Transcript minutes_fullx - Indico

30kA Energy Extraction
Semi-Conductor Switch
PRR: 29-02-2016
SM18 - CLUSTER D
Summary of questions and
suggestions
Prepared by Mateusz Bednarek
TE-MPE-EE
Welcome and Introduction Talk
Time: 10’
Speaker: Knud Dahlerup-Petersen (CERN)
• Presentation of review members
• EE switch overview at CERN
• Planning
• No questions/suggestions yet…
30 kA General Layout - Basic Concepts
Time: 30’
Speaker: Gert-Jan Coelingh (CERN)
• Davide asks when is the 1ms time coming from and where is the difficult/easy
border.
• Marta says that it could be possible to go for 10 ms as the QPS has the evaluation time of
10 ms.
• Gert – it is not necessarily much more complicated to design this switch, each solution has
its + and -.
• Gilles: Why Infineon?
• Infineon IGBT because of a low forward voltage
• Davide: Marta, do you have some experience with failures of actual switches?
• Marta – thyristor based switch did not give any failures, no magnet burnt, no special
maintenance
• Mechanical was dismantled, slow reaction time, Marta prefers Thyristor based switches.
• Andrzej: in late 90s we had a driving electronics failure,
• Knud: 20 years, after several 1000 cycles, inductors had to be exchanged due to lorentz
forces.
• Mechanical needs more maintenance.
• Davide: indeed it’s good to go with a new development, at one point we will really need
these fast switches, the question is if it is the right moment to do this development.
• Gert: we should have discussed this 3 yr ago…
7.5 kA and 30 kA; Topology - Critical Design Issues
Time: 30’
•
IGBT/thyristor – big difference between the reliability of the switching process
•
•
•
•
•
By the Vce long term evolution – in case of aged IGBT, we change the module preventively.
Davide: What if one module does not open – how do you avoid this situation
•
•
•
•
•
Alexandr: IGBT is a fail-safe device – if contact /connection to the gate is lost – it opens.
Alexandr: Thyristor needs to be opened ACTIVELY – in case of a bad contact it will remain closed!
Gert – the key point of EE is the trade off between availability/reliability
IGCT has a driver built in – there is not much gain In terms of fail safe design.
Marta: how do you trace the ageing
•
•
Speaker: Alexandre Erokhin (AGH University of Science and Technology (PL))
Redundancy is only done at the level of controls, not the power electronics.
Fuse will be the backup for the power electronics.
Failure is unlikely of the power electronics, more likely of controls
Overcurrent will not kill the IGBT in this device
Tony: IGBT fails in short circuit mode
•
•
•
Short circuit mode is not controllable, acts as a fuse, but in a timescale of many ms… the time value is not guaranteed
Fuse would solve it, but we need an active fuse (pyro) – ordinary fuse is too slow
Gert: magnet is self protected below 10-15 kA.
•
•
•
•
•
We can put 2 systems in series, but we pay heavy kW (extra ~40) at 30 kA. It’s just during the test though
Marta: in the next 10 yrs we will go to max. 23 kA
Gert: We can put more systems in ||, this allows for a standard 900 A fuse that will be OK. Still fuses dissipate significant Joule heat…
Gert: This system will open about 500 times per year, this makes a small number of commutations
Arend: We have always been using 2 switching elements in series, this would be the first one where we don’t do it.
•
•
At 20kA the 2kA fuse will open in 10-20ms
Gilles: We use the IGBT in a mode for which it is not designed. Don’t be based on datasheets… This is just a marketing info. Did
you benchmark other manufacturers? Infineon is not fair when talking about the temperatures – different reference points are
used, comparison based on a datasheet is not reliable.
