Abstract

In this presentation the application of a multivariate analysis technique for gas turbine engine health monitoring purposes will be discussed. Gas turbines constitute complex systems that require adequate monitoring to ensure availability, reliability and timely maintenance. A variety of engine health monitoring techniques has been proposed the last two decades to accomplish these demanding tasks. Generally, the proposed techniques apply algorithms of different complexity and reasoning logic to detect anomalies on the acquired operational data and assess engine condition. The data typically gathered and analysed for engine health monitoring purposes consist of gas path measurements, e.g. pressure and temperature at different locations, vibrations and operational data of the oil system. These data sets are often highly dimensional, correlated and non-casual in nature, making data analysis a very challenging task. The univariate statistical process monitoring chart is a well-established methodology to detect an abnormal behaviour within the engine by comparing analysed parameters against upper and lower limits established from periods of normal operation. However, the application of such methods could result to an unacceptable rate of missed events or false alarms since the correlation of the variables is ignored while the impact of influencing environmental factors is not accounted for. These issues can be addressed by applying a multivariate analysis technique capable of exploring the interrelation of the analysed parameters. The benefits of this approach will be demonstrated through test cases examples.