You are on page 1of 5

UNIVERSITY OF PORTSMOUTH SCHOOL OF EARTH & ENVIRONMENTAL SCIENCES 1GS314 Risk Analysis and Assessment Risk Analysis and

Assessment: Risk Modelling Introduction The classical approach in engineering or applied geological design is to consider the Factor of Safety of the particular structure being designed, for example a slope or landfill retaining wall, where the ' disturbing forces' are compared with the 'restoring forces' and failure is assumed to occur when the Factor of Safety is less than 1. The value of a Factor of Safety that is considered acceptable for a design is usually established from previous experience of successful designs. Rather than base a design on a single calculated Factor of Safety an alternative approach, which is frequently used to give a more rational assessment of the risks associated with a particular design, is to carry out a sensitivity study. This involves a series of calculations in which each significant parameter is varied systematically over its maximum credible range, in order to determine its influence on the Factor of Safety. This type of analysis is routinely carried out on many designs but this design process involves a considerable amount of judgement based upon experience built up from careful observations of actual performance of parameters being utilised in the calculation. When no such experience is available new tools need to be considered to assist the designer in making engineering decisions. The application of probability to design and the use of probability theory offer a means of assessing risk in a rational manner, even when the amount of data is very limited. Risk Modelling of a Potential Rock Bolt Failure The practical will demonstrate how a simple probabilistic or stochastic approach can be adopted to consider the probability of failure of a roof slab in a tunnel supported by a pattern of rock bolts.

Rock Bolt

Rock Bolt Spacing Grid

Rock Bolt Spacing Grid

S S In essence the risk analysis can be considered as encompassing four stages:


/opt/scribd/conversion/tmp/scratch32206/67707724.doc

1. Developing a model - by defining the problem or situation. 2. Identifying uncertainty - in the variables and specifying their possible values with a probability distribution. 3. Analysing the model via a simulation - to determine the range of all possible outcomes. 4. Making a decision - based on the results and experience. The above figure shows a roof slab of thickness t being supported by a rock bolt in a pattern spaced on a grid of S x S. The unit weight of the rock material is 2.7 tonnes/m3 and the initial spacing is 1.5m. The practical will illustrate the application of probabilistic techniques previously outlined to the assessment of the risk of failure of the roof slab: 1. A series of pull-out tests on 17mm diameter expansion-shell rock bolts have been carried out. The rock bolts have nominal pull-out strengths of 8 tonnes but the test results show variation from this value. The distribution of the results from the testing programme has been found to be Normal Download the spreadsheet ROCKBOLT.XLS and save to the n: drive. Load @Risk as instructed and open the downloaded spreadsheet. Note that Excel has been loaded with the @Risk add-in. Verify that the rock bolt strength data could be considered as being normally distributed. Note the names of the other suggested distributions from the fitting process. Using appropriate functions determine the sample mean and sample standard deviation of the test data, together with the minimum and maximum value of the tests. Note that to enter a formulae or function in Excel the = sign should be used. 2. The average thickness of the roof slab being supported has been estimated at 1m. Short of drilling dozens of holes to measure the variation in the value t over a representative area of the roof, there is no way of determining a distribution for this variable in the same way as was possible for the rock bolt capacity. This is a common problem with geoscience data sets where it may be extremely difficult or even impossible to obtain reliable information on certain variables and the only effective solution is to use educated guesswork.

/opt/scribd/conversion/tmp/scratch32206/67707724.doc

In the case of the roof slab it would not be unreasonable to assume that the thickness t is normally distributed about the mean of 1m. Obviously t cannot be less than zero since negative values would be nonsensical. In order to avoid this problem the normal distribution has to be truncated. An arbitrary minimum value of 0.25 should be used to truncate the lower end of the normal distribution of the slab thickness since smaller values will produce very high Factors of Safety. It is unlikely that t would exceed say 2m, and hence this can be used for the upper limit of the truncated normal distribution. For want of any better information it will be assumed that the standard deviation of the slab thickness is 0.5m. In other words 68% of the slabs will be between 0.5m and 1.5m thick, while the remainder will either be thicker or thinner. (Note that this is geological judgement being input into the risk evaluation). You should now enter this truncated normal distribution for the roof slab thickness in the form of an @Risk function in order to define the distribution to be used in the risk assessment and modelling. The function that you should use is: =RISKTNORMAL(mean, sd, min, max) 3. The overall demand on the rock bolt should now be defined: Input the formulae for the demand (D) on the rock bolt due to the slab. ie. Slab Thickness * Rock Unit Weight * Bolt Spacing 2 where the slab thickness will be sampled from the previously entered truncated normal distribution function. 4. The distribution of the values that will be used for the bolt capacity should now be defined: Again enter the bolt capacity as a truncated normal distribution ie. using the =RISKTNORMAL function as before. 5. The calculation for the Factor of Safety should now be constructed. Note that both bolt capacity and bolt demand will be sampled from the truncated normal distributions. Enter the formulae for the Factor of Safety where: F = Bolt Capacity / Demand on Bolt 6. To undertake the simulation in @Risk the various functions and required outputs need to be defined. This is achieved in @Risk via the icon with the single red arrow on the small menu bar. Add the following cells that contain the data that will be simulated by @Risk. Do this via a single mouse click on the required cell: Slab Thickness Bolt Capacity
/opt/scribd/conversion/tmp/scratch32206/67707724.doc

