Machine learning based multi-parameter droplet optimisation model study.

Area of Science:

• Industrial Printing
• Fluid Dynamics
• Computer Vision

Background:

• Continuous inkjet technology is vital for industrial printing due to its speed, precision, and versatility.
• Accurate droplet generation is crucial for achieving desired printing outcomes.

Purpose of the Study:

• To propose and validate a novel parameter optimization method (BO-GP) for continuous inkjet devices.
• To enhance droplet generation accuracy and efficiency.

Main Methods:

• The study combines Bayesian optimization (BO) with computer vision (CV) to create the BO-GP method.
• Experiments were conducted on millimetre-scale and microfluidic inkjet devices using droplet image datasets.
• The BO-GP method was iterated for 50 rounds to converge on optimal control parameters.

Main Results:

• The BO-GP method successfully converged to optimal control parameters and constructed the Pareto frontier.
• Optimized minimum objective function values were reduced from 0.378 to 0.331 (millimetre-scale) and 0.073 to 0.046 (microfluidic).
• The Pareto solution remained stable, and optimal control conditions were derived in just 1 hour.

Conclusions:

• The BO-GP method significantly improves droplet generation accuracy and efficiency in continuous inkjet printing.
• This approach offers a rapid and effective solution for optimizing parameters to achieve high roundness, yield, and uniformity in droplets.