Prediction of the distribution of airflow within the cotton canopy using fluid–structure interaction simulation and machine-learning methods.

Assisted airflow can change the porosity within the canopy to achieve whole-plant coverage and uniform deposition of droplets. However, it is difficult to simulate the temporal and spatial distribution characteristics of the airflow field within the canopy because of the interaction between the assisted airflow and the leaves. This paper proposed and developed a method to predict the airflow field using fluid–structure interaction (FSI) simulation and machine-learning (ML) methods. Firstly, a 3D virtual plant model was established to describe the leaf structure in the canopy. Secondly, the flow field of leaves interacting with airflow was simulated by co-simulation with structural explicit Finite Element (FE) solver and lattice Boltzmann (LB) solver. Indoor artificial cotton plants were constructed to verify accuracy of the simulation. Finally, artificial neural network (ANN), random forest (RF), and support vector regression (SVR) models were trained based upon the results of a large amount of simulation training and test datasets. The training results showed that the RF and FSI simulation results match the coefficient of determination R2 of 0.9860 and root mean square error (RMSE) of 0.3017 well. The novel combination of computational fluid dynamics and ML improves simulation eēēfficiency, reduces time and computational cost, and provides a basis to select reasonable assisted airflow parameters before precision spraying. • Airflow velocity field in canopy predicted by using FSI and machine-learning. • Interaction between leaves and airflow is co-simulated by FE and LB methods. • Co-simulation results verified by measured data. • Machine-learning methods compared in approximation of co-simulation results. • This approach can be useful for optimising spray parameters. [ABSTRACT FROM AUTHOR]