Analysis and prediction of the performance of free- piston Stirling engine using response surface methodology and artificial neural network. (April 2021)

• Record Type: Journal Article
• Authors: Ye, Wenlian; Wang, Xiaojun; Liu, Yingwen; Chen, Jun
• Is Part Of: Applied thermal engineering. Volume 188(2021)
• Publication Date: 2021-04
• Subjects: Free-piston Stirling engine -- Response surface methodology -- Artificial neural network -- Performance prediction; Heat engineering -- Periodicals; Heating -- Equipment and supplies -- Periodicals; Periodicals; 621.40205
• DOI: 10.1016/j.applthermaleng.2021.116557
• Languages: English
• ISSNs: 1359-4311

Abstract

Highlights: Two methods are used for modeling of free piston Stirling engine's performance. The Stirling engine's performance increases during the increase of T h and k d. Two models for the performance predicted are acceptable and accurate. Response surface methodology model is more accurate than artificial neural network. Abstract: Free piston Stirling engine is a popular area of research in high-efficiency thermal power conversion technology. However, owing to its strong coupling, nonlinearity, and parameter interactions, building an effective model to predict the performance is of great importance. This study was to investigate and derive the prediction models of a nonlinear free-piston Stirling engine using response surface methodology and artificial neural network. The interactive influences of thermodynamic and dynamic parameters which have significant effects on the amplitudes of the displacer and piston, operating frequency, and output power were illustrated in detail. Also, error analyses were then performed between the simulated and predicted values for both methods by comparing the mean absolute percentage errors, mean-squared errors, and correlation coefficients. The results indicated the correlation coefficients for the four output parameters from the response surface methodology as 0.9998, 0.9998, 0.9999, and 0.9994, and approximately 95% of the output parameter data were predicted with < 5% errors during verification, indicating that the response surface methodology model had good predictability compared with the artificial neural network model. Therefore, this research provides an effective approach to predict performances and can be applied to optimise the Stirling engine performance accurately and quickly.