In this paper we propose the concept and methodology of object model-based expert systerns. A systematic modeling method and a model-based reasoning scheme are shown. The object model is useful to represent the knowledge about structures and behaviors which are known as one of deep knowledge representations for complex structured objects. The key points of the object model-based knowledge system are that 1) representation of knowledge and interpretation of it are separated from each other so that the representation could be used for solving a variety of problems via associated interpretations, and 2) the knowledge model for a specific object can be smoothly constructed by means of a generalized scheme of the object. Rule-based systems, i.e., surface knowledge systems, are convenient to represent expertise, but their performance and flexibility are limited. Advanced knowledge-based expert systems to solve complex problems need to handle both types of knowledge. The object model is represented by means of a two dimensional hierarchy where a whole-parts (PARTOF) relation is represented in a horizontal hierarchy and an abstract-specific (ISA) relation in a vertical hierarchy. A trouble-shooting expert system combining IF-THEN rules and an object model for electro-mechanical devices have demonstrated interesting characteristics. The long time goal of this work is to provide a useful technology for developing a multi-objective shared-use engineering knowledge base which will become a key technology in the 1990's. The system was developed using the frame-based knowledge engineering environment ZERO.

View full abstract

[in Japanese]

View full abstract

The FINDER system employs geometric probability, Bayesian statistics, and the normal probability density function to integrate spatial and frequency information to produce a map of probabilities of target centers. Target centers can be mineral deposits, alteration associated with mineral deposits, or any other target that can be represented by a regular shape on a two dimensional map. The size, shape, mean, and standard deviation for each variable are characterized in a control area and the results applied by means of FINDER to the study area. The Kushikino deposit consists of groups of quartz-calcite-adularia veins that produced 55 tonnes of gold and 456 tonnes of silver since 1660. Part of a 6 by 10 km area near Kushikino served as a control area. Within the control area, data plotting, contouring, and cluster analysis were used to identify the barren and mineralized populations. Sodium was found to be depleted in an elliptically shaped area 3.1 by 1.6 km, potassium was both depleted and enriched locally in an elliptically shaped area 3.0 by 1.3 km, and sulfur was enriched in an elliptically shaped area 5.8 by 1.6 km. The potassium, sodium, and sulfur content from 233 surface rock samples were each used in FINDER to produce probability maps for the 12 by 30 km study area which includes Kushikino. High probability areas for each of the individual variables are over and offset up to 4 km eastward from the main Kushikino veins. In general, high probability areas identified by FINDER are displaced from the main veins and cover not only the host andesite and the dacite-andesite that is about the same age as the Kushikino mineralization, but also younger sedimentary rocks, andesite, and tuff units east and northeast of Kushikino. The maps also display the same patterns observed near Kushikino, but with somewhat lower probabilities, about 1.5 km east of the old gold prospect, Hajima, and in a broad zone 2.5 km east-west and 1 km north-south, centered 2 km west of the old gold prospect, Yaeyama.

View full abstract

GGCS is used to compare the target area with typical gold deposits based on similarity of geological conditions for mineralization. It serves as a consultant system providing not only the type of classification but also information about the gold deposits which most likely occur in the target area. About 90 deposits are included. The system is written in the PROLOG language. GGCS gives a practical example to show a method of arrangement of geological attributes in a geology consultant system.

View full abstract

GEOEXPLORER is a knowledge-based system under development, to support exploration and assessment of supergene deposits. Some existing expert systems, such as PROSPECTOR have been designed mainly for regional resource assessment or for mineral deposit prospecting. In contrast to that, GEOEXPLORER shall not only deal with problems of searching for mineral deposits (deposit prospecting), but it shall also cover the evaluation of the spatial occurrences as well as of the economic values of deposits (deposit appraisal) . Besides, an interface to an external data base is taken into account. This paper gives an outlook of the development of GEOEXPLORER.

View full abstract

Knowledge representation is still one of the KEY problems to the success of a practical expert system for geological exploration. Three aspects are discussed in this paper: (a) synthetic-uncertain-inference-model instead of a single inference model is more suitable for describing greatly diverse domain expert's knowledge; (b) strong correlation is often a basic feature among pieces of geological evidence, and can be directly treated under a new form of Bayesian model; (c) inversion of parameters describing knowledge makes knowledge engineering much easier. Finally, a prototype system expressing the above ideas is presented.

