Improvements On Non-Stationary Signal Processing Methods And Their Applications In Earthquake Engineering

Since 1980s, signal processing has entered into an unprecedented period of rapid development with the emergence of new theories and methods. As a fundamental theory and a universal tool, signal processing has more closely interpenetrated and combined with other disciplines than ever, and in many applications some traditional methods have established their overwhelming superiority. However, non-stationary signal processing is still an important issue, of which the difficulty is to make a reasonable tradeoff between the precision and resolution. Such a problem exists both in parametric and nonparametric methods.Vibration signal processing is the most typical application in civil and earthquake engineering, of which the major purpose is to extract physically meaningful information from a signal as much as possible. While in the real world, signals are often nonlinear and non-stationary and their length is often very short, which makes the processing of such signals a difficult task. Especially, in earthquake engineering ground motions and structural responses are two familiar types of vibration signals. Due to the different characteristics of them and the associated systems, two related applications, i.e. earthquake ground motion and structural damage detection, have different requirements for signal processing.Firstly, for earthquake ground motions the non-stationary properties in amplitude and frequency contents are also important properties besides the conventional properties of amplitude, frequency and duration, and instantaneous spectrum is regarded as a proper notion to describe such non-stationary properties. Though there are several methods available for instantaneous spectrum estimation, the contradiction between precision and resolution is not yet reconciled, and the modeling of instantaneous spectrum is even more rarely studied. In this context, to find a method for the study of modeling of instantaneous spectrum is an efficient way to fulfill the requirement for ground motions in structural seismic design and analysis.Secondly, in structural damage detection vibration based methods have been the research focus during the past decade. To reduce the limitations of some existing methods, continuous efforts have given rise to some improved methods. In addition, some more recent research has shown that extracting physically meaningful information from vibration signals to detect structural damages may be the new trend, which undoubtedly requires some signal processing methods to extract useful information from vibration signals in an efficient and real-time manner through which the performance (state) of structures can be accurately evaluated.For above reasons, improved signal processing methods and their applications in non-stationary properties of ground motions and structural damage detection are systematically studied in this thesis, and major works and research findings are summarized as follows:1. Non-stationary signal processing methodsFirst, three improved parametric methods are proposed which not only perfectly reconciles the contradiction between precision and resolution but also provides a basis for the modeling of instantaneous spectrum of ground motions. Then, an EMD and VARMA model based method is also proposed which offers advantages in accuracy and frequency resolution and produces more physically meaningful and readable Hilbert spectrum.2. Non-stationary properties and synthesis of ground motionsTwo types of envelope models are used to study the non-stationary properties in intensity of ground motions. Based on the 5750 strong ground motion records collected, the multidimensional correlation and attenuation relations of envelope parameters are investigated and the recommendations of envelope parameters are put forward. An exponential decay model of instantaneous frequency is proposed and the multidimensional correlation and attenuation relations of model parameters are analyzed, then the recommendations of model parameters are set forth. A simplified model of instantaneous spectrum in the form of the product of independent functions in time and frequency domains is proposed, which not only produce a reference index for selecting reasonable ground motions for structural seismic design and analysis but also provides a basis for synthesis and simulation of ground motions allowing for the non-stationary properties in amplitude and frequency contents at the same time. Based on the improved signal processing methods and the instantaneous spectrum model, several methods for synthesis and simulation of ground motions are proposed, which can provide reasonable ground motions for structural seismic design and analysis.3. Structural damage detectionIn the sense of extracting complete information from vibration signals, three improved methods for structural damage detection are proposed. Since time varying damage indices are used in these improved methods, they have the advantage in applicability, sensitivity and robustness, and can detail the time sequence of occurrence, relative severity and cumulative process of structural damages at multiple locations in an efficient manner, which cannot be done by using the conventional damage detection methods.In this thesis, signal processing is used as a link between earthquake ground motion and structural damage detection which are two unrelated but important research fields in earthquake engineering. The research findings presented here not only help to provide more reasonable ground motions for structural seismic design and analysis on the basis of a more comprehensive understanding of the properties of ground motions, but also provide a novel approach for structural damage detection which can be easily combined with other applications such as structural health monitoring and vibration control.Finally, the weakness of the present study is briefly discussed and some recommendations for future research are provided.