Piping structural health monitoring: ANN methods for internal semi-elliptical crack detection

Abstract

This study investigates the detection and characterization of longitudinal internal semi-elliptical cracks in circular steel pipes using artificial neural networks (ANNs) within a structural health monitoring framework (SHM). Internal semi-elliptical cracks, which are prevalent in pipes and pressure vessels and challenging to detect with conventional non-destructive testing methods, are modeled to evaluate their impact on natural frequencies and mode shapes. Finite element simulations are validated against analytical and numerical data to ensure accuracy. The results reveal that an increase in the crack's large diameter, while maintaining a constant depth, significantly reduces the natural frequencies. ANN models employing cascade forward back propagation (CFBP) and feed forward back propagation (FFBP) algorithms are developed to predict the shape and location of cracks based on natural frequency data. The models are trained and tested using 78 datasets derived from vibration responses of six geometric crack ratios at varying positions, employing the k-fold cross-validation method with k = 6. The CFBP algorithm demonstrates superior performance, achieving a high correlation coefficient of 0.98898 in crack detection. The novelty of this work lies in addressing the complex effects of semi-elliptical cracks on structural responses and leveraging natural frequency shifts as diagnostic parameters. This approach bridges a critical research gap by providing a robust methodology for detecting concealed internal cracks in pressure vessels, advancing SHM practices.