Study On Optimal Sensor Placement For Dynamic Monitoring And Seismic Damage Identification Method Based On Deep Learning For High Arch Dams

The core of structural health monitoring(SHM)includes monitoring,diagnosis and evaluation.The evaluation is the ultimate goal of the SHM.The research on structural damage mechanism is the premise of the system dynamic research.The sensor monitoring means(optimal sensor placement)is the data base of the research,and the structural status diagnosis is the key link to complete the evaluation.Focusing on the first two links of SHM,this dissertation systematically summarizes the optimal sensor placement methods and damage identification methods in the field of SHM.On this basis,the damage characteristics and seismic safety performance of high arch dams under earthquakes,the consideration factors of optimal sensor placement and the robustness of damage sensitivity characteristics to failure state are studied.In this dissertation,the damage characteristics and equivalent damage model of high arch dam under earthquakes are discussed.An objective function of sensor optimal placement and swarm optimization intelligent algorithm considering multiple factors are developed.A feature extraction method based on unsupervised deep learning and a fast damage identification model after earthquake are constructed.By embedding domain adaptive method,the instability of detection performance caused by model error and load environment change is alleviated.Specifically,the main research contents and conclusions are listed as below:(1)The damage characteristics and damage evolution laws of different parts of the high arch dam under earthquakes are systematically explored,and a regional damage distribution model based on seismic damage characteristics is proposed.In view of the unclear research status of damage characteristics and seismic safety performance index of high arch dam under earthquakes,a series of deterministic seismic dynamic analysis is carried out based on a 300 m grade arch dam,and the validity of two kinds of evaluation indexes of bearing capacity based on weighted average damage volume ratio and weighted average damage area ratio are discussed.Taking the arch crown beam,which is the most active vibration of arch dam,as the analysis target,this dissertation discusses the damage sensitivity characteristics of high arch dam in different regions under earthquakes,and verifies its applicability to different earthquakes;Aiming at the problem that the damage homogenization covers the real damage existing in the previous zoning method of arch dam,this dissertation puts forward a regional damage distribution model based on seismic damage characteristics.This model extends the previous identification target in the process of damage identification of arch dam from simple damage to damage center and damage area,which provides a new extension idea for the damage identification method of arch dam.(2)Aiming at the sensor optimal placement system with increasing monitoring demand,an information matrix integrating multiple monitoring targets is constructed,and a normal cloud hybrid frog leaping algorithm(NCM-SFLA)adapted to the optimization target is proposed.A comprehensive information matrix including modal independence,damage sensitivity and modal strain energy information is deduced,and a distance coefficient is introduced to solve the problem of information redundancy between sensors.The maximum parametric number of information matrix and the minimum condition number of sensitivity matrix are used as guidelines to achieve a balance of sensitivity and robustness of the algorithm.In order to enhance the optimization ability of the information matrix,NCM-SFLA is proposed by introducing the initialization method of chaos and elite reverse strategy,the local search strategy with scaling factor,and the normal cloud generator to update the optimal solution of the sub population.Taking a high arch dam project as an example,the objective function and intelligent algorithm developed in this dissertation are compared with the current mainstream objective function and intelligent algorithm,and the superiority of the algorithm are demonstrated.These results provide a more effective solution and diversified choices for the optimal placement of sensors in large space structures.(3)Guided by the direct extraction of damage sensitivity features from the acceleration response signal of the structure,a feature extraction method based on unsupervised learning model and a fast damage identification model are constructed.Aiming at the problems of low measurement accuracy and poor recognition robustness of damage sensitive features of existing artificial design in actual arch dam engineering,a denoising contractual sparse deep autoencoder(DCS-DAE)model is proposed by exploring the mapping relationship between monitoring data and structural state.This model integrates the advantages of denoising autoencoder,compressive auto-encoder and sparse auto-encoder.Besides,the generalization ability of network feature extraction is increased,and the risk of over fitting phenomenon is reduced by introducing Dropout technique.On this basis,based on the principle of reconstruction error and small probability,combined with box plot and WKNN algorithm,a multi-objective DCSDAE anomaly detection framework is constructed.Finally,the effectiveness and noise resistance of the proposed method are verified by an example of an extra high arch dam.The construction of this model only needs the vibration information of the structure under normal operating conditions.These results provide a solution with higher stability and robustness for the damage identification problem under strong noise pollution conditions.(4)In order to meet the requirements of variable water level detection after earthquake,a DCS-DAE model based on domain adaptive is proposed by combining the idea of transfer learning,which improves the anomaly detection performance of DCS-DAE model under variable load environment.The core idea is to fuse the idea of maximum mean difference(MMD)to constrain the consistency of the data probability distributions of the source and target domains in the feature space with the DCS-DAE model and unify them under the same framework.This fusion not only makes the model have the feature extraction ability of DCSDAE model,but also overcomes the problem that the DCS-DAE model can not be applied to other similar scenes due to the lack of consistency constraints between the source domain and the target domain in the feature space of the objective function.According to the situation of variable water level detection after earthquake,four working conditions considering the uncertainty of structural modeling and the variability of water level are designed,and the effectiveness of the proposed method is verified.From the perspective of feature design,the generalization performance of DCS-DAE anomaly detection model is enhanced,and then the anomaly detection model can be "play by ear" and "draw three inferences from one instance".This model is a useful attempt for the real-time monitoring of the structure under the change of load conditions in the actual operation of the project.