Considering that the early crack signal of aero-engine turbine blade is weak and difficult to identify

a diagnosis method was proposed based on ensemble empirical modal decomposition (EEDM) energy entropy fusion of three-dimensional tip clearance. Three-dimensional tip clearance signals of turbine blade were collected and decomposed by using the EEMD to obtain corresponding three-dimensional intrinsic mode functions (IMFs) and calculate the EEMD entropy value of each dimensional signal in order to form different high-dimensional vector sets representing the crack fault. Several stacked autoencoders (SAE) were constructed to learn the high-level features from each high-dimensional vector set and further extract the in-depth feature expression acquired by each SAE. The support vector machine (SVM) algorithm and genetic algorithm (GA) were used to fuse these different high-level features in order to confirm adequately the early crack fault of turbine blades. Results show that the proposed method can fully dig out characteristic information of the early crack

and effectively improve the diagnosis accuracy

yielding an average accuracy of 98.415%

and its standard deviation is only 0.697%

showing its good stability

adaptivity and generalization ability.