Integrated Experimental and Machine Learning-Based Predictive Analysis of FDM Parameters for Conductive PLA-Based Composites
Polymers for Advanced Technologies, vol.37, no.5, 2026 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Volume: 37 Issue: 5
- Publication Date: 2026
- Doi Number: 10.1002/pat.70626
- Journal Name: Polymers for Advanced Technologies
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Chemical Abstracts Core, Chimica, Compendex, INSPEC
- Keywords: conductive polylactic acid, conductivity, deep kernel learning, fused deposition modeling, machine learning
- Kocaeli University Affiliated: Yes
Abstract
This study investigates the effects of fused deposition modeling (FDM) parameters on the tensile strength and electrical conductivity of conductive polylactic acid (PLA)-based composites. An integrated experimental and machine learning (ML)-based framework was used. Printing speed (PS), infill density (ID), layer thickness (LT), and nozzle temperature (NT) were determined as printing parameters. A full factorial design was employed, yielding 81 unique parameter combinations. Experimental results showed that both tensile strength and electrical conductivity were affected by the selected printing parameters. Tensile strength ranged from 16.44 to 49.14 MPa, and electrical conductivity ranged from 0.01332 to 0.10442 S/cm. Overall, lower speed, higher density, lower thickness, and higher temperature improved performance. Analysis of variance (ANOVA) revealed that ID and LT were the most influential factors, with statistically significant effects (p < 0.05) for both response variables. ML models were developed to establish predictive relationships between printing parameters and output responses. For this purpose, a random forest (RF), k-nearest neighbors (KNN), and deep kernel learning (DKL) inspired model were trained and evaluated using five-fold cross-validation. For tensile strength estimation, the RF and KNN models achieved R2 values of 0.8582 and 0.8899, respectively, while the DKL-inspired model provided the highest accuracy with R2 = 0.9999. Similarly, for electrical conductivity estimation, the RF and KNN models produced R2 values of 0.9001 and 0.8810, respectively, while the DKL-inspired model again showed the best performance with R2 = 0.9999. Sample-based error analysis further confirmed the superiority of the DKL-inspired model. For this model, the mean percentage error was 0.14% for tensile strength and 0.28% for electrical conductivity. The results demonstrate that the combined use of experimental design, statistical analysis, and ML provides an effective approach to understanding the multifunctional performance of FDM printed conductive PLA-based composites.