Evaluating Damage-Sensitive Features for Stiffness Loss Detection in an Aircraft Component Using Machine Learning Classifiers

The ultimate goal of this study is to produce a data-driven monitoring system that provides advantages over current methods for identifying structural damage on an aircraft component. Two types of damage-sensitive features (DSFs) were tested in an experimental setting to evaluate their ability to detect a small change in flexural stiffness in a particular, frame-like aircraft component. The vibration-based monitoring system utilized two DSFs: one based on spectral distancing metrics and another based on spectral peak characteristics. This system was developed to discover damage classified as the loss of stiffness at connections by comparing experimental aircraft component data to a healthy baseline. The DSFs were fed to a support vector machine redundant learning model to provide a damage decision. The results of this decision were compared to the known damage case of the test to evaluate the true positive rate and true negative rate of the classification method for the model. The DSF based on spectral distancing metrics had 70.4% true positive and 81.3% true negative rates for a 2.5% reduction in bending stiffness at the location of most likely damage. The second DSF based on spectral peak characteristics had 100% true positive and 100% true negative rates for the same reduction in bending stiffness. This system shows potential for a simple, accurate, efficient, and cost-effective inspection technique for this particular type of damage, meant to simulate crack propagation.