A study conducted by Deepak Kumar et al. from the Department of Civil Engineering at the National Institute of Technology, Patna, India, focuses on the classification of soil liquefaction using advanced deep learning techniques. Soil liquefaction occurs when soil pore spaces are fully saturated, leading to a reduction in shear strength and the soil’s behaviour resembling that of a liquid. The phenomenon was initially introduced by Mogami and Kubo in 1953, subsequently explored by Ishii et al. in 1963 through shaking table experiments.
Several previous studies have employed the Standard Penetration Test (SPT-N) distribution curve to evaluate in-situ liquefaction of sand. Tsuchida and Hayashi, in 1972, established correlations between peak ground acceleration, critical N values, and liquefaction potential, further developing the foundational work of Seed and Idriss from 1967, which focused on shear stress and dynamic strength. Subsequent models have been created by various researchers, including Iwasaki and Robertson. A comprehensive review of liquefaction potential evaluation methodologies was conducted by Youd and Idriss in 2001, with refinements introduced by Boulanger and Idriss in 2015. Recent contributions have featured works by Liao et al. (1988) and Çetin (2000), who developed probabilistic regression models, and by Goh (1994) and Hanna et al. (2007), who leveraged artificial neural networks for assessing liquefaction susceptibility. Moreover, Taha and Firoozi (2012) explored the estimation of clay cohesion using artificial intelligence
techniques.
In response to the limitations of previous studies—such as overfitting and slow convergence—there remains considerable potential to enhance existing models by applying more advanced machine learning techniques. Consequently, the authors implemented an Emotional Neural Network (EmBP), an advanced artificial neural network (ANN) variant, for the classification of soil liquefaction. EmBP incorporates artificial hormones as signals to regulate the training operations of individual neurons in a feedback loop. Its architecture consists of (i) an input layer, (ii) a hidden layer, and (iii) an output layer, with deep learning emphasizing precise weight assignments across multiple layers for hierarchical representation learning. This method employs non-linear transformations to extract high-level features from data, serving as a multistage process that refines information through successive filtering.
The study utilized a dataset of 109 samples, with input parameters including peak ground acceleration (PGA) and cone resistance to predict the status of soil. The dataset was partitioned into 70% for training and 30% for testing. The EmBP model was compared with traditional deep learning techniques, where a four-layer network was selected for deep learning tasks using cross-entropy loss and the Adam optimizer, while the EmBP model was fine-tuned through a trial-and-error approach. Based on evaluation metrics, the EmBP model demonstrated marked superiority with an accuracy of 89%, in contrast to the deep learning model, which achieved an accuracy of 79%.
This research underscores the applicability of deep learning and EmBP techniques for assessing seismic hazards in geotechnical and earthquake engineering. The developed models exhibit commendable performance, with the EmBP model slightly outperforming the deep learning approach. Notably, the findings indicate that only two parameters are sufficient for calculating soil liquefaction susceptibility, and no charts or tables are required to predict seismic liquefaction potential.
In conclusion, the study presents robust models for determining the seismic liquefaction potential of soil. The findings advocate for the application of deep learning and EmBP techniques, encouraging further exploration in various related fields of civil engineering.
References
Kumar, D., Samui, P., Kim, D., & Singh, A. (2021). A novel methodology to classify soil liquefaction using deep learning. Geotechnical and Geological Engineering, 39, 1049-1058.
Mogami, T. (1953). The behavior of sand during vibration. In Proc. 3rd ICSMFE (Vol. 1, pp. 152-155).
Ishii, Y., Tsuchida, H., & Furube, T. (1963). Study on earth pressures and pore-water pressures of saturated sands during vibration. Report of Port and Harbour Research Institute.
Kishida, H. (1966). Damage to reinforced concrete buildings in Niigata city with special reference to foundation engineering. Soils and Foundations, 6(1), 71-88.
Koizumi, Y. (1966). Changes in density of sand subsoil caused by the Niigata earthquake. Soils and Foundations, 6(2), 38-44.
Tsuchida, H., & Hayashi, S. (1972). Estimation of liquefaction potential of sandy soils. Publication of: Mcgraw Hill Book Company, 14.
Seed, H. B., & Idriss, I. M. (1967). Analysis of soil liquefaction: Niigata earthquake. Journal of the Soil Mechanics and Foundations Division, 93(3), 83-108.
Iwasaki, T. (1978). A practical method for assessing soil liquefaction potential based on case studies at various sites in Japan. In Proc. of 2nd Int. National Conf. on Microzonation, 1978 (Vol. 2, pp. 885-896).
Iwasaki, T., Tokida, K. I., Tatsuoka, F., Watanabe, S., Yasuda, S., & Sato, H. (1982, June). Microzonation for soil liquefaction potential using simplified methods. In Proceedings of the 3rd international conference on microzonation, Seattle (Vol. 3, No. 2, pp. 1310-1330).
Robertson, P. K., & Wride, C. E. (1998). Evaluating cyclic liquefaction potential using the cone penetration test. Canadian geotechnical journal, 35(3), 442-459.
Seed, H. B. (1982). Ground motions and soil liquefaction during earthquakes. Earthquake engineering research insititue.
Seed, H. B., Idriss, I. M., & Arango, I. (1983). Evaluation of liquefaction potential using field performance data. Journal of geotechnical engineering, 109(3), 458-482.
Seed, H. B., & Idriss, I. M. (1971). Simplified procedure for evaluating soil liquefaction potential. Journal of the Soil Mechanics and Foundations division, 97(9), 1249-1273.
Ye, L., Gao, L., Marcos-Martinez, R., Mallants, D., & Bryan, B. A. (2019). Projecting Australia’s forest cover dynamics and exploring influential factors using deep learning. Environmental Modelling & Software, 119, 407-417.
Idriss, I. M., & Boulanger, R. W. (2010). SPT-based liquefaction triggering procedures. Rep. UCD/CGM-10, 2, 4-13.
Juang, C. H., Ching, J., Luo, Z., & Ku, C. S. (2012). New models for probability of liquefaction using standard penetration tests based on an updated database of case histories. Engineering Geology, 133, 85-93.