Layout-aware Extreme Learning Machine to Detect Tendon Malfunctions in Prestressed Concrete Bridges using Stress Data

In the past few years, several works focused on the integration of methodologies within the field of Structural Health Monitoring to build reliable automatic damage-assessment procedures. Within this context, only a few papers specifically refer to the automatic assessment of tendon malfunctions in prestressed concrete (PSC) structures, despite the key role that this construction paradigm plays in modern infrastructure networks. This paper describes a novel Extreme Learning Machine (ELM) framework characterized by a layout-aware weight generating procedure (LA-ELM), that analyzes stress data to accurately detect and localize damages affecting the prestressing system of a target PSC bridge. A comprehensive computational study is conducted, testing the proposed methodology of three structural specimens, and comparing the proposed LA-ELM with classical Machine Learning algorithms. The numerical results evidence that the proposed methodology achieves remarkable accuracies in short computational times, and the LA-ELM obtains statistically significant improvements compared to the classical ELM implementation.

Automated summary

This paper introduces a novel Extreme Learning Machine framework for accurately detecting and localizing damages affecting the prestressing system of a prestressed concrete bridge, showing remarkable accuracy and efficiency in computational experiments.