Manufacturing quality prediction using smooth spatial variable selection estimator with applications in aerosol jet(R) printed electronics manufacturing

Additive manufacturing (AM) has advantages in terms of production cycle time, flexibility, and precision compared with traditional manufacturing. Spatial data, collected from optical cameras or in situ sensors, are widely used in various AM processes to quantify the product quality and reduce variability. However, it is challenging to extract useful information and features from spatial data for modeling, because of the increasing spatial resolutions and feature complexities due to the highly diversified nature of AM processes. Motivated by the aerosol jet(R) printing process in printed electronics, we propose a smooth spatial variable selection procedure to extract meaningful predictors from spatial contrast information in high-definition microscopic images to model the resistances of printed wires. The proposed method does not rely on extensive feature engineering, and has the generality to be applied to a variety of spatial data modeling problems. The performance of the proposed method in prediction and variable selection through simulations and a real case study has proven to be both accurate and easy to be interpreted.