Framework for risk‐based planning of operation and maintenance for offshore wind turbines

Abstract

and Risø-DTU, DenmarkFor offshore wind turbines, costs to operation and maintenance (OM) are substantial. This paper describes a risk-based life cycle approach for optimal planning of OM. The approach is based on pre-posterior Bayesian decision theory, and can be used both to overall, initial planning of OM, and to sequential optimal decision making on planning of OM taking into account new information. Deterioration mechanisms such as fatigue, corrosion, wear and erosion are associated with significant uncertainty. Observations of the degree of damage can increase the reliability of predictions, especially in connection with condition-based maintenance. The approach can be used for gearboxes, generators, fatigue cracks, corrosion, etc. This paper also describes how probabilistic indicators can be used to quantify indirect information about the damage state for critical components, e.g. gearboxes.Operation and maintenance (OM) for offshore wind turbines contribute with a substantial part of the total life cycle costs, and can be expected to increase when wind farms are placed at deeper water depths and in harsher environments. This paper describes how a risk-based life cycle approach can be formulated for rational and optimal planning of operation (services, inspections, etc.) and maintenance (including repair and exchange) for offshore wind turbines. For other offshore installations such as oil and gas installations, cost-effective procedures for risk-based inspection planning have been developed during the last 10-15 years and are used at several locations worldwide, see e.g. Moan, 1 Faber et al. 2 and Sørensen and Faber. 3 These procedures are based on pre-posterior Bayesian decision theory, see e.g. Raiffa and Schlaifer 4 and Benjamin and Cornell. 5 This paper describes how procedures based on a similar theoretical basis can be applied for wind turbines, especially offshore wind farms. For wind turbines, the main aspects related to OM are availability, reliability and cost reductions.Maintenance activities can be divided in corrective and preventive (time-tabled or conditioned) maintenance. Conditioned maintenance using observations from, e.g. condition monitoring and inspections, should optimally be based on risk and pre-posterior Bayesian decision theory. This paper presents the basic principles and how they can be applied for wind energy to optimal planning of OM. Further, illustrative examples are given.

Damage tolerance and structural monitoring for wind turbine blades

“…Failure of a wind turbine blade has small to minor consequence as the risk for human lives is small, especially offshore since persons are not close to the wind turbines. The optimal target reliability level can therefore be determined by cost-benefit analysis, where all the cost during the wind turbines design life is taken into account [13]. Partial safety factors can be calibrated to obtain the desired target reliability level for the structure [14].…”

Towards farm-level health management of offshore wind farms for maintenance improvements

“…This has to be accomplished using methodologies like risk-based maintenance (RBM) [49,50] or opportunistic maintenance methods [51][52][53]. ECN's OMCE [54] and SINTEF's NOWICOB [55] are notable tools for assisting maintenance planning by estimating cost of maintenance scenario that have provision to include the constraints.…”

Observation-based Decision-making for Infrastructure

“…Pure prior and posterior decision problems, as for instance calibration of code format or service life extension, may be solved using standard approaches of Quantitative Risk Analysis (QRA) and Structural Reliability Analysis (SRA). However, efficient treatment of the broader pre-posterior decision class, which includes the problem of inspection and maintenance planning, necessitates dedicated tools (Faber, 2003;Sørensen, 2009), such as Bayesian Probabilistic Networks and Influence Diagrams (Heckerman, 1995). In this work, we elaborate on a candidate approach, namely Partially Observable Markov Decision Processes (POMDPs).…”