CONE: AN EXPERT SYSTEM FOR INTERPRETATION OF GEOTECHNICAL CHARACTERIZATION DATA FROM CONE PENETROMETERS

The general purpose of this project is to pursue the application of knowledge-based techniques in the domain of geotechnical engineering. The specific task involves the interpretation of data from an electronic cone penetrometer (CPT), a field exploration device used in geotechnical engineering to provide soil stratigraphy information and other characterizations of soil behavior. CPT readings are affected by many variables, including: soil type, density or consistency, stress conditions, stress-strain behavior, soil fabric and mineralogy. At the present state of geotechnical knowledge, these factors cannot be related in a purely algorithmic way to the engineering parameters required for design.;In the development of CONE, the interpretation task was divided into sub-tasks that embody the level of abstraction that an expert human interpreter uses to analyze the CPT log. This hierarchical decomposition was used as the basis for system organization. In implementing CONE, the decision was made to focus on two principal CPT interpretation tasks: soil classification and shear strength determination.;In geotechnical engineering, more than in most other fields of civil engineering, it is necessary to reason with incomplete data using empirically-based methods of analysis. There is a need to quantify the uncertainty that occurs in both the raw data and the analytical methods. There are two distinct sources of uncertainty present: statistical variability in the data and vagueness of methods. There are several methods for quantifying uncertainty in KBES, including certainty factors, adapted Bayesian methods, the Dempster-Shafer theory of evidence, and fuzzy logic. Fuzzy logic is employed in CONE. One of the most significant aspects of the use of fuzzy logic in CONE is that a consistent framework was developed to represent both linguistically-based (soil classification) and numerically-based (shear strength) data.;The following specific accomplishments of this system should be noted. CONE works in a realistic environment, in that it operates on the same raw data as a human interpreter; it achieves reasonably high performance (80% range correct classification); it uses multiple, conflicting sources of expertise and resolves these conflicts to arrive at its "best" interpretation; it employs fuzzy logic as an unifying representational and inferencing concept; and it represents traditional statistical information using fuzzy sets.