View full abstract

We have attempted to develop a system to evaluate the favorability of the uranium exploration projects using the“Data directed Numerical Method (DdM) ” (McCammon et al., 1986) . The system was implemented on the Macintosh personal computer and was designed so that the users can easily make the most of an iterative trial and error process. The Arnhem Land area of the Alligator Rivers Region, Northern Territory, Australia, was chosen as a prospecting area for the favorability assessment of uranium deposits. The Alligator Rivers area is one of the famous uranium provinces in the world. The genesis of the known uranium deposits of the area was studied in an extensive literature survey to find the common geological factors. All information on the distribution of lithologies and structure within the area was obtained from large scale geological maps. Five exploration parameters were selected from a statistical analysis of the relationship between the exploration parameters and the presence of the known uranium deposits and occurrences. The favorability evaluations were performed for two different types of uranium deposits, namely the Jabiluka-Type and the Nabarlek-Type.

View full abstract

In mineral exploration, several layers of geoscience information such as geophysical, geochemical, remotely sensed or lithological data are studied for their individual characteristics and interrelationships. These geologic data sets may then be integrated into a single map indicating potential areas for further investigation. Selection of the target areas depends upon not only the spatial data but also the combination rules applied to them. Multivariate regression analysis and two probabilistic procedures, Bayesian and Dempster-Shafer methods, both used in artificial intelligence methodology are reviewed and discussed, including the implications of the assumptions required in each method. We illustrate the techniques by using a simple artificial example.

View full abstract

There are several quantitative methods of integrating geological and geophysical data sets for a specific exploration target. In this study, a“fuzzy set approach”using algebraic sum and γ-operator is investigated and tested with nine sets of real geological and geophysical data from northern Manitoba, Canada. The final possibility distribution maps using both operators have successfully outlined favorable areas for base metal deposits. However, interpretation of the compensation level associated with the two operators tested require further investigation.

View full abstract

In mineral exploration, most geological field data, including geochemical, geophysical and remote sensing data, have to be plotted onto a base map for the final analysis and interpretation. Recent application of computer based GISs (Geographical Information systems) is basically developed without proper understanding of spatial characteristics or quantitative modelling framework of the data. Many of the available geological data sets are incomplete and imprecise for most exploration areas and exploration targets. In this paper, new information representation approaches with consideration of spatial and terrestrial spectral characteristics of exploration data are proposed. They include the basic probabilistic approach, the evidential belief function approach (Dempster-Shafer method) and the fuzzy logic approach. Even though basic concepts of these approaches are different, proper interpretation of the final results are expected to yield similar conclusion and more importantly they are expected to provide more accurate final results than the conventional qualitative methods. No real example is given in this paper.

View full abstract

The topographic features and the boundary of the geological units can be observed in the structural features of satellite images in the brightness changes of the spectra. Many of the interesting geological structures are very small in size compared with the minimum scale of the satellite images which is represented as pixel which is the spatially minimum size of observation, though the scales are not enough to detect the detailed structures related to mineral deposits. Pattern recognition by computer is a possible method to study digital images to detect structures on the surface. In this study, the structures in images are divided into linear, textural, and spectral patterns, then integrated together in the island arc areas in a tropical region of Indonesia in order to analyze the relationship of the images to mineral deposits.
Lomblen Island in the Lesser Sunda Islands was chosen to be studied because of some peculiar ring structures and the characteristic arrangement of the long axis of the island in the direction northeast to southwest. The ring structure is possibly a volcanic cauldron which are related to the epithermal ore deposits in many areas.
Four types of edges are generally considered as step, tilt, roof, and line. The form of line is suggested by the arrangements of three or more pixels with characteristic intensities along straight or quasi-straight lines. The parallel arrangement of pixels with distinctively different intensities is easily detected as a line. Because the digital image often includes noise, the effect of noise is frequently eliminated from the original image using averaging of intensities between the neighboring pixels or other processes. Some topographic features are observed by the areal change of the brightness of sunlight. The areal change of images can be measured by their textural quantity. There are many methods of texture extraction. The rudimentarily textural quantity is the standard deviation of the brightness in a specific area. The standard deviation of brightness indicates topographic changes which respond to the contrast or change of brightness in images. The density calculation has the effect to average the areal value, the result in this study is ambiguous like the standard deviation image.
We found mineralized zones in separated areas of northern Lomblen Island. In northwest point of the ring structure, the silicified veins contains up to 2.0 ppm silver in a manganese oxide zone. In the northeast point of the ring structure, quartz and barite veins contain up to 4.2 ppm gold with several tens of ppm silver and copper and zinc sulfide. There is up to 687 ppm gold in the weathered quartz vein. High gold samples in quartz veins contain about 20 ppm gold locally. According to gravity measurements, the rim of the ring structure has very low gravity values whereas the central mountain area of the ring structure has high gravity. We suggest that the area of the ring structure could be a resurgent cauldron which are frequently related gold-silver deposits.

