Power consumption estimation for mask image projection stereolithography additive manufacturing using machine learning based approach

Abstract

Listen

Translate article

Abstract Additive manufacturing (AM) or 3D printing has been implemented in a wide range of areas, owing to its superior capabilities of fabricating complex geometries with high design freedom compared to traditional manufacturing. In recent years, the potential environmental impacts that can be caused by AM processes and materials have attracted increasing attentions. Research efforts have been conducted to study and attempt to enhance the environmental performance of AM. In current literature on AM energy consumption, most studies focus on the production stage and investigate the relation between energy consumption and process parameters (i.e., layer thickness). In this work, multiple geometry characteristics (e.g., surface areas and shapes) at each printing layer are studied and linked with the power consumption of mask image projection stereolithography using machine learning based approach. The established models will be able to provide AM designers with a useful tool for estimating power consumption based on layer-wise geometry information in the design stage and promote the awareness of cleaner production in AM. In this work, effective features are selected and/or extracted from layer-wise geometry characteristics and used to train and test machine learning models. According to our results, the shallow neural network has the lowest averaged root-mean-square error (RMSE) of 0.75% considering both training and testing, and the stacked autoencoders (SAE) structure has the best testing performance with RMSE of 0.85%.