Font Size: a A A

Neural Networks Method On Deformation Prediction And Back Analysis Of Concrete Face Rockfill Dam

Posted on:2006-03-06Degree:DoctorType:Dissertation
Country:ChinaCandidate:F Z XiaFull Text:PDF
GTID:1102360182465735Subject:Water Resources and Hydropower Engineering
Abstract/Summary:PDF Full Text Request
The development of face rockfill dam can be divided into four phases: from 1870 to 1940 is the forepart which characterized by throw rockfill; from 1940 to 1965 is the second phase of transition from throw rockfill to rolled rockfill; 1965 to 1988 is the third phase of resuscitation; from 1988 till now is the last phase of modern when face rockfill dam becomes one of main type of dams. Concrete face rockfill dam (CFRD) is developing very fast in the construction of modern water resources and hydropower projects because of its high security, low cost and wide application. Now CFRD is one of the most widely used dam type.The development of CFRD is late in China, but it has a high start level and speed of growing.In the nearly 20 years it has been widely spread. The Shuibuya CFRD that is being constructed is the highest CFRD in the world. Now the CFRD in China is in the frontline of the world not only in the amount and scale but also in the technique level.The settlement of dam top, the distortion of faceplates and the displacements of joints are all connected with the deformation of dam. The excessive deformation of dam will cause the rupture of faceplates and joints failure that make dam create large amount of leakage and will influence the function of dam. The excessive deformation of dam will also cause the great settlement of dam top and less freeboard. This can generate flooding of dam top and eroding of dam that will possibly lead to the burst of dam. So the deformation of dam is always the main problem in the design of CFRD.Nowadays the main calculation method of the maximum settlement deformation of dam is engineering analogue method. In calculating the maximum settlement deformation of dam, this method considers only two influence factors: dam height and deformation modulus of dam, and simply thinks the maximum settlement deformation of dam is the proportion of the square of dam height, and the inverse proportion of the deformation modulus of dam. The shape of riverbed at dam site has great influence to the deformation of dam, engineering analogue method does not consider this influence. The deformation of dam is very complicated, it is the nonlinear function of the dam height, deformation modulus of dam, shape of riverbed and other factors. Sothe calculation result of engineering analogue method has great difference with the real observation result, and it can not meet the demand of engineering design. H.Imaizumi half experienced formula is used to calculate the maximum deflection of faceplate, this method, same as engineering analogue method, also only considers the influence of the dam height and deformation modulus of dam. It thinks the maximum deflection of faceplate equals the square of the dam height divided by the deformation modulus of dam after reservoir has stored water. This calculation result differs greatly from real observation.Recently, using the finite-element method (FEM) to analyze the strain-stress of CFRD is developing very rapid, there is quite a few calculation programs of 2-D FEM and 3-D FEM. In our country much research works on the FEM analysis of CFRD have been done. For high CFRD, the FEM analysis of strain-stress had been done. The calculation models keep improving, and the material parameters decided by test can be back analyzed with observations, so it is more near to the real conditions. And the effect of the path of stress, creep and rain to the deformation of rockfill materials is further considered. Now the most widely used model in all the world is Duncan E-B model. In our country Shenzhujiang's double bending faces elastic-plastic model and the K-G model of Tsinghua University are usually applied in the strain-stress analysis of CFRD. But the calculation results of FEM also differ greatly from the real observations, the main reasons are: the physical parameters of material are taken from experiment, and the different condition of the experiment and the construction field makes the parameters have errors. The real materials of dam is not uniform, the real filling procedure is not continuous, the model without considering of creep is not accord with real conditions, but the model considering of creep is difficult to find the creep parameters accurately. So nowadays determinations of the material parameters usually apply back analysis with real observations. The back analysis always uses the optimization method, which is to choose the material parameters that can minimize the square sum of the weighted errors between calculation values and observations of stresses and displacements. The disadvantage of the optimization method is the large amount of structural analyses, especially for nonlinear problems, and it is easy to get into local minimize, this brought great difficulties for back analysis. To cut down the difficulties of back analysis, just a few material parameters are picked up to make back analysis.From the above we can see, the present calculation methods of CFRD can not give accurate judgment for the deformation of dam and can not meet the demands of engineering design. Zhushuqiao CFRD and Tianshengqiao CFRD both have created great deformation of dam that led to the rupture of faceplate because of the wrong judgment to the deformation of dam in the period of design. So to find new methods to predict the maximum settlement deformation of dam and the maximum deflection of faceplate, and back analysis of dam material parameters is quite essential and has meanings in theory and engineering.The study of artificial neural network (ANN) can be cast