Machine learning-based prediction of dynamic mechanical properties for 3D-printed PETG specimens

The fused filament fabrication (FFF) is a widely used technique for 3D printing of thermoplastics. The mechanical properties of 3D-printed samples may decrease with an increase in operating temperature. In this study, the degradation of the mechanical properties of 3D-printed polyethylene terephthalate glycol samples has been studied using dynamic mechanical analysis. The effect of process parameters such as layer thickness and operating frequency has been analysed in detail. The storage and loss modulus at the glass transition temperature decrease with an increase in layer thickness. Temperature and frequency-dependent analytical models and machine-learning (ML) algorithms, such as K-nearest neighbours, random forest (RF) and gradient boosting, are used to estimate the modulus variation as a function of temperature and loading frequency. The predictability of analytical models and ML models has been assessed. The fitting coefficients of the analytical model are evaluated as a function of temperature and frequency. Also, it is observed that the RF algorithm predicts the dynamic mechanical behaviour of 3D-printed samples with better accuracy at known frequencies as well as at unknown frequencies. For a layer thickness of 0.17 mm at 1 Hz frequency (known frequency), the RF algorithm demonstrated better performance indices with the highest R²-value of 0.983 compared to other ML algorithms. Similarly, for a layer thickness of 0.17 mm at 9 Hz frequency (unknown frequency), the RF algorithm predicted the modulus values with the highest R²-value of 0.967 compared to other ML algorithms.