Flexible resistance strain sensors gain tremendous attentions in emerging fields such as wearable sensing, electronic skins and human–machine interfaces due to its excellent stretchability and sensitivity. However, due to the coupling of strain along both principal and perpendicular axes, conventional uniaxial strain sensors face challenges in recognizing multi-directional strain which naturally occurs in practical applications. In this work, we present the omnidirectional strain sensors comprised of multiple-stacked oriented conductive layers and harness the artificial neural network (ANN) to assist recognizing strain information with high accuracy. Each conductive layer consists of a patterned multi-walled carbon nanotubes mixture coated on a soft substrate, exhibiting remarkable sensitivity with a gauge factor of 4.18 along the principal direction. When centrally aligning two or three layers along different directions, the assembled structures can monitor the multi-directional electronic resistance changes under deformations, enabling the recognition of both strain direction and amplitude. The ANN model used three-dimensional resistance data as input to perform a multi-class classification task, with strain amplitude and direction as the outputs. The hidden layers employed the ReLU activation function, while the output layer utilized the SoftMax activation function. The model was compiled with the sparse_categorical_crossentropy loss function and optimized using the Adam optimizer. Its performance was evaluated based on prediction accuracy. The strain prediction accuracy is above 98 % with the assistance of an ANN model. The triple-layer configuration can perceive the human wrist motions such as rotation and bending, exhibiting high potential in applications such as multidimensional wearable electronics.
Read full abstract