Real-Time Damage Detection of Engine Blade using Computer Vision and Machine Learning

In the process of engine maintenance, the most important thing is to find out where the damage is among a large number of blades. So far, the visual inspection method with industrial videoscope is mainly used for that. However, unfortunately, there are too many blades in engine to detect the damage precisely for an inspector. In this paper, we introduce a new method that can help the inspector to detect the damage without any omission and it is developed as a software for in-situ application. Our research has three steps: (1) pre-process step, (2) classification step and (3) tracking step. In the first step, we apply diverse image processing techniques to the digital images that are from the videoscope. As results of this step, scale invariant features are extracted and they are matched to the features from the damaged blade image according to the similarity. In the next step, it seeks damaged parts using the results of previous step and classifier which is already learned in order to classify the blade images into two groups, normal and damaged. The last step is optional because it operates when damage exists. After finding damage, it should track the damage as the videoscope camera moves. Now, our research is at the second step that only can recognizes the damaged parts in the videoscope images and the last step will be conducted at further research