Distributed Optical Fiber Damage Identification Method Based on Deep Learning

Abstract

The service environment of aerospace composite structures is harsh. In order to ensure the safe operation of structures, it is necessary to develop structural health monitoring technology. Distributed optical fiber sensor based on back Rayleigh scattering is widely used in the field of structural health monitoring because of its advantages of easy embedding and strong anti-interference ability. How to identify structural damage from complex optical fiber data is one of the research difficulties of health monitoring. Based on this problem, a deep learning method for damage identification is proposed. One dimensional convolution neural network is used to identify debonding and crack damage in composite laminates. In order to verify the reliability of the method, the cantilever loading test of prefabricated damaged phenolic resin laminates is set up. The embedded distributed optical fiber sensor can well monitor the strain change characteristics of the damaged area. The parameters of the network structure are adjusted by using the experimental data, and the convolution core size and the number of convolution layers are finally determined. The experimental results show that the trained one-dimensional convolutional neural network can identify the damage features from the complex strain curve and accurately locate the damage features. In the current research, this method can accurately identify 3cm2 debonding damage and 20 mm long crack damage, and the positioning accuracy is less than 4mm.