Design of coaxial coils using hybrid machine learning

Area of Science:

• Quantum Physics
• Experimental Physics
• Computational Physics

Background:

• Uniform magnetic fields are critical for quantum experiments.
• Existing analytical coil design methods face practical limitations, such as placement constraints within magnetic shields.
• These limitations hinder the direct application of conventional designs in many quantum setups.

Purpose of the Study:

• To develop a novel coil design method for quantum experiments.
• To address practical constraints in coil placement and magnetic field generation.
• To leverage hybrid machine learning for efficient and robust coil optimization.

Main Methods:

• Developed a hybrid machine learning algorithm combining an artificial neural network and a differential evolution (DE) learner.
• Employed the DE learner to assist the machine learner in optimizing coil configurations.
• Simulated and experimentally validated the designed coil systems.

Main Results:

• Numerical simulations demonstrated the design of coaxial coil systems with relative field inhomogeneity of approximately 10^-6 in a 25 mm central region.
• An experimentally constructed coil system achieved a field inhomogeneity of 0.069%, primarily limited by machining precision.
• The hybrid machine learning approach proved more efficient and robust than single DE learner methods and analytical proposals.

Conclusions:

• The developed hybrid machine learning coil design method effectively overcomes practical limitations in quantum experiments.
• The method offers high-precision uniform field generation, crucial for advancing quantum technologies.
• This approach demonstrates convenience and broad potential for diverse quantum experimental applications.