An HPC-Based Data-Driven Process for Design Exploration and Optimization of Motor Drives

A simulation-based data-driven design process for knowledge extraction and optimizing the size and the parameters of an integrated motor drive is proposed. With the importance of simultaneously optimizing the motor and inverter in a variable-speed application, this work provides a transition from the classical design process by introducing and investigating a data-driven design process. In addition to outlining its implementation, several practical issues associated with using a high-performance computing service are addressed, such as utilization factor and speedup ratio. In this paper, a case study of an interior permanent magnet motor coupled with a two-level inverter modulated by space vectors is simulated with the help of MATLAB/Simulink, MagNet, ThermNet and Microsoft Azure. Around 36000 motor-drive models with different geometries were created and analyzed while incorporating the electromagnetic-thermal behavior of the motor components and inverter switches. Following the design sampling and exploration, valuable knowledge is extracted to aid future procedures. Finally, a neural network model fitted to more than 20 objectives is used to optimize the motor-drive operation in transient, rated and flux weakening modes using an evolutionary optimization algorithm. With the aid of the proposed approach, existing design exploration and optimization tools can be converted to an automatic and informative application-oriented design package.