A novel, data-driven framework for predicting and optimizing the mechanical performance of 3D-printed polylactic acid (PLA) composites reinforced with date pit (DP) particles under controlled annealing conditions is presented, enabling simultaneous property prediction and design optimization from a minimal experimental dataset.
Abstract
This study presents a novel, data-driven framework for predicting and optimizing the mechanical performance of 3D-printed polylactic acid (PLA) composites reinforced with date pit (DP) particles under controlled annealing conditions. This work uniquely integrates both effects into a unified machine learning (ML) surrogate modeling approach, enabling simultaneous property prediction and design optimization from a minimal experimental dataset. Three ML algorithms, Gaussian Process Regression (GPR), Random Forest Regression (RFR), and Decision Tree Regression (DTR), were systematically developed to predict Young’s modulus (E), ultimate compressive strength (UCS), and Shore D hardness as functions of DP weight fraction (0–10 wt.%) and annealing duration (0–20 h at 100 °C). Among the models, GPR demonstrated superior generalization and uncertainty-aware prediction capability, achieving test R2 values up to 0.936 and a MAPE below 2% for structurally critical properties. Beyond prediction accuracy, the study introduces a key novelty through the integration of Partial Dependence Plot (PDP) analysis, which reveals physically interpretable relationships between processing parameters and mechanical behavior. A non-monotonic annealing optimum at 5 h was identified, governed by competing crystallization enhancement and thermal degradation mechanisms, alongside a consistently positive reinforcement effect of DP loading. The optimized condition (10 wt.% DP, 5 h annealing) yielded improvements of 37.5% in stiffness, 42.4% in strength, and 21.6% in hardness compared to neat PLA. A five-fold cross-validation and split ratio sensitivity analyses (70/30, 80/20, 90/10) were conducted to assess generalization reliability. One-way ANOVA confirmed highly significant effects of DP loading on all three properties (p < 0.001 for E and hardness; p < 0.01 for UCS) and a significant effect of annealing time on UCS (p < 0.01). Permutation feature importance analysis confirmed DP weight fraction as the dominant predictor for all three outputs.
Mechanical anisotropy in Z-direction printed polyether ether ketone (PEEK) components fabricated by fused filament fabrication (FFF) remains a critical limitation for load-bearing applications. Traditional response surface methodology (RSM) assumes polynomial relationships that inadequately capture the nonlinearities governing FFF of high-performance semi-crystalline polymers. This study aims to develop and validate a machine learning-based framework to optimize FFF process parameters for enhanced Z-direction mechanical properties and reduced porosity in 3D-printed PEEK.
A central composite design across 15 printing conditions provided the experimental data set. Four ML algorithms were evaluated through leave-one-out cross-validation. Feature importance was assessed via permutation importance, SHAP values and partial dependence plots. Multi-objective optimization through NSGA-II generated a Pareto front, and consensus optimal parameters were validated through ten independent tensile specimens.
Random forest achieved superior predictive accuracy (R² greater than 0.94, MAPE equal to 3.9%). Layer thickness was identified as the dominant parameter, with a model-predicted transition at approximately 0.22 mm where porosity increases in an accelerated manner, presented as a hypothesis requiring future experimental validation. NSGA-II generated 295 Pareto-optimal solutions. Validated parameters (408°C, 57 mm/s, 0.098 mm) yielded prediction errors below 4.1%, achieving 62% tensile strength improvement and 76% porosity reduction relative to the least favorable condition.
This work presents the first experimentally validated multi-objective ML optimization framework for Z-direction mechanical property enhancement of FFF-printed PEEK. It identifies a critical layer thickness threshold governing porosity transitions, and generates a multi-solution Pareto front enabling parameter space exploration that RSM-based approaches cannot provide for this material system.
