Research On Damage Identification Of In-service Industrial Buildings Based On ANN And Vibration Characteristics

In recent years,the safety of in-service industrial buildings has attracted much attention.Therefore,based on the field measured data,this paper establishes a finite element model to compare the frequency response function(FRF)with artificial neural network technology(ANN)and principal component analysis(PCA)Combined,the structural damage of an industrial frame transfer station was located and predicted quantitatively.Because the damage identification technology of vibration is a method for the full-scale,and its principle is that the damage will change the physical properties of the structure(such as quality,stiffness and damping)and its dynamic properties.Therefore,it is workable to detect the damage by the dynamic quantity that is tested by the structural vibration.As for the damage identification of vibration characteristics is pattern recognition in essence,and its purpose is to find the difference between two or more signals,such as the difference of damage before and after structural,or the difference of damage degree and location.However,ANN can carry out the pattern recognition,classification,signal processing and system identification,so it is an ideal tool to supplement the damage detection technology that is based on vibration.And the PCA also has the function of pattern recognition,and it also can carry out the data reduction and noise filtering.Based on all these characteristics,to combine these two techniques can overcome the limitations of the method based on vibration developed before,and it is helpful to provide more accurate and reliable damage identification results.The specific research of this paper is as follows:(1)Firstly,to conduct vibration measurement for the industrial frame transfer station,That is to choose 14 measuring points on site and install sensors at those points,and then carry out the vibration measurement under four operating conditions(static,normal operation,overload operation,and equipment restart),each measurement includes 8192 data points and five measurements are carried out.Then transform the acquired time history data into the frequency domain to calculate the frequency response function.After that,to reduce the noise,filter and simplify the frequency response function data set by PCA,and select the most important 10 principal components as the neural network input.(2)At the same time,combined with the original design drawings and site conditions to establish the benchmark finite element,which is used to analyze the damage condition of structural members under different working conditions.Reducing the stiffness as a damage scenario to introduce the damage index DI and use the PCA to select the principal component,in this process,to analyze the influence of noise pollution on the selection of principal component.Then use the ANN train,verify and test the collected numerical simulation data,so as to establish the damage mode database.(3)Finally,to substitute the damage index DI of unknown damage frequency response function measured on site into the trained neural network,so as to carry out the pattern matching and finally to predict the damage location and degree of the main component.The main contribution of this paper is by using the neural network to locate and quantify the damage of the components of industrial buildings in service,which will be very helpful to the safe and effective operation of industrial buildings in service.