Peak-over-threshold extreme value analyses
TWI used the peak-over-threshold (POT) method to model extreme observations (severe defects) in the corrosion data of the conductors. First, the corrosion data was mapped into a matrix (Figure 2) for statistical evaluation. TWI then used statistical
methods to determine a severe-defect threshold (u). Any defect depth readings (exceedances) that exceeded u (Figure 3) were fitted with generalised Pareto distribution (GPD), allowing the maximum defect depth to be estimated.
DBSCAN de-clustering
The growth of one defect is highly likely to influence the growth of neighbouring defects. These defects are said to be “locally dependent”. TWI used density-based spatial clustering of application with noise (DBSCAN) to de-cluster the corrosion data and filter out the dependent observations, such that the remaining exceedances were approximately independent. Figures 4 and 5 show the defect data before and after de-clustering, respectively.
Stochastic defect depth simulation
Because localised corrosion exhibits stochastic behaviour, it is inappropriate to use a constant corrosion rate to predict future defect depths on a component. TWI therefore applied a stochastic simulation using geometric Brownian motion (GBM) and POT to model the corrosion growth beginning from the conductor’s commissioning date. Once the simulation showed a maximum defect depth had grown beyond the maximum allowable depth, it was considered at the end of its useful life (Figure 6).
Probabilistic remaining useful life
TWI also developed a probabilistic remaining useful life assessment model for the client. It integrated the limit state function of conductors by using Monte Carlo and importance sampling methods to approximate the target probability of their failure. The remaining useful life was identified by the difference between the current and the target probability of failure, as shown in Figure 7.
As a result of this work the operator was able to extend the life of its offshore conductors and predict where failures were most likely to occur, enabling optimal use of maintenance and inspection resources.