Modelling the Collapse of Metastable Loess Soils

H. Miller, Y. Djerbib, I.F. Jefferson and I.J. Smalley
Nottingham Trent University, England


Hydrocollapse of loess continues to cause major geotechnical problems the world over. Modelling the behaviour of these soils is still in its early stages. Computer models are often used to model saturated soil behaviour but for collapsing unsaturated loess soils such models are difficult to implement due to the complex nature of collapse. One method that will help overcome difficulties such as the role of clay bonding or soil fabric, is to consider collapse from a particle packing perspective, rather than as a soil suction problem. The way the particles pack together decides whether a soil is metastable or not. Using this philosophy, work is currently being conducted to develop a constitutive model to be incorporated in CRISP90 in order to analyse the behaviour of collapsing loess soils. This paper will discuss the development of this model and the methods used to validate it, which include the use of an artificial loess soil manufactured in the laboratory. This model material has been shown to reproduce loess behaviour very well and enables a full range of reproducible and repeatable tests to be conducted.

1. Introduction

Large areas of the earth's crust are covered with loess. China, America, Eastern Europe and Russia all have deposits ranging in depths from 1m to 100's of metres. For example, around 14 % of the total territory of (the old USSR) is covered by loess where the deposits are around 20m and deeper (Abelev, 1988). Consequently, many residential buildings in cities and towns and big industrial enterprises have been erected on loess around the world.

Loess can cause a number of problems associated with its sudden settlement. An example is a three storey building in Xining, Qinghai, destroyed beyond repair due to subsidence of the foundation soils upon wetting (Qian et al., 1988). Problems result because loess undergoes structural collapse when wetted. This happens when the initial dry density is low and initial water content is low (Dijkastra et al. 1995). There is still some argument as to why loess collapses. To elucidate this problem it is necessary to examine both macroscopic and microscopic aspects of loess collapse. (Feda, 1994).

Increasingly, numerical models such as finite element (FE) models are used to examine the behaviour of foundations built on soil. However, loess collapse is still causes a problem as it can not be modelled using conventional techniques. This paper will describe the combined computer and physical model approach currently being used to explore this complex problem.

2. The collapse process

Before any models are developed it is first necessary to establish the typical characteristics of loess. From this a universal standard can be produced, which can not only aid the validation of any FE models produced enable the growing body of literature to be better utilised.

2.1 Mineralogy

The structure of loess is dominated by 20-60m quartz particles of 8:5:2 aspect ratio Rogers et al. (1994). Typically, quartz is the most abundant mineral in loess material, feldspar is also present. Of the clay type minerals, mica is more abundant than montmorillinite (average 15%), which is more abundant than illite and kaolinite (average 5%). (Northmore et al., 1996).

2.2 The collapse mechanism

Upon deposition, a loose, open structured, metastable soil is formed, composed of quartz particles separated by coatings or aggregates of clay and carbonate particles. In its dry state the structure has significant strength and can withstand high loads.

Upon saturation, however, the bonding disintegrates and a denser structure is achieved by sudden collapse of the soil particles - often known as hydrocollapse. Saturation can occur through infiltration due to pooling of water from above, leakage from pipes and guttering or through rising ground water levels. The collapse of the internal structure occurs when the stresses between particles exceed the bond strength provided by bridging bonds (Holtz and Gibbs, 1951). This kind of soil is considered to be unstable as a foundation material because of the potential for large settlement. In these cases, if differential settlement occurs it can be severely damaging for structures built on loess.

2.3 Collapsibility controls

In certain cases destruction of capillary forces may account for the collapse caused on wetting. (Alonso et al., 1987). However, there is still considerable discussion as to the relative roles of suction and clay bond degradation in the collapse process of loess. Added to this, there is a considerable difficulty in obtaining reliable suction measurements in silty soils (Fredlund & Rahardjo, 1993). The main concern of the authors is that suctions may take a long time to equilibrate and even when they do the suctions developed in the clay bonds may be different to the suction developed between the silt grains. To date there is no way of measuring the differences. The suctions measured would therefore give an average value and may not be reliable. Bond degradation probably governs the collapse processes in loess and therefore experiments will be focused on finding a rule that explains the behaviour in terms of the packing parameters such as clay content and void ratio.

The basic mechanism of collapse involves particle rearrangement from metastable open structure to a denser stable hydrocollapsed structure. Whatever the mechanism includes, ultimately it is the way the particles are packed that controls the structure and hence the susceptibility of loess to collapse. This idea was actively promoted by Smalley in the 1960's (Smalley, 1964) and subsequently developed by Rogers et al. (1994). Since these crude hand generated solutions, computer models have been developed using FORTRAN and similar languages (see Dibben et al., in press) Thus by considering collapse from a particle packing perspective it is possible to avoid problem associated with relative roles of clay degradation or suction in the collapse process. Such an approach could provide useful insight at both macroscopic and microscopic levels, particularly when developing computer models.

3. Artificial Loess

A considerable body of research has been generated to investigate collapse problems. Unfortunately, the results are often variable due to location and depth of the deposit and therefore are difficult to compare directly. An artificial deposit has been produced that has the same general properties of collapse as loess deposits from around the world. The artificial loess provides results which are repeatable, reproducible and controllable.

