ABSTRACT
Several new approaches are currently under development aiming to provide a real-time characterization of excavated material from soft ground tunnel drives on the machine conveyor belt using digital methods of image and video analysis. This comprises both rule-based algorithms and methods based on artificial intelligence (AI). One of the main goals is to predict geotechnical (index-) parameters in order to answer the question of reusability of typically conditioned excavated material from EPB drives.
Against the background that powerful digital camera systems are now available at low cost, it was investigated what added value they can provide in combination with rule-based algorithms with regard to the characterization of excavated material in the form of non-cohesive soils. The aim was to use image analysis to optically determine the grain size distribution curve. It was shown that a distinction can be made between sands and gravels. Further findings can be derived about grain shapes and other morphologies.
In contrast to the rule-based algorithm, which works with individual images, an approach currently under development based on AI methods, processes entire video sequences, where not only individual soil grains are analysed, but the “overall visual impression” is evaluated. The main advantage of this approach is that it facilitates mapping of grain size distribution over the entire excavated material on the moving conveyor belt. The ultimate aim is to link or predict relevant video features with soil properties. Emphasis is put on reliably differentiating coarse-grained soils in terms of their granulometric properties, and to characterize fine-grained and mixed-grained soils in terms of consistency and homogeneity.
This paper summarises the various approaches and highlights advantages and disadvantages identified during laboratory experiments, examinations on a real-scale TBM conveyor belt, and on-site tests.
