| Autor: |
Wahl, Tobias, Gharaei, Ali, Thalmann, Tobias, Kuffa, Michal, Wegener, Konrad |
| Jazyk: |
angličtina |
| Rok vydání: |
2022 |
| Předmět: |
|
| DOI: |
10.3929/ethz-b-000604811 |
| Popis: |
Excessive heat accumulation at the workpiece surface can cause grinding burn, which is often indicated by workpiece surface color changes. Grinding burn accelerates rail surface deterioration and can lead to cracks. Data-driven approaches based on in-process measurements require extensive data labeling either based on laboratory inspections or human visual inspection. Additionally, such approaches may not be suitable for complex processes like rail grinding, which would require an excessive amount of process data for training. This paper proposes an approach for automated grinding burn detection using a camera. Grinding experiments have been conducted with a resin bond corundum wheel on a rail steel workpiece (58CrMoV4). Image classification techniques using transfer learning are applied to identify visible grinding burns. Despite limitations in the provision of a controlled set-up for the camera in rail grinding, it is shown that trained image classifiers can be a promising replacement for human visual inspection. They can increase productivity in process monitoring and facilitate continuous learning for grinding burn identification. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
