Raman-spektroskopische Klassifikation von Mineralen und Gesteinen unter Einsatz von Machine Learning - Mit Fokus auf sensorgestützte Charakterisierung von (Tunnel-) Aushubmaterialien

In Austria, (tunnel) excavated materials are classified as waste unless they are utilized at the site of generation in accordance with Section 5 of the Waste Management Act. Excavated materials account for 57 % of the total waste produced - 38.1 million tons - making them the largest waste stream in Austria in terms of quantity. The legal framework, as well as low disposal costs, favor the landfilling of such materials. At the same time, primary raw materials are mined in Austria to produce aggregates, fillers, and construction materials. The NNATT (long title: Sustainable utilization of excavation materials in civil and tunnel engineering using sensor-based technologies) research project aims to identify mineral secondary raw materials using sensor-based methods and subsequently put them to sustainable use to conserve primary resources, shorten transport routes, and reduce landfill volumes. To ensure precise classification, various methods of optical spectroscopy are combined within the framework of sensor fusion. This work focuses on the fundamental characterization of minerals and rocks using a Raman system developed specifically for the project. The focus is on granite, sandstone, carbonate rocks, as well as rock-forming minerals. High sampling rates are intended to enable material identification on a continuously moving conveyor belt. Initial experiments on minerals under static conditions (measurements on stationary material) suggest that data measured by the system used in the project can only be compared to literature values or reference databases to a limited extent. In contrast, evaluation using machine learning models shows that the classification of minerals and rocks under static measurement conditions, using a specially constructed database, exhibits high test accuracy and reproducibility. Measurements under dynamic conditions (measurements during material transport) cannot be evaluated by the models previously used successfully.