Predicting Mechanical Strength and Optimized Parameters in FDM-Printed Polylactic Acid Parts Via Artificial Neural Networks and Desirability Analysis

• Published in: Management systems in production engineering Vol. 32; no. 3; pp. 428 - 437
• Main Authors: Abdulridha, Hind H., Abbas, Tahseen F., Bedan, Aqeel S.
• Format: Journal Article
• Language: English
• Published: Gliwice Sciendo 01.08.2024; De Gruyter Poland
• Subjects: ANN; ANOVA; Design of experiments; FDM; Neural networks; PLA; Polylactic acid; process parameters; Variance analysis
• ISSN: 2450-5781; 2299-0461; 2450-5781
• DOI: 10.2478/mspe-2024-0040

Fused deposition modeling (FDM) is a commonly used additive manufacturing (AM) technique in both domestic and industrial end-product fabrications. It produces prototypes and parts with complex geometric designs, which has the major benefits of eliminating the need for expensive tooling and flexibility. However, the produced parts often face poor part strength due to anisotropic fabrication strategies. The printing procedure, the kind of material utilized, and the printing parameters all have a significant impact on the mechanical characteristics of the printed item. In order to predict the mechanical properties related to printed components made with the use of FDM and Polylactic Acid (PLA) material, this study concentrates on developing a prediction model utilizing Artificial Neural Networks (ANNs). This study used the Taguchi design of experiments technique, utilizing (L25) orthogonal array as well as a Neural Network (NN) method with two layers and 15 neurons. The effect of FDM parameters (layer thickness (mm), percentage of infill density, number of top/bottom layers, shell thickness (mm), and infill overlap percentage) on ultimate tensile and compressive strength (UTS and UCS) was examined through analysis of variance (ANOVA). With an ANOVA result of 67.183% and 40.198%, respectively, infill density percentage was found to be the most significant factor influencing UCS and UTS dependent on other parameters. The predicted results demonstrated valuable agreement with experimental values, with mean squared errors of (0.098) and (0.326) for UTS and UCS, respectively. The predictive model produces flexibility in selecting the optimal setting based on applications.