HiFiMagnet robust optimal operation and control

Optimal-operation application: determine coil-group currents and cooling controls that reach a target magnetic field with minimum power and thermal stress while satisfying temperature, voltage, power, total-current and actuator constraints.

Quick Facts

Type

Extended Mini App

Status

Planned

Work Packages

WP1, WP2, WP3, WP5, WP6

Frameworks

Feel++, PETSc/TAO, HPDDM, Sage-HPC

1. Overview

Optimal-operation application: determine coil-group currents and cooling controls that reach a target magnetic field with minimum power and thermal stress while satisfying temperature, voltage, power, total-current and actuator constraints.

1.1. Repository

github.com/numpex/apps-feelpp

2. Technical Stack

2.1. Frameworks Used

  • Feel++

  • PETSc/TAO

  • HPDDM

  • Sage-HPC

2.2. Parallel Frameworks

  • MPI

  • Kokkos

  • GPU - HIP

3. Methods & Algorithms

3.1. WP1: Discretization

  • nonlinear thermo-electric finite elements

  • coupled PDE-scalar formulation

  • integral total-current boundary condition

3.2. WP2: Model Order Reduction & SciML

  • ROM

  • hyper-reduction

  • surrogate-assisted optimization

  • FOM/ROM switching

3.3. WP3: Solvers

  • nonlinear block solvers

  • repeated state/adjoint solves

  • PETSc/HPDDM preconditioning

3.4. WP5: Optimization

  • PDE-constrained optimal operation and control: target-field tracking

  • power and thermal objectives; adjoint gradients; bound/general constraints; multiple coil-current and cooling controls; deterministic and robust formulations

3.5. WP6: Uncertainty Quantification

  • uncertain material and cooling parameters

  • stochastic and robust optimization

  • expectation and CVaR objectives

4. Data Flow

4.1. Inputs

  • Gmsh

  • JSON config

  • nonlinear material laws

  • target field

  • current/cooling bounds

  • uncertainty scenarios

4.2. Outputs

  • Optimal control vector z* (coil-group currents and cooling controls)

  • optimal state/adjoint fields

  • magnetic field

  • power and peak-temperature predictions

  • voltage and constraint margins

  • KKT and objective histories

  • Pareto/robustness reports

5. Benchmarking

5.1. Metrics

  • Taylor Gradient Test

  • Target Field Error

  • Electrical Power

  • Peak Temperature

  • Constraint Violation

  • Kkt Residual

  • Time To Optimum

  • Pde Solves To Solution

  • Tao Iterations

  • Strong Scalability

  • Weak Scalability

  • Scenario Throughput

5.2. Benchmark Scope

  • Multi Node

  • Multi Gpu

  • Full System

  • Solver Scaling

  • Ensemble Mode

6. Timeline

Milestone Date

Specification Due

N/A

Prototype Due

N/A

7. Team

Partners: Unistra, LNCMI/CNRS

Responsible: C. Prud’homme; V. Chabannes

WP7 Engineer: Javier Cladellas (UNISTRA)

8. Notes

Mathematical specification. Control z=(I1,…,Ing, cooling flow rates or heat-transfer coefficients); for each circuit the shared forward operator imposes total current Ig and computes ΔVg. A first objective is Jctrl(z)=1/2||CB B(y,z)-Btarget||² + βP(y,z) + γR(z), subject to F(y;z)=0, Tmax(y)≤Tlim, ΔVg(y)≤Vlim,g, P(y,z)≤Plim and zmin≤z≤zmax. WP6 extension: minimize Eξ[Jctrl(z,ξ)] plus a CVaR penalty on thermal/voltage risk. Outputs are z* and engineering performance/constraint margins—not calibrated material parameters. TAO is the deterministic reference; Sage-HPC can compare surrogate-assisted, Bayesian and learning-based strategies on the same contract.