Search

This article develops a physics-guided data-driven model for the height evolution of parts printed in droplet-based additive manufacturing. The proposed model is a convolutional recurrent neural network (ConvRNN) whose structure is derived based on the physical understanding of mass conservation during the height evolution. Because of this physics-guided model structure, the model parameters obtained are invariant to the geometry of the printed part and thus portable from one geometry to another, the conditions on physical stability of the evolution translate directly to training stability of the neural network, and the data required to train this model are much less compared to a pure black-box model. These aspects of the model are validated experimentally on an inkjet 3-D printing setup. The proposed model outperforms a black-box off-the-shelf multilayer perceptron (neural network) by using about two orders of magnitude less data for training, at the same time delivering $1.7\times $ smaller rms error on test data. The proposed model is also compared with a state-of-the-art reduced order linear model and shows $1.4\times $ smaller rms error on test data. Finally, experimental results also underline that the model parameters learned are geometry invariant, that is, the model parameters trained on one geometry can be used to predict the height map evolution for other geometries without relearning. [ABSTRACT FROM AUTHOR]

In this article, we develop a predictive geometry control framework for jet-based additive manufacturing (AM) based on a physics-guided recurrent neural network (RNN) model. Because of its physically interpretable architecture, the model’s parameters are obtained by training the network through back propagation using input–output data from a small number of layers. Moreover, we demonstrate that the model can be dually expressed such that the layer droplet input pattern for (each layer of) the part to be fabricated now becomes the network parameter to be learned by back-propagation. This approach is applied for feedforward predictive control in which the network parameters are learned offline from previous data and the control input pattern for all layers to be printed is synthesized. Sufficient conditions for the predictive controller’s stability are then shown. Furthermore, we design an algorithm for efficiently implementing feedback predictive control in which the network parameters and input patterns (for the receding horizon) are learned online with no added lead time for computation. The feedforward control scheme is shown experimentally to improve the RMS reference tracking error by more than $\boldsymbol{30\%}$ over the state of the art. We also experimentally demonstrate that process uncertainties are compensated by the online learning and feedback control. [ABSTRACT FROM AUTHOR]

This article develops and experimentally validates a distributed predictive control algorithm for closed-loop control of inkjet 3-D printing to handle constraints, e.g., droplet volume bounds, as well as the large-scale nature of the 3-D printing problem. The large number of decision variables, i.e., droplet volumes at each grid point, in high resolution inkjet 3-D printing makes centralized methods extremely time-consuming, thus, a distributed implementation of the controller is necessary. First, a graph-based height evolution model that captures the liquid spreading dynamics is described. Based on this model, a scalable closed-loop control algorithm using distributed model predictive control (MPC) that can reduce computation time significantly is designed and experimentally implemented. The performance and efficiency of the algorithm are shown to outperform open-loop printing and closed-loop printing with existing centralized MPC methods. [ABSTRACT FROM AUTHOR]