![](https://congressarchive.cimne.com/sim-am_2023/assets/logo-sim-am-03_00000002.jpg)
Additive Manufacturing Process-Aware Topology Optimization Using Deep Learning Surrogate Model-Based Constraint Functions
Please login to view abstract download link
In this work, a gradient-based discrete variable optimization strategy is used to include AM processing effects in topology optimization using a data-driven surrogate model. The model was previously developed to predict the residual stress of an LPBF AM part based on the inherent strain method in seconds of computational time, compared to days using finite element analysis. The model was trained for a specific material and process parameter set for a variety of geometries, making the model geometry agnostic. As this model was trained on discrete 0/1 elements, a gradient-based discrete optimizer was used to perform the topology optimization, rather than the computationally intensive genetic algorithm. This novel approach requires minimal additional cost to the overall optimization strategy. The example parts shown have similar end-use performance as those without the process constraint and can be printed without issue.