电工技术学报2025，Vol.40Issue(23)：7449-7461,13.DOI:10.19595/j.cnki.1000-6753.tces.241953
基于增强特征传递结构神经网络的超磁致伸缩电声换能器输出特性分析
Analysis of Output Characteristics of Giant Magnetostrictive Electroacoustic Transducer Based on Enhanced Feature Transfer Structure Neural Network
摘要
Abstract
The giant magnetostrictive electroacoustic transducer(GMT)represents a sophisticated underwater device that utilizes the magnetostrictive effect of rare earth giant magnetostrictive materials(GMM)under alternating magnetic fields to achieve efficient electroacoustic energy conversion.This technology demonstrates significant application potential in underwater sonar systems.However,the dynamic performance of GMT is substantially influenced by complex nonlinear electric-magnetic-mechanical-acoustic coupling effects.Traditional frequency-domain methods often fail to accurately represent the input-output relationships,making precise time-domain characteristic analysis crucial for optimal transducer design.Although the time-domain model has the problems of nonlinearity,strong coupling and low computational efficiency,the time-domain characteristics of GMT can be analyzed quickly with the help of deep learning methods.In this context,this paper proposes an enhanced feature transfer structure(EFTS)based on the time-domain model of GMT. Firstly,based on the Jiles-Atherton hysteresis model of GMM,a time-domain finite element model(FEM)of GMT is established,incorporating the coupling of electromagnetic,mechanical,and acoustic multiple physical fields.The dynamic output characteristics of the transducer are analyzed by the FEM simulation results,and the accuracy of the dynamic model is validated based on existing research.Secondly,to address the issue of long computation time and low efficiency in the GMT FEM model,an EFTS deep learning model is constructed.The neural network is densely connected through a Dense block structure,enabling the reuse of feature information in the channel dimension,thereby accelerating the model's convergence speed.Thirdly,the GMT EFTS deep learning model is trained to achieve rapid analysis of GMT output characteristics under multi-parameter nonlinear coupling,leveraging small-sample data from the FEM model.Comparative analysis demonstrates that the predicted values of TCR by the EFTS neural network align well with the simulation results,with an RMSE of 0.017,indicating that the proposed method can efficiently and accurately obtain the output characteristics of GMT.Finally,the structural parameters of GMT are optimized based on EFTS neural network and particle swarm optimization algorithm.The optimized structural parameters meet the design requirements and can be used as the final design scheme. To further validate the feasibility of the proposed design method,a GMT prototype was developed based on the optimized design scheme,and a lake-based experimental platform was constructed.Experimental results demonstrate that at a target water depth of 60 meters,the transducer exhibits a resonant frequency of 450 Hz and achieves a maximum transmit current response(TCR)of 182.79 dB,accompanied by excellent waveform quality.Furthermore,the EFTS neural network prediction results show strong agreement with the experimental data.The feasibility of this method is verified,which provides a new solution for fast and accurate design of GMT. In summary,the following conclusions can be drawn from the comparative analysis between the model and experimental results:(1)Based on the input and output data of GMT dynamic model,the EFTS neural network model of the transducer is constructed in this paper,which can predict the output characteristics of GMT quickly and accurately and obtain the corresponding design scheme.(2)The GMT EFTS neural network significantly enhances computational efficiency,reducing the single computation time from 14 854 seconds to merely 0.012 9 seconds.(3)The EFTS neural network exhibits high calculation accuracy,with a maximum error of 0.419%and an average error of 0.26%when validated against experimental data,confirming its reliability and precision.关键词
超磁致伸缩电声换能器/多物理场耦合/增强特征传递结构/发射电流响应Key words
Giant magnetostrictive electroacoustic transduction/multiple physical field coupling/enhanced feature transmission structure/transmit current response分类
信息技术与安全科学引用本文复制引用
徐修贤,高兵,吴泽伟,何志兴,赵能桐..基于增强特征传递结构神经网络的超磁致伸缩电声换能器输出特性分析[J].电工技术学报,2025,40(23):7449-7461,13.基金项目
国家自然科学基金项目(52377010)、国家自然科学基金重大科研仪器研制项目(52127901)和国家自然科学基金青年科学基金项目(52407203)资助. （52377010）