Artificial loess samples have been produced by adapting work undertaken by Assallay et al., 1998 and Dibben et al. 1998. This artificial loess allows direct control of the material constituents, thus providing a good method of evaluating the effect of varying the constituents, and hence provides a control method of computer validation. Two methods to generate artificial loess are currently being developed.

3.1 Ballotini balls

An artificial loess has been made by mixing glass ballotini balls with kaolinite. Ballotini balls are uniform spherical glass objects. They are sieved through a 63 (m sieve to create typical loess size particles the smaller balls are removed through a 20(m sieve. Initial studies by Assallay et al., 1998 illustrate the loess like structures formed with these balls. These structures illustrated that kaolinite was the best clay mineral binder to use, producing a greater control on collapse.

3.2 Crushed sand

Another method of producing artificial loess is to use crushed sand. Impurities are removed from the sand by washing with water. The sand is then crushed in an end runner mill which reduces the sand size to silt size particles. The ground material is passed through a 63(m sieve to retrieve the smaller fraction. It is deflocculated and the fines removed by sedimentation. The resultant material (63-20(m) is then mixed with kaolinite to produce the bonding needed to create a metastable structure. The effect of varying the clay content has been examined the results of which are shown in Figures 1 and 2.

4. Hydrocollapse Testing

Artificial and remoulded natural loess soils are prepared using methods recently established (see Dibben et al, 1998). Once formed, standard double and single oedometer collapse tests have been performed (between stresses of 5 and 1600kPa).

Single oedometer tests are carried out by loading the sample to a certain pressure (between 5 and 1600 kPa), saturating the sample and then continuing the loading up to 1600kPa. Double oedometer tests are carried out on two samples in parallel one a saturated sample and the other an unsaturated sample these are also loaded to 1600kPa.

Two main curves are observed one for unsaturated behaviour and one for saturated behaviour. (See Figure 1.) The same characteristic collapse behaviour is observed in the artificial loess as it is in the natural deposit. See also, Assalay et al.,1998 and Dibben et al., 1998)

Figure 1: Oedometer tests on artificial loess from crushed sand : kaolinite 85:15. The collapse pressure due to saturation is 200kPa.

Ballotini balls have the advantage that they are rounded, and hence are more representational of natural loess deposits that contain weathered quartz particles as the main building blocks. However they do not have the same specific gravity as the silt particles and therefore exhibit different void ratios for the same mass of soil.

Figure 2 shows how hydrocollapse varies with clay content at 200kPa for the artificial loess samples (crushed sand with kaolinite). Clearly, only a relatively small amount of clay content is required to yield significant collapse. Too much clay and the soil behaves in a typically plastic 'clay like' fashion exhibiting no collapse. Too little clay content and insufficient binder exists to produce the metastable structure required for collapse to occur. The maximum collapse occurs at 25% clay content which corresponds to approximately 18% clay mineral (kaolinite) content. However, it should be noted that the samples were produced all to have the same mass and therefore, different void ratios result. To make the experiments more realistic the samples will be produced with the same compactive effort which will emulate deposition of loess soils and the effects of overburden pressure. This will enable a true comparison of the effect of the clay content to be assessed. A similar set of results is observed when Ballotini balls are used as the primary mineral constituent. However the magnitude of collapse observed with these samples is lower, due mainly to the spherical shape of the primary particles.

Figure 2: % hydrocollapse for different kaolinite contents.

5. Computer modelling

Classical soil mechanics has been developed predominantly from the study of saturated soils. More recent developments in fundamental frameworks of soil behaviour have also been limited to saturated soils. However, loess is an unsaturated soil and new frameworks need to be developed for such soils.

Most approaches use the unsaturated formulation of critical state theory which relies on measurement of suction changes throughout the experiment. Relatively little experimental data on the behaviour of unsaturated soils are available, due to technical difficulties related to both control and measurement of suction (Cui and Delage, 1996). Laboratory equipment for controlling and measurement of suctions has been developed at Imperial College but measurement of suction in the field is still unreliable. If a procedure could be developed based on a more empirical approach without the need for the measurement of suctions, designing for unsaturated soils would be made easier.

A finite element computer package, CRISP90 (CRItical State Program), based on the critical state theory of soils will be used to investigate the collapse behaviour of loess subjected to load and/or wetting. The program was written with a provision to incorporate new soil models to suit the problem under study.

Loess displays two different kinds of behaviour. The first is a partially saturated soil. However as the degree of saturation is increased and collapse takes place, the behaviour is typically that of a saturated clayey silt. In its first state, prior to collapse, it is analogous to cemented soil with the bonding between the grains of silt provided by the clay/carbonate elements. Comparison can therefore, be made with the behaviour of weak concrete or rocks.

Nesnas (1995) developed an elasto-plastic constitutive model to predict the behaviour of partially saturated soils including collapse. This model was based on the Barcelona model (a suction based model) Alonso et al., 1990. This model incorporated in the finite element code of CRISP can be modified further and applied to study the behaviour of loess.

