Regularized interpolation in 4D neural fields optimizes geometric fidelity in 3D printing

While additive manufacturing (AM) enables unprecedented design freedom, process-induced geometric defects often limit its use in high-precision applications. Consequently, the ability to predict and correct these deviations by dynamically adjusting process parameters remains a critical challenge often necessitating extensive manual iteration, particularly given that current machine learning approaches fail to model the continuous causal link between specific settings and resulting spatial deviations. To address this, we leverage a coordinate-based neural field trained on computed tomography (CT) scans of parts printed under varied process settings, augmented with a Jacobian-based regularization that constrains sensitivity to process parameters and promotes physically consistent geometric changes. By treating this neural field as a differentiable “world model”, we perform data-driven inverse optimization to derive ideal process settings tailored to specific local and global geometric requirements. Experimental validation demonstrates that applying these optimized parameters significantly enhances the fidelity of as-built components, effectively bridging the gap between digital design and physical output while minimizing material waste.