Judgment Of Clay Dispersion Based PCA-BP Artificial Neural Network Model

Dispersive soil as a kind of special soil is concerned in the geotechnical engineering field in recent years. The main characteristic of dispersive soil is its low water erosion resistance. Engineering practices and experiment researches indicated that permeability coefficient of dispersive soil is relatively low and therefore dispersive soil has good anti-seepage performance. However, once water conservancy projects that were constructed using dispersive soil have cracks, soil particles around cracks will also be washed away by water erosion, thereby resulting the piping of impervious bodies and even destructions for water conservancy projects. From this perspective, dispersive soil is the main reason for engineering accidents of reservoirs, dams, roadbed and channel projects. Geological surveys showed that abundant dispersive soil is existed in many areas in China. So applications of dispersive clay as damming soil and subgrade foundation are inevitable.Based on the research status around the world, 53 groups of soil sample for 6 water conservancy projects were studied. Physical and chemical indexes as well as mineral contents of 53 groups of soil sample were analyzed. The basic principles and applicable conditions of 5 experimental methods for determining the clay dispersion were discussed in details. Based on that, different weights for 5 experiments were assigned employing statistical analysis method and quantitative standard for clay dispersion was also given. Then influencing factors for clay dispersion were studied using integrated grey correlation analysis and correlation analysis methods to explore the mechanism of clay dispersion. Finally, two clay dispersion judgment models that had different topological structures were developed on the basis of grey correlation analysis and principal component analysis, which created a new approach for the judgement of clay dispersion at the first time. The main results are as follows:(1)53 groups of soil sample were low liquid limit clay. The corresponding particle composition was given priority to powder. The plastic index ranged from 8.7 to 21.5. PH values varied from 7.98 to 9.62, indicating alkaline. The content of sodium ion was relatively high within positive ions. Clay mineral composition was mainly illite and its mass fraction ranged between 30% and 69.6%. Small amounts of illite were existed as illite and smectite forms. Non-clay minerals were mainly composed of quartz and plagioclase.(2)Quantitative standard for judging clay dispersion was given employing statistical analysis method for the first time. Results showed that soil samples were dispersive soil when the dispersive weight of the soil sample was over 44%. Soil samples were dispersive soil when the dispersive weight of the soil sample was equal to 44% and the transitional weight was greater than or equal to 39%. On the opposite way, soil samples were transitional soil. When dispersive weight of the soil sample was less than 44% while the sum weight of the transitional and dispersive weight was greater than or equal to 44%, soil samples were transitional soil. Otherwise, soil sample belonged to non- dispersive soil.(3)This paper introduced the basic concepts and the mathematical model of the grey relational analysis and further clarified the dispersion mechanism of soil. The results indicated that the correlation degree among the percentage of sodium, the percentage of exchangeable sodium, p H value and dispersive degree was larger than 0.9. That is, the effect of these three indicators on clay dispersion was the largest. Correlation degree among organic material, sodion(Na+), clay content and bicarbonate ions ranged from 0.8 to 0.9, indicating that the relationship between these four indicators and clay dispersion was relatively close. The correlation degree between smectite and clay dispersion was the lowest, with the value of 0.730.(4)Correlation analysis results indicated that the correlation among organic material, clay content and dispersive degree was negative. That is, the larger organic matter and clay content were, the lower dispersive tendency of soil was. The percentage of sodium, the percentage of exchangeable sodium, p H value and dispersive degree had significant positive correlation. Namely, the higher the content of sodium ions and exchangeable sodium in the pore water and p H value were, the greater the dispersive likelihood of the soil was.(5)We determine the dispersion of clay by using the artificial neural network, which is a new approach for the judgement of clay dispersion at the first time. Based on the correlation analysis and principal component analysis, two clay dispersion judgment models(i.e., traditional BP neural network model and PCA-BP neural network model) with different topological structures were developed. Results showed that relative errors between the predicted results and expected outputs for two neural network models were less than 5%, indicating that the evaluation results of two neural network models were relatively reliable and two models could be used for evaluating clay dispersion. Compared with the traditional BP neural network model, PCA-BP neural network model had advantages of reducing the complexity of the network structure and speeding up the convergence rate of the corresponding network as well as improving the predictive accuracy of the network. The method has certain actual meanings to save manpower, material resources and finacial resoures.