This article presents a novel second-order output spectrum–based method for detecting and localizing multiple bolt loosening faults in complex structures with a sensor chain. This new method is developed based on a recently developed nth-order output spectrum estimation algorithm using only proper data obtained from a chain of sensors. The properties of the second-order output spectrum transmissibility of the sensor chain along physical structures with bolts are studied systematically. In the new method, the noise effect on accuracy of the estimation of fault indicators is investigated as well. The new method also considers more general nonlinear restoring forces not only due to faults but also due to inherent nonlinearity in structure when deriving the properties of the second-order output spectrum transmissibility. Extensive simulation and experimental results demonstrate that the second-order output spectrum of the studied structure can be estimated efficiently, and the second-order output spectrum transmissibility can be used as an effective and reliable damage indicator for the detection and localization of multiple bolt loosening faults in complex bolted structures.

Complex grid-like structures such as steel truss bridges and steel truss roofs commonly exist in civil engineering applications. Since these structures are usually subject to dynamic operational forces, faults like chemical corrosion, fatigue crack, and bolt loosening arise, and seriously affect structural health. To consider some issues in existing frequency domain fault evaluation methods like applicability for complex structures, requirement of baseline data, neglect of nonlinear boundaries, and localization of local faults, a novel method using nonlinear vibration features and Euclidean distance measurement is presented in this paper. Firstly, complex grid-like structures are decomposed into a series of simple T-type substructures according to structural characteristics and load paths. Secondly, related T-type multi-degree-of-freedom model is built by simulating potential faults and nonlinear boundaries as related nonlinear damper-spring connections. Thirdly, exciting the model and extracting output responses only, novel and local fault features, which are functions of structural properties, nonlinear fault-induced loads and transmissibility functions, are defined. Then, utilizing defined features from the local substructure to be evaluated only, a novel fault index is defined with Euclidean distance measurement. Finally, corresponding evaluation method is developed, and its effectiveness and applicability are demonstrated by comparative studies on a lab bolted grid-like structure.