Automated detection of hot-gas path defects by Support Vector Machine based analysis of exhaust density fields

Abstract

Defects in the hot-gas path of aero engines have been shown to leave typical signatures in the density distribution of the exhaust jet. These signatures are superposed when several defects are present. For improved maintenance and monitoring applications, it is important to not only detect that there are defects present but to also identify the individual classes of defects. This diagnostic approach benefits both, the analysis of prototype or acceptance test and the preparation of Maintenance, Repair, and Overhaul. Recent advances in the analysis of tomographic Background-Oriented Schlieren (BOS) data have enabled the technique to be automated such that typical defects in the hot-gas path of gas turbines can be detected and distinguished automatically. This automation is achieved by using Support Vector Machine (SVM) algorithms. Choosing suitable identification parameters is critical and can enable SVM algorithms to distinguish between different defect types. The results show that the SVM can be trained such that almost no defects are missed and that false attributions of defect classes can be minimized.