Modeling Data-Limited Systems

In a field such as geomechanics, where data is not always available, the methodology used in numerical modeling should be different from that used in a field such as mechanical engineering. [Starfield1988] provide suggestions for an approach to modeling that is appropriate for a data-limited system. This paper should be consulted before any serious modeling with a numerical modeling code such as PFC, FLAC3D, or 3DEC is attempted. In essence, the approach recognizes that field data (such as in-situ stresses, material properties, and geological features) will never be known completely. It is futile to expect the model to provide design data, such as expected displacements, when there is massive uncertainty in the input data. However, a numerical model is still useful in providing a picture of the mechanisms that may occur in particular physical systems. The model acts to educate the intuition of the design engineer by providing a series of cause-and-effect examples. The models may be simple, with assumed data that is consistent with known field data and engineering judgement. It is a waste of effort to construct a very large and complicated model that may be just as difficult to understand as the real case.

Of course, if extensive field data is available, then this may be incorporated into a comprehensive model that can yield design information directly. More commonly, however, the data-limited model does not produce such information directly, but provides insight into mechanisms that may occur. The designer can then do simple calculations based on these mechanisms that estimate the parameters of interest or the stability conditions.

It may be possible in some cases to use numerical models in a more predictive manner by performing one or more calibration steps where existing conditions are simulated and model parameters are adjusted to match to observed results. The calibrated model can then be used with some degree of confidence to make predictions regarding proposed future scenarios.