Automatic Differentiation for sensitivity calculation in electromagnetism: Application for optimization of a linear actuator

Automatic Differentiation (AD) is introduced as a powerful technique to compute derivatives of functions given in the form of computer programs in high level programming languages such FORTRAN, C or C++. The paper applies AD to compute error-free gradients of sizing models describing electromagnetic devices. Then, the obtained gradients are exploited by optimization algorithms. The goal is to find the device geometry that minimizes an objective function by respecting some additional constrained parameters and performances. When algorithms are used for model formulation, evaluating their derivatives becomes a complicated task. The paper proposes a model of a linear actuator dealing with implicit equations solved by numerical methods.