1. ANALYSIS OF THE PRODUCTION DATA AFTER THERMAL CYCLE SECOND PART
CERN, 20th June 2016
Quench Behaviour Team meeting
ANALYSIS OF THE PRODUCTION DATA AFTER THERMAL CYCLE SECOND PART
E. Todesco
CERN, Geneva Switzerland
Based on production data
(thanks to project engineers and SM18 staff)
Based on the HC data of 2015 and 2008
(thanks to MP3 teams)

2. DATA ANALYSIS: AFTER THERMAL CYCLE
About 100 magnet tested, 35 to 60 per series
The most interesting case (3000 series) has the fewest 
We consider only the first quenches
At 6.5 TeV the 3000 series anomaly is very clear
Situation is getting more even at 7 TeV, where all firms basically behave in the same way (statistical error at 7 TeV is about 13%)
This is not our best estimate for 7 TeV

3. DATA ANALYSIS: AFTER THERMAL CYCLE
Dismantled magnets:
Among the magnets tested after thermal cycle, 13 were decollared and coils were replaced
In this case the number increased by 500, i.e was decollared and became 1526
Out of these 13 magnets, 12 are in the LHC
Out of these 13 magnets recollared, 4 have been tested after thermal cycle: 1626, 1643, 2525, 3636
So I remove from the analysis the data relative to dismantled magnets (i.e. 1026, 1143, 2025, …)
Ie we consider all problems related to these magnets as understood and solved, and the training data not significative

4. AFTER THERMAL CYCLE VERSUS HARDWARE COMMISSIONING
Comparison between a naive scaling of the After Thermal Cycle (ATC) data and our forecast
For the 2000 series we have similar numbers
For the 1000 series, ATC would suggest a much larger quenching w.r.t. the 5% I took in a rather arbitrary way (no data available for Gaussian extrapolation)
Will we have a 1000 series problem at 7 TeV ?
For the 3000 series the HC data are much worse
Let us look in more detail in the sample

5. SAMPLING IN THE 3000 SERIES
Very unlucky ! The most interesting batch 3157 to 3208 had only one magnets tested after thermal cycle
Removing the initial 100 magnets, we move towards values that are more in agreement with what observed in the HC
From 11%±10% on the whole sample, to 20%±18% on the sample excluding the first 100, to be compared to 34% observed in HC
At 7 TeV this gives 40%±22%, at the edge of the 67% expected from Gaussian extrapolation
So ATC data could be compatible with what observed in HC and what expected based on extrapolation

6. SAMPLING IN THE 2000 SERIES
For 2000 series, large sampling at the beginning, and between
Data are compatible with the HC to 6.5 TeV, and with extrapolation to 7 TeV (we gave 25% based on Gaussian fit)
Quite uniform performance

7. SAMPLING IN THE 1000 SERIES
For 1000 series, large sampling at the beginning, and between
Worrying data for the batch
Very scarce sampling afterwards
Assuming the first 52 as representative, we will have a ~20% quench probability
So probably the estimate (a priori) of 5% was too low

8. CONCLUSIONS Data after thermal cycle have special features
Special sampling, clearly biased, but it can be biased on both sides
Large portions of production poorly sampled, in particular the 3000 batch that went in 45
We made the following postprocessing
Removed second and third quenches (we had many of them)
We remove the magnet that have been disassembled (13 out of 58 tested ATC)
We smooth the nonuniform sampling in time
After this non trivial massage, we see
ATC data can be compatible with HC data to 6.5 TeV (3000 series at the limit)
ATC data in line with the Gaussian extrapolation for 2000 and 3000
ATC data suggest to take a 20% probability for the 1000 series in going to 7 TeV

9. CONCLUSIONS
Revised estimate
Plus the second quenches - add ¼ to 1/3?