•
•
Mitsubishi are technically very good, they are conservative
Aleksandr: Mitsubishi have bad features, ABB make old generation IGBTs, Infineon: open module measurements and Vce measurements
correspond to the datasheet
Electronics for Control & Interlocks
Time: 30’
Speaker: Arend Dinius (CERN)
• Tony: output protection on 15V power supply is needed. Concept drivers are
fragile. Put a TVS…
• Piyush: Are heat sinks grounded? -> yes
• Is there a central support at cern for NI devices? -> yes
• Marta: temperature of water should be monitored
• Gilles: ageing process seems to be slightly confused.
• Vce should have a trend…
• Why other drivers?
• On concept drivers: Breakages, no reasonable support, feedback of the status is not suitable
• Gilles: Possibility of our own driver, up to many W it is not a problem to have a twisted pair
• Aleksandr doesn’t like simple solutions on drivers
• Can be quickly changed so not a big issue
• Gilles – the other manufacturer (Amantus?) is not good for IGBTs, the feedback is good, but driver itself
is not good, you may find yourself bound to this manufacturer, which should be avoided in this kind of
development
• Horst: Vce is measured on all IGBTs, they are identical, why do you measure them?
• We measure in fact the Vce on the wafer. They are not the same, it differs by connections
and shunts.
• We have a negative feedback for balancing the current in IGBTs
Software and Signals Conditioning
Time: 30’
Speaker: Mateusz Jakub Bednarek (CERN)
• Marta: is the matrix expected to be delivered by SM18?
•
No, it’s our internal matrix
• Piyush: how do you set the pre/post trigger? Do you capture the trigger timestamp?
•
We capture the end timestamp of the buffer. Temperature, DIO, AIN buffers are internally synchronised
• What is the accuracy of the timestamp?
•
We use NTP => few ms, but it doesn’t really matter
• Andrzej: What is the bandwidth of the AD215?
•
~150 kHz
• Piyush: Why do you need the external watchdog? Should it provoke FPA or SPA?
•
Pal: it will indeed most likely provoke SPA
• Piyush: PM data buffer length and trigger location:
•
We read 1s with the trigger in the middle. We can change the trigger location. We may envisage increasing the length
as we have enough memory left. It’s just a question of defining of what is really needed.
• Alexandr: Fast transients are measured by scopes during the prototype tests. We don’t need them for the PM
• Offline from Tony:
•
•
•
•
•
Simulation would be useful to test the behaviour of the FPGA program
Simulation has to be done by another person than the program!
Tester will be nice, but not necessarily sufficient.
Good subject for a tech. student?
External code review would be useful
Mechanics, Rack, support and Integration within SM18 Facility
Time: 20’
Speaker: Gert-Jan Coelingh (CERN)
• Are busbars connected with proper isolators to the rack
structure?
• Yes
General water system and calculations (Cooling & Thermal)
Time: 20’
•
Piyush: Good to monitor the flow for future long term analysis
•
•
First we had a failure due to evacuation pipe with the stagnant water, then we insulated the pipe and the HV test was OK
Piyush: What is the conductivity of water?
•
•
We are already looking into that, it would be a bonus.
Piyush: Hipot?
•
•
Speaker: Stephen Pemberton (CERN)
It’s good enough.
Marta: SM18 foresees 200 l/min budget.
•
It’s enough, we are just right.
•
Marta: If you go away from Elettas, take into account the maintenanceof the new device
•
Davide: 20-25 bar test is possibly not enough
•
Gilles: hard to find a flowmeter insensitive to the magnetic field, fulfilling all other requirements, Gilles would like to replace
his Elettas, but he couldn’t find anything better at the end.
•
•
•
•
•
•
In the prototype we will keep investigating better solutions
For SM18 we may need to stay with Elettas
Marta: Don’t put something untested…
Davide: flow interlocks are needed if the thermal runaways are too fast to be detected with temperature sensors
Davide: UK company Apollo does nice devices: to be investigated
Horst: do you use Constaflow devices?
•
•
No.
The water has to be very clean to avoid failures.
•
Knud: it is strange that there is no automatic water de-gas valve.
•
Davide: what is the delta T
•
•
•
delta T is max 10 deg, but it is in practice 5 deg.
