Soft computing-based predictive modeling of flexible electrohydrodynamic pumps

Abstract

Flexible electrohydrodynamic (EHD) pumps have been developed and applied in many fields due to no transmission structure, no wear, easy manipulation, and no noise. Physical simulation is often used to predict the output performance of flexible EHD pumps. However, this method neglects fluid–solid interaction and energy loss caused by flexible materials, which are both difficult to calculate when the flexible pumps deform. Therefore, this study proposes a flexible pump output performance prediction using machine learning algorithms. We used three different types of machine learning: random forest regression, ridge regression, and neural network to predict the critical parameters (pressure, flow rate, and power) of the flexible EHD pump. Voltage, angle, gap, overlap, and channel height are selected as five input data of the neural network. In addition, we optimized essential parameters in the three networks. Finally, we adopt the best predictive model and evaluate the significance of five input parameters to the output performance of the flexible EHD pumps. Among the three methods, the MLP model has exceptionally high accuracy in predicting pressure and flow. Our work is beneficial for the design process of fluid sources in flexible soft actuators and soft hydraulic sources in microfluidic chips.