Evaluation and prediction of thermal defects in SLM-manufactured tibial components using FEM-based deep learning and statistic methods.

Selective laser melting (SLM) is crucial for fabricating complex geometries, such as the tibial component of an artificial knee joint. However, its inherent high-temperature gradients often induce thermal defects and deformations, compromising product accuracy. This study addressed these challenges by using a multifaceted approach in which finite element method (FEM) simulations, Taguchi method analysis, and deep neural network (DNN) prediction were combined. Initially, a transient thermal model was developed using the FEM to analyze the thermal behavior of a tibial component during and after SLM; the model was then validated through experiments. Subsequently, the Taguchi method was employed to evaluate the thermal influence of various SLM parameters on deformation and stress. Furthermore, a K-means-based exploration method was developed and used to identify critical thermal areas affecting a component's size and quality. Finally, a DNN model was developed for rapid prediction of thermal deformation by leveraging an FEM proxy modeling methodology to facilitate efficient monitoring in a digital-twin framework. The FEM analysis revealed average deformation errors of 3.26% during SLM and 0.05 mm after SLM. The optimal SLM parameters for minimizing thermal deformation and stress for a tibial component were identified. The DNN model, trained on the proxy database, achieved error margins of only 5.47% and 4.62% in comparison with the FEM and experimental results, respectively, but took substantially less computation time than did the FEM. This study integrated FEM simulations, the Taguchi method, and DNN prediction to enhance the accuracy of SLM manufacturing of tibial components. [ABSTRACT FROM AUTHOR]