Study On Soft Computing And The Application To Bridge Monitoring And Assessment

One of the challenges in health monitoring and assessment of large scale bridges is to process measured data effectively and make reliable evaluations for the service states. Specifically, the primary difficulties include ill-posedness of the inverse problem, the computation complexity of combinatorial optimization, uncertainties in evaluations, non-stationarity in signal, etc. It is critical to develop new methods that can solve the above problems effectively and practically with pressing needs, since the techniques based on the dualistic logistic, linear system and conventional numerical analysis have many deficiencies.Soft computing (SC), integrating artificial neural networks (ANN), swarm intelligence optimization (SIO) and uncertainty reasoning (UR) has more advantages in data inversion, global nonlinear optimization, decision-making analysis, etc. Based on the studies of SC, decision making and parameter identification are systematically researched combining bridge health monitoring and assessment.The main contents and details are as follows:(1)The basic concepts and characteristics of SC methods are introduced. As the important member of SC, the methods of SIO are also discussed. Comparison and analysis on the methods of SIO by numerical experiments are conducted based on the proposed benchmark for optimization.(2) Improve the conventional BP (back propagation) ANN by the proposed self-adaptive ratio and the operator for local optimum check with the purpose of enhancing the identification precision. Based on this improvement, the ANN is established to identify impact force during the ship-bridge collision on the pier of the Nanjing Yangtze River Bridge (NYRB). By the established ANN, the magnitude, direction and location of impact force during the collision can be identified effectively due to the ability of memory, associability and anti-jamming of the proposed ANN. The results of numeric studies show that the proposed ANN has rapid convergence speed and high precision for the identification of the impact force.(3) A new coding approach named "number set coding" and the corresponding crossover and mutation operator in genetic algorithm (GA) for the optimization of senor locations are proposed. Combining with the clone mechanism in artificial immunity system (AIS), the methods aiming at optimizing the placement of sensors are developed which can optimize the placement of sensors globally for both single-objective and multi-objective optimization. With the proposed methods, the optimal placement of monitoring sensors for the 1/8 scale structure of Ting Si He River Bridge (TSHRB) and NYRB are obtained under the defined single-objective and multi-objective. The results show the flexiblility due to shorter code length of "number set coding" and strong global convergence of the proposed method. For single-objective criterion, the optimal locations of sensors obtained accumulatively and congregated in certain regions. For multi-objective criterion, The Pareto solution representing multi-scheme is the convenient to compare the plans of sensor placements for decision makers.(4) The analytical hierarchy processing (AHP) with the three points interval number (TPIN) is developed to evaluate the state of bridges. The indexes system for consistency check and the optimization model for the optimal weight are established. In order to obtain the optimal weight and the corresponding index of consistency from the defined optimization model, a self-adaptive particle swarm optimization (PSO)-simulating annealing (SA) algorithm with self-adaptive adjustability for inertia coefficient in PSO and strong local search ability of SA is proposed. Then the proposed AHP with TPIN (TPIN-AHP) is employed to evaluate the steel truss of NYRB. The results show the TPIN-AHP is simple, flexible and able to consider the uncertainty synthetically. The method to check the consistency of judgment matrix with TPIN and the indexes of the reference numbers of mean random judgment matrix has promising engineering applications.(5) The ACO (ant conolt optimization)-Chaos Pursuit algrithom is proposed to decompose signal sparsely with the application to parameter identification and time-frequency analysis. Based on the decomposition of the simulating signal, the result shows that the speed for signal decomposation of proposed algrithom is 15 times faster than the conventional Matching Pursuit (MP) algrithom in the aspect of efficiency of the proposed algrithom. For modal parameter identification, the experiment on the 1/8 scale structure of TSHRB is executed, and the results show the proposed method is stable, efficient and has high precision for identification by comparison with theoretic analysis. The self-adaptive spectra for time-frequency analysis of measured signal from NYRB monitoring system are obtained by sparse decomposition with the proposed method. By comparison with the wavelet spectra and the Wigner-ville distribution (WVD) spectra of the same signal, it shows that the self-adaptive spectra obtained by the proposed algorithm for sparse decomposing has good time-frequency aggregation.