Ultrasonic waves time series analysis for damage detection and assessment in composite and metallic structures -

Abstract

Ultrasonic guided waves (Lamb waves) are widely used for damage detection and structural health monitoring in mechanical, aerospace and civil structures. In this work, various signal processing and feature extraction techniques are applied on Lamb waves for damage assessment in both metallic and composite structures. Waves are excited at different frequencies and the fundamental symmetric (S0) mode was chosen to monitor different types of damage including flaws in welded metal structures, and barely visible indentation damage (BVID) in composite structures resulting from low speed impacts. The attenuation of the signal amplitude was first used as the feature to define a damage index (DI) which was used later to identify flaws in friction-stir-welded AZ31B magnesium alloy plates. Four plates of various weld qualities and a non-welded plate were used for investigation, where the DI was proposed as a comparison between the responses in the welded plates and the non-welded one. The S0 mode was separated using the improved complete ensemble empirical mode decomposition with adaptive noise “improved CEEMDAN” technique. On the other hand, computed tomography (CT) scanning was used as a non-destructive testing (NDT) to determine the actual weld qualities, and the method was further verified using finite element analysis (FEA). The proposed approach was highly sensitive to internal flaws within the weld and further had a great potential to classify the quality of weld in the various tested specimens. In addition, the S0 mode was used to assess barely visible indentation damage (BVID) in a carbon fiber reinforced epoxy (CF-EP) composite structure using symbolic time series analysis (STSA). STSA transforms a time series into a symbol sequence according to a pre-constructed symbol space using a set number of partitions. The symbolization technique allows keeping only the probability vectors of symbols’ occurrence as a feature. This reduces data and computational time significantly. A quasi-static loading