Speaker: Staderini, Vanessa (AIT - Austrian Institute of Technology)
“Visual quality inspection of die-cast objects with complex shapes presents several challenges in the manufacturing industry. The most common approach is manual inspection but this is time intensive and prone to human errors. As consequence, we want to develop an automatic system to compute an inspection planning with a focus on simultaneously achieving high speed, complete coverage and high spatial resolution. We adopt a model-based approach as the cad model is always available in industries and we aim to reduce the gap between simulations and reality. For example, material properties are often overlooked during the inspection planning phase, hindering the accurate assessment of object quality.
At present, we proposed a simple method to generate candidate viewpoints based on Poisson disc sampling, improving the coverage achievable with a small number of optimal viewpoints. In addition, we introduced two matrices to ensure high spatial resolution and consider the properties of materials integrating the BRDF during viewpoint selection. Finally, we developed a clever algorithm to modify viewpoints that cannot be reached with a given kinematic setup, ensuring that the new viewpoints maintain similar visibility properties to the original ones.
Unfortunately, many issues are still unresolved, such as how to best parametrize the BRDF for different materials without time intensive experiments and how to reconstruct the mesh of the inspected object from partial point clouds while preserving the finest details.
In addition, the speed needs to be improved. One solution could be the integration of several sensors. Indeed, the combination of various sensors, such as cameras, lasers and structured light systems, can provide a more complete view of the object, enabling faster inspection without compromising accuracy.”