The Internet belongs to everyone. Let’s keep it that way.

Protect Net Neutrality
Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Monitoring the LHC magnets

No description
by

matej mertik

on 26 November 2017

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Monitoring the LHC magnets

Monitoring the LHC magnets
1. additional monitoring/security layer - "Watchdog" Layer
2. additional detection layer which can improve maintainance of the LHC
RMSE-based and Grid-based algorithm
Visualization of the Grid-based algorithm
DEEP LEARNING as an EXTENSION
Further steps
Goal of the research
Our approach ...
Modeling LHC magnets voltage time series
Other use cases of Deep Learning in the area of the LHC
real-time anomaly detection and prediction
using advantage of
huge amount of data
generated by the equipment for
better maintenance and protection
of the machine
examine deep learning
algorithms to address the task of
better
understanding anomalies
and
long and short time behavior
of the magnet
monitoring the equipment
associated with LHC
RNN models,
Network compression for low latency
FPGA (Nallatech, p395_mac_ab),
HLS OpenCL,
optimized modules VHDL
Using Deep Learning in LHC
Using DL as additional operating level
for better understanding
anomalies of the electronics

and as a

security layer
Parameters, cost and complexity of an FPGA implementation
Research map
DATA
ALGORITHM
VISUALIZATION
Logging
Service
LSTM, RMSE
Concept
POST Mortem
HiLumi
LSTM + window
GRU + window
HiLumi
GRU + window
Visualization
I
II
III
IV
Building
blocks
for data
exploration
*
M. Wielgosz, A. Skoczeń, M. Mertik, Recurrent Neural Networks for anomaly detection in the Post-Mortem time series of LHC superconducting magnets (Feb 2017). arXiv:1702:00833
M. Wielgosz, A. Skoczeń, M. Mertik, Using LSTM recurrent neural networks for detecting anomalous behavior of LHC superconducting magnets (Nov 2016). arXiv:1611:06241.
M. Mertik, M. Wielgosz, A. Skoczeń, A Conceptual Development of Quench Prediction App build on LSTM and ELQA framework (Oct 2016). arXiv:1610:09201.
L. Barnard, M. Mertik, Usability of visualization libraries for web browsers for use in scientific analysis, International Journal of Computer Applications 121 (1)
Mertik, M., Dahlerup-Petersen, K., Data engineering for the electrical quality assurance of the LHC - a preliminary study. International Journal of Data Mining, Modelling and Management.
Research papers
Thank you for your attention
Disscusion
Prediction one step ahead
Accuracy for different parameters
DL Framework: Keras with Theano backend

Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz with 32GB DDR3 1600MHz memory

Post-Mortem: Nvidia Tesla K80
Setup
Data
Time series used in the study
Resistive voltage extracted from voltage measured on the superconducting LHC magnet Measurement and extraction are performed by dedicated hardware - quench detector

In case of
RMSE-based
approach
data from CERN Accelerator Logging Service (CALS) was used
– sampling period 400ms

In case of
grid-based
approach
data from CERN Post Mortem System (PM System) was used
–sampling period 4ms

Between many sets of LHC magnets the study focuses on the class of magnets with maximal current on the level of
600 A

blue - real, green - predicted
0
0.25
0.5
0.75
1.0
Principal Deep Learning Research Engineer - Cadence Design Systems
Faculty of Computer Science, Electronics and Telecommunications, AGH University of Science and Technology, Krakow, Poland

Alma Mater Europaea Maribor, Web and information technologies, Slovenia
Faculty of Physics and Applied Computer Science, AGH University of Science and Technology, Krakow, Poland
Monitoring the LHC magnets
Maciej Wielgosz, PJASS TE-MPE-EE CERN (2016-2017),
Matej Mertik, SASS TE-MPE-EE CERN (2014-2016)
Andrzej Skoczen, PJASS TE-MPE-EE CERN
DS@HEP 2017
Intelligent, Predictive and Proactive System
Developing Intelligent, Predictive and Proactive System
CERN Control Systems:
Cryogenics
Vacuum
Machine Protection
Power Converters
Examples:
Anomaly detection on Beam Screen
Faulty Cryogenics Valve Detection
Reliability, Availability, Maintainability and Safety (RAMS) studies for the Future Circular Collider (FCC)
[1] Antonio Romero Marín, Manuel Martin Marquez: Fault Detection using Advanced Analytics at CERN's Large Hadron Collider, accessed 27.4.2017 on INDICO CERN
[2] Linda Tsan, CERN Big Data Exploration for FCC, Customers, Engineered Systems, June 2, 2016


Predict faulty valves before they actually fail
Why?
How DL?
Modeling
Large Open source DL community
DL for LHC
Other applications of DL
Further
research

Deep Learning - DL
DL-based solution integration
Minjae Lee, Kyuyeon Hwang, Jinhwan Park, Sungwook Choi, Sungho Shin:
“FPGA-Based Low-Power Speech Recognition with Recurrent Neural Networks”, 2016; [http://arxiv.org/abs/1610.00552 arXiv:1610.00552].

Concept
Concept
Threshold-oriented
lots of anomalies for training

PLATFORM
CPU
CPU + GPU + FPGA
CPU + GPU
CPU + GPU
Window-oriented
small amount of anomalies for testing

M. Mertik, M. Wielgosz, A. Skoczeń: Deep Learning for Engineers (Book in preparation)
Full transcript