CNN–based crack detection in oxide layers of hot rolled steel sheet samples for the validation of a pickling process model

Cracks in the oxide layer of steel sheets after hot rolling play an important role during the oxide layer removal with acid in the following pickling process. The time required to remove the oxide layer should increase with the crack distance as the acid is supposed to undercut the oxide layer. In order to validate a corresponding mathematical model, hot rolled steel sample surfaces are analysed in a microscope in a first step. The cracks in the microscope images are segmented using a CNN–based algorithm for semantic segmentation, followed by a post–processing step to determine distances between neighboring cracks. The approach allows an automated crack distance determination over a region 300 times larger than the typical crack distance of approximately 30 µm. In a laboratory pickling simulator, the oxide layer of the samples is removed in a second step. During this process, the sample surface is observed by a camera, allowing to identify the locally varying time for the removal of the oxide layer. In a final step, the local distribution of the crack distances is compared to the local distribution of the pickling time, which should correlate according to the mathematical model.