Rapid co-kriging based multi-fidelity surrogate assisted performance optimization of a transverse flux PMLSM

A Review of Transverse Flux Machines Topologies and Design

“…In the field of TFMs, the most common approach are 3D simulations, but also surrogate 2D models have been used instead [26,31,63,78,138]. Among the previous references, only in [78,138] have 2D-and 3D-FEM results been compared, showing a good agreement.…”

“…Among the previous references, only in [78,138] have 2D-and 3D-FEM results been compared, showing a good agreement. Authors in [31] propose a multi-fidelity model that combines both 2D and 3D-FEM simulations in order to obtain reasonable accuracy and computation time.…”

“…The operation, design and modeling of rotating, linear and tubular machines is quite similar (e.g., torque ripple in rotating machines versus force ripple in the linear case), that is why this review is mainly focused on rotating machines. Nevertheless, linear [4,9,10,21,[27][28][29][30][31][32][33][34][35][36][37][38][39][40][41] and tubular [11][12][13]33,[42][43][44][45][46][47][48][49][50][51][52][53][54] TFMs have specific applications that are presented at the end of this section. In any case, commercial exploitation of TFMs is currently limited to very few manufacturers [55][56][57].…”

“…Different algorithms have been applied for TFMs optimization, such as genetic algorithm [31,88,93,98,100,102,112] (NSGA-II in [88,93,100]), particle swarm optimization [15,100] (MOPSO in [100]), imperialist competitive algorithm [70] and Hooke-Jeeves algorithm [63]. Before the experimentation, to reduce the number of decision variables the Taguchi method can be applied [100], it consists of four stages [69]: identification of the design variables, develop the experimental matrix and conduct experiments, analysis, analysis of means.…”

“…Taguchi method has been applied in some publications about TFMs [67,69,75,77,100]. For the function approximation, the Kriging method has been used [31,100,112], sometimes with variations [31]. Other Gaussian Process Regression (GPR) algorithm has been applied in [88], GPR is a kind of machine learning method based on Bayesian statistics.…”