Residual displacements are inevitable [1] and a large challenge in metal additive manufacturing [2]. Process simulations based on the finite element method can calculate these displacements. The research focus is split between improving the virtual model of the process and improving its efficiency [3]. For an example of the latter, the inherent strain simulation is a widely used method for calculating the resulting deformation by inserting layer-wise process-specific strain [4–8]. This work is the basis for improving the computational efficiency of most finite element simulations in additive manufacturing by utilizing machine learning algorithms in inherent strain simulations. The mechanical equilibrium is calculated by an implicit equation, solved using a newton-method [9]. The algorithm is initialized with a test solution for which the displacement of zero in new nodes is assumed. However, since the displacement in the new layer behaves like the previous layer, a node-based neural network trained on the current simulation can predict these test solutions. Thus, it potentially improves the efficiency of the newton-solver in future work. A linear regression slope of up to 0.99 was obtained, with 1.00 being an exact prediction. The self-similarity in displacement behavior between layers provides a good use case for neural network-based predictions. The implicit solver in the finite element method will require fewer iterations to solve the equilibrium due to a better test solution without loss of accuracy. The node-based approach will generate enough training data within the simulation itself. This preliminary work will be the starting point for machine learning applications to enhance finite element simulations in additive manufacturing simulations.