NEURAL NETWORKS FOR MODELING AND CONTROL OF PARTICLE ACCELERATORS - - PowerPoint PPT Presentation

▶
neural networks for modeling and control of particle
SMART_READER_LITE
LIVE PREVIEW

NEURAL NETWORKS FOR MODELING AND CONTROL OF PARTICLE ACCELERATORS - - PowerPoint PPT Presentation

Feb 25, 2024 281 likes •639 views

NEURAL NETWORKS FOR MODELING AND CONTROL OF PARTICLE ACCELERATORS Auralee Edelen 2016-02-02 Advisors: Sandra Biedron and Stephen Milton Published Work Edelen, A. et al. Neural Networks for Modeling and Control of Particle Accelerators.

slide-1
SLIDE 1

NEURAL NETWORKS FOR MODELING AND CONTROL OF PARTICLE ACCELERATORS

Auralee Edelen

2016-02-02 Advisors: Sandra Biedron and Stephen Milton

slide-2
SLIDE 2

Published Work

  • Edelen, A. et al. Neural Networks for Modeling and Control of Particle
  • Accelerators. Submitted to IEEE Transactions in Nuclear Science, Jan. 2015.
  • Edelen, A., et al. Initial Experimental Results of a Machine Learning-Based

Temperature Control System for an RF Gun. Paper for the 6th International Particle Accelerator Conference (IPAC), Richmond, VA, May 3-8, 2015.

  • Morin, A., et al. Trajectory Response Studies at the Jefferson Laboratory

Energy Recovery Linac and Free Electron Laser. Paper for the 16th Annual Directed Energy Symposium, Huntsville, AL, March 10-14, 2014.

slide-3
SLIDE 3

Control Challenges

SLAC JLAB Fermilab LBNL Visualization Group Fermilab www.tka-architects.com

slide-4
SLIDE 4

Inspiration from Operators

Model learning Planning Diagnostic Analysis Optimization Prediction Learning Control

Control Room Photo: Reidar Hahn, FNAL

slide-5
SLIDE 5

Gradient descent Conjugate gradient Newton method Quasi-Newton methods Simulated annealing Evolutionary algorithms Swarm intelligence Machine Learning Mathematical Optimization Computational Statistics Artificial Intelligence Model-free Methods Learning Theory Supervised Learning Unsupervised Learning Reinforcement Learning Regression Classification Clustering Dimensionality reduction Biological Sciences and Psychology (inspiration!) Reactive search optimization Fuzzy Logic Model-based Methods

A trip to the zoo…

Decision Trees Neural Networks Support Vector Machines Ensemble Methods ICA, PCA Intelligent Control Adaptive Control Nonlinear Control Optimal Control Robust Control System Identification Stochastic Control

slide-6
SLIDE 6

In general: greater theoretical understanding + increased computational capability + advantageous co-developments in related fields + feedback from a wider variety of application attempts

à greater overall technological maturity

But still difficult in the context of nonlinear control à à we need R&D!

Many Failures Early On à So Why Try Again Now?

  • J. Schmidhuber

Shutterstock IBM, ANL

Google

  • B. Rhoads, UCSB
slide-7
SLIDE 7

Let’s develop and test some AI-based solutions for control problems in accelerators!

  • Explore the tools and techniques
  • Examine some real-world problems, focusing on process control
  • Need to test on an actual machine; not just in simulation
  • Have at it!

Central Focus:

slide-8
SLIDE 8

Some Tools

slide-9
SLIDE 9

Neural Networks

  • What are they
  • How do they learn?
  • When are they useful?
  • What are the disadvantages?

à How can we use these things in particle accelerators?

x1 x2 xn

f

w1 w2 wn

y

a neuron (“node”)

. . .

xn x1 x2

a neural network

. . .

slide-10
SLIDE 10

Learning Paradigms

slide-11
SLIDE 11

Model Predictive Control

Basic concept: use a predictive model to assess the outcome of possible future actions

slide-12
SLIDE 12

Model Predictive Control

Plant Model Plant Cost Function Constraints Solver um(k – 1)… um(k – Nm) Optimization of Controlled Variable Trajectories ucv (k) yr (k)… yr (k + Np) ucv (k)… ucv (k + Nc – 1) yp (k)… yp(k + Np) Reference Trajectory Measured Variables Future Inputs Predicted Outputs

Nm previous measurements Np future time steps predicted Nc future time steps controlled

!!! !! ! + ! − !! ! + !

