A hybrid machine learning approach for the quality optimization of a 3D printed sensor

Abstract

Sensors play a crucial role in train condition monitoring as it can offer real-time data of the train for health status estimation, fault diagnosis and decision-making of maintenance. In order to improve the performance of the sensors for data acquisition, an aerosoljet 3D printing technology is adopted to print customized sensors in this research. Compared with conventional bulk sensors, the customized sensors have a smaller size, higher accuracy, faster response time and could be printed onto the surface directly. However, as the line morphology of printed patterns has significant influence on the electrical properties, we need to investigate the influence of the process parameters on the line morphology and optimize the printed line quality. In this paper, we consider sheath gas flow rate and carrier gas flow rate as the key process parameters. The line roughness and line overspray are considered as the line quality indices. Latin hypercube sampling is adopted to fully explore the entire design space. And, a hybrid machine learning approach is proposed to analyze the relationship between line morphology and the process parameters, and finally an optimal operating window is identified based on the proposed approach.