A novel active learning stochastic Kriging metamodel for improving reliability and stability of additive manufacturing processes

Abstract

Instability in the manufacturing process may lead to significant differences in product quality characteristics under the same set of process parameters, thus directly affecting the reliability and consistency of the product. Reducing quality variation by optimizing process parameters is the key to improving process stability. Process instability results in quality characteristic data with a low signal-to-noise ratio, thereby affecting the optimization and control of process parameters. Furthermore, high manufacturing costs also restrict the sample size available for training models. To identify more robust process parameters at a lower cost and ensure product quality consistency, this study proposes a novel active learning method based on the stochastic Kriging model. Firstly, we considered the necessity of replication and the heterogeneity of variance, establishing a heteroscedastic Gaussian process model that allows for fast inference. Secondly, we proposed an integrated mean squared prediction error (IMSPE) optimization strategy based on an active learning method to balance replication and exploration with fewer samples. Finally, numerical examples demonstrated the effectiveness and advantages of the proposed modeling and active learning methods. A 3D printing case further confirmed that the proposed method outperforms other competing methods regarding model prediction performance and the robustness of optimal process parameters.