It makes some 70 kW! It’s a lot!
It’s not continuous though.
Dump Resistor: Requirements and Design
Time: 20’
Speaker: Szymon Michniuk (AGH University of Science and Technology (PL))
From kW to MJ.
• Andrzej: Is the cooling forced?
• Szymon: No, it’s natural convection
• Possibly the position of thermostats needs to be changed? Or the threshold lowered.
• Marta: fast recovery is not needed: the cryo will anyway be slower. The data analysis will
take time. The max number of quenches per day is around 2.
• With these dimensions even 3 quenches in a row can be done.
• Piyush: what is the thermal coefficient of R?
• Szymon: 2 % over 100 deg.: it is not a problem, we can assume that it is constant
• Andrzej: where is the hotspot?
• Szymon: In the middle of the disk, the hottest is the middle disk.
• Pt100 are at the connection flags
• There is some overrating of the resistor.
• Knud: will the SM18 operation do the R crosscheck at low current?
• Normally yes.
• Marta: The proposed solution is nice. A documentation of configurations is needed. The GND
connection is very important! Shouldn’t be forgotten or unbalanced…
• Piyush: The middle of the R will be grounded. Will you monitor the earth current?
Quality Control, Test Program and Results
Time: 30’
Speaker: Alexandre Erokhin (AGH University of Science and Technology (PL))
Mainly the results of tests
•
Marta: Are the busbars silver plated? => yes
•
Marta: which is the company that made the busbars?
•
Finnish company, some works were outsourced and the quality was not good
•
Fittings leaked, possibly damaged during the transport: for the future: secure packages and fully test them.
•
Marta: will epoxy coating cause overheating?
•
•
Piyush: will you be checking the V-drop class of you IGBTs during the production of the series?
•
•
No, the busbars run fairly cold, thin, hard epoxy won’t make things worse.
Yes, we will measure them and match them
Davide: Are the tests on a real SC magnet foreseen?
•
•
•
We have a large dummy load, we don’t need a real SC magnet
Everything that is important happens within the first ms
Still would be good to connect to a real magnet, but it is not obvious where to do it and how to ensure the infrastructure. It will be done, but
later. At the moment no need. The opening is identical for normal conducting. magnets. LI2 must simply be much larger than CU2.
•
Gilles: indeed it’s very hard to make a real measurement of the wafer. The wafer can be used as a temp sensor. The wafer has
to be characterised first. You get the temperature of the part that you really need.
•
Gilles: Why do you use 1700 V IGBT? 1200 V is lower Vce, cheaper.
•
•
•
•
Gilles: Why do you need the shunt? Especially if you select Vce class of IGBTs.
•
•
We need to re-discuss it, we were conservative at the time.
It’s to keep some margin.
Piyush: 1700 V gives more flexibility.
We will decide on it. Normally we might remove it. It helps slightly with the current sharing.
Piyush: Signal conditioning calibration: verivolt may be indeed a good idea.
Final discussion
• Marta: is June deadline maintained?
• Depends on the industry once the orders go out.
• Marta: Not fully redundant design of the EE, is it OK or not?
• Piyush had no problems yet with his 6 IGBTs. We were totally independent
and we went in the same direction.
• Andrzej: it is not a fully non-redundant topology: there is still some R in the
circuit.
• Marta: should we launch a development of the backup system?
• Andrzej: make a risk analysis and tell us.
• Davide: in the worst case scenario will we destroy only the magnet or much
more? We need to see if it is worth making a heavy fully redundant system.
• Marta: we need to build the test station in a way that people that will come in
10-20 years time from now will not burn their fingers with our ‘trap’ system
that is not fully safe. They may not be aware of that. They may assume the
the EE system must be reliable if it is called ‘EE’.
• Gert: For 2-3 years the cluster D will not have any back-up, later it is more
likely that it gets it.
• Piyush: What really kills IGBTs are HV spikes. Right away. If you
master this, your device will be very stable.
• You need to monitor transients. That’s what is important.