Imbalanced Classification of Fatigue Crack for Aluminum Plates Using Lamb Wave

Abstract

Data-driven methods are widely used in structural health monitoring (SHM) systems, and most of them focus on feature parameters extracted from damaged structure. However, structure is usually in healthy situation, which produces much more healthy data compared with damaged data. The classification of severity of damage using both healthy and damaged data is an imbalanced classification problem. This paper presents a damage classification method using Lamb wave and Synthetic Minority Over-sampling Technique with Iterative-Partitioning Filter (SMOTE-IPF) algorithm. A fatigue test with Lamb wave detection is conducted and three damage sensitive features, namely, normalized amplitude, correlation coefficient and normalized energy are extracted from signals as dataset. Generation of minority examples and filter of noisy examples are implemented with SMOTE-IPF method. Cross validations are performed on the proposed model using feature parameters and the length of fatigue cracks. The metric parameter for classifier performance is calculated to verify the performance of the proposed method for crack size identification.