Prediction of dielectric behaviour of FDM printed-acrylonitrile butadiene styrene (ABS) reinforced with 10% carbon fiber: a machine learning approach

Polymer composites with particular functional properties for electrical purposes can be made using fused deposition modelling (FDM) and additive manufacturing (AM). In order to experimentally analyze the dielectric behaviour of ABS reinforced with 10 weight% carbon fibre (CF), this study looked at the effects of infill density, printing speed, and infill pattern on the dielectric constant, quality factor (Q-factor), loss tangent (tan δ), and breakdown voltage. Five ensemble machine learning models—Gradient Boosting (GB), AdaBoost (AB), Light Gradient Boosting Machine (LGBM), Extra Trees (ET), and CatBoost (CB)—were employed to predict the reported dielectric properties after the printing settings were optimized using a Taguchi L9 experimental design. The dielectric constant ranged from 1.280 to 4.071, while the breakdown voltage varied between 2.58 and 5.83 kV. The most important element was found to be infill density, which accounted for (49.1–61.0%) variation in the dielectric properties. Printing speed (18.0–36.7%) and infill pattern (13.9–27.9%) came next. The Gradient Boosting method fared better than the other models in terms of prediction accuracy, with an R2 of around 0.999 and incredibly low prediction errors. The durability of the suggested model was further confirmed using independent train-validation-test evaluation, 10-fold cross-validation, and external validation using additional specimens. The findings demonstrate that optimizing FDM process parameters provides an effective way to customize the dielectric performance of carbon fiber-reinforced ABS composites and underscore the potential of machine learning as a reliable tool for predicting dielectric properties in additively manufactured polymer composites.