Predictive models for 3D inkjet material printer using automated image analysis and machine learning algorithms

Additive manufacturing (AM) is a smart manufacturing process to fabricate components with high precision, minimal post-processing, and increased component complexity in a variety of materials. This research focuses on developing automated image analysis and predictive models for a widely used 3D material inkjet printing (IJP) process. The interplay of four input process parameters, which include frequency, voltage, temperature, and meniscus vacuum, on the output metrics of the inkjet printer was evaluated using statistical measures (ANOVA). Droplet types were classified as no drop, satellite drop, and normal drop using four machine learning classifiers, including random forest, support vector classifier, k-nearest neighbor, and decision trees. Hyperparameter tuning was performed for each model for over 486 data points. Regression predictive models were developed for both ink droplet velocity and volume with three linear models (linear, lasso, and ridge regression) and four non-linear models (random forest, decision tree, support vector regression, and k-nearest neighbor). Mean squared error and the coefficient of determination, r-squared value, were used to evaluate the performance of the predictive models. For the drop type classification models, k-fold of 5 yielded the highest accuracy for the RF, KNN, and DT models of around 98%. Similarly, for the regression based predictive models RF, DT and KNN accuracy results ranged from 97 to 99%. All the machine learning models were validated with experimental data with high prediction accuracies accuracy. This research serves as a foundation for developing design guidelines for 3D material inkjet printing with applications in biosensors, flexible electronics, and regenerative tissue engineering.