Abstract
This study describes an intelligent method for modeling and optimization of piezoelectric ejection system design for additive manufacturing. It is a combination of neural network (NN) techniques and finite element simulation (FES) that allows designing each parameter of a piezoelectric ejection system faster and more reliably than conventional methods. Using experimental and literature data, a FE model of the droplet ejection process was developed and validated to predict two indexes of droplet ejection behavior (DEB): jetting velocity and droplet diameter. Two artificial neural network (ANN) models based on feed-forward back propagation were developed and optimized by genetic algorithm (GA). A database was established by FE calculations, and the models were trained to establish the relationship between the piezoelectric ejection system design input parameters and each DEB indicator. The results show that both NN models can independently predict the droplet jetting velocity and droplet diameter values from the training and testing data with high accuracy to determine the optimal piezoelectric ejection system design. Finally, the accuracy of the prediction results of the FES and ANN-GA models was verified experimentally. It was found that the errors between the predicted and experimental results were 4.48% and 3.18% for the jetting velocity and droplet diameter, respectively, verifying that the optimization method is reliable and robust for piezoelectric ejection system design optimization.