From: Charles L. Neumeyer
Sent: Saturday, April 17, 2010 4:38 AM
To: Mike Williams
Cc: Robert D. Woolley; Lawrence E. Dudek; NSTX Center Stack Upgrade Team
Subject: Re: NSTX CSU Project Meeting Notes from 4/14/10
Dear All, 

Coil currents are not predictable with sufficient confidence (due to plasma uncertainties) or sufficient reliability (due to possibility of operator error, software failure, other control system failure) to protect the machine. This is the reason, in the first place, for real time protection!

Chas


On Apr 16, 2010, at 11:06 PM, Mike Williams wrote:

Although the exact coil currents cannot be predicted, for the reasons Bob has stated, the anticipated coil currents can be pre-programmed and the shot aborted if conditions vary beyond some acceptable envelope. We have not taken this approach in the past but could consider it going forward if the DCPS became too complicated.
Mike


On Apr 16, 2010, at 4:49 PM, Robert D. Woolley wrote:

Hi Larry,
No, the current waveforms for a particular shot will not be known in advance of that shot, and that obviously also apples to the entire set of shots in a day's run or a week's run.  We can project what we think the OH current waveform, and various other coil current waveforms, will likely be if the shot goes off with everything nominal, but that projection is based an assumptions about plasma performance which are not reliable enough for coil protection since they routinely are violated when the plasma does not behave as anticipated, as happens frequently. 
 
For instance, if the plasma's loop voltage during startup is a little bit higher than normal then the peak OH current will as a result also be higher than normal;  this can occur if the plasma's energy confinement is poorer than normal (e.g., if lithium is not completely coating plasma-facing surfaces or e;g;, if NB's fault or fire late, etc.).   The higher peak current may endanger equipment through excessive forces/stresses.  Since peak coil currents for planned operations cannot be accurately known before a shot, the DCPS cannot accurately protect equipment based on prior plans for that shot.
 
An additional issue is that mis-operations can occur during a shot due to actual failures of control equipment.  Such miisoperations due to failures also cannot be anticipated before a shot.
 
-Bob Woolley
 
From: Lawrence E. Dudek
Sent: Fri 4/16/2010 4:21 PM
To: Robert D. Woolley
Cc: NSTX Center Stack Upgrade Team
Subject: Re: NSTX CSU Project Meeting Notes from 4/14/10

Bob,
You just did the numbers below to map the whole universe of combinations, which is not practical.  But does it make sense to perform the algorithms ahead of time for a shot, a day's run or weeks run?  
Larry


On Apr 16, 2010, at 3:47 PM, Robert D. Woolley wrote:

Hi Larry,
 
In your meeting notes (copied below) you say about the DCPS:  
 
"The advantages and disadvantages of running the algorithms in real time is being reviewed and compared to running algorithms ahead of time." 
 
This comment surprises me, since at the meeing I explained why running the algorithms ahead of time makes no sense. 
 
Repeating myself here, there are two possible conceptual approaches to running DCPS algorithms ahead of time, one approach in which the detailed current waveforms for a proposed shot are investigated immediately in advance of that particular shot, and another approach in which all possible current combinations are mapped in advance of NSTX CS upgrade commissioning into a comprehensive computer-based map showing the acceptable safe regions in the 15-dimensional "current hyperspace" (i.e., 12 PF currents, 1 OH current, 1 TF current, and 1 plasma current) that together detemine electromagnetic forces. 
 
The first approach to running DCPS algorithms ahead of time does not work because the actual currents flowing in a particular shot will not be accurately and reliably known before the shot has occurred.  Coil and plasma currents are determined by real-time feedback control, which is subject to changes due to various influences.  For some coils the coil currents themselves are not preprogrammed variables subject to servo-feedback control; instead some plasma current or shape characteristic is being feedback controlled.  For instance, the OH coil current waveform vs. time will be remarkably different for shots whose preprogrammed setup parameters differ only in their gas fueling profiles or in their neutral beam plasma heating profiles.   Furthermore, unplanned misoperations in gas fueling or in neutral beam heating may cause the OH coil current or OH-current-depndent forces to surpass limits that cannot be anticipated by running the DCPS algorithms ahead of time.
 
The second approach to running DCPS algorithms in advance is precluded because it is not practical to make a computer-basd map with enough detail to store the acceptability results of all possible combinations of 15 currents, each one of which can vary continuously over its range from its negative to its positive maximum value.  For instance, if 100 coarse "bins" were used to discretize each current signal (i.e., each 2% of full scale) then 
       (100)^15 = 1,000,000,000,000,000,000,000,000,000,000 bits
would need to be stored just to hold the pre-computed map of safe vs. unsafe combinations of currents.  Such computer memory capacities do not exist today.  In addition, the second approach would entirely ignore the need to track in real-time the heating/cooling internal states of coils, which depend on details of the coil current histories. 
 
I'm curious who is doing the review that your comments state is in progress, and what is the technical concept being considered in that review for running DCPS algorithms in advance.
 
-Bob Woolley
 
 
From: Lawrence E. Dudek
Sent: Thu 4/15/2010 4:25 PM
To: NSTX Center Stack Upgrade Team
Subject: NSTX CSU Project Meeting Notes from 4/14/10

Notes from 4/14 Meeting:
Update on the TF/PF Structure [Mangra (see VGs) ]
The design of the outer TF leg supports and the Outer PF supports was presented at the project meeting. The latest refinements include reduced parts weight while increasing strength and improved assembly features.

PF 4/5 Lead area stress, Existing PF4/5 Support Weld Algorithm, Lower Umb. Lid [Titus (see VGs) ]
Titus presented the status of the PF4/5 lead block stress analysis. 
The strategy for the implementation of the DCPS is being looked at to minimize the required effort while providing the most useful protection for the hardware.  The advantages and disadvantages of running the algorithms in real time is being reviewed and compared to running algorithms ahead of time.

Diagnostics Impacted by Upgrade [Boales (see VGs) ]
A review of the diagnostics on the NSTX is being conducted to determine what impact the new centerstack and higher power operations may have. Of concern so far are the shutters and activation of G10 materials


Viewgraphs will be posted by B. Simmons at:

Outstanding Action Items (last update 4/15/2010)
Action: Structural design criteria has a table with likelihood of failures that is in the NSTX (orig.) GRD.  Structural design criteria needs work to be released such as references 
STATUS: Revision being prepared for signatures. (Status 3/15)

Action: The idea of a Design Load Requirements (Criteria) Document is a good one. Action: Mark Smith to pull this together for the Structures– Need this BEFORE the PDR

Action:  We have a set of Calculations signed off, but not all were checked for the CDR.  These calcs need to be updated and checked (by PDR?). ACTION: Titus


________________

Larry



__________________________________________________
Michael D. Williams
Associate Laboratory Director for Engineering and Infrastructure
Princeton Plasma Physics Laboratory
P.O. Box 451
Princeton, New Jersey 08543
Phone: 609-243-2866
Fax: 609-243-2800