Frequency domain methods for fault diagnosis in complex structures with inherent nonlinearities

Abstract

Structural faults including bolt self-loosening and fatigue crack commonly exist in industrial structures since they are frequently subjected to dynamic loads in their life cycles. These faults could cause various problems such as slip and impact in bolted joints and fracture break at the root of cracks, and then affect the reliability and integrity of structures. Therefore, structural health monitoring (SHM) and non-destructive evaluation (NDE) methods evaluating structural states are important to guarantee the smooth and normal operation of industrial structures. Although vibration based methods utilizing transmissibility function (TF) provide some obvious advantages compared with other fault diagnosis methods, there are still some problems that need to be solved to improve their sensitivity and reliability. For example, inherent nonlinearities could also affect structural dynamic characteristics, TF based methods may provide false position information of multiple faults, simple chain-type multi-degree-of-freedom (MDOF) model may not be available for pipeline and frame structures, and diagnostic methods should be applicable for both initial and serious faults. Considering above issues in turn, some methods and procedures are introduced and applied, and then a systematic second-order output spectrum (SOOS) based method with a local tuning approach (LTA) is summarized for correctly detecting and localizing bolt loosening and fatigue crack faults in complex structures even with inherent nonlinearities. In the systematic method, complex structures are simplified as chain-type or ring-type structures first, and then a general ring-type MDOF model with nonlinear damper-spring connections simulating faults and inherent nonlinearities is built to perform dynamic analysis. In conjunction with the Volterra series and the nth-order output spectrum estimation (nth-OSE) algorithm, output spectra are theoretically derived, and are effectively estimated using measured data only. To eliminate effects of inherent nonlinearities and introduce additional equations, extra SOOS from the health structure is utilized and a LTA with three alternative ways (extra mass, extra connection and extra system) is adopted to change local structural properties respectively. Based on this, new nonlinear features for reference and diagnosis are analytically derived, and then corresponding relative errors between them are calculated as damage indicators for initial and serious faults respectively. Finally, experiments on a satellite-like structure with inherent nonlinearities and bolt loosening faults are implemented. Through comparison studies with related vibration based methods, the results demonstrate that the proposed systematic SOOS based method with LTA could accurately detect and localize bolt loosening faults irrespective of the type of structures (chain-type or ring-type), the number of faults (single or multiple) and the severity of faults (initial or serious).