In-service fatigue crack monitoring through baseline-free automated detection and physics-informed neural network quantification

Abstract

Online monitoring and quantification of fatigue cracks are essential for ensuring engineering structural integrity. Current structural health monitoring (SHM) methods, which have demonstrated potential to be applicable in service are either baseline or can only be applied on ground, which increases maintenance costs and risks of undetected rapid crack propagation. This paper proposes a reliable in-service method for online crack detection and growth assessment, providing early warning for maintenance. This novel approach extracts the third harmonic parameter ${\hat{\gamma }}^{\prime }$, defined as the ratio of the fundamental frequency amplitude ($A_1$) to the cube of the third harmonic amplitude ($A_3$), from the fatigue response. A dynamic piecewise linear (DPL) method is then employed for automatic online crack detection. Results from four specimens demonstrate the method’s capability for real-time detection of cracks below 2 mm during operation. Additionally, a physics-informed Long Short-Term Memory (PI-LSTM) model is developed to quantify the crack online, achieving an average RMSE of 0.498 mm on six datasets, outperforming traditional methods like pure LSTM and Paris’ Law with RMSE values of 3.205 mm and 3.641 mm, respectively. This study provides a cost-effective, reliable solution for in-service crack monitoring using external excitation signals, enhancing structural maintenance and safety.