基於物理信息神經網路進行壓電能量擷取振子之參數反算研究

Parameter Identification of a Piezoelectric Harvester Based on Physics Informed Neural Network

戴煒宸(Wei-Chen Tai)

指導教授 ： 舒貽忠

國立臺灣大學/工學院/應用力學研究所/碩士(2022年)

https://doi.org/10.6342/NTU202200931

摘要

本研究主要利用類神經網路(Artificial Neural Network, ANN)以及基於物理信息神經網路(Physics Informed Neural Network, PINN)進行壓電能量擷取振子之參數反算研究。當輸出功率之解析解為已知時，類神經網路可藉由內含壓電參數之解析解與不同電路負載下的實驗數據所構成的損失函數進行訓練。但是絕大部分物理模型的解析解難以取得，因此必須採用基於物理信息神經網路。該神經網路損失函數包含了基於物理模型的微分方程以及實驗所得的時域電壓訊號。研究結果顯示，類神經網路能夠替代人力且大幅降低曲線擬合時間，並識別出機械阻尼比、外力振動大小以及壓電電容值。此外，基於物理信息神經網路在含有雜訊模擬訊號的訓練下，能夠準確的識別出機械阻尼比、外力振動大小以及力電耦合係數，展現了其在有限的實驗數據下的優勢，並將於未來的研究中證實。

關鍵字

類神經網路 ； 基於物理信息神經網路 ； 參數反算問題 ； 壓電能量擷取振子 ； 振動反算問題

並列摘要

The thesis studies the inverse parametric identifications in a piezoelectric energy harvester based on the artificial neural network (ANN) and the physics informed neural network (PINN). The former is suitable when the analytic estimate of power is available in terms of device parameters. Thus, ANN can be trained by assigning the loss functions consisting of the analytic formula of power estimate as well as the experimental data of power against various electric loads. The latter approach is suitable when the formula of power estimate is unavailable. Under this circumstance, the loss functions contain two parts: the first is the model-based differential equations and the second is the experimental data of time waveforms of voltage signals. The results show that the ANN approach is capable of replacing the curve fitting which requires significant labor efforts in identifying mechanical damping ratio, voltage source magnitude and piezoelectric capacitance. In addition, the PINN approach trained on noisy simulated data accurately identifies the parameters of mechanical damping ratio, voltage source magnitude and electromechanical coupling factor. It provides an advantage of extracting limited amount of experimental data and will be demonstrated in the future work.

並列關鍵字

artificial neural network (ANN) ； physics informed neural network (PINN) ； inverse parametric identification ； piezoelectric energy harvester ； vibration inverse problem

參考文獻

M. I. Jordan and T. M. Mitchell. Machine learning: Trends, perspectives, and prospects. Science, 349: 255-260, 2015.

G. Bonaccorso. Machine learning algorithms. Packt Publishing Ltd, 2017.

A. Singh, N. Thakur and A. Sharma. A review of supervised machine learning algorithms. International Conference on Computing for Sustainable Global Development, pp. 1310-1315, 2016.

S. Ray. A quick review of Machine learning algorithms, International Conference on Machine Learning, Big Data, Cloud and Parallel Computing, pp. 35-39, 2019.

T. O. Ayodele. Types of machine learning algorithms. New Advances in Machine Learning, 3: 19-48, 2010.