Regression-Based Surrogate Model for Rapid Prediction of Temperature Evolution in a Microscale Selective Laser Sintering System

Abstract

Abstract Existing metal Additive Manufacturing (AM) tools suffer from limitations on the minimum feature sizes of their producible parts. The Microscale Selective Laser Sintering (μ-SLS) system directly addresses this restriction with a minimum part resolution on the order of a single micrometer. However, the production of parts at the micrometer scale is unreliable due to unwanted heat transfer in the nanoparticle powder bed. As a result, Finite Element thermal models are developed to predict the temperature evolution within the particle bed during sintering. These thermal models are not only computationally expensive, but also must be integrated into an iterative control framework to optimize the digital mask used to selectively control laser power, making it critical to enable quick temperature predictions. Therefore, this paper proposes a regression-based Machine Learning model to facilitate accurate and rapid predictions of the thermal evolution within the μ-SLS particle bed and subsequent optimization of the system input parameters. The regression model presented in this work uses an “Element-by-Element” approach, where machine learning models are trained on individual finite elements to learn the relationship between the conditions experienced by the element at the current timestep and the element’s temperature at the subsequent timestep. An existing bed-scale thermal model of the μ-SLS system is used to generate element by element training data for various regression models. Key features for each element include the temperature and applied volumetric heat generation of the selected element and its surrounding elements at the current timestep. Additional features are included that represent the spatial distribution of laser energy relative to the selected element. This feature vector is used to predict the temperature of the given element at the next timestep. Using this dataset, a feed-forward Artificial Neural Network with 3 hidden layers is used to predict the temperature evolution of a 2D powder-bed over a 2 second sintering window with high accuracy.