Elsevier

Applied Ocean Research

Volume 59, September 2016, Pages 408-416
Applied Ocean Research

Uplift capacity prediction of suction caisson in clay using a hybrid intelligence method (GMDH-HS)

https://doi.org/10.1016/j.apor.2016.07.005Get rights and content

Highlights

  • In this study, a novel hybrid intelligent method which is called GMDH-HS has been developed for uplift capacity prediction.

  • Developed method and GMDH1 and GMDH2 methods have been used for suction caisson uplift capacity prediction.

  • The results of GMDH-HS were compared with other methods such as IFRIM, MARS, FEM, ANN, LGP, GEP and TGP.

  • The predicting results of GMDH-HS method indicated that it is a valuable technique for the prediction of suction caisson uplift capacity.

Abstract

Suction caissons are widely used for offshore facilities foundation or anchor system. They should be very stable and also to provide stability of main massive structures those are upon them. Suction caisson uplift capacity is the main issue to determine their stability. During recent years, many artificial intelligence (AI) methods such as artificial neural network (ANN), genetic programming (GP) and multivariate adaptive regression spline (MARS) have been used for suction caisson uplift capacity prediction. In this study, a novel hybrid intelligent method based on combination of group method of data handling (GMDH) and harmony search (HS) optimization method which is called GMDH-HS has been developed for suction caisson uplift capacity prediction. At first, the Mackey-Glass time series data were used for validation of developed method. The results of Mackey-Glass modeling were compared to conventional GMDH with two kinds of transfer function called GMDH1 and GMDH2. Five statistical indices such as coefficient of efficiency (CE), root mean square Error (RMSE), mean square relative error (MSRE), mean absolute percentage error (MAPE) and relative bias (RB) were used to evaluate performance of applied method. Then the GMDH-HS method has been used for suction caisson uplift capacity prediction. The 62 data set of laboratory measurements were collected from published literature that 51 sets used to train new developed method and the remaining data set used for testing. Not only the results of suction caisson uplift capacity prediction using GMDH-HS were evaluated with statistical indices, but also the results were compared to some artificial methods by previously works. The results indicated that performance of GMDH-HS was found more efficient when compared to other applied method in predicting the suction caisson uplift capacity.

Introduction

Suction caissons are open bottom structures which can be used as anchor foundation for offshore structures such as highway bridges, harbors and wind turbines. These structures have a cylinder section tube, which is hollow or filling with sand. Suction caissons firstly mentioned by Senpere and Auverange (1982) as mooring anchor, because they reflected the multiple advantages than other alternatives. Suction caisson advantages are: 1) simple design 2) simple installation 3) fast installation 4) high resistance against vertical loading [1]. Uplift capacity in suction caissons in sand and clay is an important issue for their stabilities. If this parameter doesn’t estimate correctly, suction caisson may be collapsed [2]. The total uplift capacity of caisson depends upon passive suction under caisson-sealed cap, self- weight of caisson, frictional resistance along the soil-caisson interface, submerged weight of soil plug inside the caisson and uplift soil [3]. Different method including upper bound analysis [4], finite element method (FEM) [5], [6], [7], [8], [9], laboratory models [10], [11], [12], [13], [14], [15], [16], [17], [18], centrifuge model [19] and prototype model tests [20], [21] have been used to calculate axial and lateral load capacity of suction caisson for static and cyclic load under different soil conditions. The upper bound method and FEM along with the laboratory and centrifuge tests are the most popular methods in predicting the uplift capacity of suction [3]. Due to the complex behavior of suction caisson during the loading process, obtaining an accurate prediction is particularly difficult when the target system is governed by dynamic process under uncertainty conditions. To deal with this problem, relatively recent works apply artificial intelligent (AI) method to the problem of suction caisson uplift capacity prediction in order to generate more accurate and reliable models. Using convectional technique is not appropriate because of the poor understanding of the relationship among variable and the complex interactions between the processes. Hence there is a need for improvement in prediction techniques AI method and especially data driven model (DDM) offers flexible and non-parametric algorithms capable of modeling the relationship among inputs and output data set.

Some researchers used AI methods such as artificial neural network (ANN) [2], tree-based genetic programming(TGP) [22], linear genetic programming (LGP) [22], gene expression programming (GEP) [22], neuro-genetic [23], multi expression programming (MEP) [24], multivariate adaptive regression spline (MARS) [3], support vector machine and ANN (SVM- ANN) [25] and intelligent fuzzy radial basis function neural network inference method (IFRIM) [1] to predict the uplift capacity of suction caissons. The results of these studies indicated that AI techniques significantly outperformed FEM, and radial basis function neural network (RBFNN) in uplift capacity prediction.

