Loading...

Application of Machine Learning in Numerical Simulation of Electrohydrodynamic Printer

Beyk Ahmadi, Alireza | 2025

250 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57958 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Moosavi, Ali; Nouri Borujerdi, Ali
  7. Abstract:
  8. Electrohydrodynamic Printing (EHD Printing) has become a popular method for producing components and devices with extremely small dimensions. This type of printing is a subset of inkjet printing and is considered one of the additive manufacturing techniques. One of the most widely used methods in EHD printing is the application of voltage in the form of pulses, which results in the generation of small droplets for constructing the desired pattern. Since this type of printing involves a combination of various physical phenomena and phases, predicting the fluid behavior and consequently the printing properties is highly complex. Therefore, a method is needed to optimize the printing process by understanding the factors influencing the quality of EHD printing. Machine learning has recently emerged as a powerful tool for predicting optimal outputs in highly complex systems. However, proper training of a machine learning model requires high-quality data linking inputs to their corresponding outputs. In this research, the behavior of an EHD printer was first simulated using a multiphysics and multiphase model in COMSOL software. After validating the simulation, it was used to generate the data required for training the neural network. Additionally, data from other reputable scientific articles were used to improve the neural network training. A total of 1100 data points were collected for predicting droplet diameter, and 600 data points were gathered for predicting droplet generation frequency. In the case of droplet diameter prediction, the mean squared error (MSE) loss value for the test data was 229.26 μm², and the coefficient of determination (R^2) was 0.9671. For droplet generation frequency prediction, these values were 0.3459 and 0.9349, respectively. Sensitivity analysis revealed that droplet diameter is most sensitive to parameters such as the applied pulse voltage frequency, pulse and base voltage, nozzle diameter, fluid density, and flow rate. Similarly, droplet generation frequency showed the highest sensitivity to parameters such as pulse voltage, applied pulse voltage frequency, base voltage, nozzle diameter, fluid density, and flow rate. Finally, the two trained neural networks were used to predict the applied voltage function to achieve the desired droplet diameter and generation frequency. The coefficient of determination and loss function values in this case also confirmed the reliability of the trained neural networks
  9. Keywords:
  10. Electrohydrodynamic Printer ; Drop-on-Demand ; Additive Manufacturing ; Numerical Simulation ; Neural Network ; Machine Learning

 Digital Object List

 Bookmark

No TOC