A machine learning approach for predicting the Johnson-Champoux-Allard parameters of a fibrous porous material

Abstract

Porous fibrous materials have been widely used as acoustic treatments for noise attenuation. Their acoustic properties are typically characterized by Johnson-Champoux-Allard (JCA) model, which includes five dominant parameters, i.e., open porosity, flow resistivity, tortuosity, viscous characteristic length, and thermal characteristic length. The JCA parameters depend on the microstructure configuration of the material, which can be attained by experimental measurements or numerically analyzing the flow field inside the microstructure, but significant efforts to predict the parameters are typically required. This study proposes a machine learning approach based on an artificial neural network (ANN) for predicting the JCA parameters of a fibrous material. Two geometric parameters that can characterize the fibrous material, i.e., the radius of the fiber and the equivalent throat size between neighbouring fibers, are set as inputs for the prediction model, while the five JCA parameters are set as outputs. The datasets for the network are prepared from finite element simulations. Results confirm that the trained model can predict the JCA parameters accurately and reliably based on the micro-structural geometric parameters. Finally, the model is further validated by the measured acoustic characteristics of a metal-based fibrous material in an impedance tube. The machine learning model opens up possibilities to facilitate the design of advanced porous materials.