Abstract
Low-coherent fiber-optic sensors combined with neural network algorithms were designed to carry out a load-sensitizing spherical bearing. Four sensing fibers were wound around the outside of the pot support of the spherical bearing uniformly deployed from upper to bottom. The upper three were configured in a distributed way to respond to the applied load as a function of the three strain sensors. The bottom one was employed as a temperature compensation sensor. A loading experiment was implemented to test the performance of the designed system. The results showed that there was a hysteresis in all the three sensors between loading and unloading process. The neural network algorithm is proposed to set up a function of the three sensors, treated as a set of input vectors to establish the input-output relationship between the applied loads and the constructed input vectors, in order to overcome the hysteresis existing in each sensor. An accuracy of 6% for load sensing was approached after temperature compensation.