Review on Structural Health Monitoring of Offshore Platform

The structural health monitoring (SHM) and damage detection of offshore platforms, one of the most common marine structures operating in a hostile environment, have gained global attention in recent years. This paper presents a review of vibration-based damage identification methods used for SHM of offshore platforms. The application and progress of these methods are discussed and some case studies are analyzed. The challenges and future work for vibration-based damage identification are summarized.

between the changes in modal parameters, the damage location and the damage size by using shifts in resonant frequencies, mode shape changes, strain mode shape, flexibility matrix or updating structural model parameters. Some of these methods have been successfully used in SHM of offshore platforms.

Mode shape changes
The underlying assumption of this method is that the modal vectors of degrees of freedom (DOFs) near the damaged members should vary more than those of the degrees of freedom away from the damaged members. Numerical data in previous studies demonstrated the usefulness of the modal assurance criteria (MAC) approach. Then many researchers have presented more sensitive indicators, such as Node line MAC and coordinate MAC (COMAC). Viero et al. [4] compared the performance of MAC and COMAC. Two indicators were applied on hydro-elastic offshore platform models which were all to scale based on the similitude theory. Zarrin et al. [5] exploited modes of vibration to improve the accuracy of static pushover analysis for the design of jacket type offshore platforms (JTOPs) under the abnormal level earthquake (ALE). Haeri et al. [6] conducted a case study where they used mode shapes as one of the damage indexes and employed the inverse vibration technique on the models located in the Persian Gulf. To illustrate the effectiveness and feasibility of the proposed procedure, three levels (5%, 10% and 15%) of random error were considered. As shown in Table 1, the coefficient of variation varied slightly. In terms of random errors, the proposed procedure was robust.

Mode shape curvature/strain mode shape(SMS) changes
When a structure experiences damage or change, the change of force distribution can be noted from the SMSs of the structure measured before and after the damage. Since force redistribution is, in general, greatest near the damaged area, the location of damage is implicitly identified by the severity of the SMS change. Researchers dramatically improved results by using measured strains instead to measure curvature directly or new indexes such as the Difference between the Real and Estimated Curvature Function, DREC [7]. Kondo et al. [8] made use of the identified modal properties. The damage region containing a damaged member was isolated from the whole structural system in terms of the curvature change ratio for detecting the local damage of flexible offshore platforms. Nonetheless, there exist several obstacles which impede the application of modal-based methods. For instance, plenty of measurement locations are indispensable to provide better accuracy in mode information. In addition to that, mode shapes and their derivatives often present larger variability of statistical data than natural frequencies.

Frequency response functions (FRFs) and their variants
Modal-based methods are susceptible to measurement noise, while extensive studies indicate that FRF based methods are more reliable. FRF data are acquired directly from vibration response signals. Another advantage of using FRF data is that it provides enough information and equations in the frequency domain. Wang et al. [9] derived an algorithm which achieved damage localization and measured the magnitude of the damage based on nonlinear perturbation equations of FRF data that were weighted by a further developed technique. Teloli et al. [10] visualized the damping ratio in the experimental FRF and combined the Bouc-Wen model with higher-order FRF data as an alternative, 3 resolving conundrum involving experimental bolted joints with data fluctuation. Transmissibility, a variant of the FRF, is found more sensitive than FRFs [11]. Amin Fathi et al. [12] applied a new Bayesian model updating framework combined with incomplete noisy FRF data in the health monitoring of an offshore platform for the first time. The accuracy between the actual and predicted damaged values of the research results was compared with that conducted by Liu et al. [13], who used the mode shapes and natural frequencies for damage detection of the fixed platform by calculating the Mean Sizing Error (MSE) and Root Mean Square Error (RMSE).

Recent development
With the advancement of mathematical heuristic methods, more complicated and developed methodologies have been studied for SHM of marine platforms, among which wavelet based methods, the use of neural networks and model updating have gained much attention worldwide. But some of these studies have employed complicated procedures which render them still far from real engineering cases.

Wavelet based methods
The wavelet transform (WT), one of signal processing-based methods, has been widely used in both civil and mechanical SHM applications. The strength of WT lies in its capability to provide multiple levels of details and approximations of the transient signal [14]. Wavelet packet transform (WPT) is another form of the WT, which can eliminate noise, process and extract sensitive features of the signal. Filho et al. [15] analyzed the signals through wavelet transform and the results were compared with those obtained by the FRF based method, verifying that wavelet transform was more sensitive than the classical frequency response methods. The WT, however, is unfit for the higher frequency region. Hence, more WT techniques such as Wavelet multi-resolution analysis (WMRA) and wavelet packet transform have been developed to overcome the limitation. Asgarian et al. [16] used WPT to analyze experimental results of a prototype scaled marine platform under several damage conditions. Lotfollahi-Yaghin et al. [17] used the wavelet packet energy rate index as damage indicator for a jacket type offshore platform in Persian Gulf .

Machine learning(ML) methods
Machine learning has proved to be suitable for a broad range of industrial applications, divided into supervised, unsupervised and semi-supervised learning modes [18]. To avoid results highly depending on the users' judgment, Yuen et al. [19] developed a practical and rational Bayesian ANN design method and apply it to a damage diagnosis method as an illustrative example. Aqdam et al. [20] presented new architecture of RBF Radial Basis Function (RBF) neural network, a new design of ANN, for the SHM of mooring lines, essential components in marine platforms. Bao et al. [21] explored the possibility to use the one-dimensional convolutional neural network (CNN) (figure 2) and optimize it through a data processing procedure. The results of this method were confirmed by using a numerical simulation and experimental study of a marine platform respectively.  Table 2, Table 3. As shown in Table 2 and Table3, the total predicted accuracy of damage location on multiple members is 100% and 60.4%, respectively. In terms of multiple full member damage scenarios, CNN method reaches a superior performance level. Nonetheless, the results are not satisfactory for very minor multiple-damage locations on local elements, especially under a random wave excitation [21].  Table 4. Prediction of damage location of case 4 under a random wave. [21] Key factors limiting the widespread implementation of neural network solutions in industrial scenarios have been the difficulty of demonstrating that the deep networks are robust and reliable enough to ensure the authenticity and accuracy of outcomes.

Model updating methods
The model updating procedure is aimed to minimize the disparities between the numerically and experimentally signals through regulating uncertainty parameters. Thus model updating procedure comes down to optimization problems [22]. SHM systems have used model updating methods based  [23], FRF data [12], dynamic strain responses [24] and so on. Mojtahedi et al. incorporated the fuzzy logic system and FE-model updating (FEMU) for health monitoring of offshore jacket platforms [25].

Conclusion
Many researchers have combined the aforementioned methods when conducting studies. Many emerging procedures for SHM of offshore platforms are still at the theoretical level, far from the practical goals. At present, most methods are based on modal parameters used for damage identification and localization. Due to the limitation of modal parameter extraction in actual operation, especially when there exists the influence of noise, the result of extraction is far from good as the experiment. The damage identification and location methods are generally limited to the simulation study of the model, and many researchers have carried out the studies on the scaled offshore platform models. Future research needs to make the SHM based on vibration measurements combined with more developed procedures and mathematical theories a viable, practical, and commonly implemented technology.