Micro-structured surfaces (riblet surfaces) reduce drag, and as a consequence, allow increasing efficiency in various flow applications. Therefore, riblet surfaces can save fuel consumption substantially in industrial applications like aircrafts and high-speed trains or to increase energy output of wind turbines and lower noise emission. In the last ten years industry did massive investments in the development, production and application of riblet surfaces. Efficiency gains depend on the quality of the riblet surface. Therefore, it is very important to make use of special inspection equipment in order to assure the quality of the riblet surface in terms of their expected reduced drag behavior. Unfortunately, the current inspection processes being used are very timeand cost- intensive. Thus, there is a need for easy to use and inexpensive inspection devices allowing to estimate the riblet surface’s properties during maintenance processes on several applications (aircrafts, wind turbines, etc.). RiSPECT proposes to apply machine learning and in particular deep neural networks for classifying and predicting the quality of riblet surfaces for the first time. The proposed approach makes use of example images obtained from available riblet surfaces and simulations for this purpose. Because of the need for having a larger number of labelled images comprising riblet surfaces with and without defects, RiSPECT will provide a database of such images that are acquired using a tailored machine vision system to be developed as part of the project. The images stored in a database rely on EASA certified riblet test bench, high precise SEM pictures and CFD simulations. Because of the fact that machine learning applications based on neural networks are vulnerable, verification & validation becomes a critical issue of RiSPECT. In particular, the focus will be on how to prevent small image changes that lead to misclassification and erroneous quality predictions. This will increase the robustness of the overall application. The main objective of RiSPECT is to come up with the foundations behind an automated system for estimating the quality of riblet surfaces based on image data that reduces the overall costs for quality assurance measures within the riblet surface maintenance. For this purpose, RiSPECT also includes a sophisticated case study to be carried out at the end of the project.
|Effective start/end date||4/11/19 → 3/11/21|
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.