Ultrasonic Testing of Corrosion in Aircraft Rivet Using Spiking Neural Network

Abstract

This paper proposes a nondestructive testing (NDT) method for the inspection of corrosion in rivets used in an aircraft. The NDT system uses an ultrasonic sensor coupling with a membrane that allows the ultrasonic wave propagates through to the inspecting rivet. The measured signal is then analyzed by a spiking neural network (SNN), a neural network that mimics the biological neurons for efficient detection of the corrosion in rivet. Compared to the conventional deep neural network, SNN is low energy consumption and can be implemented on a compact SNN accelerator chip, making them better run on a compact NDT system and general edge computing applications. We have tested the proposed SNN model on different sizes of corrosion in rivets (i.e., 30–70% of cross-section area) and at different depths from the surface (i.e., 1.0–2.0 mm). The proposed SNN model achieves about 95.4% accuracy with a small number of rivet samples (i.e., four rivet with corrosion) for training.