In the first part of this presentation, John Coggan of Camborne School of Mines discussed the findings from two sites in Cornwall - Delabole Quarry and a china clay pit. Doug Stead, also of Camborne School of Mines, then concentrated on the Frank Slide in Alberta, Canada and included examples of landslides in Europe.
The theme of the discussions was ultimately how modelling could assist engineers in understanding failure mechanisms and primarily involved back analysis of existing failures rather than prediction. Both presentations discussed the understanding of geological controls in the landslide, identification of failure mechanisms, monitoring of displacements, limit equilibrium analyses and numerical modelling. Stead then discussed in detail the issues concerning the development of complex numerical models including the choice of input parameters and the implications of scale effects.
Delabole slate quarry, Cornwall
Delabole slate quarry has been the subject of detailed investigation since a failure of the west face in 1967. Coggan described the geology of the failure and presented an update of deformation monitoring since the event.
There are a number of discontinuities throughout the rock mass including slatey cleavage and clay lodes.The failure was progressive and the geometry of the failure surface allowed back-analysis both for limit equilibrium and numerical methods.Modelling the rock mass in UDEC was carried out to ascertain the different mechanisms operating in the failure process.These included:
a chisel effect at the top of the failure mass block rotation as the chisel block forced the underlying blocks away from the face translation of the toe block flexural toppling within the mass Tension crack development was monitored before the failure, enabling the displacement trend to be followed. After the failure, additional monitoring was established immediately behind the crest.Movements continued at a rate of between 50mm and 100mm per decade in tension cracks that are, in some instances, now 2m to 3m wide.While the general trend is progressive widening, the rate ofmovement is seasonal and appears to correlate with rainfall.
As well as the movements behind the crest, monitoring stations were established on the quarry face. These provided information that allowed comparison with the UDEC model of the existing slope. The displacements from the model compared favourably with those from the face monitoring. The displacements throughout the body of the rock mass indicated that the mechanism was one of deep-seated flexural toppling with displacement consistently greater on the top of each block compared with the base.
China clay pit, UK
This failure was a structurally controlled flow slide which occurred in 1998. The height of the face was 50m (failure height was about two-thirds of slope height) and the runout of the failed mass was 155m from the crest. A digital elevation plan of the china clay pit and a laser profile of the slope failure demonstrated the benefits of these techniques in appreciating the extent and shape of the surface being studied.
The geology is a mixed weak rock with significant variation in alteration grade. Rock mass structure such as jointing was present and it was this that controlled the extent and geometry of the failure.Failure was preceded by high rainfall and water seepage was observed in the lower section of the face before failure. The potential for slope instability had been identified prior to the failure and a piezometer had been installed behind the crest of the slope. Manual readings taken at monthly intervals prior to the failure had shown a seasonal reduction in water level. Back analysis of the failure was carried out using deterministic analysis and numerical modelling. A Mohr-Coulomb FLAC model was used in the analysis, to study strains, in particular shear strain increment.
Additional FLAC modelling using the Mohr-Coulomb ubiquitous constitutive criterion modelled the main discontinuities and relic features. The modelling was considered to be successful as the pattern of shear strain increment indicated by the FLAC model essentially followed the failure surface.
The Frank slide, Canada
The Frank slide is on Turtle Mountain in Alberta, Canada. It occurred on 29 April 1903, when a 760m high failure of 30M. m 3of material slid down the mountain and across the Crowsnest River for 3.5km, killing 70 people. Frank, a coal mining town, was at the base of the mountain. It had been established in 1901 to extract coal from the sedimentary sequence close to the toe and lower slopes of the mountain. Coal production was 15,000t in the first year, 160,000t in the second year and 100,000t in the first four months of the third year, immediately before failure.'Bumps'were heard in the mine during extraction. Study of the failure analysed the extent to which the coal mining activities may have contributed. Monitoring of the slope since the failure showed essentially negligible displacement.
Stead described the geological conditions believed to exist at the time of failure. These comprised a regularly bedded limestone sequence dipping into the slope of the mountain which had been thrust over a shale sequence including the coal beds, which formed the lower slopes of the mountain. The failure was assumed to be on the cross joints, raising concern as similar geological conditions existed elsewhere in the area. This model was eventually shown to be incorrect and a subsequent model indicated that the mountain was an anticlinal structure in the limestone, with the thrusted blocks again forming the lower slopes.
