Data-Driven Process-Product Optimization Under Uncertainty for Large Scale Additive Manufacturing

Design of parts for Large Scale Additive Manufacturing (LSAM) represents a complex problem because of the interplay of process, structure, property, and performance (PSPP). This study seeks to address this problem by developing a multilevel optimization approach that combines data-driven and physics-based models in a stochastic (Bayesian) framework. Model training is done by performing inference on experimental process, structure, and property (PSP)data that include flexural testing and microscopy. The framework’s usefulness is demonstrated via a coupled process-product design optimization under uncertainty on a nine-bar frame example problem. The PSP relationships are informed by a Bayesian model for predicting the probability distributions of properties. Stochastic modeling of PSPP relations and their integration in a multilevel optimization serve as a crucial step in the design and fabrication of reliable LSAM produced structures. The work also represents a step towards integrating Multidisciplinary Design Optimization (MDO) with experimental data.