Date of Award
Engineering and Applied Science - Electrical
Abdul Rahman Alsamman
Md Tamjidul Hoque
Replacing a portion of high-energy Permanent Magnets (PMs) with low-energy PMs, generally known as hybrid PM machines, is an effective solution to lower the manufacturing cost in PM machines. However, partial removal of high-energy PMs without proper design adjustments could lower the overall torque capacity and introduces operational expenses. In addition, the hybrid structure requires a coordinated distribution between the two types of PMs to ensure a smooth operation. Such sophisticated design considerations could impose a high computational burden and may not be easily achievable with classical design methods. This dissertation presents a semi-analytical and deep-learning-based design methodology to facilitate design, development and optimization of PM machines. The ultimate goal is to lower the manufacturing cost of PM based electric machine systems, while keeping the operational quality intact. This includes basic performance measures of the machine such as Back electromagnetic force (EMF), power factor, cogging torque and electromagnetic torque. Cogging torque causes major operational setbacks for PM machine operation, particularly in applications where a quiet performance is desired. For this reason, this dissertation presents a heuristic optimization framework to optimize the cogging torque in Surface-mounted PM (SPM) machines consisting of a hybrid magnetic structure (i.e., rare-earth and ferrite magnets). To avoid excessive computational time and volume associated with Finite Element (FE)-based optimization solutions, the analytical approach is paired up with the optimization algorithm to determine the optimal design while FE is utilized for verification and validation purposes. Next, a novel topology of a hybrid PM machine is designed and proposed by coupling FE with deep neural network (DNN) algorithm. Finally, the DNN (prediction model) successfully predicts the machine's performance for any random set of parameters, as confirmed via FE. Then the prediction model is used to optimize the machine performance using a heuristic optimization algorithm.
Poudel, Bikrant, "Machine Learning Based Design Methodology for Electric Machines" (2022). University of New Orleans Theses and Dissertations. 3050.
Available for download on Thursday, December 16, 2027