A. El Magri, H. Vanaei, Zakariae Bouchkara et al.· Rapid prototyping journal· 0 citations
This study presents an experimental and machine learning (ML) based investigation of the mechanical performance of laminated polymer structures fabricated by fused deposition modeling (FDM). The specimens were designed as a three-layer configuration consisting of polyethylene terephthalate glycol (PETG), thermoplastic polyurethane (TPU), and polylactic acid (PLA) to combine rigidity, elasticity, and dimensional stability. Mechanical performance was evaluated through tensile and flexural tests under different processing conditions. A Taguchi L16 orthogonal array was employed to investigate the effects of four printing parameters: nozzle temperature (NT), infill pattern (IP), wall thickness (WT), and printing speed (PS), each at four levels. Experimental results were analyzed using signal to noise (S/N) ratios and analysis of variance (ANOVA). Tensile strength ranged from 23.19 to 36.22 MPa, while flexural strength varied between 10.29 and 53.40 MPa. ANOVA revealed that WT was the most influential factor affecting tensile strength (
p
= 0.003), whereas NT had the greatest effect on flexural strength (
p
= 0.029). To enhance predictive capability, four ML algorithms were developed and compared. These were artificial neural network (ANN), support vector regression (SVR), random forest (RF), and extra trees (ET). The ET model achieved the highest predictive accuracy, with R
2
values of 0.875 for tensile strength and 0.906 for flexural strength. The results demonstrate that ensemble tree-based models capture nonlinear relationships between FDM parameters and mechanical responses more effectively than ANN and SVR. The integration of Taguchi design, statistical analysis, and ML provides an effective framework for predicting and optimizing the mechanical performance of FDM-printed PETG/TPU/PLA laminated polymer structures.
Arif Karadağ, Emin Ağrali, O. Ulkir· Journal of Thermoplastic Com...· 0 citations
This study proposes a data-driven surrogate modeling framework for predicting
solidification time and mold thermal stress during low-pressure die casting
(LPDC) of aluminum alloy wheels. The methodology employed an optimal Latin
hypercube design (OLHD) to sample key parameters including cooling channel
geometry and process conditions. A sequential simulation methodology combining
ProCAST and Abaqus was implemented to generate a comprehensive dataset of
solidification times and thermal stress distributions. Based on this dataset,
surrogate models were developed using Support Vector Regression, Kriging, and
Polynomial Response Surface Methodology, with their hyperparameters
automatically tuned through Bayesian Optimization (BO). The optimized models
were rigorously evaluated using four statistical metrics: Coefficient of
Determination (R2), Mean Squared Error (MSE), Mean Absolute Error (MAE), and Root
Mean Squared Error (RMSE). The evaluation results show that the BO–SVR model
demonstrated superior prediction accuracy for both output responses and
exhibited exceptional nonlinear fitting capability. This work establishes an
effective modeling approach for simultaneous quality and efficiency optimization
in wheel manufacturing.
Fan Fuhao, Yunlang Zhan, Zhenfei Zhan et al.· SAE technical paper series· 0 citations
A comprehensive data-driven framework integrating ensemble machine learning models with systematic hyperparameter sensitivity analysis and explainable artificial intelligence techniques is proposed, demonstrating that the XGB model significantly outperforms the other approaches, achieving superior accuracy and robust generalization.
Qaim Shah, Waheed Ali Khoso, Fawad Iqbal et al.· Discover Artificial Intellig...· 0 citations
This article presents how ML approaches may speed up material discovery, reduce trial-and-error, and enable individualized design solutions for 3D-printed polymers across a variety of contexts.
N. Senthilkumar, S. Gopalakrishnan, S. Gopinath et al.· Interactions· 0 citations
The demand for sustainable and high-performance materials in additive manufacturing has accelerated the development of reinforced biopolymer composites for fused filament fabrication (FFF). In this study, polylactic acid (PLA) was reinforced with molybdenum disulfide (MoS
2
) and silicon carbide (SiC) powders to fabricate multifunctional composite filaments tailored for 3D printing applications. Experimental datasets were generated by systematically varying filler composition, extrusion temperature, and screw speed during filament production. The experimental testing results of PLA–MoS
2
/SiC filaments were obtained through tensile testing. The porosity trends are consistent with the ANN-based predictions and sensitivity analysis. Lower porosity correlates with higher predicted strength and composition plays a dominant role in reducing void formation. To further enhance predictive capability and process optimization, an artificial neural network (ANN) model was developed for estimating key mechanical properties based on the input processing parameters. The ANN model demonstrated effective optimization yielding high correlation coefficients of 0.95 for training, 0.98 for validation, 0.98 for testing, and 0.96 overall. The optimal validation performance was achieved with a mean squared error (MSE) of 5855.29 at the three epoch, with the training process completing in just 3 epochs. These outcomes confirm that the ANN model exhibits strong stability and reliability in predicting the process parameters. Overall, this work illustrates the potential of ANN-based data-driven modelling to accelerate the design of sustainable PLA-based composite filaments and supports their suitability for next-generation 3D printing applications.
Zeba Jahan, R. Tyagi, Nitesh Kumar· Journal of Thermoplastic Com...· 0 citations