Group method of data handling (GMDH) is one of DDMs that belong to self-organizing modeling approach. It was introduced by Ivankhenko in 1968. GMDH is similar to ANN but neither the number of neurons nor the number of layers in the network, nor the actual behavior of each created neuron is predefined [26]. This method is self-organizing, thus number of neurons, layers and the behavior of each neuron are adjusted during the process of training. Therefore complex system modeling in GMDH is more common than other artificial methods. In this method in training period, weights are calculated by least square estimation (LSE) method. Recently, some researches have been used GMDH for solving engineering some problems [27], [28], [29], [30]. Although several researchers applied ANN and GMDH in geotechnical issue [31], [32], [33], [34] but based on our knowledge it has not been used in suction caisson uplift capacity prediction.

Another AI method that has been used for optimization of engineering problems is harmony search (HS) algorithm. This method inspired by the improvisation process of musicians proposed by Geem(2011). In this algorithm, each musician (decision variable) plays (generates) a note (a value) for finding a best harmony (global optimum) all together. HS algorithm can be used for handling both continuous and discrete variables. This method has many successful applications in many aspect of optimization related problems such as optimal reservoir operation [35], design of water supply system [36], truss structure problems [37] and so on. HS presented a number of advantages over their optimization techniques. These advantages include less computational effort to find a solution derivative information is not needed, fast convergence rate, capability of significantly converge to the optimal solution.

In this research, based on combination of GMDH and HS algorithm, a new hybrid intelligent method called GMDH-HS is developed and used for suction caisson uplift capacity prediction. The code of new developed method is written in the MATLAB software. The developed method is validated using Mackey-Glass time series data and compared to two kinds of GMDH method, GMDH1 and GMDH 2. The results of developed model were evaluated using five statistical indices including coefficient of efficiency (CE), root mean square error (RMSE), mean square relative error (MSRE), mean absolute percentage error (MAPE) and relative bias (RB). Next, the GMDH-HS is used for the suction caisson uplift capacity prediction. Results of this method are evaluated with statistical indices. Also the results of GMDH-HS in uplift capacity prediction were compared to GMDH1, GMDH2 and the results of another works previously works [1], [2], [3], [22], [23], [24], [25], [38]. Finally, for better comparison between previously works and develop methods, a ranking system was applied.

Section snippets

Group method of data handling (GMDH)

In this section a brief definition of GMDH, problem solving by this method, formulating, and its structure and algorithm will be presented. GMDH is an inductive concept based on the perceptron theory, which has been developed for systems recognition, modeling, and prediction of sophisticated systems. GMDH is a combination of N-Adaline and compared to perceptron, its structure is more precise, since it has used data classification both usefully and uselessly, and needs less observational data.

Verification of GMDH-HS

The chaotic Mackey-Glass differential delay equation is recognized as a benchmark problem that has been used and reported by a number of researchers for comparing the learning and generalization ability of different models. The Mackey-Glass chaotic time series generated from the following equation:dx(t)dt=ax(tτ)1+x10(tτ)bx(t)where a, b and τ are Mackey-Glass constants and x(t) is the time-series output at time t. Mackey-Glass time series prediction is complicated due in addition of time

Results

In this section, at the first results of GMDH1 and GMDH2 methods in suction caisson uplift capacity prediction are presented. Next, the results of uplift capacity in suction caisson using the GMDH-HS method are explained. The performance of each method for both training and testing period is evaluated using statistical indices and some figures. Then the results of developed method compared to another works that have been used AI method in suction caisson uplift capacity prediction. Finally a

Conclusion

In this study, a hybrid intelligent method based on combination of group method of data handling (GMDH)” and “harmony search (HS)” which is called GMDH-HS has been developed for uplift capacity prediction in clay. In GMDH-HS, neurons and layers were built based on GMDH and weights in each neuron were calibrated using HS optimization method. At the first, developed methods were validated using Mackey- Glass times series and its statistical indices were compared with GMDH1 and GMDH2. Five

References (48)

  • R.E. Abdel-Aal et al.

    Modeling and forecasting the mean hourly wind speed time series using GMDH-based abductive networks

    Renew. Energy

    (2009)
  • R.E. Abdel-Aal et al.

    Modeling and forecasting the mean hourly wind speed time series using GMDH-based abductive networks

    Renew. Energy

    (2009)
  • M.G. Omran

    Ma 490 hdavi M. Global-best harmony search

    Appl. Math. Comput.