!! !! !!!

(output variable targets) !!,! !! ! + ! − !!,!"# ! + !

! !!!! !!! !!" !!!

! (controllable variable targets) !∆!,! !! ! + ! − !! ! + ! − 1

! !!!! !!! !!" !!!

(movement size)

slide-13
SLIDE 13

Reinforcement Learning

slide-14
SLIDE 14

Real-World Problems

slide-15
SLIDE 15

At Fermilab…

High-intensity RFQ for the PIP-II Injector Experiment (PXIE)

— Time delays — Large, dynamic frequency response — Tight tolerances — Coupling — Recursive behavior — Three controllable parameters

Photo: J. Steimel Photo: P. Stabile

RF electron gun at the Fermilab Accelerator Science and Technology (FAST) facility

— Long, variable time delays — Tight tolerances — Recursive behavior — Two controllable parameters

slide-16
SLIDE 16

At Fermilab…

Photo: E. Harms

RF electron gun at the Fermilab Accelerator Science and Technology (FAST) facility

— Long, variable time delays — Tight tolerances — Recursive behavior — Two controllable parameters

Photo: P. Stabile

!!Type Photoinjector !!Number!of!cells 1½ !!RF!Mode TM010,π !!Loaded!Q ~11,700 !!RF!Frequency 1.3 GHz !!Frequency!Shift 23 kHz/°C !!Macropulse!Duration 1 ms !!Repetition!Rate 1−5 Hz !!Bunch!Frequency! 3 MHz !!Design!Gradient 40−45 MV/m !!Power!Source 5 MW Klystron

FAST!RF!Gun!Parameters

!!Gun!Parameters !!Nominal!Operating!Parameters

slide-17
SLIDE 17

Gun Water Cooling System

slide-18
SLIDE 18

Time%Elapsed%[minutes]% Temperature%[°C]%

%

impulse response from a 20-second decrease in the heater power setting from 7 kW to 2.5 kW

Water Temperatures — Open Loop

slide-19
SLIDE 19

Time%Elapsed%[minutes]% Temperature%[°C]%

%

1-°C step change in temperature set point à Oscillation is NOT due to poorly tune PI gains!

Existing feed-forward/PI Control of the Gun Temperature

slide-20
SLIDE 20

Initial Solution

  • Neural network model
  • Model predictive control

à Serves as a simple benchmark for future studies

slide-21
SLIDE 21

model predicted next value of T02 T01 ([t-d1 ], . . ., [t-d1 -n]) TOUT ([t - d2 ], . . ., [t - d2- n2]) T01 ([t-d1 ], . . ., [t-d1 -n]) valve ([t - d3 ], . . ., [t - d3 - n3 ]) T01 ([t - d1 ], . . ., [t - d1 - n1]) heater ([t - d4 ], . . ., [t - d4 - n4])

d - delay tme n - number of previous samples

Neural Network Modeling

slide-22
SLIDE 22

Benchmark Controller

slide-23
SLIDE 23

MPC Benchmark Controller

Note: different horizontal and vertical scale than for PI loop ~5x faster settling time no more overshoot still needs work… (esp. T02-to-TCAV model)

slide-24
SLIDE 24

Time%Elapsed%[minutes]% % Requested%Heater%Power%[kW]% Time%Elapsed%[minutes]% % Requested%Control%Valve%Posi9on%[%%open]%

MPC Benchmark Controller: Actions

Requested by Controller Actual Read-backs

slide-25
SLIDE 25

Of course, there’s more to the story….

0.5 1 1.5 2 0.5 1 1.5 2 2.5 3

Average RF Power [kW]

TOUT−TIN TCAV−TIN TOUT−TIN expected

Average'RF'Power'[kW]' Steady'State'Temperature'Difference'[°C]'

!

!""# = !!"#[℃]!!!"[℃] !×! !"#$![!"#] !"#$%!!""#$%&!!"#"$%&'! !"#!℃

!"

!

