Reliability analysis of settlement of pile group
- PDF / 2,671,005 Bytes
- 17 Pages / 595.276 x 790.866 pts Page_size
- 30 Downloads / 365 Views
TECHNICAL NOTE
Reliability analysis of settlement of pile group Manish Kumar1 · Pijush Samui1 · Deepak Kumar2 · Wengang Zhang3 Received: 16 June 2020 / Accepted: 23 September 2020 © Springer Nature Switzerland AG 2020
Abstract Considering the highly variable nature of soil, reliability analysis of pile foundation is being explored in the modern scientific era. The paper investigates the application of relevance vector machines (RVM), generalized regression neural network (GRNN), genetic programming (GP) and adaptive-network-based fuzzy inference (ANFIS) in reliability analysis of settlement of pile group. The simulation is checked using Monte Carlo simulation (M-C). The performance of models is ascertained using various performance parameters and Taylor diagrams. The normality and homogeneity in performance of the models is tested by carrying out Anderson–Darling (AD) test and Mann–Whitney U (M–W) test, respectively. The paper concludes that the performance of RVM, GP and ANFIS were excellent while that of GRNN was poor. Keywords Pile foundation · Reliability · FOSM · GP · GRNN · ANFIS · RVM
Introduction Safety is a key concern in any project. Importance of addressing safety and risk factors was addressed long back [1]. Observing the growing scarcity of space and demand of skyscrapers has shifted the focus of scientists toward advancements in pile foundation. The soil by very nature is variable in terms of material parameters (inherent variability), sampling, testing and equipment errors (measurement variability) and variabilities related to premises involved in the model (transformation uncertainties) [2, 3]. We need to ascertain that ‘how much safe is safe enough’ or ‘how much risk is allowable’; so that both safety and economy aspects are addressed.
* Manish Kumar [email protected] Pijush Samui [email protected] Deepak Kumar [email protected] Wengang Zhang [email protected] 1
Department of Civil Engineering, NIT Patna, Bihar, India
2
NIT Patna, Bihar, India
3
School of Civil Engineering, Chongqing University, Chongqing, China
Traditionally, Factor of safety approach has been used; however, it results in a conservative analysis and hence economically unsound. The probability of failure too remains high [4]. Reliability analysis has been successfully deployed over the years to overcome these shortcomings. First-Order Second Moment Method (FOSM) is the most effectively used First-Order Reliability Method (FORM) [5, 6]. However, FOSM has disadvantage of being invariance problem i.e., with change in the specific form of the limit state function, the value of the reliability index varies. Additionally, results tend to be time consuming as well as inaccurate if partial derivatives of distribution functions with respect to the random basic variables don’t follow normal distribution. Truncation errors too tend to be significant. Despite many successful implementations, ANN models [7–9] are found to have shortcomings such as slow convergence speed, overtraining, getting trapped in local minima
Data Loading...
