Keywords: geotechnical, ground, foundation, structure, neural network, rock engineering system, geohazard.

Consideration is given to determining the importance of factors affecting a ground-foundation-structure system. A novel strategy has been developed for analysing the relative importance of factors; this employs both artificial neural network (ANN) and Rock Engineering System (RES) methodologies. This yields a Global Relative Strength of Effect (GRSE). The revelation that the values of GRSE can be incorporated in a back-propagation neural network is explored, producing a computer- based system for analysing ground investigation information.

Two questions need to be addressed when determining the relative importance of factors: (1) which factors significantly influence the behaviour of a ground-foundation- structure system? (2) how can the degree of influence of individual factors be quantified?

A ground-foundation-structure engineering system typically incorporates a number of, possibly many, factors. These exhibit a multiplicity of interactions which are likely to be complex, time-dependent and difficult to describe explicitly in algebraic terms. A feature of current practice is that data and information obtained from ground investigation and structural survey studies rarely explicitly set out to define, rather they reflect, the mechanisms. This creates a challenge for prediction and assessing behaviour of the ground-foundation-structure system using traditional techniques. Attention has thus turned to investigating the potential of novel IT techniques: the ANN and RES.

The systems methodology is typified by RES, developed by Hudson, which considers the need for an all-encompassing procedure to deal with complex ground engineering problems. The Interaction Matrix is a basic device employed by the RES for representing the influential parameters and their interactions. It provides a tool which is useful for identifying the relative importance of these parameters and evaluating how they influence each other within a given ground system, i.e. identifying parameter dominance and intensity. This is implemented by coding the components (factors) of the Interaction Matrix using semi-quantitative methods so generating a cause vs. effect plot using group consensus. The relative degrees of influence of each factor on the whole system can thus be deduced from an analysis of the coded interaction matrix.

A feature of the RES methodology is that the weights are primarily knowledge driven, and thus the relative importance of individual factors can be determined without the need for a large database. A preliminary Ground-Foundation-Structure Information Infrastructure (GFSII) has been drawn up using the Rock Engineering Mechanisms Information Technology (REMIT) developed within RES. Determination of the relative importance of the main groups of parameters, which consist of ground, foundation, structure and environment, is illustrated. Despite a paucity of objective data, the relative importance has been determined by utilising experience and existing practice.

The ANN provides a computational mechanism which is capable of representing a mapping of one multi-factor information system with another, given a set of data representing that mapping. The Relative Strength of Effect (RSE) and the Global Relative Strength of Effect (GRSE) are measures obtained when back-propagating the neural network. The larger the absolute value of RSE, the greater the effect which the corresponding input unit will have on the output being determined. The GRSE appears as a linear RSE which is related just to the weights themselves. GRSE reflects the relative dominance of inputs on outputs within an interval, as opposed to acting at a point as does the RSE. Hence GRSE is a "global" parameter, and can be used to guide a hierarchical analysis of the relative strength of effect of the input parameters.

Both RSE and GRSE are dynamic parameters which change with input. They provide a quantitative measure of the relative dominance of inputs compared with the output. Consequently they can be used to evaluate the relative dominance of the factors considered.

An illustrative example is provided which describes how to determine the importance of factors in the Ground Group of the GFSII. These influence the shrink- swell behaviour of soils. 36 soil samples were used as input for the ANN modeling. Seven factors were considered: Clay Content, Liquid Limit, Plasticity Index, Cation Exchange Capacity (CEC), Coefficient of Linear Extensibility (COLE), Swelling 2:1's (i.e. the smectite content plus half the vermiculite content), and Swell Index. By means of the GRSE, the order of relative importance were determined as being: COLE Liquid Limit CEC Swelling 2:1's Swell Index Plastic Index Clay Content.

A strategy has been developed which analyses the relative importance of factors which influence a ground-foundation-structure engineering system employing both ANN and RES methodologies. ANN is mainly employed in situations where significant quantities of data are available; RES is primarily used to obtain a pragmatic solution in cases where the data are sparse. Used together, these two methodologies are able to compensate for each other's weaknesses and so derive a more realistic description of likely ground conditions.

The application of such tools is underpinning a UK study of geohazards influencing shallow foundations for low-rise structures in the East Midlands. An index is being developed that takes into consideration the relative importance of those factors contributing to the potential swell-shrink behaviour of the ground, and will eventually lead to development of a hazard index system which can assess the impacts of geohazards on low-rise structures.

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 £78 +P&P)