Research On Damage Identification Of Existing Steel Truss Bridge Based On Deep Learning

Bridges are important infrastructure to support social and economic prosperity and quality of life.However,they are susceptible to various kinds of damage(deterioration,corrosion,fatigue,creep,shrinkage,etc.)during service.It is becoming more and more important to properly inspect,monitor and maintain them.How to identify bridge damage timely and accurately is of great significance.Deep learning has attracted much attention due to its great success in image recognition,natural language processing and other fields.Compared with traditional machine learning methods,it can extract high-dimensional features of data.In this paper,a damage identification method based on in-depth learning is developed and validated by steel truss bridges.The main work is as follows:(1)The basic theory of deep learning is introduced,and the working principle of deep belief network and stack auto-encoder used in this paper is introduced in detail.(2)Finite element model modification of steel truss bridge is completed by response surface and multi-objective particle swarm optimization.(3)Damage identification of bridges can be divided into two categories: one is to know the potential location of bridge damage beforehand,to lock the real bridge position through deep belief network,and then to analyze the extent of damage using regression model;the other is to unknown damage location and extent of bridge damage.The damage location and extent of structure are taken as output vectors at the same time.The non-linear mapping relationship between mode shapes and structural damage is established directly.A damage identification method based on depth belief network is proposed and verified by steel truss bridge.The research shows that depth belief network has obvious advantages in accuracy and noise resistance compared with traditional BP neural network.(4)A stack auto-encoder for bridge damage identification is constructed.The stack autoencoder is applied to two kinds of bridge damage identification problems mentioned above.The research shows that the stack auto-encoder is superior to the traditional BP neural network in noise resistance and accuracy,and superior to the deep belief network in computational efficiency.The experimental results in this paper show that the damage identification method based on depth belief network and stack self-encoder can effectively carry out damage identification research.From the results of numerical research,it can be seen that the calculation efficiency of the bridge damage identification method based on stack self-encoder is higher than that based on depth belief network.