Transcript Siemens Openlab Workshop-5.Nov.2015x

CERN openlab
technical workshop
›
5th Nov 2015
Siemens collaboration
Filippo Tilaro,
EN/ICE CERN
Analytical framework for the
CERN control system
Scalable and fault-tolerant !!!
Data Analysis Framework
Data Processing Modules
MOON
Supervision
Analysis
layer
FFT
memory and
configuration
Machine
Learning
Neural
(Monitoring)
Network
+600 Application
Services
(R)
CEP
Expert
(Java)
TN
DIM/CMW
Patterns
Process
layer
PLCs
(LabView)
Visualisation
OPC
(WatchCAT)
• + 2000 PLCs
• + 650 FECs
Data collection & feedback
High Voltage
Fieldbus
Field
layer
5th Nov 2015
Sensors
&
Actuators
Siemens CERN openlab
Historical
Data +50M channels
2
Analytical areas of interest
›
Online monitoring
 Continuous service to analyse the system status
and inform operators in case of fault detection
 Oscillation analysis for cryogenics valves
›
Fault diagnosis
 Anomaly
bysystem
sensors
data
mining
“Forensics”detection
analysis of
faults
that
have
already happened in the past. In some cases rootcause analysis
›
5th Nov 2015
Engineering design
 Analysis of historical data to draw conclusions
aboutsupervision
system behaviours which could be helpful to
 PID
improve / optimize the system under analysis
Siemens CERN openlab
3
LHC Cryogenics system and valve oscillation
›
›
›
›
›
Keep magnets under Superconductivity
Liquid helium bathing the LHC’s magnets
cooled down to 1.9K
34000 physical instrumentations and
channels
 12136 AI, 4856 AO,4536 DI,1568 DO
 4000 analogical control loops
Wrong valve oscillation
More than 120 PLCs
 Siemens S7-416-2DP
 30000 conceptual objects/parameters
Valve oscillation can affect:




Control system stability
Maintenance (stress on the equipment)
Performances & Safety
Increase in the data load
5th Nov 2015
Siemens CERN openlab
4
Oscillation analysis for cryogenics valve
›
›
High computational cost
 Time window length
 Number of channels
 Filters degree
Running system:
 ~3000 channels under analysis
daily
 Algorithm outcome validated by
system experts (in terms of false positives)
5th Nov 2015
Siemens CERN openlab
5
PID supervision
›
›
›
›
In collaboration with the University of Valladolid (Ph.D. R. Martinez)
Based on: “Performance monitoring of industrial controllers based on the
predictability of controller behaviour”, R. Ghraizi, E. Martinez, C. de Prada
CERN control systems contains hundreds of thousands of control
loops in operation
v
PID performance has an impact on:
 Process security
 Quality of physics
 Maintenance (stress of the equipment)
w
SP
Controller
u
MV
y
Process
CV
Issues:






Many sources of faults / malfunctions
External disturbances / factors
Bad tuning
Wrong controller type / structure
Slow degradation
System status dependency
5th Nov 2015
Siemens CERN openlab
u
q
w
TC
TT
6
T
Evaluation of PID supervision
›
›
›
Analysis algorithm:
 Based on performance Harris
index and error prediction
 No a priori knowledge about
the system under analysis
Running system:
 More than 3000 control loops
under analysis daily
 Algorithm outcome validated by
system experts (in terms of
false positives)
Bad
High computation cost:




Order of the regression model
Prediction level
Size of the process data history
Time window size
5th Nov 2015
Siemens CERN openlab
Good
7
Integration of data analytical solutions
into Siemens’ Smart Data Technologies
›
›
›
›
New version of WatchCAT
Code name “WatchCAT”
 Extended support for R and
Octave
 Still a prototype version
Plug and play architecture
 R implementation of the
oscillation detection algorithm
 Octave implementation of the
algorithm for evaluation of PID
supervision
Evaluation of the new
version:
 Improved performances and
memory allocation
 Feedback to Siemens
5th Nov 2015




Data Fusion of events & sensors
Complex Event Processing
Automated Learning of fault patterns
Logical Reasoning for Fault
Detection & Isolation
 Fault prediction based on
recognizable patterns
Siemens CERN openlab
8
Cloud-based analytical solutions &
Siemens’ Smart Data Technologies
›
›
Code name “ELVis”
Jupyter & Dockers
Slave
Slave
Master
Slave
 Cloud-based BIG Data Analytics for
Time Series Sensor Data
 Real-Time Stream Processing at
customizable KHz-Rates
 High Performance Online
Visualization in Rich Web-based UI
 Intelligence for Sensor Data Validation
 Job-based Online Data Analysis
5th Nov 2015
Slave
NFS
Scripts
Data




Siemens CERN openlab
Sharing code and results
Distribute computational load
Interactive analysis
Multi-language development
environment
9
CERN & Siemens collaboration :
Status & Next Steps
›
›
›
›
›
›
5th Nov 2015
New analytical algorithms designed, implemented and
integrated into Siemens analytical framework
 Signal oscillation detection
 Evaluation of PID supervision
 Anomaly detection by sensors data mining [on going]
Evaluation of Siemens diagnostic tools
Cloud computing
 ELVis as a Storm based solution
 Jupyter and Docker as a cloud development environment for code
sharing
“Formalizing expert knowledge in order to analyse CERN
control system”, ICALECPS 2015
Continue the integration of CERN specific extensions &
data analysis algorithms / solutions
Extensive deployment of Siemens cloud-based solutions
for Big Data analytics as a Service
Siemens CERN openlab
10
Any Questions
Thank you for your attention!
5th Nov 2015
Siemens CERN openlab
11