Optimization of Infill Parameters for Fused Filament Fabrication (FFF) of Polyether Ether Ketone (PEEK) Using Adaptive Neuro-Fuzzy Inference System (ANFIS)

In this work, the infill parameters of fabricated samples of additively manufactured PEEK using FFF is analyzed and an optimized model of ANFIS tends to predict the tensile strength of printed part. The goal is to get the best “trade-off” between minimizing the amount of printing time and printed material and maximizing tensile strength. This is done through experimentation understanding the contribution of the combination of three infill type (IP): Gyroid, Line, and Tri-hexagon and three infill density (ID): 30, 50, and 70% to the tensile strength of the printed part. An Adaptive Neuro-Fuzzy Inference System (ANFIS) is used to model the non-linear relationship between the two input and output factors. ANFIS combines the strength of both neural networks and fuzzy logic properties, the purpose is to leverage the learning capabilities of neural networks in defining complex rules for the fuzzy inference system. One-hot encoding is used to convert IP, categorical to numerical data. The generated FIS structure with grid partition of [3, 3] with gaussian membership function with a total of 15 rules. The FIS model is trained, tested, and checked with an RMSE of 2.2383, 1.0298, and 1.4846 respectively and coefficient of correlation of 0.8958. The ANFIS surface plot suggests two optimal points, both for Line IP and with two ID of 35% and 55% yielding approximately around 40 MPa and 50 MPa.