Development of an AI‐Based Damage Detection System Including Classification Thresholds for Steel Structure Defects
TL;DR
A research project at the steel and light metal construction laboratory of the University of Applied Sciences Munich develops an automated system for condition assessment and image‐based damage detection using machine learning.
Abstract
Infrastructure and structural buildings degrade over time due to usage and environmental influences, leading to various types of damage. Regular inspections enable early detection and assessment, allowing preventive maintenance before safety is compromised. Manual documentation and evaluation require significant time and personnel. To improve efficiency, a research project at the steel and light metal construction laboratory of the University of Applied Sciences Munich — together with laboratories for building materials, construction informatics, geoinformatics, glass, and façades — develops an automated system for condition assessment and image‐based damage detection using machine learning. Damages relevant to steel construction are recorded and compiled into a catalog to create a dataset for training automated detection methods. The dataset consists of annotated images produced in cooperation with engineering companies specializing in structural inspections. A large‐scale test setup enables generating additional damage images under controlled conditions. Various digital measurement methods are tested to ensure high accuracy in damage classification. Laser scanning and photogrammetry are used, for example, to analyze crack detection and propagation under fatigue loads.