Strain monitoring-based diagnosis and prognosis for hidden cracks in metallic lap-joint structures

A strain monitoring-based approach for hidden crack diagnosis and prognosis in aircraft metallic lap-joint structures has been developed. Fatigue experiments employing the marker load method were conducted to acquire accurate strain/hidden crack information for aluminum alloy lab-joint specimen. The Cumulative Sum (CUSUM) is utilized to determine the existence of cracks, paired with the K-Nearest Neighbors (KNN) for crack localization; the data-driven Multi-Output Gaussian Process Regression (MOGPR) algorithm is applied to establish the relationship between crack sizes (length/depth) and strain characteristics, with validated higher identification precision and generalization performance than Single-Output Gaussian Process Regression (SOGPR). A Dynamic Bayesian Network (DBN) is constructed by integrating the MOGPR model with Paris formula to achieve real-time crack propagation prediction. Experimental verification confirms that this approach effectively detects and locates fatigue cracks, with high accuracy and robustness in crack size identification and propagation prediction. Predictions align well with monitored data, offering an innovative framework for health assessment and preventive maintenance for hidden cracks in metallic lap-joint structures, with significant potential for industrial applications in sectors like aerospace, where it can enhance safety and reduce maintenance costs.