Diagnosing damage to steel structures based on a deep transfer learning algorithm Available to Purchase

Accurate diagnosis of the presence of defects or damage in structures is crucial in modern maintenance methods to improve structural elements’ reliability and efficiency in accomplishing the required tasks. The strategy proposed in this work includes using deep transfer learning technology in Convolutional Neural Networks (CNN) to diagnose damages in identical structures, where three types of structures were used: the cantilevered beam, the simply supported beam, and the fixed-fixed beam. In addition, the Continuous Wavelet Transform (CWT) method was adopted to generate images of damage to structures based on the difference deformation of mode shape and consider it as input to the Convolutional Neural Networks. The findings indicated that the amalgamation of CWT and CNN methodologies yields a notable diagnostic accuracy of 100%. Moreover, the employment of deep transfer learning technology facilitates the acceleration of the training procedure, consequently diminishing the time and computational resources needed for training.