ML Based Damage Detection in Simply Supported Beams Using Free Vibration Data

Abstract Monitoring structural damage in terms of material degradation is extremely important for enhancing the residual life of civil infrastructure. For the past two decades vibration based techniques have gained importance in the field of structural engineering in damage detection. The changes in the dynamic characteristics are observed due to changes in the mass and stiffness of the structure/structural component. The present paper focuses on detection of damage through vibration analysis in simply supported beams and predicting the damage extent using machine learning technique. The numerical model is validated against the standard Euler-Bernoulli beam subjected to free vibration. The same finite element model is subjected to induced damage in terms of change in material properties. Modal characteristics of all the damage scenarios i.e. reduction in Young’s modulus by 10% -80% with increment of 10% at a time are captured and a machine learning algorithm (Support Vector Machine) is developed for the prediction of damage extent. It is concluded that the SVM algorithm is effective in the case of natural frequencies as input features rather than statistical parameters as input features in predicting the damage extent in the beams.