Optimising remedial outcomes for gas turbines through large scale data analysis

Abstract

An investigation into approaches to model and predict the costs, risks and outcomes, relating to a common failure mode within a large population of remotely monitored engines will be presented. This investigation will cover the relevant aspects of lifecycle management, customer operating cycle, visibility of classified issues remotely using low sample rate remote monitoring systems and identification of the most appropriate repair regime. Unlike previously identified studies, the datasets used are larger comprising sensor readings from over a 120 gas turbine units (distributed across the world) which total 100 million observations taken across 117 parameters. These are cross referenced with relevant service events (preventative, corrective, fault investigation and observational reports) and service execution data (spares and manpower required) gathered over a period of up-to 11 years. We have now succeeded in identifying behaviours in a particular problem domain using techniques such as K-means, that indicate non-optimal decisions may have been made (with the benefit of hindsight). This indicates that the intended outcome of the overall research to produce of quantitative models of the different strategies and applying them to optimize for the best outcome both in terms of and customer and supplier objectives in a justifiable form is possible.