarian point would be well served and for most divers, who needs to progress to a theoretical model?

[...]

It may well be that a new mathematical construct that isn't based in the physical world is a better fit to reality. Someone would give it a name and it would become the new gold standard.

No, that would by the nature of being empirical with a random equation fitted be a weak model.

Risking to go a little off on a tangent here, but this is one of the fundamental principles of science. For centuries, the geocentric universe was a "gold standard" and perfectly able to predict phenomena within the limits of observation. As technology progressed and more measurements were made, the fundamentally flawed "gold standard" needed more and more fudging to function. Then came the heliocentric model, a lot of things fell into place and the predictive value of the model increased immensely. Now, by just trial and error one might stumble over a model which reflects reality, but the chances of that are pretty slim.

So, in science, one would strongly prefer to base one's model on fundamental principles if at all possible. Purely empirical models just don't cut it.



--
Sent from my Android phone
Typos are a feature, not a bug

 

[Patoux01]

No, that would by the nature of being empirical with a random equation fitted be a weak model.
[snip]
So, in science, one would strongly prefer to base one's model on fundamental principles if at all possible. Purely empirical models just don't cut it.

Ever heard of machine learning? It can make quite a few nice things that would be impossible using what you define as "fundamental principles". It would be interesting to see what algorithm would come out of such design process, as it would be easy to gather a massive amount of data and "safe" (ie that doesn't kill the user) profiles nowadays...



(ah, and if I understood correctly, the Haldanian models come from... experiments on goats and "oh, that doesn't seem to kill them, okay I'll use half pressure as SAD")

 

[Storker]

Ever heard of machine learning?[/SIZE]

Of course. It's great engineering, but unlike a true model it doesn't add a grain of understanding of a system. It's basically a "black box" approach to modeling. And while it's an excellent solution to systems which are way too complex to be modeled from first principles, we know pretty well the fundamental mechanisms of oversaturation, gas going out of solution and bubble growth. Science just hasn't quite determined which factor(s) is (are) rate-determining.


--
Sent from my Android phone
Typos are a feature, not a bug

 

[lowviz]

Of course. It's great engineering, but unlike a true model it doesn't add a grain of understanding of a system. ...

Thank you.

I was attempting to find two extreme approaches to the same end. I seem to have been able to convey the idea of one of the two extremes.

But I'm not discussing the relative merits of modeling approaches. I'm trying to gain a fundamental understanding of the two camps in this interesting thread.

Let me restate my exact same question. I’ll reword both my premise and the question.



Again we have access to the fictional perfect database that covers every dive and ascent plan that has ever been attempted in real life. Given that, I could see two extreme camps of modelers emerging.

One camp would only be concerned with a set of "rules" that would get them out as quickly as possible without bending them. The other camp would only be concerned with the physiology of the exact same situation. Both seek the same goal, get out quickly and safely. But they have very different objectives along the way.

I see no conflict, I see two clearly different ways to obtain the same goal. Which approach is more valuable? Depends on who YOU are.

So my question is again: “Would the two camps in this discussion be willing to place themselves on a line between:

extrema 1) "Just don't ever bend me."

extrema 10) " Why and how will that bend me?"

 

[Storker]

Well, on average camp 2 would probably be less out of their depth if they tried to extrapolate their model outside the realm of previously available data. You're free to choose your preferred approach. I know which one I prefer.

 

[lowviz]

Not something between both extremes?

 

[Storker]

No, since they're not two extremes of a spectrum. They are two fundamentally different approaches to the problem.


--
Sent from my Android phone
Typos are a feature, not a bug

 

[victorzamora]

No. Deep stop profiles typically finish with about the same maximum supersaturation pressure as the shallow stop equivalent profiles. (shown above)

Also this can be checked with individual cell pressure levels in MultiDeco - not significant as suggested above.

This, Ross, is the root of my issues with your posts. I know my opinion means nothing to anybody but myself, but I can't be alone here. You state things as fact because MultiDeco spits out some charts based on an algorithm to theoretically track loadings based on a theory that has been proven to be incomplete (at best). MultiDeco is a great program, and I applaud your efforts in bringing the features you do to the table. VPM is an algorithm that has returned many divers safely to the surface. What neither of them can do is claim to actually track supersaturation (and other factors) of divers doing dives. It can calculate projected, theoretical values based on a large series of assumptions. Deco studies, like this one and the NEDU deep stops one and many other ones can be used to test algorithms and assumptions those algorithms are based on. What you can't do is use the algorithm to self-validate itself.

This study has a lot of very interesting implications if true, many of which would invalidate (word choice is probably poor) many algorithms and assumptions we use today. For practical purposes, I think little has changed. I'll still continue to dive the algorithm and conservatism that gets me out of the water feeling good. However, long-term consequences are yet to be seen and this could be a huge step forwards. If this study is true, it seems the "Helium penalty" might need to be reduced to nothing in Buhlmann and the "Depth penalty" (my term) needs to be increased. This might mean a new algorithm can emerge, which is exciting, even if it's only a refined variant of what we know and dive today.