Deep and Machine Learning-based Methods for Defect Classification in Jet Engines

In this paper, the utility and accuracy of Machine Learning (ML) and Deep Learning (DL) methods are investigated for detecting defects in civil aircraft engines. Rather than to disassemble jet engines, the approach investigated in this study utilizes images of the exhaust of jet engines and infers defects in the turbine and burner section. While the proposed DL methods make use of one or two cameras, the ML methods depend on data obtained by extracting the density fields of the Hot Gas Path (HGP). The HPG data are computed from images acquired by an array of cameras. The corresponding ML features are crafted from these density fields. The proposed algorithms employ optimized hyperparameters and separate training as well as validation data sets. The study illustrates the potential of DL methods and the resulting simplification in the necessary instrumentation to accomplish near perfect defect classification outcomes.