One-dimensional convolutional neural network for damage detection of jacket-type offshore platforms

Abstract

Vibration-based damage detection techniques play an important role in health monitoring of offshore structures. This study explores the possibility to use the one-dimensional convolutional neural network (CNN) to extract the damage sensitive features automatically from the raw strain response data of a structure under a certain excitation without requiring any hand-crafted feature extraction. The validity of the proposed method is verified by using a numerical simulation of a jacket-type offshore platform under a regular and a random wave excitations in different directions, respectively. The damage localization and damage severity identification are conducted with considering several damage cases including different damage locations and the effect of noise. The data preprocessing procedure based on convolution and deconvolution for noisy data is proposed to enhance the capability of feature extraction and noise immunity of CNN. Furthermore, the experimental studies of a jacket-type offshore platform model subjected to a sinusoidal excitation, a white noise excitation and an impulse excitation are respectively investigated to check the applicability of the method, in which the major damage and minor damage on single and multiple locations are involved. Results indicate this approach has an excellent performance on structural damage detection.