Numerical And Experimental Study On Damage Detection Of Large Span Space Structure

Large span space structure with advantages such as big span, light weight and goodcomprehensive economical indexes is popular and widely used in the stadium,exhibition hall, waiting hall and other types of building. However, the structuralformation and force conditions of large span space structure are complicated. At thesame time, because of the effect of environmental erosion, material aging, unevensettlement of foundation, fatigue effect and so on, large span space structure issusceptible to be damaged. The damage will even cause a great loss of lives andproperty. In order to avoid such loss, health monitoring of large span space structures isnecessary. The damage detection is the core of health monitoring. In this paper, taking aK6single-layer spherical reticulated shell model an example, based on time seriesanalysis and neural network, the damage cases and the location of damaged element aredetected, the main contents are as follows:(1)According to the present research situation on damage identification ofstructures at home and abroad, the common damage identification approaches aresummarized and their advantages and disadvantages are pointed out.(2) The damage detection method of large span space structure based on time seriesanalysis and neural network is researched. Firstly, according to the change of AR timeseries model coefficient, the structure’s damage is detected. Then, based on the neuralnetwork, the location of damaged element is recognized. The study shows that thismethod is simple, feasible and accurate. Besides, this method has a certain ability toresist noise.(3)By numerical simulation, taking a K6single-layer spherical reticulated shellmodel for example, the damage of a K6single-layer spherical reticulated shell isdetected based on time series analysis and neural network. Firstly, the accelerationresponse is obtained by the finite element analysis by adding5percents of noise;Secondly, AR time series model is built. According to the change of the modelcoefficient, the damage is detected. Lastly, based on the neural network, the location ofdamaged element is recognized. The results of numerical simulation show that: the damage detection method of large span space structure based on time series analysis andneural network is feasible, accurate and has capability in noise tolerance.(4)Using a test program of single-point excitation and multi-point response, animpulse response experiment on a K6spherical reticulated shell model is conducted.The model’s scale is1:10. The vertical acceleration response of every note is acquiredafter a vertically downward single-point excitation is applied on the vertex of thereticulated shell model by a rubber hammer head. Considering the number of thesensors is limited in the practical engineering, based to the symmetry of reticulated shellmodel and the analysis result of particle swarm optimization, the acceleration responsesof6notes are selected as research object. In view of the18damage cases, based on timeseries analysis and neural network, the damage detection method of large span spacestructure is vertified. The results of experiment show that:①The natural vibrationfrequency of the experimental model is similar to the result of finite element analysis. Itshows that the experimental model is reliable to some extent.②The location ofdamaged elements on K6single-layer spherical reticulated shell model can berecognized accurately by the damage detection method of large span space structurebased on time series analysis and neural network. It shows that this method is feasible,accurate and has capability in noise tolerance.③The number and location ofacceleration sensor and the number of the hidden unit in neural network have greatinfluence on the accuracy of damage detection.