Choose timezone
Your profile timezone:
6th Quench Behavior Team meeting
→
Europe/Zurich
Minutes of the 6th meeting of the Quench Behavior Team
Date: 2015-11-12, 13.30-15.00
Place: 927
Presents: P. Hagen, E. Todesco, D. Tommasini, A. Verweij, F. Savary, P. Ferracin, J. C. Perez, R. Schmidt
Shape of quench distributions and guess for training above 6.5 TeV [E. Todesco]
Ezio shows the latest results about a study on the shape of distributions of training quenches. Hardware commissioning data are first analysed. The second quench is removed, and the 3000 series quenches distribution is studied. At first, the 45 sector data are removed. The remaining 100 quenches are compatible with a Gaussian distribution with average 11.6 kA and sigma 730 A (slide 4). Therefore the training curve is a Erf (error function, integral of a Gaussian).
There is some surprise among the colleagues to see the good agreement of an Erf (curve having only two free parameters) with the HC data – one can claim that this result is somewhat unexpected.
The 3000 series in 45 sector are also compatible with a Gaussian distribution, with smaller average 10.75 kA and smaller sigma of 450 A. The agreement is within the larger statistical error (we have only 62 magnets instead of 350 as in the previous case).
The 2000 series shows an excess of low current quenches in the tail, even when the 78 sector is removed (slide 8 and 10). The fit has to rely on few data, and therefore looks rather shaky. No attempt on 1000 series is done since the data set is too small.
Extrapolating the curves one obtains ~450 quenches to get to 7 TeV (11850 A plus 100 A margin), plus the second quenches on which no information is available.
If the management decides to push some sectors to 7 TeV, the proposal would be to try with 45, where most of the 3000 series already quenched (75% of them), so after ~15-20 quenches one could see if the second quench is below 7 TeV or above. And one could try with sector 12, having only nine 3000 series magnets, to test the behavior of 2000 and 1000 series. According to the extrapolation this would imply making a total of 50 quenches in 45 and 12 (about 25 per sector, plus the second quenches if visible).
This test, if approved by the management, could be done in different moments. The most conservative option (Ezio) is to have it just before LS2 (end of 2018). Rudiger suggest to study also the possibility of having it in the YETS (end of 2016) to leave some time to react to the outcome of the test. The option of doing it in spare time (Arjan) during normal operation looks not optimal since the time overhead of ~50 quenches (3 weeks) is very limited, so there is no need to complicate operation with a mix regime of HC and operation.
To further explore the shape of the distribution of quenches for the whole range, and not only in the tails, the analysis of the production data are shown. 1000 series first quench has a Gaussian distribution until 11.8 kA (slide 12), and a perfect fit using a 30% lower sigma above the average (slide 13). The same applies to 2000 series, whereas 3000 series first quench does not look to be in agreement with a Gaussian distribution (slide 15). This is a further sign of the 3000 series anomaly during production. It is suggested (Rudiger) to split the worse batch if one manage to recover a Gaussian fit as done for the HC. The second quenches are also analysed, showing a good Gaussian fit for all cases.
An interesting discussion starts on the meaning of a Gaussian distribution associated to quenches. It can be seen as
- Quench is due to several different causes combining as in the central limit theorem (Ezio’s viewpoint)
- Quench is due to a single cause, having a Gaussian distribution (Davide interpretation).
There are minutes attached to this event.
Show them.