人工神经网络和电阻抗谱法压电材料快速表征

压电材料作为重要的功能材料,广泛应用于社会的各领域,但其弹性常数的偏差会导致应用过程中出现错误的设计,弹性常数的正确表征对压电器件的正确设计尤为重要。与其他测量方法相比,电阻抗谱仅需要阻抗分析仪即可实现测量,通过测量阻抗谱反演获得压电材料的弹性常数。传统电阻抗谱法通过不断修正材料参数,使得测量阻抗谱和计算阻抗谱最大程度吻合,该过程需要多次迭代,计算量大,耗时较长。该文提出采用神经网络建立阻抗谱到弹性常数的正向模型,测量得到阻抗谱后仅需一次正向计算即可得到弹性常数。使用Comsol和Matlab联合仿真建立数据集,引入丢弃法避免模型过拟合,利用Pytorch建立模型,经过训练后,最大谐振频率偏差从初始2.8%降至0.8%。该技术为压电材料弹性常数精密测量提供可靠的理论与实践途径。

Fast Electrical Impedance Spectroscopy-Based Characterization of Piezoelectric Material Using Artificial Neural Network

As important functional materials, piezoelectrics are widely used in various fields. However, the deviation of their elastic constants results in erroneous designs during application processes. Accurate characterization of the elastic constant is crucial for the correct design of piezoelectric devices.In contrast to other measurement methods, electrical impedance spectroscopy can only be carried out using an impedance analyzer, and the elastic constants of piezoelectric materials can be obtained by the inversion of the impedance spectroscopy data.In traditional electrical impedance spectroscopy,the measured impedance spectra are coincided with the calculated impedance spectra to the best possible extent by constantly modifying the material parameters. However, the process requires many iterations, making it tedious and time consuming.This study developed a forward model that yields elastic constants from impedance spectra by harnessing an artificial neural network. Upon measuring the impedance spectrum, the elastic constant can be obtained by only one forward calculation.COMSOL and MATLAB co-simulation were used to generate the data-sets.The discard method was employed to avoid overfitting of the model, and Pytorch was used for implementation. The resonant frequency error was reduced from the initial 2.8% to 0.8% after training.The proposed technique affords a reliable theoretical and practical approach for accurately measuring the elastic constants of piezoelectric materials.