Deep learning methods for damage detection of jacket-type offshore platforms

Abstract

Recently, big data and machine learning based damage detection methods to support risk management of offshore facilities have received great attention, compared to traditional modal parameters-based methods. This paper illustrates the application of deep learning methods in damage detection of offshore platforms using measured vibration response of the structures subjected to random excitations. The numerical example of a jacket-type offshore platform under random wave excitation is applied to verify the applicability of convolutional neural network (CNN), long short-term memory (LSTM) networks, and CNN-LSTM method. The comparison of the three approaches are conducted in terms of accuracy and efficiency of damage localization and severity estimation for the simulated damage cases. In addition, the random decrement technique (RDT) for data preprocessing is used to improve the capability of damage detection of the three deep learning methods in noisy conditions. Moreover, the proposed RDT combined with the deep learning methods are applied to laboratory tests of a jacket platform model under random loading produced by a shaking table. Minor and major damages at different locations are discussed. Results show that the proposed combination method has an outstanding performance in structural damage detection even in noisy conditions, and also has great potential application in industrial process safety and operational risk management.