STEMMER IMAGING

Marble is one of the most decorative construction materials in the world and highly valued as a high-class finish in buildings such as hotels, airports or upscale residences. No marble slab is identical to another, as inclusions, cracks or so-called veins shape the appearance of the naturally formed stone and always create a unique look. Depending on the number and characteristics of such flaws, marble is divided into three quality classes, which generate different prices on the market.

The inspection results obtained by the machine vision module used in the very beginning are the basis for all subsequent steps performed on the BMM. Precise positioning of the imperfections is an essential prerequisite for the laser treatment, which should only remove the structures that actually reduce the value.

In order to minimize visual differences between the laser-treated and the natural areas, the colour blending applied must match the colour surrounding the lasered structures as closely as possible. It is important that structures and veins crossing a retouched area are visible without any discontinuity.

Here, again, the image processing quality with regard to the recognised colours and structures is decisive for the quality of the outcome. The sealing and polishing of the marble slabs at the end of the process is also based on the position data of the imperfections detected in the first module.

The image acquisition module is based on two trilinear RGB line scan bars from Mitsubishi Electric with widths of 600 mm and 900 mm respectively and built-in lighting.

A Teledyne DALSA Xtium CameraLink frame grabber is used to transfer the image data to a customized PC, where the partial images captured by the two line scan bars are combined and evaluated with the help of STEMMER IMAGING's machine vision programming library Common Vision Blox (CVB).

Hellmuth continues: "This solution is ideally suited to identify inclusions, veins or cracks in the marble slabs with an accuracy of 40 µm, as well as to analyse the colours and structures surrounding the areas to be mended."

Felix Hellmuth highlights the importance of the machine vision software used: "This software is self-learning in order to constantly improve the quality of image recognition and evaluation. We put a lot of effort into its programming – the results are quite convincing, so the hard work was worth it."

About 200 to 300 images are required to train a new type of marble, which are later used by the AI-based software to automatically create suggestions for processing the slabs. Due to the manual adjustments of these suggestions, the software continuous to learn during operation thus enabling an ongoing improvement of the processes as well as an adaptation to the individual taste of the user or the customer.