Prediction of viscoelastic and printability properties on binder-free TiO2-based ceramic pastes by DIW through a Machine Learning approach

Abstract

Printability, a key factor in the success of Additive Manufacturing (AM) processes, relies on various aspects, including material properties and process parameters. This study aims to explore the relationship between the rheological properties of binder-free TiO2-based ceramic pastes and their printability using a machine learning approach, specifically utilizing a feedforward Deep Neural Network (DNN) model. The main objective of this thesis is to develop a predictive model that can accurately estimate the printability of binder-free TiO2-based ceramic pastes based on key rheological properties. To achieve this, a comprehensive dataset consisting of 25 samples of rheological properties and corresponding printability assessments was collected and used for model training and evaluation. The methodology involved the collection of rheological and viscoelastic data for a range of TiO2-based ceramic pastes including shear viscosity (η) values as a function of shear rate (γ_ ), and storage (G0) and loss (G00) modulus values as a function of oscillatory stress (σ). Rheological properties (γ_ and η) were then utilized as input features for the feed-forward DNN model, while the viscoelastic data (G0 and G00) was set as the target properties. The model was trained using a portion of the dataset, and the remaining set was utilized for testing. No validation set was defined due to the small size of the dataset. The results obtained from the developed model not only demonstrated a correlation between the rheological properties and printability of the studied ceramic pastes but also revealed a relationship between the rheological and viscoelastic behavior of the samples, providing deeper insights into this ceramic system. The predictive capabilities of the model were evaluated based on performance metrics such as Mean Squared Error (MSE) and Relative Error ( ). The discussion section provides a detailed analysis of the findings and explores the limitations of the study, such as the sample size and potential influence of the experimental work on the model’s performance. This thesis contributes to the field of AM by presenting a machine learning-based approach for predicting the printability of binder-free TiO2-based ceramic pastes. The DNN model proves to be a valuable tool for assessing the printability of the pastes. Moreover, it provides further insights into the ceramic system by correlating rheological and viscoleastic properties. This research opens up possibilities for optimizing paste formulations and process parameters to enhance the additive manufacturing of ceramics. Furthermore, the model here proposed may serve as starting point for exploring alternate machine learning algorithms, such as transfer learning, generative networks and even Bayesian optimization.