Multi-Objective Yield Optimization for Electrical Machines Using Gaussian Processes to Learn Faulty Design

This work deals with the design optimization of electrical machines under the consideration of manufacturing uncertainties. In order to efficiently quantify the uncertainty, a hybrid Gauss-Process regression (GPR) model is employed. In contrast to classic Kriging or Bayesian optimization approaches, we train a GPR surrogate for the performance feature specifications, not for the objective function. A multi-objective optimization problem is formulated, maximizing simultaneously the reliability, i.e., the yield, and further performance objectives, e.g., the costs. A permanent magnet synchronous machine is modeled and simulated in commercial finite element simulation software. Four approaches for solving the multi-objective optimization problem are described and numerically compared, namely:
$\varepsilon$
-constraint scalarization, weighted sum scalarization, a multi-start weighted sum approach and a genetic algorithm. We show that the efficiency gain thanks to our hybrid GPR model enables even computationally heavy multi-objective optimization for real-world applications.