Deep Transfer Learning-Based Demagnetization Analysis for Linear Oscillating Actuator Considering Circumferential Segmented Structure

The linear oscillating actuator (LOA) achieves high efficiency and features a simple mechanical structure because it doesn't require the conversion of rotational motion into linear motion. Therefore, the LOA is an appealing option for devices such as compressors, linear pump and automobile active suspension due to its high efficiency and power density. The stability of permanent magnets (PMs) can be impacted by different factors such as temperature, electromagnetic fields, and other external influences. In more severe cases, these factors can result in the occurrence of irreversible demagnetization, causing permanent damage to the magnetic properties of the PM. The irreversible demagnetization of permanent magnets impacts the electromagnetic functionality of the LOA, making it necessary to account for it during the design stage. However, the intricate configuration of the LOA such as divided outer stator and PMs diminishes accuracy of the 2-D finite element analysis (FEA). Despite its high computational cost for calculating accurate demagnetization ratio (DR), 3-D FEA is essential. Thus, we propose a demagnetization analysis based on transfer learning (TL) to reduce the computational cost associated with accurately calculating the 3-D FEA-based DR. This approach takes into account the permeance in the stator core and circumferential leakage flux. With TL, the parameters of pre-trained models learned from a source dataset in different but similar domains are transferred to learn the target dataset and effectively enhance the performance of neural network. The TL is conducted with a substantial dataset from 2-D FEA-based demagnetization ratio (DR) anda limited dataset from 3-D FEA-based DR. TL is a cognitive learning approach that utilizes knowledge acquired from a source task to enhance learning in a related but different target task were compared.