Spatially Resolved Strain Measurement at Meter Scale Using a Carbon Fiber Based Strain Sensor and Artificial Neural Networks

Abstract

Life cycle optimization, maintenance planning and adaptive control systems in fiberreinforced structures such as aircraft wings require the monitoring of loads and stresses during operation. State of the art systems using strain gauges can measure strains at limited numbers of discrete points, while systems based on fiber optic time domain reflectometry require complex and cost intensive evaluation units. A novel sensor based on electrical time domain reflectometry (ETDR) allows to acquire information about the spatial distribution of strain along a fractured carbon fiber (CF) embedded in a composite structure. This sensor concept has been investigated in previous studies with specimens up to 60 mm in length. Based on this work, a demonstrator with an improved sensor layout and two embedded sensors of 1 m length is developed. A shallow feed-forward network and a convolutional neural network are compared regarding their ability to infer strain profiles from measured ETDR reflectograms. The simultaneous evaluation of two sensors with a convolutional neural network allowed the inference of strain distributions with a good generalization ability.