Abstract In recent years, the effectiveness of geochemical exploration techniques have improved significantly, mainly due to major progress in multi-element analytical methodology and the computer-processing of data. In France, these improved techniques have made it possible to analyze and interpret the results from more than 250,000 samples. The necessity to select mineral targets as early as possible has shown the need for reliable selection procedures for geochemical anomalies, taking account of highly varied nonnumerical factors such as sample type, morpho-climatic an geological settings, and anthropogenic context. Such a procedure is provided by serge (Systeme Expert en Reconnaissance d'Anomalies Geochimiques) developed jointly by the Bureau de Recherches Geologiques et Minieres and Paris-Sud University. serge simulates the reasoning process of a geochemist in ranking multi-element geochemical anomalies. serge is composed of three distinct parts: a knowledge base, an inference engine, and user interfaces. The knowledge base contains the expert knowledge, which is expressed according to the production rules of the type “If X then Y”, for example, “if Cr grades are high, then a pollution risk exists”. These rules are close to natural language and can easily be modified and applied. At present, serge comprises a knowledge base of about 150 rules applicable to base-metal anomalies in Brittany. Such rules concern mainly the morphology of the anomaly, the sample type, the contrast, the contamination risk, the landscape, and the accompanying validating or penalizing elements. The inference engine manages all of the knowledge base and functions backward-chaining and forward-chaining modes without using a probability coefficient. The engine explains its reasoning, either during dialogue with the user by explaining the problems it is trying to resolve, or at the end of its diagnosis by justifying its modus operandi. The user interfaces serve to “feed” the system. Two types of data input are possible: (1) manual input, where the user answers questions asked by serge ; and (2) semi-automatic input, where serge searches image files for the data it requires, such as geochemical contrast, geological context, and type of anomaly. Tests show that serge presents a performance that falls between that of highly qualified experts and of junior geochemists. Therefore, in contrast to the expert system prospector , whose aim was to help all specialists contributing to mineral exploration, serge is designed to provide specific assistance by ranking geochemical anomalies, which were previously defined using an algorithmic processing chain.
Read full abstract