Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Finding solutions to crack identification problems using artificial neural networks

Stamatelou Efthalia

Full record

---

Summary

In the present dissertation, the solution of an inverse problem of crack identification in engineering is presented using a multilayer neural network that is trained with backpropagation. Examples will be cracks in static or harmonic excitation (ie a problem that lies in the field of elastodynamics). More specifically, this problem concerns the identification of the existence and characteristics of a hidden crack, which is located within an elastic structure. For this purpose we use structural response measurements, within the available limits for specific loads. The direct problem is to be solved by methods of computer engineering (boundary element method) and the inverse problem is being solved through the application of a backpropagation neural network. As for the implementation of the neural network, we are going to use Python due to its good performance and the ability it gives us to optimize the required computing processes. Existing libraries are considered suitable for any general work related to data processing and we are also provided with sufficient coverage in case there is a need to integrate ML into other software. In addition, TensorFlow and Keras are used, which are open source libraries for high performancenumerical calculations developed by the Google Brain team. They can train and run deep multilevel neural networks and are widely used in the field of research and application of deep learning. The method is general and can be used for the solution of inverse and parameter identification problems for every problem of mechanics of continuum for which similar data are available. The implementation of the neural network will be presented step by step in the form of a manual and finally the results obtained will be analyzed.

Available Files

Services