Machine learning based multi-parameter droplet optimisation model study

Abstract

Continuous inkjet technology, as a key technology in the field of industrial printing, is favoured for its excellent printing speed, precision and versatility. In order to achieve the accurate generation of ideal droplets in continuous inkjet devices, this paper proposes a new parameter optimisation method, BO-GP, which combines the Bayesian optimisation algorithm with computer vision, and after 50 rounds of iterations, it can converge to the optimal values of the control parameters, and successfully constructs the Pareto frontier of the control parameters. In this paper, experiments were conducted on two different device droplet image datasets, a millimetre-scale inkjet device and a microfluidic device, respectively. Compared with the original BO in Loop method, the optimised minimum objective function value is reduced from 0.378 to 0.331 in the millimetre-scale device, and from 0.073 to 0.046 in the microfluidic device. Moreover, the Pareto solution of the 10 sets of predicted parameters output using the BO-GP method tends to be stable with fluctuations around 0.1, and it takes only 1 h to derive the control conditions for achieving high roundness, high yield and uniform size droplets.