• Da Silva, Lfm
• Lopes, Am
• Ramalho, Gmf
• Barbosa, Mrsp

Abstract

As the aerospace industry develops, there is a need for applying new materials and construc-tion techniques, able to create lighter and more efficient aircrafts. Most advances also imply severe regulations that require novel methods suited to monitor critical components. One method that goes beyond simple nondestructive testing is structural health monitoring (SHM), more specifically Lamb waves (LW)-based SHM. Indeed, LW have shown great promise in nondestructive in situ testing, but require computationally expensive calculations, so that precise results can be obtained. An opportunity to overcome LW drawbacks arises with the use of machine learning (ML) algorithms. In this article, the performance of conventional feedfor-ward and convolutional artificial neural networks for damage classification in aluminum sheets is compared, and a novel methodology to classify damage is proposed. The ML techniques adopted require large sets of prior data, which are generated by numerical simulations utilizing the finite element method. The damage classification pipeline comprises (i) generating LW by one actuator, measuring the structure response using a set of sensors, (iii) extracting features from the raw signals and training the ML algorithms, and (iv) assessing the classification accuracy. The methodology has the advantage of being baseline free, easily extendable for automatic feature extraction and testing, and adaptable to different types of damage and structures, as long as the algorithms are trained with suitable data.

Topics

• impedance spectroscopy
• simulation
• extraction
• aluminium
• machine learning