A statistical model for the design of rotary HTS flux pumps based on deep-learning neuron network

Rotory high temperature superconducting (HTS) flux pumps can consistently generate a DC voltage by rotating magnets over superconducting tapes, and thus energize the circuit if a closed loop is formed. The voltage output is a crucial factor to reflect the performance of such an HTS flux pump, which is determined by a set of design specifications, and some of them have been investigated extensively in the current literature. However, no work has been done yet to study the HTS dynamo output voltage by efficiently integrating all the design parameters together. In this paper, a well-trained deep-learning neuron network (DNN) with back-propagation algorithms has been put forward and validated. The proposed DNN is capable of quantifying the output voltage of an HTS dynamo instantly with an overall accuracy of approximately 98% with respect to the simulated values with all design parameters explicitly specified. The model possesses a powerful ability to characterize the output behavior of HTS dynamos by considering multiple design parameters, e.g., airgap, superconductor tape width, operating frequency, remanent flux density, rotor radius, and permanent magnet width, which have covered all the typical design considerations. The output characteristics of an HTS dynamo against each of the design parameters have been successfully demonstrated using this model. Compared to conventional time-consuming finite element method (FEM) based numerical models, the proposed DNN model has the advantages of automatic learning, fast computation, as well as strong programmability. Therefore the DNN model can greatly facilitate the design and optimization process for HTS dynamos. An executable application has been developed accordingly based on the DNN model, which is believed to provide a useful tool for learners and designers of HTS dynamos.