An evolutionary modelling approach to predicting stress-strain behaviour of saturated granular soils. Issue 8 (5th November 2018)

• Record Type: Journal Article
• Authors: Ahangar Asr, Alireza; Faramarzi, Asaad; Javadi, Akbar A.
• Is Part Of: Engineering computations. Volume 35:Issue 8(2018)
• Page Start: 2931
• Page End: 2952
• Publication Date: 2018-11-05
• Subjects: Evolutionary-based data mining -- granular soils -- triaxial stress-strain behaviour; Computer-aided engineering -- Periodicals; Computer graphics -- Periodicals; 620.00285
• DOI: 10.1108/EC-01-2018-0025
• Languages: English
• ISSNs: 0264-4401

Abstract

Abstract : Purpose: This paper aims to develop a unified framework for modelling triaxial deviator stress – axial strain and volumetric strain – axial strain behaviour of granular soils with the ability to predict the entire stress paths, incrementally, point by point, in deviator stress versus axial strain and volumetric strain versus axial strain spaces using an evolutionary-based technique based on a comprehensive set of data directly measured from triaxial tests without pre-processing. In total, 177 triaxial test results acquired from literature were used to develop and validate the models. Models aimed to not only be capable of capturing and generalising the complicated behaviour of soils but also explicitly remain consistent with expert knowledge available for such behaviour. Design/methodology/approach: Evolutionary polynomial regression (EPR) was used to develop models to predict stress – axial strain and volumetric strain – axial strain behaviour of granular soils. EPR integrates numerical and symbolic regression to perform EPR. The strategy uses polynomial structures to take advantage of favourable mathematical properties. EPR is a two-stage technique for constructing symbolic models. It initially implements evolutionary search for exponents of polynomial expressions using a genetic algorithm (GA) engine to find the best form of function structure; second, it performs a least squares regression to find adjustable parameters, for each combination of inputs (terms in the polynomial structure). Findings: EPR-based models were capable of generalising the training to predict the behaviour of granular soils under conditions that have not been previously seen by EPR in the training stage. It was shown that the proposed EPR models outperformed ANN and provided closer predictions to the experimental data cases. The entire stress paths for the shearing behaviour of granular soils using developed model predictions were created with very good accuracy despite error accumulation. Parametric study results revealed the consistency of developed model predictions, considering roles of various contributing parameters, with physical and engineering understandings of the shearing behaviour of granular soils. Originality/value: In this paper, an evolutionary-based data-mining method was implemented to develop a novel unified framework to model the complicated stress-strain behaviour of saturated granular soils. The proposed methodology overcomes the drawbacks of artificial neural network-based models with black box nature by developing accurate, explicit, structured and user-friendly polynomial models and enabling the expert user to obtain a clear understanding of the system.