Minutes of the 3^rd meeting of the Quench Behavior Team

Date: 2015-08-13, 10.00-11.30

Place: 927

Presents: L. Bottura, G. De Rijk, P. Hagen, S. Le Naour, J.-P. Tock, M. Modena, D. Tommasini, E. Todesco

Some additional data analysis [E. Todesco, P. Hagen]

Ezio presents some additional analysis on the hardware commissioning data done with Per. All observations relative to quenches needed to reach 6.5 TeV plus 100 A (11080 A).

• Reminder of the 2015 data: in average, 14% of the dipoles had one quench: 1% of 1000 series, 6% of 2000 series, and 34% of 3000 series. There is a strong sign on inhomogeneity also along the production of 2000 and 3000 series (nothing new, this is already know since few months).
  • Sector 78 and 81, gathering the initial part of the 2000 series, had 25% and 13% quench probability. Then situation improved in the second part, all below 10%.
  • Sector 45 has 74% probability of quench for the 3000 series, but only 15% for sector 78 that contain the initial part of the production.
• We have one sector (56) that has been powered twice to 6.5 TeV plus 100 A. In 2008 it required 24 quenches, and 17 in 2015. Is this difference significant, indicating that some memory is kept ? Associating the error of a binomial distribution, we find 24±8 and 17±7 (with two sigma error), so the data are compatible with a stable situation where there is no degradation and no improvement.
• Using the 56 sector data, we can distinguish between two radically different situations:
  • The 34% probability associated to 3000 series is made of 34% of the magnet quenching with probability 1, and 66% of magnet not quenching. In this deterministic case exactly the same magnets would quench in 2008 and in 2015.
  • The 34% probability is uniformly spread between all magnets. The behavior at each thermal cycle is uncorrelated with the previous one, so the probability of having magnets quenching in 2008 and 2015 should be the square of 34%, i.e. 12%.

In 56 we had about 20% of quench probability in 2015, and 30% in 2008 so we could expect 20% * 30% = 6% of the 84 3000 magnets of 56 quenching both in 2008 and 2015. We observed only 3 magnets quenching in both campaigns, so we deduce that data are compatible with the second situation.

This point has pretty strong implications: no point in replacing a magnet that is quenching since next time it will be another one. One this topic, suggesting the existence of an intrinsic random part in the training behavior, there is a wide debate. Davide points out that one should examine what happens to higher levels of the energy. It could happen that this random behavior is related to the fact of being rather close to the “wall”, and that going to 7 TeV all magnet will quench with probability one, so the random part disappears. It is suggested to make an extrapolation of the distribution to see where we will be at 7 TeV.

• The third point of the analysis concerns the secondary quenches. We selected all the 3000 series magnets having secondary quenches (induced by the quench of a neighbor magnet) at a current larger than 6 kA. All together we had 68 secondary quenches above 6 kA in 2015. With a generic quench probability of 3000 series of 34%, we would expect 23±8 quenches (two sigma), against 7 observed. So the secondary quenches above 6 kA reduce the quench probability from 34% to about 10%.

If we consider only the quenches above 7 kA, we had 42 magnet with secondary quenches, so we expect 14±6 quenches, but we observe only one magnet quenching in this subset. So the quench probability would be reduced to 2.5%.

This analysis would suggest that inducing quench above 7 kA could significantly reduce the training in the 10-11 kA range. Luca suggests to check this hypothesis at SM18.

• If the above argument is valid, within the same magnet the training of one aperture would also train the other one. So one would have very little cases of quenches in the same magnet with two different apertures. Today we have a few cases of double quench, but we do not know if it is the same aperture or not, since it has not been stored in the database. For the future quenches it has been suggested to MP3 to store also the quenching apertures. The analysis of the quenching aperture of the 200 past quenches will be done by Ezio in the next weeks.
• Finally, the 3000 production is analyzed. Gerard already showed at LMC the fluctuations of performance during the production. Ezio associates the statistical error, showing that with a binning of 26 one can see a uniform good initial production from 3001 to 3100, with a quench probability below 10%, and then a progressive degradation towards 70% around magnet 3150-3200, followed by a slow recovery, and a final degradation. Paolo Fessia points out that a qualitative behavior similar to this plot was observed during the production. Ezio observed that the degradation in the first quench of the virgin training takes place before, around 3050, and Paolo reminds that in that period he was seeing the first quench as rather random, and uncorrelated with sequent performance, whereas the second virgin quench was more representative of final performance. It is suggested to make some correlations with the first and the second virgin quench.

Brainstorming on tasks [all]

• Michele proposes to take the task of writing a short mote to point out the differences in the assembly procedures. It is also suggested by Davide and Luca to have a special session dedicated to discuss the differences between the three manufacturers.
• For the plateau quenches, Paolo suggest to see if there is a correlation with the high internal splice resistances. Per will recover it and check with Gerard.
• It is suggested to carry out the analysis of secondary quenches also on the 2008 hardware commissioning data.
• Gijs and Rudiger agree to take the task of identifying and proposing to the management of safe ways of training faster the magnets.
• Rudiger suggests to add a task of studying a strategy of the powering test above 6.5 TeV to be proposed before LS2.