Experimental and intelligent modelling for predicting the amplitude of footing resting on geocell-reinforced soil bed under vibratory load
School of Engineering
The reliable estimation of displacement amplitude is the prime factor for the design of geo-structures supporting the vibration loads. Currently, very subtle knowledge is available to compute the displacement amplitude of footing resting on the geocell-reinforced bed subjected to vibration loading. The advent of artificial intelligence modelling has antiquated many traditional approaches. Thus, in the present study, a novel hybrid paradigm has been developed by combining the artificial neural network (ANN) with dragonfly optimizer (DFO), abbreviated as ANN-DFO. To train and test the model, the reliable database was developed by conducting extensive field vibration tests. To establish the specific prediction target for the proposed model, displacement amplitude (da) was considered as an output index. A combination of parameters involving properties of foundation bed, geocell reinforcement, and dynamic excitation has been considered as input variables. Along with the standalone ANN model, the predictive veracity of the ANN-DFO was also compared with three other robust machine learning models, namely, Gaussian process regression (GPR), random forest (RF), and M−5 rules. Primarily, the forecasting ability of the developed models was assessed based on rigorous statistical criteria and a multi-criteria approach. Moreover, the model is also validated against the entirely new independent data which is not part of the original dataset. From the results, ANN-DFO model has shown a superior performance in predicting the displacement amplitude of footing resting on geocell-reinforced beds as compared to the other benchmark models. Finally, the strength of input variables on the estimation of displacement amplitude was highlighted using the sensitivity analysis.