Neuro-Fuzzy Model Evaluation for Enhanced Prediction of Mechanical Properties in AM Specimens

Abstract

This paper explores the integration of adaptive neuro-fuzzy inference systems (ANFIS) with additive manufacturing (AM) to enhance the prediction of mechanical properties in 3D-printed components. Despite AM’s versatility in producing complex geometries, achieving consistent mechanical performance remains challenging due to various process parameters and the anisotropic behavior of printed parts. The proposed approach combines the learning capabilities of neural networks with the decision-making strengths of fuzzy logic, enabling the ANFIS to refine printing parameters to improve part quality. Experimental data collected from AM processes are used to train the ANFIS model, allowing it to predict outputs such as stress, strain, and Young’s modulus under various printing parameters values. The predictive performance of the model was assessed with the root mean square error (RMSE) and coefficient of determination (R2) as evaluation metrics. The study initially examined the impact of key parameters on model performance and subsequently compared two fuzzy partitioning techniques—grid partitioning and subtractive clustering—to identify the most effective configuration. The experimental results and analysis demonstrated that ANFIS could dynamically adjust key printing parameters, leading to significant improvements in the prediction accuracy of stress, strain, and Young’s modulus, showcasing its potential to address the inherent complexities of additive manufacturing processes.