Shivanshu Siyanwal

Magnet design · 0.5 T

Bitter Solenoid Simulator (0.5 T)

A real-time simulator for a 0.5 T water-cooled copper Bitter solenoid design, with shim coils and Swiss-roll RF metamaterial models.

Public repository Electromagnetism / Thermal-hydraulics / Optimisation / Numba

Interactive abstract

Play with the idea

This interactive diagram needs JavaScript. The schematic and write-up below explain the same idea.

Layman view

In plain words

Draft — to be verified. Written in plain language from the paper and project notes on this site.

An MRI scanner needs a strong and very steady magnetic field. One way to make a strong field is to push a large electric current around a loop. A Bitter magnet does this with a stack of thick copper plates, cut so that the current circles the central bore, with many small holes through the stack so that cooling water can flow along it.

The difficulty is heat. Doubling the current doubles the field, but it quadruples the heat the copper gives off, so the water has to carry that heat away. The simulator on this page ties together the field, the electricity, the heat, the water flow and the mechanical forces, and searches for the design that uses the least power while staying inside its limits.

A separate layer looks at what an MRI scanner needs on top of the magnet itself: extra coils that even out the field, a check on how steady the field stays, and the radio-frequency signal received with Swiss-roll metamaterials, which are small rolled-up resonant structures.

Expert view

The work, step by step

  1. 01

    Problem

    Design a 0.5 T water-cooled copper Bitter solenoid and evaluate it in real time on modest hardware (a 32 GB Ryzen 5 laptop, Kaggle or Colab), using only Python-3.6-compatible NumPy, SciPy, Numba and Matplotlib APIs that existed before July 2018.

  2. 02

    Approach

    The continuum simulator couples closed-form and exact elliptic-integral field models (checked against an independent straight-segment Biot–Savart engine), Bitter-plate electrics, thermal and hydraulic models of the axial cooling holes (Dittus–Boelter, with a Gnielinski check) and Lorentz-force mechanics. Differential evolution followed by SLSQP minimises electrical plus pump power subject to 0.5 T at the centre, a copper hot spot of at most 85 °C, a supply of at most 8 V, an inner radius of at least 50 mm, at most 100 ppm inhomogeneity, a pressure drop of at most 5 bar and Re ≥ 10⁴. A separate fMRI layer adds Z2/Z4 shim loops, B0 stability and a quasi-static RF SNR model of Swiss-roll metamaterials. A mesoscopic emulation layer adds tolerance and contact-resistance scatter, and particle estimators re-derive the continuum numbers as a cross-check.

  3. 03

    Results

    From the repository's generated results (differential-evolution run, seed 1).

    The optimum has an inner radius of 50.0 mm, an outer radius of 300 mm and a 1.10 m stack. It carries 2576 A at 4.54 V, dissipating 11.7 kW plus 81 W of pump power, with 99.95 ppm peak-to-peak homogeneity over a 30 mm-diameter sphere and a 25.9 °C copper hot spot. The active limits are field homogeneity and size, not cooling.

    Two Z2/Z4 shim-loop pairs bring the homogeneity over the same sphere to 0.26 ppm.

    With the simulator's assumed inputs, the field at the centre varies by about 4.4 ppm over 10 minutes with a current-regulated supply, against about 600 ppm with a voltage-regulated one, so the magnet needs a current-regulated supply and tightly held cooling-water temperature.

    In the quasi-static RF model, a Swiss-roll slab improves SNR over the same coil held off over an air gap, but does not beat a coil placed directly on the tissue.

  4. 04

    Validation

    From the repository's validation suite.

    On the axis, the loop field model matches the closed-form fields of a uniform thick solenoid and of a Bitter (J ∝ 1/r) coil to within 2 × 10⁻¹³ (relative), and it reproduces the long-solenoid limit, B = μ₀nI, to 2 × 10⁻⁶. The independent straight-segment Biot–Savart engine converges at second order (measured order 1.99998, theory 2) and agrees with the thick-solenoid closed form to 1.1 × 10⁻⁴.

    A pytest suite covers analytic limits, convergence, energy balance, the particle emulation and agreement between the browser demo and the Python model.

Schematic

How it works

⊗⊙B₀borewatercurrent around the borefield: Biot–Savartelectrical: plate resistancethermal: Joule heathydraulic: coolant in axial holesmechanical: Lorentz forceoptimiser: differential evolutionthen SLSQP, within the limits
Schematic. Left: a side-on cross-section of a Bitter solenoid. Current runs around the bore through stacked copper plates (⊗ into the page, ⊙ out of it), water is pumped through axial holes, and the field B₀ points along the axis. Right: the five coupled models of the simulator, and the optimiser that works on them. No values are shown.

Contribution

My role

In summer 2018 (May–July) I was an intern in Dr. Ananth Venkatesan's Mesoscopic lab at IISER Mohali, where I worked on a theoretical design for a water-cooled solenoid reaching 0.5 T, studying Bitter and Swiss-roll magnet models.

More threads