Intelligent condition monitoring via sparse representation and principal component analysis for industrial gas turbine systems

This paper proposes an intelligent condition monitoring methodology based on sparse representation and principal component analysis (PCA), for application to key constituent systems of industrial gas turbine units. The contribution and novelty of the presented methods are i) To detect sensor faults, a method based on the recognition results of PCA, is described; ii) A condition monitoring method based on sparse representation data mining techniques, is proposed; (iii) Even in the presence of measurements from faulted sensors that can still provide some information but may be subject to drift or bias, for instance, it is shown that the condition of an operational unit can be assessed. Experimental results based on data from a 14MW SGT-400 industrial gas turbine are used to demonstrate the efficacy of the developed procedures, although it should be noted that the proposed methodologies are much more widely applicable to many other industrial and commercial systems.