Development of Fault Specific Soft Sensors With Application to Gas Turbine Diagnosis

The paper presents a method of diagnosis for a gas turbine installation monitored for a number of parameters by using tools specific to multivariate statistical analysis. Data is acquired periodically and organized as individual observation vectors. The objective is to detect the occurrence of a fault, to identify and discriminate the type of fault and eventually to find its cause. The author developed a new concept of soft sensor (similar to the sensor fusion in other works, see [1]) based on the intrinsic properties of the data historically acquired when the process was deemed to be in normal operating conditions, and on the current observation vector. This soft sensor was applied to an example of detecting the occurrence of a fault and characterizing it in the p-dimensional space of observation vectors associated with a proper metric.