VPM-B Oddity

[Rick Walsh]

Thanks for the report and xml file

On 7 April 2016 at 05:06, John Van Ostrand <john at vanostrand.com> wrote:

> On Tue, Apr 5, 2016 at 10:55 PM, <john at vanostrand.com> wrote:
>>
>>
>>
>> John,
>>
>> On 6 April 2016 at 12:02, John Van Ostrand <john at vanostrand.com> wrote:
>>
>>> I have an example where VPM-B +0 gives an earlier and longer deco
>>> obligation than VPM-B +1 and +2 for the same imported dive.  I found
>>> another dive where +0 and +1 were almost identical.
>>>
>>> That is indeed odd; +0 should not be more conservative than +1 or +2.
>> Similar obligations are less surprising.
>>
>>
>>> I was comparing VPM-B to Cochran's algorithm on a dive were the Cochran
>>> gave me a minor deco obligation. To do that I'd change to the VPM-B algo in
>>> the Plan dive function and it would show the obligation for a dive in the
>>> profile display for imported dives.
>>>
>>> I'm not familiar with the Cochran.  Do you know what the deco model and
>> conservatism you were using?
>>
>> It's using a modified Thalman. They've added bubble mechanics to it and
>> other modifications. It's proprietary. Conservatism is 20% on one dive and
>> 30% on another, Greater means more conservative.
>>
>> Personally I'm interested in how it compares but this bug report doesn't
>> involve the Cochran deco algo.
>>
>>
>>> I tried pulling out the XML for just the two dives in question but I was
>>> unable to duplicate the problem. When I included previous dives from the
>>> trip it was reproducible. It seems to be an issue when repetitive dives are
>>> included. The SIT was 18 hours.
>>>
>>> It sounds like it's somehow related to tissue loading from previous
>> dives, but I can't work out what.  The VPM-B algorithm tracks tissues
>> mostly like Buhlmann, but there are a few funny quirks/features.
>>
>>
>>> I can provide an XML file to whomever is interested in investigating.
>>> The algo implementation is way beyond me.
>>>
>>> If you send the file to me and let me know which dives are giving the
>> odd results, I'll have a look when I get a chance, but don't hold your
>> breath.
>>
>> I'll do that in the morning.
>>
>> A few questions:
>>
>> What gasses were you using?
>>
>> Ean32 in the most significant case, ean31 in another.
>>
>> Were the gas changes (if any) picked up when downloading from the DC?
>>
>> Yes. Ean32 on the entire dive.
>>
>> Does the ceiling look normal with the Buhlmann model with varying
>> gradient factors.
>>
>> I looked at them in the past and they looked reasonable under Bühlmann‎.
>>
>
> For those following this. This might be a display problem more than a bug
> with the algo. Turning "3m incremental stops" on makes it break. So it's
> the 3m ceilings that appear broken not the actual ceilings.
>
> I worked out what was causing the oddity.  The ceiling was calculated as
it is supposed to be with +1 conservatism, but with +0, you managed to
reveal a hidden bug.  VPM-B has an iterative algorithm (CVA) to determine
the shortest time to surface allowable.  Each iteration uses the deco time
from the previous iteration as an input, until the deco time converges.
The problem was a negative deco time was being calculated for one of the
iterations, which doesn't make sense and made the calculations do silly
things.

I've sent a patch that corrects this.

Rick
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.subsurface-divelog.org/pipermail/subsurface/attachments/20160407/a7b59fb2/attachment.html>