Keywords: ANNs, GRNNs, neural, soil replacement, plate bearing test, compressibility, assessment.

Soil replacement comprises one of the possible solutions for founding on some problematic soils, e.g., soft clays, expansive or collapsible soils. The selection of the type, thickness and extension of the replacement soil has been among the challenges facing geotechnical engineers, lately. The quality of soil replacement constitutes a vital item to ensure the quality of the newly-formed foundation soil. In case of ill-installed replacement soil, more damaging problems, other than the cause for using replaced soil, may result. Thus, judging the quality of the replacement soil is a fundamental requirement before proceeding in the foundation works. Several types of field testing may need to account for the quality of the replaced soil. The sand cone test is usually chosen by consultants as an adequate one to ensure the suitability of the replaced soil. Nevertheless, in some cases, the sand cone test is not the appropriate one. The reason may be the type of the used soil, where an excessive amount of large gravel may limit the use of such a test. In addition, for one reason or another, more than a single layer may have been replaced without testing them appropriately. Other causes may also limit the use of the sand cone. Several alternatives are then available; the plate load test, or alternatively, any of the field soundings, e.g., CPT, flat dilatometer, Pressuremeter, etc. Thus, the plate bearing test presents a viable solution that is a good substitute to the sand cone test. Another problem, then, arises, namely, the acceptance criterion of the replaced soil via this test.

In this paper, a large number of plate bearing tests, namely, 349 tests, have been used in judging the efficiency of soil replacement. The data used has been collected from four Egyptian cites; 319 from Port Said, 18 from Isamilia, 7 from Suez and 5 from Delingat. Different thicknesses and materials have been utilized within these sites. The criterion for accepting (or not accepting) the soil replacement layers has been their compressibility measured through the plate bearing tests.

In this paper, one of the widely available soft computing techniques, namely, artificial neural networks (ANNs) is used for the assessment of the quality of replaced soil. A model using general regression neural networks (GRNNs) has been prepared and used to assess the soil replacement and proved to be successful. The input for the ANNs model included eleven variables; the position of the tested point within the site; the value of the allowable settlement; the value of the allowable stress; the recorded settlement at the allowable stress; the settlement at double the allowable stress; the settlement at triple the allowable stress; the percentage of gravel to sand; the dry/wet condition; the ratio of the settlement at twice the allowable stress to that at the allowable stress; the ratio of the settlement at triple the allowable stress to that at the allowable stress; and the geographic location of the site. On the other hand, a single output, which is the acceptability of the layer, is anticipated. Very promising results have been obtained from this study.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £95 +P&P)