Analysis indicated that the slope was at limit equilibrium.FLAC modelling of the failure started with the simple bedded sequence (no discontinuities) using initially a Mohr-Coulomb failure criterion and then the Mohr-Coulomb ubiquitous model. The modelling was developed to investigate the impact of the coal extraction by increasing the height of the mine opening, in increments of 10m, until significant movement or failure occurred, at about 50m.UDEC modelling, which included the anticlinal structure and orthogonal cross joints, identified the initiation of tensile failure at the crest of the slope while modelling the coal extraction.
Canadian surface coal mining
Surface coal mining and sedimentary bedding parallel to slope surfaces in high mountain slopes are common in Western Canada. Failure and slope distress is indicated by buckling and toeing out on low angled thrust planes.Modelling shows that the impact of changes in the slope curvature and the effect of rolling of the beds is important and can affect the stability of the slope.The aspects that are considered important are :
slope geometry joint orientation insitu stress; sensitivity of K (s h/s v); as K increases, failure occurs water - high water pressures produce failure.
Issues in numerical modelling
Conventional limit-equilibrium approaches must often be accompanied by numerical methods because:
Slope deformations are often complex and cannot be readily accommodated in simple models.
Time effects (creep).
An example of a failing mass sliding on a basal seat-earth showed a failing mass separating into a number of major blocks, producing graben type structures.This represented the type of complex behaviour and extreme internal deformation that can occur.
Stead listed the modelling input requirements:
slope geometry (2D or 3D? ) geology (rock/soil/mixed) discontinuities (orientation, spacing/persistence), block size could change the mode of failure from toppling to planar failure material properties (intact/discontinuity/mass scale effects); rock mass classification must be capable of being related to the modelling constitutive criteria insitu stress; rarely measured but may be important groundwater pressure/seismicity; not always known Followed by an appreciation of possible complex slope deformation due to:
weak/highly weathered materials deep-seated failures: toppling, active/passive (bilinear) and creep (time dependent/crack propagation; some slopes may exhibit progressive failure development) buckling/ploughing rigid block overlying soft strata (bearing capacity failure) flows (rheological models) Stead discussed examples of failures that exhibit some of these points.
Vaiont slide Differences in interpretation of the geometry of the failing mass were discussed. Two thousand people died when a landslide behind the Vaiont arch dam in Italy caused a massive wave of water to flow over the top of the structure, engulfing the town of Langaronne below. One investigator believed that before the event the failing slab was only 15m thick, while another estimated that a volume of 200M. m 3was involved. Issues of the risk to people therefore need to be considered. Is the failure a brittle or ductile one? Specifically, what is the speed and extent of movement? This has implications in the time taken for energy release.
Spis Castle, Slovakia This is an example of a slow failure with competent blocks overlying a weak layer. UDEC analysis indicates the likely movement of the blocks over the weak layer.
Randa rock slides, Switzerland The Randa failures are an example of rapid failure. Three separate failures occurred over three weeks, the first with a volume of 20M. m 3.It occurred over a period of seven hours and included several individual events. The geology was gneiss under high stress and water pressure.A post-failure diversion tunnel excavated through the landslide indicated evidence of high stress in the abutments and low stress in the landslide area.
Volume of landslide Finally, Stead presented a plot of run-out geometry against volume of landslide which presented a range of data from Hong Kong (including retaining walls and fill slopes), coal mine waste rock, chalk, kaolinised granite and rock avalanches.The plot indicated that as the volume increased, the distance the flow extended increased.
The futureThe possible future of landslide modelling was summarised as:
improved modelling of failure mechanisms eg particle flow codes with the capability to simulate the propagation of fractures integrated modelling - risk assessment numerical modelling 2D to 3D: increased data demands and improved modelling constraints application of coupled FEM/DEM codes use of parallel computers in slope analysis, to include more detail and run much larger jobs Discussion
The chairman of the meeting, Ruth Allington, asked Stead for an insight into parallel computing and, on the issue of the dip of the coal seam - which appeared to be very steep - how this affected the failure of the material above. Stead said that parallel computing allowed the connection of a large number of computers or processors (PCs) to run a single model, increasing the computing power avai lable.
