머신러닝-SHAP value를 활용한 탄소 나노 전왜성 센서의 형상설계 및 온도보상 연구

Abstract

A carbon nano composite having piezoresistivity can be fabricated with nano carbon isotopes such as carbon nanotubes (CNTs), and graphene as fillers in an epoxy matrix. A piezoresistive composite strain sensor is one of the promising applications of the carbon nano composite using the excellent properties of the fillers, and there have been many studies. The piezoresistive composite strain sensor may be able to design its sensitivity by the fabrication parameters such as filler weight percent (wt%) and the geometric pattern of the sensor. However, it is difficult to accurately predict the sensitivity because variables in the composite fabrication process may induce nonlinear effects on sensor characteristics. The piezoresistive sensor has suffered a sensitivity design difficulty due to a lack of piezoresistivity analysis. The piezoresistive sensitivity designs seem to depend on experience, and systematic study is needed to solve the uncertainty of its fabrication process. Still, only a few studies have attempted to design the piezoresistivity of the sensors. This study showed that filler content and patterns of the composite could determine the sensitivity of the CNT/epoxy strain sensor based on the tunneling resistance model. Through this, it was shown that a gauge factor could be designed by the composite geometrical pattern. Data from 1,200 sensor samples were analyzed to experimentally investigate the contributions of the carbon nano composites' gauge factor via the filler contents, length, and width variations. The data were quantitatively analyzed via the lightGBM model and SHAP (SHapley Additive exPlanation) value. The analysis revealed that the filler content of the composite contributes the most to the gauge factor. From the view of the geometric pattern of the sensor, the length does not affect the gauge factor. However, the pattern width is related to the piezoresistivity, and a narrower width can improve the sensitivity. Another experimental study was also conducted to compensate for the temperature dependence of the carbon nano composite piezoresistive sensor. The electrical resistance of CNT/epoxy-based piezoresistive sensors varies with environmental temperature, and the variation may deteriorate its reliability for applications. The temperature compensation effect for the sensors was experimentally studied via comparisons with the signal outputs from 3 Wheatstone bridge types, such as principle half bridge, full bridge, and quarter bridge signals, respectively. The experiment confirmed that the voltage signal changed to 6 mV/V in the quarter bridge, and the Wheatstone bridge in half and full bridge showed output voltage variations within 1 mV/V under temperature changes. Through these results, this study presented a new sensitivity design strategy for carbon nano composite piezoresistive sensors using machine learning. It also provided temperature compensation to improve sensor reliability for signal processing.