back to 1890, but the real widely study of it began at the end of the 50s and the middle of the 60s of the 20th century. In the 70s, it once fell back to a low tide, but it recovered at the middle of the 80s. From then on, the study and application of ANN begin to boom. The academic of our country attend to this field at the middle of the 80s, but the following is a rapid development and till now, our country has reached the top of this field. Now the study of ANN is moving toward the intelligence calculation and the biology intelligence.The theory of ANN has been applied to many areas such as model recognize, optimization calculation, nonlinear signal management, intellective control, artificial intelligence and so on. The application area almost contains all fields, for example: mechanical engineering, electronic engineering, electric engineering, water resources engineering, chemical engineering, agriculture, environment, economics, military, healthy, meteorology, architecture, minerals and so on.The amount of ANN models can be near a hundred, but just a few are constantly used, BP ANN is the most widely used type of the feed forward networks. Any feed forward ANN network with three layers can make nonlinear mapping of any precision, so it can approach any multidimensional nonlinear function, this offer a new approach to study the deformation of CFRD.In this paper, the BP ANN method to calculate the maximum settlement deformation of dam and the maximum deflection of faceplate of CFRD is presented for the first time. Two BP ANN models with three layers have been built separately, the input layer contains the dam height, deformation modulus of dam, the proportion between the height and wide of the riverbed and the instance that do not consider this proportion. The output layer is the maximum settlement deformation of dam or the maximum deflection of faceplate. The element amount of implied layer can be gainedby calculation experiments. The samples to train the network are got by collecting domestic and foreign observations. To improve the distributing state of the samples, logarithm transformation has been made to the datum. The network model takes Sigmoid function as the transfer function, to do standardization in proportion treatment to the datum of samples make them within the value field of Sigmoid function. Concerning the value of Sigmoid function varies very slow when near 0 and 1, to reduce the training time of the network, in this paper the datum of samples have been transformed to [0.2, 0.8]. Because the amount of samples that can meet the demands are less, only one sample is taken as inspection sample and the others are taken as training samples each time, and by interchange one by one to check the stability and reliability of the network structure. The ANN functions supplied by MATLAB have been used to build BP ANN and write programs. The calculation results indicate that the method raised in this paper is effective and reliable, the accuracy of calculation is obviously higher than those methods that are being used, so it is a method worth spreading.In this paper, BP ANN is used to make back analysis of the material parameters of Duncan E-B model of CFRD for the first time. The widely used Duncan E-B model does not consider creep of dam, but the observations contain creep. So before back analysis, deformation decomposition must be made to separate creep from total deformation. A unitized model has been built in this paper, to divide the total deformation into the components of filling weight, water pressure and creep. The components of filling weight and water pressure use polynomial model, and the component of creep can be simulated by the four elements model. By regression analyses with the observations, determine the undetermined parameters of the models. Based on the deformation decomposition, BP ANN model for the back analysis of dam material parameters is built. The input layer parameters contain the settlements of the observation points and the output layer is the corresponding dam material parameters. This is a multi-output problem, in this paper it is transformed into many single output problem, and so it is simplified. The elements of implied layer still can be gained by calculation experiment. The relationships between the material parameters and the corresponding settlements of the observation points can be got by using 2-D FEM and the datum are taken as training samples. Also, standardization in proportion treatment is done to transfer the datum within [0.2, 0.8]. By training thenetwork, the mapping relationship between the deformation of dam and material parameters can be gained. The loads component of observations are used for simulations of the network, the corresponding outputs are the material parameters of back analysis. Back analysis of the material parameters of Duncan E-B model is carried out for some CFRD. The calculation results indicate that the settlements of dam calculated with the material parameters that are determined by back analysis using the method recommended in this paper is quite near to the observations. The method is simple, effective and reliable, it demands less structural analysis works and can do back analysis to all the material parameters, evidently it has more advantages than optimization method.BP ANN methods to calculate the maximum settlement deformation of dam and the maximum deflection of faceplate and do back analysis of dam material parameters, presented in this paper, supply new calculation methods for the design of CFRD, by further improvements it can be applied in engineering design.
Keywords/Search Tags:concrete face rockfill dam(CFRD), maximum settlement deformation of dam, maximum deflection of faceplate, Duncan E-B model, back analysis, deformation decomposition, BP neural networks method, regression analysis
PDF Full Text Request
Related items