Determining effective electromechanical coupling coefficient of ring-shaped piezoelectric transformer using coupled artificial neural networks

• Main Authors: Supatutkul C., Laosiritaworn Y.
• Format: Journal
• Published: 2017
• Online Access: https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84982281227&origin=inward; http://cmuir.cmu.ac.th/jspui/handle/6653943832/41433
• Institution: Chiang Mai University

© 2016 Taylor & Francis Group, LLC. In this work, two coupled Artificial Neural Networks (ANNs) were proposed to determine the first two higher-order resonant frequencies in radial vibrational modes and their associate effective electromechanical coupling coefficient (k eff ) of the ring-shaped piezoelectric transformer. In the first network, the inputs to the ANN were composed of ring width-to-thickness ratio (w/t), ring thickness (t) and the modes of the first two higher-order resonances (R1 and R2). Then, the output from the first network, i.e. the universal resonant frequency (the resonant frequency (f r ) multiplied by ring thickness (t)) in logarithmic form (ln(f r t)), was later supplied to the second network along with w/t and the order of the vibrational modes (R1 and R2) to predict ln(k eff ). The data for this coupled ANNs modeling was derived from Finite Element calculation analysis on the ring-shaped piezoelectric transformer. From the results, while it was unlikely to obtain reliable predicted results with just one ANN, the use of coupled ANNs provides the prediction of universal resonant frequencies and the coupling coefficient with high accuracies even using only a few basic parameters, which are the ring geometries and the resonant modes (R1 or R2). Specifically, from the best ANN architectures, the mean absolute errors (MAE) are 0.000952 and 0.001558 in predicting ln(f r t) for R1 and R2, and are 0.002564 and 0.001047 in predicting ln(k eff ) for R1 and R2 respectively. Furthermore, the R-squared's of the scattering plots between predicted and targeted outputs are all higher than 0.98, which confirms the trustworthy of the model. Therefore, from this coupled ANNs design, it becomes possible to model complicated process with just only a few basic parameters. Explicitly, in this work, important parameters required for constructing highly efficient piezoelectric transformer were achieved without the needs to go through lengthy complex impedance spectrum analysis as necessitated in the conventional processes.