Application of artificial neural network for predicting the dynamic performance of a free piston Stirling engine. (1st March 2020)

• Record Type: Journal Article
• Authors: Ye, Wenlian; Wang, Xiaojun; Liu, Yingwen
• Is Part Of: Energy. Volume 194(2020)
• Publication Date: 2020-03-01
• Subjects: Free piston stirling engine -- Artificial neural network -- Dynamic performance prediction; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2020.116912
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: In this study, an artificial neural network model is built to predict the dynamic performance of a beta-type free piston Stirling engine. The influences of six input dynamic parameters on operating frequency, amplitude ratio and phase angle are analyzed. The operating frequency is significantly affected by the spring stiffness and the mass of the pistons. However, the relationships of the dynamic parameters are comprehensive, which are determined by multiple parameters. Then, a number of dynamic output parameters are used as training and testing data. The best results are obtained by 6-6-1, 6-6-1 and 6-10-6-1 network architectures for the operating frequency, amplitude ratio and phase angle respectively. For these network architectures, the back propagation algorithm, namely Levenberg-Marguardt is applied. Stirling engine's dynamic performance predicted with the network model is compared with the actual values. After training, correlation coefficients (R 2 ) values for training and testing data are close to 1. The mean relative errors of the operating frequency, amplitude ratio and phase angle are 0.85%, 2.78% and 3.19% for the training process. These results show that the artificial neural network model is an acceptable and powerful approach for predicting the dynamic performance of the beta-type free piston Stirling engine. Highlights: Parameter correlated effects of a free piston Stirling engine are analyzed. Artificial neural network is proposed to predict the FPSE′ dynamic performance. A number of dynamic parameters are used as training and testing data. ANN models for the dynamic performance predicted are acceptable and accurate.