Crack Detection on a Structural Beam: A Simplified Analytical Method Based on Artificial Neural Network Model

Abstract

Early detection of cracks in beams, machines, and other structural elements with applications in the maritime industry, oil and gas sector, mining sector, and other systems or facilities have become a major concern in building and construction and other engineering fields. Cracks frequently appear first on concrete surfaces and structures under load conditions and serve as an indicator of further deterioration. Fatigue can have a momentous impact on crack. It is, therefore, significant to detect cracks at an early stage in order to avoid catastrophic consequences. As a result, researchers have devised a variety of methods to achieve this goal. This study considers the use of a simple cantilever beam to detect the presence of a crack from measured vibration data. For this experiment, a beam with cracks in various locations and another beam without cracks were considered. Some of the variables considered at different crack locations include the acceleration responses, crack locations, mode shapes, and natural frequencies. The experimental framework was built around the data by using an ANN feed-forward backpropagation algorithm with a sigmoid function of layers 2–10-4–4, divided into 70% training, 15% validation, and 15% testing sets using a data portioning approach. The statistical indices showed a mean squared error (MSE) value of 0.0000312, root mean squared error (RMSE) value of 0.000111, and coefficient of determination (R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ) value of 1.0, and regression coefficient (R) value of 1.0. The result shows a very low MAE and RMSE. It was concluded from the result of the model that the presence of crack on the test sample is minimal. ANN proved to be a satisfactory technique for the analysis of stochastic datasets.