Machine learning approach to impact load estimation using fiber Bragg grating sensors

Automated detection of damage due to impact in composite structures is very important for aerospace structural health monitoring (SHM) applications. Fiber Bragg grating (FBG) sensors show promise in aerospace applications since they are immune to electromagnetic interference and can support multiple sensors in a single fiber. However, since they only measure strain along the length of the fiber, a prediction scheme that can estimate loading using randomly oriented sensors is key to damage state awareness. This paper focuses on the prediction of impact loading in composite structures as a function of time using a support vector regression (SVR) approach. A time delay embedding feature extraction scheme is used since it can characterize the dynamics of the impact using the sensor signal from the FBGs. The efficiency of this approach has been demonstrated on simulated composite plates and wing structures. Training with impacts at four locations with three different energies, the constructed framework is able to predict the force-time history at an unknown impact location to within 12 percent on the composite plate and to within 10 percent on a composite wing when the impact was within the sensor network region.