Analyzing additive manufacturing process capability based on geometrical and dimensional tolerances using machine learning algorithms

Additive manufacturing has garnered attention in recent years owing to significant technological advancements. However, the intricate error mechanisms in both the physical and digital chains of the additive manufacturing process lead to the geometric inaccuracies of the final product. This poses a significant challenge in designing and setting up tolerance for additive manufacturing products. This research introduces a novel algorithm to predict geometric distortions in additive manufacturing specimens produced using the laser powder bed fusion (LPBF) method. Initially, the manufacturing procedure of cylindrical parts with varying diameters is simulated, considering different values for layer thickness, laser power, and scanning speed as variable parameters during the manufacturing procedure while adhering to appropriate tolerances. Subsequently, the geometric distortions resulting from the manufacturing process are calculated based on finite element simulation results. In the next step, an artificial neural network algorithm accurately predicted the created geometric errors in the parts with ۸۸% accuracy. Through data calculation aided by tolerance values, which stem from changes in the part's shape through the production procedure, the quality and capability of the manufacturing are predicted using the k-nearest neighbor (KNN) algorithm for all specimens. The presented model demonstrated an ۸۹.۲۸% accuracy in predicting the manufacturing process capability for cylindrical samples. This predictive algorithm for estimating geometrical distortions and process capability in additive manufacturing offers opportunities for enhancing efficiency and reducing production costs.

Keywords:

Process capability analysis , Geometric distortions , Laser powder bed fusion (LPBF) method , Artificial neural network , KNN algorithm