!""# = ! !" ≈ ! !"!"#.

slide-26
SLIDE 26

FAST Next Steps

  • Neural network model predictive control
  • Extension to direct resonance control
  • Reinforcement learning control
slide-27
SLIDE 27

At Fermilab…

High-intensity RFQ for the PIP-II Injector Experiment (PXIE) RF electron gun at the Fermilab Accelerator Science and Technology (FAST) Facility

— Long, variable time delays — Tight tolerances — Recursive behavior — Two controllable parameters — Time delays — Large, dynamic frequency response — Tight tolerances — Coupling — Recursive behavior — Three controllable parameters

Photo: J. Steimel Photo: P. Stabile

slide-28
SLIDE 28

PXIE RFQ

High-intensity RFQ for the PIP-II Injector Experiment (PXIE)

— Time delays — Large, dynamic frequency response — Tight tolerances — Coupling — Recursive behavior — Three controllable parameters

!!RF!frequency 162.5 MHz !!Q-factor ~13,900 !!Loaded!Q ~7,000 !!Physical!Length 4.45 m (2.4 wavelengths) !!Vane-to-Vane!Voltage 60 kV !!Estimated!Power!Dissipation < 100 kW !!RF!Repetition!Rate pulsed − CW !!Current 0.5 − 10 mA (nominal 5 mA) !!Input!Energy 30 keV !!Output!Energy 2.1 MeV

!!Beam!Parameters

PXIE!RFQ!Parameters

!!RFQ!Design!Parameters

Constructed by LBNL

Photo: J. Steimel

slide-29
SLIDE 29

PXIE RFQ

4"inner"channels" 8"outer"channels"

Vane channels Wall channels

All images courtesy LBNL, D. Li, A. Lambert

3-kHz max. freq. shift 0.1-°C water stabilization

slide-30
SLIDE 30

Expected Frequency Response

ANSYS simulation data courtesy A. Lambert, LBNL

slide-31
SLIDE 31

Water System

slide-32
SLIDE 32

ACNET&user&interface&

&

Opera&onal*Mode*Request*

*

Present*Opera&onal*Mode* Error*Read4backs* Heartbeat* PLC&

*

Secondary*Temperature*Sensors* Flow*Valve*Read4backs* Pressure*and*Flow*Readings*

*

Flow*Valve*SeCngs* PLC*Cntrl.**vs.*Resonance*Cntrl.* Cryo5con&

*

Cri&cal*Temperature*Sensors* LLRF&

&

Resonant*Frequency*Devia&on** RF*Amplitude,*Rep.*Rate,*Pulse* Length* CW*vs.*Pulsed*Indicator* Other*State*Informa&on*(TBD)* * PLL*(“SEL”)*vs.*GDR*switch* (Op&onal)*feed4forward*amplitude*

ACNET& Erlang& CLX&Machine& Resonance&Controller& Data*Reading*and* Organiza&on* Opera&onal*Module* Selec&on* Control*Calcula&on*in* Corresponding*Module* Error4checking* Data*Organiza&on*and* Sending* UDP&& Protocol&Defini@on& (from&protocol&file&and& protocol&compiler)&

to#resonance#controller# from#resonance#controller#

Resonance&Control&System&Architecture&and&Component&Interfaces&

slide-33
SLIDE 33

Resonance Controller State Flow

Requested state S0: No control S1: PLC has control S2: PID temperature control (user set point) S3: PID temperature control (default set point) S4: MPC temperature control (user set point) S5: MPC temperature control (default set point) S6: PID resonance control S7: Resonance control S8: Startup with RF S9: Experimental Is RF On/Off? Resonance control anticipatory state Resonance control with LLRF in SEL No control PLC control Temperature Control

  • Water system settled?
  • f0 within tolerance of 162.5 MHz?

Resonance control with LLRF in GDR RF still on? count1++ count1 > threshold? close flow control valves count2++ count2 > threshold? Resonance control anticipatory state Resonance control w/RF settings RF on? Settled? S6,S7 Off On GDR conditions met Check GDR conditions GDR conditions not met S2,S3,S4,S5 S0 S1 no no yes yes no yes yes no yes no S8

slide-34
SLIDE 34

Initial Frequency Response Measurements

initial transient steady state Time 7.2 min 115.4 min ∆fres 46.13 kHz 21.15 kHz TT101 22.280 oC 25.796 oC 25.649 oC TT102 31.489 oC 28.126 oC 26.141 oC TT103 30.941 oC 27.639 oC 26.284 oC flow path time delay [s] TT101 → TT103 1.0 TT103 → TT102 17.0

slide-35
SLIDE 35

Next Steps for PXIE

  • Finish and test control framework
  • Characterize and debug water system
  • Characterize RFQ and water system under RF power
  • Implement basic controllers for contracted work
  • Design and test neural network controllers