Machine learning models to predict the relationship between printing parameters and tensile strength of 3D Poly (lactic acid) scaffolds for tissue engineering applications

Abstract

3D printing is an effective method to prepare 3D scaffolds for tissue engineering applications. However, optimization of printing conditions to obtain suitable mechanical properties for various tissue engineering applications is costly and time consuming. To address this problem, in this study, scikit-learn Python machine learning library was used to apply four machine learning-based approaches which are ordinary least squares (OLS) linear regression, random forest (RF), light gradient Boost (LGBM) and extreme gradient boosting (XGB) models to understand the relationship between 3D printing parameters and tensile strength of poly(lactic acid) (PLA). 68 combinations of process parameters for nozzle temperature, printing speed, layer height and tensile strength were used from investigated research papers. Then, datasets were divided as training (80%) and test (20%). After building the OLS linear regression, RF, LGBM and RF models, the correlation heatmap and feature importance of each printing parameter for tensile strength values were determined, respectively. Then, the tensile strength was predicted for real datasets to evaluate the performance of the models. The results demonstrate that XGB model was the most successful in predicting tensile strength among the studied models with an R2 value of 0.98 and 0.94 for train and test values, respectively. A close R2 value for the train and test also indicated that there was no overfitting of the data to the model. This study can be extended for independent variables including nozzle pressure, strut size and molecular weight of PLA and dependent variables such as elongation and elastic modulus of PLA which may be a powerful tool to predict the mechanical properties of scaffolds for tissue engineering applications.

XGB Modeling Reveals Improvement of Compressive Strength of Cement-Based Composites with Addition of HPMC and Chitosan

“…Figure 2 shows that R 2 for the XGB model is 0.99 (~1). This is an excellent fit of the test values to the model trained by the training dataset [42]. R 2 for testing and training are also very similar, which shows that the model works similarly for training and testing data.…”

“…Figure 2 shows that there is a degree of overfitting of the model for the dataset. Despite this, the RMSE and MAE values are quite low even for test data (MAE = 0.4, RMSE = 0.7); therefore, the model can still provide satisfactory knowledge for analyzing the importance of each factor and the mechanical behavior of the composites [42]. Figure 2c shows that SHAP values are positive for all the features.…”

Enhancing quality control in bioprinting through machine learning

“…Finally, the material and printing parameter optimization can be conducted considering not only shape fidelity but also the overall a priori performance of the construct from functional and biological perspectives [80,85,94]. For instance, Kondiah et al employed an ANN regression model to optimize the ink formulation based on the drug loading efficiency for a model drug in bone TE applications.…”

Micromechanical Models for FDM 3D-Printed Polymers: A Review

“…According to Ahn et al [18], the raster orientation and air gap have a large effect on the tensile strength, while the bead width was found to have only a small influence. A recent study using machine learning (ML) models to predict relationships between printing process parameters and mechanical properties [46] found that larger infill densities (related to the air gap) have a positive effect on tensile strength, while higher printing speeds negatively affect this mechanical property. Furthermore, compressive strength was found to be unaffected.…”