Bolt Demand Factor of Safety 3

Via the @Risk icon with the two arrows (red and blue) list the output variables that you have selected. Verify that all the required variables have been defined then close the window. 7. The sampling technique and other simulation settings should now be set via the Simulation Settings icon. Simulation refers to a method whereby the distribution of possible outcomes is generated by letting the computer recalculate your worksheet over and over again each time using different randomly selected set of values from the probability distributions of your data. Set the sampling technique to Latin Hypercube and set the number of iterations to 25000. Set Update Display On. Set Random Generator Seed to 1. Turn Monitor Convergence On. Execute the simulation via the Simulation icon. You will now see the previously defined distributions being sampled by @Risk to provide the randomly selected input data, from the various parameters and their distributions, for the Factor of Safety calculation. Each sampled piece of data will be used to calculate a new Factor of Safety. This will be done 25000 times. 8. On completion of the simulation process a series of results, statistics and graphs are available for further analysis and interpretation. Data is available for the calculated Factor of Safety, the slab thicknesses used as well as the bolt capacity and bolt demand data ranges. View the results window and summary statistics data for the variables in the Factor of Safety modelling. Note that these can be cut and pasted into the main spreadsheet for printing. 9. The final Factor of Safety distribution can be shown graphically, again for further interpretation and analysis. In the left hand window labelled Outputs highlight the Factor of Safety output and click on the graph icon. Choose the Fitted Distribution graph type. Set appropriate axes and titles and print the graph. What is the most probable Factor of Safety for the roof slab? Explore the other outputs and types of graph that you can generate. Choose the Ascending Cumulative Line graph and determine from the graph the probability that the Factor of Safety will be less than or equal to 1. Rerun the simulation with the other key distributions determined from the distribution fitting analysis and consider the sensitivity of the model to a change in distribution type. ie. How does this effect the probability of the Factor of Safety being less than or equal to 1? Is the new model more or less conservative than with the original normal distribution?
/opt/scribd/conversion/tmp/scratch32206/67707724.doc

Conclusions The graph shows the resulting distribution of the Factor of Safety. The data is lognormally distributed. from the statistical record produced by @Risk it can be seen that approximately 30% of the 25000 cases sampled have a Factor of Safety of less than 1 ie. the probability of failure of this rock bolt design is 30% for the assumed conditions. In order to establish whether a 30% probability of failure is acceptable, consider the consequences of one bolt in a pattern failing. The closest four bolts to this failed bolt would suddenly be called upon to carry an additional load of 20 to 25% over the load which they are already carrying. This is equivalent to increasing the bolt spacing to about 1.65m and substitution of this value back into the @Risk analysis shows that the probability of failure increase to about 50%. This suggests that an expanding domino type failure process could occur and that the original Factor of Safety is not adequate. Decreasing the bolt grid spacing to 1.25m in the @Risk analysis shifts the entire Lognormal distribution to the right so that the minimum Factor of Safety for the assumed conditions is found to be 1.04. The probability of failure for this case is zero. This decrease in bolt spacing would be a prudent practical decision in this case. Re-run the simulation with the Truncated Normal Distribution and a decreased bolt spacing to observe this effect. This simple example demonstrates that the use of probability theory produces a great deal more information than a simple deterministic Factor of Safety calculation. Even with the minimal amount of input data the shape of the probability distribution curve and the estimated probabilities of failure for different bolt spacings can give the designer a feel for the sensitivity of the design and suggest directions in which improvements can be made. References Hoek, E (2000) Factor of Safety and Probability of Failure. Practical Rock Engineering, Ch 8 http://www.rocscience.com/library/pdf/SL_4.pdf Hoek E. et al (1995). Support of underground excavations in Hard Rock. Balkema. @Risk 4.5 for Excel. Pallisade Corporation, New York. http:/www.palisade.com/ DPG @Risk 4.5 Version 1.0 December 2006

/opt/scribd/conversion/tmp/scratch32206/67707724.doc

You might also like