Data will be built up from both oedometer tests and a physical model. These models will provide data for calibrating and validating the computer model. The oedometer tests will use both natural and artificial deposits. The natural deposits have been obtained from sites in Southern England, where collapse has resulted in costly engineering problems, e.g. Torquay (Cattell 1997). The artificial loess will also be used to investigate the effect of material variations. Specifically this will allow examination of bonding mechanisms that occur in natural loess systems.

An initial step in implementing the new model will be to introduce threshold stresses at which the deposit will collapse. The lessons learned in the laboratory will be used to calibrate the numerical model. This will be coupled with a code based on a particle packing perspective following on the work of Dibben et al. (1998).

This will overcome the difficulties inherent with critical state models for 3-phase systems which rely on suction measurements. Such measurements are notoriously difficult in silty soils. Ultimately, it is planned to use a scale laboratory model foundation to examine the effects of various water infiltration systems, which will be compared directly to FE analysis. This will be used to validate such models, and hence allow further elucidation of the complex problem of loess collapse.

Application for such a model may include assessment of radioactive waste repositories built in loess, which is currently under consideration in Bulgaria (Jefferson and Smalley, 1997). The FE model developed could help to predict the overall behaviour of foundations and other infrastructure built on or in loess and similar soils.

6. Conclusions

This project is currently developing a FE model to assess the behaviour of collapsible loess soils. However, to overcome the arguments of which factor controls the collapse mechanism, collapse is being modelled on a particle packing basis for it is ultimately the way the particles pack that dictate the degree of collapse that occurs. This model will be calibrated using an artificial loess developed to enable full control over the material constituents and hence the collapse that occurs. The artificial loess will be used in the further development of the model, aiming ultimately to enable the full assessment of foundations built on loess and other collapsible soils.


Abelov, M.Y., (1988), Loess and its Engineering Problems in the USSR, Proc. of the Int. Conf. Engineering Problems of Regional Soils., 11-15 August, Beijing, China, 3-6. Beijing: Chinese Institute of Soil Mech. and Found. Eng.

Alonso, E.E., A. Gens, & D.W. Hight, 1987, General report. Proc. 9th Conf. Soil Mech., Dublin 3, 1087-1146.

Alonso, E.E., A. Gens, & A. Josa, (1990), A constitutive model for partially saturated soils, Geotechnique, 40: 405-430.

Assallay, A.M., I. Jeffeson, C.D.F. Rogers, & I.J. Smalley, (1998), Fragipan Formation in Loess Soils: Development of the Bryant Hydroconsolidation Model, Geoderma. 83, 1-16.

Cattell, A.C., (1997), The development of loess-bearing soil profiles Permian Breccias in Torbay. Proceedings of the Usher Society, Vol 9, p168-172.

Cui, Y.J. & P. Delage, (1996), Yielding and plastic behaviour of an unsaturated compacted silt, Geotechnique, 46: 291-311.

Dibben, S.C., I.F. Jefferson, & I.J. Smalley, (1998), The "Loughborough Loess" Monte Carlo Model of Soil Structure. Computers and Geosciences. (in press).

Dijkastra, T.A., I.J. Smalley, & C.D.F. Rogers, (1995). Particle packing in loess deposits and the problem of structure collapse and hydroconsolidation. Engineering Geology, 40: 49-64.

Feda, J. (1994), Mechanisms of Collapse of Soil Structure, In: E. Derbyshire (ed.), Genisis and properties of collapsible soils,NATO SeriesC: Mathematical and Physisical Sciences Vol 468. 149-172. Dordrecht: Kluwer.

Fredlund, D.G., & H. Rahardjo, (1993), Soil Mechanics for Unsaturated soils. New York. John Wiley and sons, inc.

Holtz, W.G. & H.J. Gibbs, (1951), Consolidation and related prperties of loessial soils, ASTM STP 126: 9-26.

Jefferson, I. & I.J. Smalley, (1997), Nuclear Waste Poser, Ground Engineering. 30(10): 26-27

Nesnas, K, (1993), A finite Element Implementation of a Critical State Model for unsaturated soil to simulate Drained Conditions. PhD University of Sheffield.

Northmore, K.J., F.G. Bell, & M.G. Culshaw, (1996), The engineering properties and behaviour of the Brick Earth of South Essex, Quarterly Journal of Engineering Geology, 29: 147-161.

Qian, H.J. & Z.G. Lin, (1988), Loess and its Engineering Problems in China. Proc. of the Int. Conf. Engineering problems of Regional Soils. 11-15 August, Beijing, China, 136-153. Beijing: Chinese Institution of Soil Mech. and Found. Eng.

Rogers, C.D.F., T.A. Dijkstra, & I.J. Smalley, (1994), Hydroconsolidation and subsidence of loess: Studies from China, Russia, North America and Europe., Engineering Geology, 37, 83-113. Schwartz, K., (1985), Collapsible Soils, The Civil Engineer in South Africa, July, 379 - 393.

Smalley, I.J., (1964), Representation of packing in a clastic sediment. Amer. J. Science. 262 pp. 242-248.