A basal reinforced piled embankment consists of a reinforced embankment on a pile foundation. The reinforcement consists of one or more horizontal layers of geosynthetic reinforcement (GR) installed at the base of the embankment. The design of the GR is the subject of this thesis
...

A basal reinforced piled embankment consists of a reinforced embankment on a pile foundation. The reinforcement consists of one or more horizontal layers of geosynthetic reinforcement (GR) installed at the base of the embankment. The design of the GR is the subject of this thesis. A basal reinforced piled embankment can be used for the construction of a road or a railway when a traditional construction method would require too much construction time, affect vulnerable objects nearby or give too much residual settlement, making frequent maintenance necessary. The GR strain needs to be calculated to design the GR. Multiplying this GR strain by the GR stiffness gives the tensile force, which needs to be smaller than the long-term GR tensile strength. The GR strain is calculated in two steps. Calculation step 1 divides the load – the weight of the embankment fill, road construction and traffic load – into two load parts. One part (load part A) is transferred to the piles directly. This part is relatively large because a load tends to be transferred to the stiffer parts of a construction. This mechanism is known as ‘arching’. The second, residual load part (B+C) rests on the GR (B) and the underlying subsoil (C). Calculation step 2 determines the GR strain on the basis of the result of step 1. Only the GR strips between each pair of adjacent piles are considered: they are loaded by B+C and may or may not be supported by the subsoil. The GR strain can be calculated if the distribution of load part B+C on the GR strip, the amount of subsoil support and the GR stiffness are known. An implicit result of this calculation step is the further division of load part B+C into parts B and C. Several methods for the GR design are available, all with their own models for calculation steps 1 and 2. The methods give results that differ immensely. The Dutch CUR226 (2010) and the German EBGEO (2010) adopted Zaeske’s method (2001). However, measurements that were published later (Van Duijnen et al., 2010; Van Eekelen et al., 2015a) showed that this method could be calculating much higher GR strains than those measured in practice, leading to heavier and more expensive designs than necessary. The objective of the present study was to establish a clearer picture of load distribution in a basal reinforced piled embankment and, on that basis, to develop and validate an analytical design model for the geosynthetic reinforcement in a piled embankment. The results were described in five papers published in the international scientific journal ‘Geotextiles and Geomembranes’. Those journal papers can be found in Chapters 2, 3, 4, 5 and Appendix A of this thesis (Van Eekelen et al., 2012a, 2012b, 2013, 2015a and 2011 respectively). Chapter 2 presents a series of twelve 3D experiments that were carried out at the Deltares laboratory. The scaled model tests were carried out under high surcharge loads to achieve stress situations comparable with those in practice. A unique feature of these tests was that load parts A, B and C could be measured separately, making it possible to compare the measurements with calculation steps 1 and 2 separately. In these tests (static load, laboratory scale), smooth relationships were obtained between the net load on the fill (surcharge load minus subsoil support) and several measured parameters such as load distribution and deformation. Consolidation of the subsoil resulted in an increase in arching (more A) and more tensile force in the GR (more B and more GR strain). The measured response to consolidation depends on the fill’s friction angle. A higher friction angle results in more arching during consolidation. One of the major conclusions based on the test series was that the load on a GR strip is approximately distributed as an inverse triangle, with the lowest pressure in the centre and higher pressure close to the piles. This conclusion was the basis for the remainder of this doctorate study and the development of the new calculation model. Chapter 3 considers calculation step 2. This chapter starts by comparing the measurements in the experiments with the calculation results of step 2 of the Zaeske (2001) model, which uses a triangular load distribution on the GR strip and considers the support of the subsoil underneath the GR strip only. It was found that Zaeske’s model calculates GR strains that are larger than the measured GR strains (approximately a factor of two for GR strains larger than 1%). Chapter 3 continues with the suggestion of two modifications to Zaeske’s step 2. Firstly, the load distribution is changed from a triangular to an inverse triangular load distribution. Secondly, the subsoil support is extended from the support by the subsoil underneath the GR strip to the subsoil underneath the entire GR between the piles. The new step 2 model with these modifications produces a much better fit with field measurements than Zaeske’s model. Chapter 4 considers calculation step 1, the arching. Additional tests were conducted for this purpose, varying factors such as the fill height. This chapter gives an overview of the existing arching models and introduces a new model. This Concentric Arches model (CA model) is an adaptation and extension of the models of Hewlett and Randolph (1988), and Zaeske (2001), which have been adopted in several European design guidelines. Some countries use piled embankments without GR. Introducing GR changes the load distribution considerably. A major part of the load is then exerted on the piles and the residual load is mainly exerted on the GR strips between the piles, with the load being distributed approximately as an inverse triangle. Chapter 4 explains the development of the load distribution as a result of continuing GR deflection; new small arches grow within the older larger ones. Smaller arches exert less load on their subsurface. This idea is related to the concentric arches of the new model, which gives an almost perfect description of the observed load distribution in the limit state situation. Furthermore, the new model describes the influence of the fill strength and embankment height correctly. Chapter 5 compares the existing, and the newly introduced, design models with measurements from seven full-scale projects and four series of scaled model experiments. Two of these seven field projects were conducted in the Netherlands and they were carried out in part for this doctorate research. One of the four experimental series – the one presented in Chapters 2 and 4 – was conducted specifically for the present research. The other measurements were reported earlier in the literature. The calculations were carried out using mean, best-guess values for the material properties. The calculation results from the CA model match the measurements much better than the results of the arching models of Hewlett and Randolph (1988), and of Zaeske (2001). The results of the CA model are also the closest match with the results of the 3D numerical calculations, as described in Van der Peet and Van Eekelen (2014). These authors also show that the new CA model responds better to changes in the fill friction angle than any of the other models considered. When there is no subsoil support, or almost no subsoil support, the inverse triangular load distribution on the GR strips between adjacent piles gives the best match with the measurements. When there is significant subsoil support, the load distribution is approximately uniform. This difference between the situation with or without subsoil support is understandable when one considers that most load is attracted to the construction parts that move least. In the cases with limited subsoil support, the load distribution that gives the minimum GR strain should be used to find the best match with the measurements. The GR strain calculated with Zaeske’s model is on average 2.46 times the measured GR strain. The GR strain calculated with the new model is on average 1.06 times the measured GR strain. The calculated GR strain is therefore almost a perfect match with the measured GR strain. The new Dutch CUR226 (2015) has therefore adopted the model proposed in this thesis.