Prediction of acoustic wave parameters of thermoacoustic prime mover through artificial neural network technique: Practical approach for thermoacoustics

Thermoacoustic prime movers are considered new alternative heat engines to traditional ones. For good performance of such a heat engine, a careful apparatus design is required. To predict the acoustic wave parameters responding to geometrical parameters of stack and resonator, is important for such a design. Artificial neural network (ANN) model is first proposed to predict the oscillating frequency and acoustic pressure amplitude, under given resonator length, stack length, stack plate spacing and thickness. ANN models for one standing wave thermoacoustic prime mover had been developed based on published experimental data, and evaluated based on some criteria such as least mean square error between the predicted and actual outputs during the testing phase. Concerning oscillating frequency, ANN model with the configuration of 4-4-4-1 was adopted whilst 4-4-1 for acoustic pressure amplitude, namely 4 neurons representing the four input design parameters, one or two hidden layers each with optimal four hidden neurons and one neuron representing the output oscillating frequency or acoustic pressure amplitude. Moreover, a statistical analysis has been conducted to show the contribution percentages of the proposed geometrical parameters of resonator and stack where it was found that the resonator length has the largest contribution effect with the approximate percentage of 76% on the two considered acoustic wave parameters. Compared to both experimental and DeltaEC model results, the determined ANN models had been proven to be desirable in their prediction accuracy with the error percentages of 2.26% and 0.78% for predicted oscillating frequency and acoustic pressure amplitude, respectively. This research work provides a promising practical modeling approach based on ANN technique for complex design problems in thermoacoustic systems.