Quantifying the value of SHM for wind turbine blades

In this paper, the value of information (VoI) from structural health monitoring (SHM) is quantified in a case study for offshore wind turbines (OWTs). This is done by combining data from an operating turbine equipped with a blade SHM system with cost information from a service provider for OWTs in a Bayesian decision framework. The reliability of the blade SHM system is evaluated based on a monitoring campaign with a 225 kW Vestas V27 wind turbine, where one of the blades was introduced to an artificial trailing edge damage of increasing size. The blade was equipped with a prototype of an SHM system, which consists of an electro-magnetic actuator that periodically impacts the blade and an array of accelerometers mounted along the leading and trailing edges of the blade. Changes in the structural integrity can be detected using conventional outlier analysis, where the current state of the blade is compared to a statistical model from the healthy state using a metric that yields a damage index representing the structural integrity. As the damage was introduced artificially, it is possible to statistically estimate the confusion matrix corresponding to different threshold values, and here we opt to select thresholds to optimize the value of SHM. Based on SHM data from the V27 wind turbine, a probabilistic model is developed for the relation between the damage level and indicator, and this is assumed to be representable for the reliability of similar SHM systems installed on OWTs. A case study is developed to quantify the value of SHM for an 8 MW OWT using a decision framework based on Bayesian pre-posterior decision analysis. Deterioration is modelled as a Markov chain developed based on data, and the costs are obtained from a service provider for OWTs. Discrete Bayesian networks are used for deterioration modelling and Bayesian updating within the decision framework. First, the value of SHM is evaluated for different interference thresholds for the damage indicator. Then, strategies are applied using thresholds for the probability of failure, which is updated using Bayesian networks with damage indicators received from the SHM system. Three sensor configurations are tested, and for the least reliable configuration, the strategy using thresholds for the probability of failure results in much higher VoI than the strategy using a threshold for the damage indicator. For the most reliable configuration, they result in similar VoI.