Study On Structural State Detection Based On Quaternion Wavelet Transform

Vibration-based state detection is one of the main method of structural health monitoring. And in most of the vibration-based method, the excitation force needs to be known or seems like white-noise. In practice, however, structural vibrations observed in operation cannot always meet above conditions. Therefore, to complete structural state detection based on response-only data is more attractive for engineering structures.The method of structural state detection based on response-only data is mainly divided into two stages, one is to explore an effective state index which reflect the vibration characteristics of structures. The other is to choose a reasonable method of state index identification method to complete the state detection. As a well-known linear system concept, the transmissibility function (TF) reflects the vibration transfer law between the two points of structure and the dynamic characteristics. And it is only one typical state index based on response-only data. The single axis sensor is used to obtain vibration signal in the process of state index selection. The transmissibility based on scalar signal will deviate from its theoretical value with changes of the installation orientations of the sensors, resulting false identification. When triple axis accelerometer is fixed to collect spatial vibration, no matter what the installation orientation of the triple sensor is, the real spatial vibration will be faithfully acquired in principle. Therefore, measured value (no noise is taken into consideration) of quaternion-based three-channel joint transmissibility (QTJT) should be consistent with its theoretical value in spite of the installation orientation of the sensors. In most state index identification methods, the training data derive from multiple samples but the testing data derive from single sample. To overcome the measurement uncertainty of testing data, it is reasonable to construct testing data based on multiple samples also. QWT reduces the influence on state detection caused by measurement uncertainty of single testing sample.In this paper, a state detect method using QTJT in conjunction with quaternion wavelet transform (QWT) is described in this paper. And an actual experiment is performed on the test platform of ballastless track structure to verify the effectiveness of the method. The main work is as follows:(1) The finite element model of ballastless track was established using ABAQUS software. The state change was obtained by loosening different fasteners one after another between two testing points. And the model is validated by a comparison between the vibration signal obtained from the model and the measured vibration signal in time domain and frequency domain, respectively. The vibration signals obtained from the model of track structure were utilized to verify the repeatability of QTJT and then it was verified that it was feasible to use QTJT during structural state detection.(2) The varying structural states were stimulated by loosening different fasteners one after another and the acceleration signal of rail under different states were obtained by using two triple axis. Then multiple QTJTs under different stateswere calculated as state index. The method based on QWT, BP neural network, non-negative matrix factorization (NMF) and Karhunen-Loeve (K-L) transform were performed for QTJTs to obtain the feature index. Finally, the training sample and testing sample were constructed using the feature index and the Euclidean distance between state feature index was defined as the state indicator to complete the structural state detection. The results indicated that the QWT is the optimal among all the methods involved comparison.(3) The forcre generated by LG-900 hydraulic rail tensor was utilized to simulate track longitudinal stress. The situation of different longitudinal stress and loosening different fasteners coupled together was studied. Mmultiple QTJT under different situations were calculated. Then QWT was used to structural state detection. Experimental results indicated that the longitudinal stress of object structure should be taken into consideration during state detection based on vibration characteristic. For the suggested method, the resolution of the longitudinal force of rail was superior to 10 Mpa which is equivalent to temperature change of 1.75? for a completely locked railway structure.