Jetting prediction model for weakly viscoelastic ink by machine learning

Abstract

Inkjet printing has considerable potential ability for additive manufacturing technology. The functional inks, which have been commonly used in the industry, lead to the high complexity of predicting the jetting behavior. The complexity is induced by the rheological properties of the ink. However, the printability map composed of Ohnesorge and Weber numbers doesn’t contain the rheological features. We present the machine learning-based predictive model of jetting behavior from the Deborah number, Ohnesorge number, and waveform parameters. We prepared the 10 weakly viscoelastic model inks and measured their storage and loss modulus, which were close to the theoretical Maxwell model linear viscoelastic fluid. The relaxation time of viscoelastic inks is obtained by analyzing the Maxwell model equation. The big datasets of the ink jetting at different waveforms were collected from the high-speed images. Deborah number contributed to increasing the prediction accuracy of learning models. Multi-layer Perceptron showed outstanding performance after comparing the prediction accuracy of machine learning models. The final predictive model also exhibited remarkable accuracy of unknown ink according to waveform parameters and the correlation between jetting behavior and ink properties were in reasonable agreement. We proposed a new printability map characterized by the Ohnesorge number and Deborah number through the predictive model for weakly viscoelastic ink.