Metal Additive Manufacturing

Our producibility-aware topology optimization (PATO) framework enables an approach to incorporate producibility of additively manufactured parts directly into the design process using a deep neural network predictor of producibility. For instance, given a heat conduction problem (left), a typical TO approach generates ‘no-go’ designs with regions of high residual stress that crack after manufacturing (top). On the other hand, PATO leverages an attention-based neural network surrogate model trained on a diverse set of optimized designs to augment the performance sensitivity field with a producibility sensitivity field (here MSSI) obtained through automatic differentiation to generate ‘go’ designs that will remain crack-free after post-processing (bottom).