Abstract
Thermo-acoustic systems enable the conversion of thermal energy into acoustic waves and vice versa. These waves are harnessed for purposes like thermo-acoustic refrigeration to induce cooling or thermo-acoustic generators to produce electricity. This conversion relies on the thermo-viscous interaction within a porous material, called stack or regenerator, where the gas medium oscillates acoustically interacting with the pore internal walls, especially in traveling-wave thermo-acoustic engines.
This research study explores the application of machine learning techniques to enhance the analysis of thermo-acoustic devices, focusing on nonlinearities. Three models namely Artificial Neural Network trained by Particle Swarm Optimization (ANN-PSO), Adaptive Neuro-Fuzzy Inference System (ANFIS), and a conventional Artificial Neural Network were applied to predict key parameters in both travelling-wave thermo-acoustic generators and a standing-wave thermo-acoustic refrigerator. Rigorous experimental analyses were conducted on physical prototypes, and the machine learning models were validated against experimental data.
For single stage and multi-stage thermo-acoustic generators, the ANN-PSO model demonstrated exceptional predictive accuracy in forecasting output voltage, with an 𝑅2 value of 0.9959. Similar success was observed in the ANN model, which has demonstrated the highest predictive accuracy in predicting working pressure and sound pressure level, affirming the efficacy of considering temperature differentials and the number of engine stages as crucial input parameters.
The study also extended its focus to a standing-wave thermo-acoustic refrigerator, where the developed models (ANN-PSO, ANFIS, and ANN) accurately predicted the Coefficient of Performance (COP), showcasing their potential to reduce the need for extensive experimental configurations. Notably, the ANN model demonstrated the highest accuracy in predicting COP. Furthermore, in the context of simulating a travelling-wave thermo-acoustic engine, the integration of DeltaEC simulation with ANN models proved successful in predicting working pressure amplitude and acoustic power, with minimal deviations from target values.
The study demonstrates machine learning's potential in improving thermo-acoustic device analysis and prediction, offering a streamlined approach for enhanced experimental practices. It significantly contributes to thermo-acoustic research, providing an efficient modelling framework.
Keywords: thermo-acoustic, generator, refrigerator, ANN, ANFIS, ANN-PSO.