Sunteți pe pagina 1din 3
TABLE 15. Summary of Testing Results Postion No. | Case A output | Case B output | Target om @ 6) 1 ‘9999957 7 2 2.00019 ° 3 ? 1 4 0999996 1 5 Boponee 1 6 dois ° 7 t 1 8 0999587 1 9 7 1 10 0.000006 ° n 0.985 1 2 0999958 t a 0999861 i 1s 0.001528 | 0.003678 o 1s ‘osesros | o9gsz2s ° 16 0999992 | o9994s 1 7 ‘0.000001 8.00088 ° 18 noes =| Gone 1 1» ° 8.000006 o 2» ogesers | Ogoust 1 Fr 0000s | o.00az05 2 ‘oococe | 8.000059 ° 2 099999 0.999962 1 ae ° 000006 ° 3 ° ° ° 26 ° ° 8 Folate importance cay Case Case B @ @ 64 49 21 13 1 0 131 212 95 29 10s 109 2483 21 tos ios testing data. Also, the relative importance of the input neurons for cate A and case B as summarized in Table 17 indicate the ‘same tends. On the other hand, the discusser's resulis were Guite different from the writer's results, The writer is unable {o fathom the reason for the discrepancy as insufficient details hhave been provided. It is unclear whether the discusser used the measured SPT N-value rather than the standardized (N;)w used by the writer and Seed et al. (1985) to account for the different energy-delivery efficiencies indifferent countries and the effective overburden pressure, It is also unclear from the 226 / JOURNAL OF GEOTECHNICAL ENGINEERING / APRIL 1096 data presented by the discusser whether convergence of the taining data had occured or whether the average Sum squared error was oscillating. 4. The writer agrees with the discusser that the simplistic technique proposed by Garson (1991) to interpret the relative importance of each input neuron is at best, very approximate Nevertheless, it does provide the user with some intelligible Insights into the internal processing of the neural network. Garson has demonstrated for a simple example that the tech nique was more accurate than using regression analysis to de- termine the relative importance of the input neurons. APPENDIX. REFERENCES ‘lod, 1 and Karam, N, (1994) Neural networks in civil enginerog. nd understanding” J. Comp. in Ch Engr, ASCE, (2), (Gare, J. H. Je (1992). "Neural networks and the applicaiity within ‘hl engineering." Pro. &dh Con: Comp. in Ch Engrg. ASCE, New York, N¥e 1198-1162, Masters, T. (1993), Practical neural network recipes in C++, Academic Pros, San Diego, Cai. BEARING CAPACITY OF RECTANGULAR FOorinGs ON GEOGRID-REINFORCED SAND" Discussion by Ali Fakher,* and Colin J. F. P. Jones* ‘The authors have produced an interesting paper on the bear- ing capacity of rectangular footings on geogrid reinforced sand. The nature of the tests raises questions associated with possible boundary effects in the tests. In addition, the discuss- fers have made a comment on the influence of scale effect and the application of the results to practice. BOUNDARY EFFECTS ‘The depth of soil in the cubical test box used by the authors was 6.7 and 84 times the footing length and width, respec- tively, while the wid of the box was about 5.5 and 6.9 times the footing length and width, respectively. These ratios are less than those that have been used by some other researchers, ‘Table 4. Have the authors made any assessment of the bound: ary effects on their experimental results or numerical studies? ‘Any information about the amount of stress atthe boundaries ‘would be of interest. SCALE EFFECTS ‘The scale effects associated with a scale model can be ap- preciated by dimensional analysis as has been demonstrated by Langhaar (1951). In addition, dimensional analysis pro- vides sealing laws that can convert data from a small model to design information for a large prototype. Dimensional "Deven 1994, Vol. 120, No 12 by Temel Yetinogle, Jonathan F 1, Wo, and Ahmet Saplamer (Pager 7206) Rex Studeny Geotech, Div. Ci, Engrg. Dep, Univ, of Neweastle Upon Tyne, UK. "Prof, Geolch. Div, Civ. Engg. Dept, Univ. of Newcastle Upon ‘Tyne, UK TABLE 4. A Selection of Model Tests to Study Bearing Capac- ly of Reinforced Soll Raton betwoon: Sie of box]Dopn of sat and sizeof and uz of Footing | “tootng | “tootng Resoarenare o | “oe o o Some 38 7, | Aiiaanas aad ania OT) Seine on Se |Omareea 90) Rictngae| rae | 47.224 |Prguay aod Lawton (1960 Sep w 10 | Surat and Spee (1989) ae 2 ‘3 | Bingon and uze 995) Sep ies | [omwrecntasen, analysis has also been used in the field of reinforced soil 10 study the bearing capacity of unpaved roads (Love 1984). The authors have