Developing Machine Learning Algorithm for Investigating Features Affecting Relative Density and Shrinkage of Printed Parts in Binder Jetting Additive Manufacturing of Microjet TP-80 Ceramic Composite Powder

Binder jetting is a versatile 3D printing technique capable of producing complex geometries with various materials, including ceramics. However, the final properties of binder-jetted parts are affected by multiple process parameters, making optimization challenging and time-consuming. To address this, we developed a machine learning model to predict and optimize the key factors affecting part quality. This study explores the application of machine learning (ML) algorithms to investigate the factors influencing relative density and shrinkage in binder jetting additive manufacturing of ceramic composite powders. The ML algorithm was designed to identify the optimal printing parameters that maximize the relative density of printed parts while minimizing shrinkage in the final parts. Importantly, we employed SHAP algorithms to determine which parameters significantly impact shrinkage and part density. By leveraging this data-driven approach, we aim to reduce the need for extensive trial-and-error experiments, thereby saving time and resources in the manufacturing process. The results demonstrate the effectiveness of ML Algorithms in predicting and optimizing binder jetting outcomes for ceramic composite powders. Furthermore, the study revealed that ensemble methods, such as Bagging and Random Forest, were most effective in predicting relative density, with Bagging achieving the lowest MSE (0.0156 0.085). For shrinkage prediction, simpler models like Linear Regression outperformed more complex approaches, achieving the lowest MSE (0.0028 0.031). Feature importance analysis revealed that Sample Geometry and Sample Actual Volume were critical determinants of relative density, while Sample Delay had the most significant impact on shrinkage.