ABSTRACT
Assessing the load bearing capacity of civilian bridges is a challenging task for routine reassessment jobs, especially in the military context. In order to ensure mobility of troops, the load bearing capacity of bridges needs to be assessed to make the safe crossing of military vehicles possible at home and abroad. For this reason, rapid assessment methods have been developed that allow for classification of civilian bridges after a quick visual reconnaissance. In the absence of verifiable calculations or drawings, the applied methods need to be based on geometrical data that is suspected to be correlated to the military load class of a structure. Current approaches are based on simplified assumptions such as correlations between dead loads and live loads or consist of simplified calculations assuming conservative material properties. In order to improve the current classification method, a Machine Learning (ML)-based methodology is given that classifies slab to a particular military load class without human intervention using a correlation between measured geometrical data and classification result. It is concluded that the usage of ML models is very promising for rough classifications of bridges with unknown military load class (MLC). However, the available training data is insufficient to train reliable models and an expanded amount of training data will be needed for deployment as a software package in the armed forces.