used a number of dimensionless parameters to represent their results. However they do not refer to any di- ‘mensional analysis. For the case of only one very wide layer of reinforcement, fa very simple analysis can be undertaken by assuming thatthe important factors in this model are 1B, Ly W, 8. Sx GeV 4.4) = 0 © where f= a function that governs the system; and G = shear ‘modulus of sol. The other symbols are the same as those used by the authors. A number of writers on dimensional analysis have stated the rule that the number of independent dimen- sionless parameters in a complete set is equal to the total num- ber of variables minus the number of fundamental dimensions in the parameters (Langhaar 1951). Since there are 10 factors in (5) and the model involves only two dimensions, force and Teogth, the system might be studied by any complete set of ‘eight independent dimensionless parameters, for example G)(@) G)2)@). 2.) where g =a function that governs the system. The dimension- Tess parameters in (6) can be identified a8 (ny, tye -y ad 1), respectively. Applications of 2 model test fora rea-ife project require that the model and the prototype should be Similar, that is (relating to the modcl and the protoype ‘hold be the same, However, tr not feasible to impose com: plete similarity in a model tex, consequently, some ofthe n- Sependent dimensionless variables may deviate fom ther cor rect values resulting in "scale effects." Consider « model footing which has dimensions m times smaller than the prototype L Bb BoE” oy where subscripts p and m refer to prototype and model, re- spectively and the soils in the model and the prototype have the same density, In order to satisfy similarity all (1) values should be the same for the prototype and the model (2) (22) aaa SE craceioo, © igte(E) (K) me SE ° o Itcan be concluded that the geogrid in the tests requires a stiffness (I/n') times that of the reinforcement used in practice. ‘A scale effect is apparent ifthe shear modulus of the soil in the model is the same as inthe prototype but the importance of this effect depends on the sensitivity of the system to the shear modulus, Even ifthe soils in the model and the prototype faze the same, the stiffness ofthe reinforcements should not be the same because in this cate the scale effect due to the stiff ness of the reinforcement is added to the scale effect through the shear modulus, The scale effect due to reinforcement stif- ress is very significant and should be avoided, particularly as the numerical study showed thatthe stiffness of the reinforce- ‘ment ig important as demonstrated in Fig. 11 of the paper. It should be added thatthe scale of the vertical axis in this figure ‘snot clear and it seems thatthe numbers 3 and 6 are missing. ‘A similar conclusion about the relationship between the stiffness of the reinforcement in the model and the prototype ‘ean be achieved by using other dimensionless parameters. The zatio of the bearing capacity of the prototype footing and the ‘model footing on unreinforced sand is equal to ae _ O.5NB, Gon" O5N,Bu ‘The coefficient of the bearing capacity, N,, in the mode! and the prototype is not exactly the same. This is partly because of the size effect of the footing which has been termed ‘de Beer's scale effect” by Kimura et al. (1985). However, we can assume that they are the same and that the scale effect from this is negligible , a0) () os = ( a) For the same settlement ratio (s/B), it can be seen that the stiffness of the geogrid in the tests is equivalent to 7° times the slffness of the geogrid in practice, In light of this fac, the authors’ conclusions in practice might be applicable. Interpre- tation of model tests without taking account of scale effects will overestimate the effect of the reinforcement. The authors’ comments on this aspect of the paper would be appreciate. APPENDIX. REFERENCES Kimura, T, Kusalabe, 0, and Satoh, K. (1985). “Geotechnical mode! tests of beasing capacity problems in a ceaifuge.” Geotechnigue, London, Ui. 380), 33-43, Langhane, 3.1. (1981) Dimensional analysis and theory of models John "Wiley & Sons, New York. NY. Love, J.P (1984), "Model testing of geogids in unpaved roads, ‘hess, Uns of Oxford, Oxore, UE JOURNAL OF GEOTECHNICAL ENGINEERING / APRIL 1996 /227 Copyright © 2003 EBSCO Publishing

S-ar putea să vă placă și