    (2008)
  • K.S. Lee et al.

    A new meta-heuristic algorithm for continuous engineering optimization: harmony search theory and practice

    Comput. Methods Appl. Mech. Eng.

    (2005)
  • C.M. Wang et al.

    Self-adaptive harmony search algorithm for optimization

    Expert Syst. Appl.

    (2010)
  • E.C. Clukey et al.

    The response of suction caissons in normally consolidated TLP loading conditions

    Offshore Technology Conference 1995 Jan 1. Offshore Technology Conference

    (1995)
  • A.J. Whittle et al.

    Formulation of MIT-E3 constitutive model for over consolidated clays

    J. Geotech. Eng.

    (1994 Jan)
  • S. El-Gharbawy et al.

    Modeling of suction caisson foundations

    The Tenth International Offshore and Polar Engineering Conference 2000 Jan 1. International Society of Offshore and Polar Engineers

    (2000)
  • L. Zdravković et al.

    A parametric study of the pull-out capacity of bucket foundations in soft clay

    Geotechnique

    (2001)
  • J. Ahn et al.

    Finite element analysis of the holding capacity of shallow suction caisson anchors

    Mar. Georesour. Geotechnol.

    (2015)
  • M. Datta et al.

    Suction beneath cylindrical anchors in soft clay

    The Sixth International Offshore and Polar Engineering Conference 1996 Jan 1. International Society of Offshore and Polar Engineers

    (1996)
  • W. Deng et al.

    A theoretical study of the vertical uplift capacity of suction caissons

    Int. J. Offshore Polar Eng.

    (2002)
  • S.N. Rao et al.

    Studies on pullout capacity of anchors in marine clays for mooring systems

    Appl. Ocean Res.

    (2006 Apr 30)
  • E.C. Clukey et al.

    A centrifuge and analytical study to evaluate suction caissons for TLP applications in the Gulf of Mexico

    Design and Performance of Deep Foundations@ Piles and Piers in Soil and Soft Rock

    (1993)
  • Cited by (40)

    • Application of artificial intelligence in geotechnical engineering: A state-of-the-art review

      2022, Earth-Science Reviews
      Citation Excerpt :

      Fig. 10 shows the number of papers published using different AI methods in shallow and deep foundation research. Intelligent prediction predominantly prevails in settlement and load capacity of pile foundations while few have emerged in bridge foundations (Akib et al., 2014; Yousefpour et al., 2014) and uplift capacity prediction of suction caisson (Shahr-Babak et al., 2016). ANN dominates the selection of AI methods overall.

    • Investigation into the influence of caisson installation process on its capacities in clay

      2020, Applied Ocean Research
      Citation Excerpt :

      The short-term bearing capacity here means the caisson foundation undrained capacity prior to soil consolidation. Many studies focusing on the bearing capacity of caisson in clay have been conducted, including field tests (Houlsby et al., 2005; Le et al., 2018), model tests (Guo et al., 2018; Zhu et al., 2018), limit analysis (Aubeny and Murff, 2005; Tang et al., 2016), finite element analysis (Gourvenec and Barnett, 2011; Fu et al., 2018), finite element limit analysis (Ukritchon and Keawsawasvong, 2016) and artificial intelligence methods (Masoumi Shahr-Babak et al., 2016; Kim et al., 2017; Derakhshani, 2017 and 2018). However, in most of the available numerical studies, caissons were assumed ‘wished-in-place’, where the process of installation was neglected (Aubeny and Murff, 2005; Monajemi and Razak, 2009; Gourvenec and Barnett, 2011; Hung and Kim, 2012; Masoumi Shahr-Babak et al., 2016; Mehravar et al., 2016; Tang et al., 2016; Ukritchon and Keawsawasvong, 2016; Kim et al., 2017; Derakhshani, 2017 and 2018; Fu et al., 2018).

    • Finite element modeling of the tensile capacity of suction caissons in cohesionless soil

      2019, Applied Ocean Research
      Citation Excerpt :

      The transient load effect, in which suction is generated for monotonic loads of a short duration, is also confirmed numerically by Thieken et al. [22] and Sørensen et al. [21]. Other investigations are presented by Masoumi Shahr-Babak et al. [11] that used GMDH-HS to analyse laboratory test results, whereas e.g. Cassidy et al. [5] use the laboratory data to fit a strain-hardening plasticity model that describes the caisson behavior. Wang and Cheng [26], Mehravar et al. [12], Ukritchon et al. [23] used numerical modeling to investigate the undrained pull-out capacity for suction caissons in clay.

    View all citing articles on Scopus
    View full text