A Study On The Identification Of Hysterically Degrading Structures With General Hysteresis

With the rapid development of globalized economy and increasing growth in mega cities and major infrastructure systems,some key issues of engineering construction of sustainable urban systems have become increasingly important.These problems mainly cover several aspects,health(safety),durability(longevity),green(low carbon),intelligence(life).It is essential to construct a sustainable development system for maintenance and evaluation of major infrastructures during the whole life cycle.System identification is increasingly becoming an important technology for intelligent structural monitoring and control.It is valuable to make use of system identification technology to study the state of the system,evaluate damage and make the system controlled,so that the structure is stable and reliable under multiple disasters and extreme loads.Chapter 1 is introduction.A comprehensive structural health monitoring system mainly includes online monitoring,real time analysis,damage diagnosis,state evaluation and resolution and so on.Several main control strategies include:active control,semiactive control,passive control,hybrid control,etc.System identification is a strategy that uses the input and output data to build system model.Different "model transparency" relies on different physical models.In this research,control parameters of Bouc-Wen class are analyzed,and a hysteretic system is established by considering strength and stiffness degradation as well as pinching behavior.Three different strategies are used for system identification.Chapter 2 first describes hysteresis behavior,i.e.the non-elasticity of the system,the genetic and memory properties of the nonlinear system.The restoring force depends not only on the transient deformation but also on the historical deformation.Generally,under cyclic loading,both mechanical and structural system may dissipate considerable energy.The energy exists in the forms of hysteresis loops,which indicates the system experiences a complete energy dissipation cycle due to friction and other factors.Bouc,Wen and Baber,Wen first proposed a smooth hysteresis model and considered the stiffness and strength reduction by modifying the model.Then Baber and Noori considered the pinching effect by introducing the spring concept,and used this model to study the stochastic systems with nonzero mean.Bouc-Wen class model includes numerous parameters.By parametric space transform and mathematical transformation deviation,an equivalent system is established.Some parameters of the model are found redundant.Local and global sensitivity are evaluated by using root mean value and Sobol index.Conclusion indicates system nonlinearity parameters are most sensitive in local sensitivity while sliding parameter is most sensitive in global sensitivity.Physical meanings of different control parameters of Bouc-Wen model are analyzed.In Chapter 3,a three story shear structure is established to model the original dynamic system.Each substructure is regarded as a lumped mass and is assumed to move laterally.The excitation of structure includes external sin wave and seismic wave.The structure is excited to produce energy dissipation due to factors such as vibration and friction.The modeling of hysteretic system mainly includes system dynamics,restoring force representation and hysteretic model equation.This chapter lays foundation of comparative study on system modeling using intelligent parameter varying,genetic algorithm and transitional Markov Chain Monte Carlo simulation.In Chapter 4,white box model is entirely derived from the first principle of physics without relying on any data,while black box model entirely relies on measurement data without modeling system dynamics.Grey box is the combination of white box and black box.Through a series of mathematical derivation,system dynamic equations including restoring forces are derived.The restoring forces are the nonlinear items including mass,stiffness and damping.The proposed intelligent parameter varying approach combines traditional parametric and nonparametric methods.It uses radial basis function as neuron activation,and it can predict inelastic and hysteretic forces and apply it to multi degree of freedom system.In use of intelligent parameter varying approach for modeling,the model structure is determined by first principle while nonlinear characteristics of system is simulated by neural network.Results show that the shape of hysteresis loops can accurately reflect the change of damage states.When SNR is large,the correlation coefficient approximates 1 regarding restoring forces of identified system and original system.In Chapter 5,Genetic algorithm is used for hysteretic system identification.Populations and individuals are produced when GA is initialized.Selection,crossover and mutation operators are then employed.The selection operator drives the search direction to the optimal individual,and the crossover operator combines the individual to produce the next generation.The mean square error of the predictive response function and the reference response function is set to be the objective function,and the purpose of the optimization is to make the objective function error the smallest.Results show higher SNR has better identification correlation.Similar optimization algorithms are used for comparative study,and these explain local and global optimum phenomenon.In Chapter 6,the design of the optimized Markov Chain Monte Carlo sampling model is used for system identification,and the resulting objective function distribution is the distribution state that we expect to approximate.The problems that the traditional sampling may exist include multimodal distribution of probability density function,spike distribution or smooth distribution.By designing a series of intermediate probability distribution sampling,the algorithm will be more effective.The objective function employs the difference of the simulated function and predicted response function.Identification results show higher SNR has a better identification correlation.Chapter 7 presents a general summation of this study.First control parameters are studied by considering the characteristics of hysteresis model.Subsequently,the hysteresis model is applied to the shear structure,and restoring forces are identified using different system identification methods.The main contributions of this research include:(1)Differential equation of Bouc-Wen hysteresis model exhibits good application value in multiple fields.(2)When different system identification strategies are employed,the choice of different parameters may be correlated with identification accuracy.(3)The variance of restoring forces for structural identification can reflect the strength or stiffness degradation,slip-lock or pinching effects.(4)Intelligent parameter varying approach has its special advantages over Genetic algorithm and Markov Chain Monte Carlo simulation in that Genetic algorithm may be trapped in local optimum,and Markov Chain Monte Carlo may not be applicable for high dimensional data.Intelligent parameter varying approach has its adaptive learning capabilities by approximately designing radial basis functions.