Details

The soil properties characteristics are the object of the current study. Determination of the soil properties characteristics is a complex and responsible engineering and geological task. Reliability of engineering construction and its cost depend on the quality of solution of this task. The article presents the results of the study of the possibility of predicting the soil properties characteristics on the example of determining the sand deformation modulus. Based on the analysis of previous studies of correlation between the soil properties characteristics, the list of independent soil properties characteristics was determined: soil genesis, static normal stress, granulometric composition, initial density and humidity of the soil sample. The main disadvantages of existing methods of predicting the soil properties characteristics were identified. The possibility of using artificial neural network for predicting the soil properties characteristics was determined. The soil deformation modulus was selected as a response (dependent variable). The presence of not only numerical but also classification features among the independent characteristics did not allow predicting the soil properties characteristics within the framework of the classical regression model. A soil information model, based on an artificial neural network, was used to solve this problem because not only continuous quantitative but also discrete classification parameters (genesis) can be used among the independent parameters of the soil information model. Laboratory studies of 655 samples of alluvial sand of the Irtysh River floodplain were performed to confirm the possibility of using the soil information model. 5895 data vectors were obtained, including information on independent and response parameters. A detailed study of two granulometric compositions demonstrated limited possibilities for using known statistical methods for determining the soil properties characteristics. In 9 out of 20 cases, the results of the studies did not follow a normal distribution. The use of the soil information model allowed to solve this problem - the absolute percent error in determining the deformation modulus did not exceed 12.55 % (mean - 5.05 %), the coefficient of determination R[2] was at least 0.83 for unloaded sand samples, and at least 0.94 for loaded ones, for all datasets - 0.97. The performed studies confirmed the prospects of using the soil information model for predicting soil properties based on its known characteristics, which reduced the cost of engineering and geological surveys while ensuring the required accuracy of determining the soil characteristics.

• Soil information model for prediction the soil properties characteristics
  • 1. Introduction
    • 1.1. Correlations between the Soil Properties Characteristics
    • 1.2. Methods for Predicting the Soil Properties Characteristics
    • 1.3. Predicting Models based on the Artificial Neural Networks
    • 1.4. Soil Information Model
  • 2. Materials and Methods
    • 2.1. Laboratory tests
    • 2.1.1. Experiment design
    • 2.1.2. Materials
    • 2.1.3. Laboratory equipment
    • 2.2. Soil Information Model
  • 3. Results and Discussion
    • 3.1. Laboratory Studies
    • 3.2. Prediction based on SIM
  • 4. Conclusions

Magazine of Civil Engineering / Санкт-Петербургский политехнический университет Петра Великого. — Санкт-Петербург: СПбПУ, 2020-. — Периодичность: 8 раз в год. — Выходит с 2020 г. — Загл. с титул. экрана. — Свободный доступ из сети Интернет (чтение, печать, копирование). — Электрон. журнал. — Текст: электронный

Magazine of Civil Engineering / Санкт-Петербургский политехнический университет Петра Великого. — Санкт-Петербург: СПбПУ, 2020-. Т. 17, № 5, 2024. — 1 файл (12,5 Мб). — Свободный доступ из сети Интернет (чтение, печать, копирование). — Электрон. журнал. — <URL:http://elib.spbstu.ru/dl/2/j25-12.pdf>.