Complementarity of Sensors and Weak Supervision for Defect Detection in Concrete Structures

The automation of concrete structure inspection methods such as the hammering test is highly desirable and critical, especially for social infrastructures such as tunnels and bridges. This is to ensure the safety of their users. Semisupervised approaches have great compatibility with critical inspection methods since they allow to greatly reduce the workload on humans while still not removing them completely from the process, and thus providing some level of reassuring confidence. However, the performance of such semi-supervised approaches is conditioned by the correctness of the provided weak supervision by human and it can easily be imagined that, in practice, weak supervision will rarely be without errors. Therefore, the present paper proposes a method to complement weak supervision using sensor-provided information in order to both increase performance and mitigate the negative impacts of human errors. Experiments conducted in laboratory conditions using concrete test blocks in various configurations showed the effectiveness of the proposed method, returning better performance and higher robustness to errors in weak supervision.