Omnidirectional strain sensing using single flexible acoustic wave device with machine-learning algorithm

Abstract

Strain sensors are crucial for development of smart systems, providing valuable feedback on the conditions of structures and mechanical components. However, there is a huge challenge for highly accurate detection of both strain intensity and direction (i.e., omnidirectional strain) using one single strain sensor, mainly because only one signal feature is commonly obtained from a single device. To overcome this limitation, we proposed a strategy to achieve omnidirectional strain detection by applying a single flexible surface acoustic wave (SAW) strain sensor, empowered by a machine learning algorithm to analyze multiple signals derived from the same device, simultaneously. Using AlN/flexible glass based SAW devices, we performed omnidirectional strain predictions using eight different machine learning models, and the data were compared with the experimental measurement results. The results showed that the extreme gradient boosting (XGBoost) model showed the highest prediction ability and the best accuracy (i.e., with its coefficient of determination larger than 0.98 and root mean square error less than 0.1) for both strain intensity and direction. This work provides an effective solution for omnidirectional strain sensing using a single device.