Session: 04-23-01: Process Development, Characterization, and Optimization for Additive, Subtractive, and Hybrid Manufacturing

Paper Number: 170729

Machine Learning-Driven Optimization of Fdm Process Parameters for 3d Printing of Pla 

Fused Deposition Modeling (FDM) is a widely used additive manufacturing technique due to its versatility and low cost, making it popular for producing functional parts in engineering and manufacturing. Polylactic Acid (PLA), a biodegradable thermoplastic, is among the most common FDM materials. However, the mechanical performance of PLA prints can vary greatly with processing conditions. The choice of process parameters in FDM such as infill, pattern, shell thickness, etc. directly and significantly affects the strength and quality of the printed part. Optimizing these parameters is critical to improve tensile strength and overall mechanical reliability of PLA components. This study is motivated by the need to maximize the mechanical performance of PLA prints through systematic parameter optimization by integrating machine learning (ML) techniques into the process optimization of FDM, thereby enabling broader use of PLA in load-bearing and high-performance applications. While previous studies have identified key parameters influencing PLA part strength, our approach introduces a data-driven optimization framework. We combine experimental design with supervised ML models to efficiently explore the parameter space, something not achievable with trial-and-error methods alone. In particular, a trained ML model is coupled with a genetic algorithm (GA) to intelligently search for the optimal set of printing parameters. This ML-driven optimization strategy represents a novel contribution – it leverages the predictive power of ML to guide parameter selection, significantly streamlining the identification of ideal FDM settings. Recent research has demonstrated the effectiveness of such an approach, where ML models paired with GA successfully pinpoint optimal print parameters to maximize mechanical properties. By adopting this framework, our study provides a robust methodology for improving the strength of PLA prints and showcases how AI can enhance engineering workflows in 3D printing.

In this research, a Taguchi experimental design (L9 orthogonal array) was employed to evaluate the tensile performance of PLA under varying print parameters. Three key FDM parameters were selected based on their expected influence on mechanical properties: infill percentage, infill pattern, and shell thickness (outer wall count). Each parameter was tested at three levels (e.g., low, medium, high), and PLA tensile specimens were printed for each combination in the Taguchi matrix. Standard tensile tests (ASTM D638-type methodology) were conducted to measure young’s modulus, ultimate tensile strength and elongation at break. The experimental data were then used to train a supervised ML model (an artificial neural network) to capture the relationship between process settings and tensile outcomes. Building on this model, we implemented a genetic algorithm that iteratively searches the parameter space to find the combination that maximizes predicted tensile strength. This coupled ML+GA framework effectively serves as an optimizer, suggesting improved settings beyond those explicitly tested. The Taguchi-based tensile tests yielded clear insights into the influence of each process parameter on PLA’s mechanical properties. Statistical analysis (ANOVA) of the results indicates that infill pattern and shell thickness are two of the most critical factors governing tensile strength, with infill percentage also contributing significantly. In particular, the infill pattern emerged as a dominant factor – certain patterns facilitate better stress distribution, leading to higher strength – followed in importance by the shell thickness (number of perimeters or wall layers) which provides structural reinforcement to the parts. For instance, specimens printed with a robust infill geometry and thicker outer shells exhibited substantially higher ultimate tensile strength compared to those with sparse infill and thin walls. Quantitatively, the best parameter combination from the Taguchi array (e.g. a high infill percentage coupled with an optimized pattern) achieved markedly improved tensile performance over the worst-case combination, underscoring the value of proper parameter selection. These experimental findings will be used to train the ML model. Overall, the early results confirm that all three parameters examined have a significant effect on PLA’s tensile properties.

Presenting Author: matthieu hallaert University of Lille - Unité de Mécanique de Lille - Joseph Boussinesq (UML)

Presenting Author Biography: PhD student from France, specialised in solid mechanics

Authors:

matthieu hallaert University of Lille - Unité de Mécanique de Lille - Joseph Boussinesq (UML)
Ahmed Ammar University of Lille - Unité de Mécanique de Lille - Joseph Boussinesq (UML)
Toufik Kanit Université de lille - UML
Ahsan Mian Wright State University