Identifying structural damage using a convolutional neural network from time-domain dynamic response data

Identifying structural damage using a convolutional neural network from time-domain dynamic response data

Stať ve sborníku v databázi WoS či Scopus

Ageing transport infrastructure brings increased economic burden and uncertainties regarding the reliability, durability and safe use of structures. Early damage detection to locate incipient damage provides an opportunity for early structural maintenance and can guarantee structural reliability and continuing serviceability. Structural Health Monitoring (SHM) is essential for assessing structural conditions, using sensor data to detect potential issues. SHM complements predictive maintenance in modern industry, reducing downtime and costs by addressing problems before they escalate. Machine learning techniques are increasingly employed to analyse vibration data, extracting valuable insights often based on prior structural knowledge, further enhancing the accuracy and effectiveness of SHM efforts. This paper describes a method for identifying the location and extent of structural damage using a convolutional neural network (CNN). The time-domain dynamic response of the structure is provided as input data to CNN. The method is used to identify damage to an existing riveted truss bridge. The effect of damage rate and location on the identification speed and solution accuracy is investigated and discussed. The method is also compared to an artificial neural network-based inverse analysis method, where the input data is the dynamic response of the structure in the frequency domain.

Ageing transport infrastructure brings increased economic burden and uncertainties regarding the reliability, durability and safe use of structures. Early damage detection to locate incipient damage provides an opportunity for early structural maintenance and can guarantee structural reliability and continuing serviceability. Structural Health Monitoring (SHM) is essential for assessing structural conditions, using sensor data to detect potential issues. SHM complements predictive maintenance in modern industry, reducing downtime and costs by addressing problems before they escalate. Machine learning techniques are increasingly employed to analyse vibration data, extracting valuable insights often based on prior structural knowledge, further enhancing the accuracy and effectiveness of SHM efforts. This paper describes a method for identifying the location and extent of structural damage using a convolutional neural network (CNN). The time-domain dynamic response of the structure is provided as input data to CNN. The method is used to identify damage to an existing riveted truss bridge. The effect of damage rate and location on the identification speed and solution accuracy is investigated and discussed. The method is also compared to an artificial neural network-based inverse analysis method, where the input data is the dynamic response of the structure in the frequency domain.

Damage identification, Convolutional neural network, Artificial neural network, FE model updating, Structural health monitoring

Damage identification, Convolutional neural network, Artificial neural network, FE model updating, Structural health monitoring

ŠPLÍCHAL, B.; LEHKÝ, D.

07.08.2025

CRC Press

London

9781003677895

Engineering Materials, Structures, Systems and Methods for a More Sustainable Future

1403

1408

6

https://www.taylorfrancis.com/chapters/edit/10.1201/9781003677895-236/identifying-structural-damage-using-convolutional-neural-network-time-domain-dynamic-response-data-šplíchal-lehký?context=ubx&refId=cb1c3bec-e03f-4a50-8413-573212493c95