PROBABILISTIC LIFING METHODS FOR DIGITAL ASSETS

Modern industrial gas turbines (IGTs) must be capable of operational flexibility to fulfil the requirements of an evolving power industry. Base load applications require turbines to operate for long periods at full load conditions whilst load-following applications require turbines to undergo repeated start-stop cycles. Traditional design and lifing approaches, which are based around an assumed worst case operational scenario and critical damage mechanism, cannot fully represent the durability of components when exposed to flexible operation. Condition based assessments, conversely, consider multiple operating scenarios and damage mechanisms to more accurately predict the durability of components. Condition based assessments are particularly powerful when applied to digital assets, where component lives can be calculated for each individual turbine based on detailed operational data. Despite the additional data available to conduct assessments of a digital asset, the information about the asset's manufacture, maintenance or environment is unlikely to be complete or perfect. This leads to uncertainty in the current and future condition of the asset, which must be accounted for when deciding upon maintenance, retirement or life extension. The uncertainty can be accounted for using bounding assumptions or safety factors, but these approaches often lead to overly conservative results and do not provide any insight into the underlying causes of the uncertainty. Probabilistic methodologies provide a means to accurately evaluate and interrogate this uncertainty, by explicitly considering the potential variation in calculation inputs and assumptions. The degradation of hot gas path components by creep-fatigue mechanisms often limits turbine life. Probabilistic creepfatigue assessment methods have been developed and are used to predict and understand the uncertainty in creep-fatigue damage. However, deploying these methods across a large fleet of digital assets, each with multiple components presents several challenges: the assessments rely on Monte Carlo sampling or other discretised calculations and hence are too computationally intensive to be used in real time on a large fleet; assets have often not been digitized for their entire operating lives, hence periods of missing data must be accounted for; finally, predicting the uncertainty of future operation requires information about the likely distribution of future operating regimes. This paper presents a methodology to effectively calculate the uncertainty on hot gas path component creep-fatigue assessments across a large fleet of IGTs. The methodology divides operational periods into two categories. In the first category a full suite of operational data is available. Damage is modelled using an emulator of a full Monte Carlo assessment. The emulator accounts for the fact that different operational profiles may result in different degradation uncertainty, and that the mode of operation of an asset may change throughout its life. In the second category no information is available. This category covers both future operation and historical operation prior to the instrumentation of the asset. These periods are modelled by considering fleet-wide statistics of degradation and the path-dependency of creep-fatigue damage progression. The predictions for both categories of operation are integrated into a system that can predict distributions of the damage accumulated within a turbine component and the future progression of this damage.