The structural failure of engineering materials and buildings represents a critical issue for infrastructure maintenance and safety. Predictive strategies based on sensor data are paramount in providing early warning and mitigating potential disasters. There have been numerous efforts in applying machine learning (ML) for structural health monitoring (SHM) [1]. This study explores the innovative use of Physics-Informed Neural Networks (PINNs) for detecting structural damage, employing simulated sensor data.

PINNs provide an alternative to conventional data-driven machine learning. Instead of relying on large data-sets, which are very scarce in the realm of large-scale structural failure, they incorporate domain knowledge about physical laws in the training process [2]. We present a SHM model which assimilates sensor monitoring data in a PINN framework to execute a live system identification, predict the structural state development and potential failure.

The proposed approach was evaluated on a T-shaped element in various structural health states. Furthermore, the fusion of sensor data and PINNs allowed for a more detailed understanding of the evolution of structural damages up to failure.

This work shows how PINNs provide a time efficient tool for the detection of structural damage from simulated SHM sensor data [3]. Future work will entail refining our model through validation with experimental data and broadening its applicability to a wider variety of structural types and failure mechanisms. It is anticipated that this research will contribute to safer and more efficient operation of our built infrastructure, heralding a new era in materials science and engineering.

References:

[1]    M.T., Ghosh; C.hen Genda “Artificial intelligence in civil infrastructure health     monitoring—Historical perspectives, current trends, and future visions”, Frontiers     in Built Environment, 2022, vol. 8, 10.3389/fbuil.2022.1007886

[2]    Z., Zhang; C., Sun “Structural damage identification via physics-guided machine     learning: a methodology integrating pattern recognition with finite element model
    updating”, Structural Health Monitoring, 2021, Vol. 20

[3]    F.G., Yuan “Machine learning for structural health monitoring: challenges and     opportunities”, Proceedings of SPIE, 2020, Keynote Paper