Predicting the dynamic tensile response of FDM materials using machine learning

Additive manufacturing (3D printing) materials have gained growing attention in engineering applications due to their design flexibility and cost-effectiveness. However, their mechanical behavior, particularly under dynamic tensile loading, remains insufficiently understood, introducing a challenge for reliable design and modelling. This study investigates the dynamic tensile behavior of 3D-printed materials across a wide range of strain rates through extensive experimental testing and machine learning (ML) modeling. The research work started by conducting extensive experimental work to investigate the 3D printing materials under various strain rates. Afterwards, employing multiple ML models and exploring their applications to predict stress-strain behavior. Six ML models were utilized: Neural Network, Decision Tree, Support Vector Regression Random Forest, Gaussian Process Regression, and k-Nearest Neighbors. Model performance was validated using unseen experimental data at strain rates of 0.5 and 100 /min. For a strain-rate of 0.5 /min, the Neural Network demonstrated an exceptional precision of 92.69%, and lowest MAPE of 7.31%. While for high strain-rate of 100 /min, the Gaussian Process Regression demonstrated the lowest MAPE of 4.49%. On the other hand, the Decision Tree and Random Forest showed the highest MAPE among the aforementioned ML models. The novelty of this work lies in integrating high strain-rate experimental data with multi-model ML evaluation to identify a reliable stress-strain model for 3D-printed materials. The presented results highlight the effectiveness of ML in predicting seen and unseen stress-strain data at wide range of strain-rates, furthermore, the findings open new avenues for data-driven materials modelling and design.