In answer to the dip of the coal seam, the coal seams and pillars did not fail.The model excavated the coal in 10m slots and it was not until 50m had been extracted that there was an impact. The assumption was that distress to the slope was the result of stress-induced fracture and displacements above the coal mine up to the toe of the failure. This exacerbated the situation but that was not to say that failure would not have occurred if there was no mining. Also, the mining caused deformation in the rock mass in the caving zone and had the effect of reducing the shear strength on the joints above.
Bob Pine of Golder Associates enquired about the insitu stresses near the coal workings in the Frank slide, and asked if Stead could elaborate on why he considered the run-out of the landslide material was so high.Stead said that in the modelling, the K value was varied from 0.5 to 1.5 to model the stress build up, but there was no evidence of failure of the coal seam. There was deformation but nothing like total closure.However, there was enough movement to allow shearing and reduction in shear strength. In the modelling, up to 200mm deformation was recorded but not overall failure or stope closure. Mining had nothing to do with the failure. The bedding planes were the pre-existing weaknesses and the main failure would have occurred without the mining, but displacements induced by the mining may have precipitated the failure. Numerous causes have been suggested for the extended run-out including:
air lubrication within the mass (generally discounted) acoustic fluidisation with fine materials at base and coarse at top the river at toe of slope dispersive phenomena Stead said that modelling using FRACMAN may provide some answers in predicting in particle size reduction from the top of the landslide to intermediate to the run-out.Perhaps the detail of the trigger zone needed to be investigated more thoroughly. He said possibly there was build-up of stress that contributed to the effect.
Michael de Freitas of Imperial College related his experiences from the geology of North Devon.
He explained that geology seemed to have gained the upper hand.Once failure occurred, a new set of joints, essentially within the same sets, developed.He said that it would appear that the rock mass was shedding stress. For example, toppling split the rock into two and then more joints appeared, the rock mass self-destructing, and the process was repeated.
Stead suggested that micro-fractures may have already existed, and that was possibly why new joints appeared to develop.Conventional modelling could not cater for this phenomenon. In UDEC, tensile failure may be indicated but it does not allow the creation of new fractures.Other numerical models have been developed to look at this type of problem including coupled finite element/distinct element codes. At Delabole, the tension cracks suggested that movements were complex but there was no definitive answer from the models. Flexural toppling is occurring but is not accelerating so continued monitoring is important to understand the rock mass.The deformations imply damage to the rock mass and internal deformation should be the future study for this slope.
Bob Pine of Golder Associates asked whether it would ever be possible to predict with parallel computers. He wanted to know if the data limitations were too severe and if engineers had to rely on limit equilibrium. Stead said that data collection must improve and that frequently there was not enough monitoring data for 2D, let alone 3D modelling.
John Harrison of Imperial College described his experiences with modelling. He had set up a number of models for UDEC and as a sensitivity tool suggested that it gave reasonable results.
However, the input parameters (normal and shear stiffness) had to be decided upon and site investigations and laboratory data did not provide this information.
Stead said industry must change the level and type of testing that is done for such studies but this was not the complete answer as scale effects had to be considered. The engineer or geologist must have awareness of input parameters and their sensitivity. Modelling had a part to play in complex situations but a lot more work needed to be done, he added. Stead reminded the audience that Evert Hoek had compared modelling to a toolbox with the different methods representing the various tools; the key being to select the correct tool. Hoek also stressed that modelling was not a substitute for good engineering judgement.
Ivan Hodgson of Scott Wilson and Kirkpatrick asked how many of the input parameters for FLAC and UDEC models were selected from the field and how many were empirical.Stead said that both were useful and reminded the audience that block size was very important and could be as important as the shear strength.
Andrew Schofield of the University of Cambridge explained that there was an error in the assumptions in Mohr-Coulomb - there was no such thing as cohesion or adhesion, he said.Changes in the density of the material being sheared occurred and once the volume change had taken place, the material disintegrated. Stead acknowledged Schofield's comments and stated that current experimental work, undertaken with Erik Eberhardt of ETH Zurich, on brittle rock failure in granite, sandstone and evaporites used a damage-based constitutive criteria. It was suggested that such approaches would be more common in the future and could simulate the progressive stress induced deterioration of intact, rock discontinuity and mass strength.
Cruden DM and Krahn J (1978). Frank rockslide, Alberta, Canada. In Voight (ed) Rockslides and Avalanches, pp97-112. Elsevier.
Findlay PJ, Mostyn GR and Fell R (1999). Landslide risk assessment: prediction of travel distance. Can Geotech J, Vol 36, pp556 - 562.