Feedforward Neural Network as An Approach for Predicting Direct Ink Writing Kinematics Using G-Code

Additive Manufacturing (AM) is a versatile manufacturing technique that can fabricate complex 3D parts. A typical AM printing process involves transforming computer-aided design (CAD) models into G-Code files that drive the kinematics hardware. A big challenge in AM, due to its complex ensemble of hardware, is ensuring consistent quality parts that meet specifications and desired properties; this can introduce deviations between as-designed and as-built parts causing downstream defects and compromised performance. Controlling kinematic hardware is one such component that influences the quality of the final printed part. The data collected from hardware can be massive which can be leveraged by deep learning models to predict kinematics behavior. In this study, we obtained the printer’s kinematics data and G-Code instruction files from Direct Ink Writing (DIW) experiments. We then extract G-Code features that serve as inputs for a Feedforward Neural Network. Our preliminary results show the model can predict X, Y, and Z-axes kinematics with a learning curve fit that suggests neither over- nor underfitting and has Mean Absolute Errors (MAE) ranging from 8.8-1.3%, showing promise in this approach. While demonstrated for DIW, this approach offers potential scalability across AM processes that use G-Code files.

This material is based upon research in the Materials Data Science for Stockpile Stewardship Center of Excellence (MDS3-COE), and supported by the U.S. Department of Energy's National Nuclear Security Administration under Award Number(s) DE-NA0004104 and partially under the auspices of Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344,LLNL-ABS-872712.