Image-based roughness estimation of laser cut edges with a convolutional neural network
Leonie Tatzel, Fernando Puente León
Abstract
Laser cutting of metals is a complex process with many influencing factors. As some of them are subject to change, the cut quality needs to be checked regularly. This paper aims to estimate the roughness of cut edges based on RGB images instead of surface topography measurements. We trained a convolutional neural network (CNN) on a broad database of images and corresponding roughness values. The CNN estimates the roughness well with a mean error of 3.6 µm. Sometimes it is more reliable than the surface measuring device because the RGB images are less prone to reflectivity problems than the measurements.
Topics & Concepts
Convolutional neural networkRGB color modelSurface finishArtificial intelligenceSurface roughnessComputer scienceLaserComputer visionArtificial neural networkProcess (computing)Surface (topology)Image (mathematics)Pattern recognition (psychology)OpticsMaterials scienceMathematicsGeometryPhysicsComposite materialOperating systemSurface Roughness and Optical MeasurementsLaser Material Processing TechniquesIndustrial Vision Systems and Defect Detection