Semi-automatic detection of cracks during bending test by using deep learning
ArcelorMittal Global R&D Maizières-lès-Metz (57)
Before being used by OEMs (Original Equipment Manufacturers) to manufacture cars, new steel grades are tested and approved based on laboratory tests. Bendability value is necessary to successfully manufacture some auto-components. It can be described as the ability of steel sheet to be bend without breaking. One of the tests used by OEMs is flanging 90°. The aim of flanging 90° according to ISO7438 is to find the minimum bending radius achievable without cracks. 90° bending tests are performed with several punch radii by clamping the sheet between a die and a blank-holder.
The crack is today identified visually by operators. As a consequence, bending cotation values could suffer scattering.
In order to tackle this issue, numerical microscope have been recently purchased to be able to better detect the cracks.
The trainee will be in charge of
– Understanding bending cotation through microscopic observations
– Design the photography support taking into account the dimensions of the samples, camera angles to access the entire sample, lighting and magnification for a better view
– Create database with image with failure, cracks, microcrack after tests (numerical microscope)
– Create Python program, using Deep Learning to detect cracks