Statistical damage classification under changing environmental and operational conditions

Stated in its most basic form, the objective of damage diagnosis is to ascertain simply if damage is present or not based on measured dynamic characteristics of a system to be monitored. In reality, structures are subject to changing environmental and operational conditions that affect measured signals, and environmental and operational variations of the system can often mask subtle changes in the system's vibration signal caused by damage. In this paper, a unique combination of time series analysis, neural networks, and statistical inference techniques is developed for damage classification explicitly taking into account these ambient variations of the system. First, a time prediction model called an autoregressive and autoregressive with exogenous inputs (AR-ARX) model is developed to extract damage-sensitive features. Then, an autoassociative neural network is employed for data normalization, which separates the effect of damage on the extracted features from those caused by the environmental and vibration variations of the system. Finally, a hypothesis testing technique called a sequential probability ratio test is performed on the normalized features to automatically infer the damage state of the system. The usefulness of the proposed approach is demonstrated using a numerical example of a computer hard disk and an experimental study of an eight degree-of-freedom spring-mass system.