Crack Identification in a Structural Beam Using Regression and Machine Learning Models

Abstract

A manufacturing fault causes a defect consisting of a crack in the structure. Identification and classification are essential challenges in scientific research because cracks can lead to catastrophic system failure. Structural fitness tracking aims to diagnose and predict structural fitness. A complete crack detection method based on free vibration is widely used to find potential cracks in systems. However, bending stiffness methods are limited in predicting the crack parameters. Therefore, the bending stiffness approach has been used in the present work to determine the crack locations and depth in the cantilever beam. A dead weight was attached to the beam's free end, and two dial gauges were used. A gauge was attached to the free end of the beam to measure the free-end deflection. Another dial indicator was installed near the crack to measure the static deflection at the crack. Numerical and experimental analyses were performed on 25 cracked specimens to measure the static deflection and stiffness at two points. Regression models were developed for the crack parameters to predict them without the need for numerical and experimental analyses. Also, the ANN model was developed for the same purpose to relate the considered input and output variables. The crack depth and location results obtained from the regression and machine learning models are consistent with the actual values. The crack parameters were predicted using static two-point bending stiffness values as input, and the results were encouraging. Therefore, the static two-point bending stiffness approach may be widely used to detect future cracks in more complex structures.