View full abstract

We have been entrusted to study the expert system for lineament extraction from opticalsensor data. In this study, we conducted the semi automatic lineament extraction experiments, using satellites digital image data. The main purpose of this study is to build an expert system to extract lineaments automatically from remotely sensed optical sensor data, using the knowledge of photogeologists which is frequently used for the interpretation of lineaments. First, we put the knowledge of photogeologists in order, and based on it, worked out the basic system design. Next, we subdivided and arranged this knowledge for the computer and defined them more clearly. A prototype of an expert system for lineament extraction was developed first. Lineament extraction conducted on TM image data of the Sichuan Province of China produced good results.

View full abstract

We tried to develop a preliminary expert system to discriminate geologic units from 1987 to 1989. The purpose of this project is to develop a computer system, with which a user discriminates geologic units outcropping in an unfamiliar area. The discrimination criteria are based on information which is compiled by a geological expert surveying the area.
In order to develop this system, Nishiura, Kagoshima Pref., Kyushu, Japan, was selected. This system is a prototype, so it was developed on the basis of only visible information of rocks. The knowledge of the experts is expressed in a table called the“Attribute Probability Distribution Table”. The system computes the probability of occurrence of each geologic unit according to answers given by a user.
The field test indicated that the system was not working well enough to be used in actual field work, and showed some problems that need to be solved. However, this work was meaningful because the system indicated an availability of computer software that could employ descriptive and qualitative factors. It was especially encouraging that in only a few cases, the geological expert agreed to revise the identification of geologic units according to the suggestion of the system.

View full abstract

This paper deals with the technological transfer of geostatistical knowledge and expertise in ore reserve assessment. The main challenge lies in the simplification, accessibility and enhancement of usability of geostatistical methods. Intelligent geostatistical computer systems may be seen as generalizations of conventional geostatistical programs, containing in addition to numerical capabilities, the knowledge and expertise required to guide, undertake, evaluate, and reason about geostatistical operations.
To provide the required knowledge transfer, the integration of knowledge-expertise for geostatistical ore reserve assessment with symbolic techniques developed in the field of artificial intelligence has been perused, and obtained by establishing a formal hierarchy-taxonomy of geostatistical conceptual units. This approach captures both the basic notions of geostatistical methods and the practical intricacies involved, thus providing the basis for the development of intelligent systems in this domain.
The feasibility and abilities of intelligent geostatistical systems are demonstrated by two experimental systems, implemented in LISP. The first, GEOSTAT-1, is concerned with the inference of variograms, and the second, GEOSTAT-2, with grade estimation using ordinary kriging. The characteristics of both systems is demonstrated by reference to a case study on an auriferous vein deposit.

View full abstract

An interactive computer system has been developed in order to assist in mineral identification under the petrographic microscope. The present reference file consists of optical and chemical properties of about 150 mineral species. The properties are name, chemical formula, refractive index, birefringence, indicatrix, optical angle (2Ω), extinction angle, elongation, pleochroism, and color in each of the X-, Y- and Z-directions of each species. The system utilizes optical properties of an observed mineral given by a user, and reduces the number of possible minerals. This procedure is essentially the same as using published mineral identification tables except that the process runs in a personal computer. The system presents a series of questions to a user, and the user selects any of the prepared answers according to his observation. The user can also choose any desired question independently of the sequence. When a user has chemical and/or mineralogical data for a sample, the user can provide the information to the system at any step of the identification process. In order to assist a user the system prepares some explanations and suggestions for the questions about some optical properties, such as refractive index, indicatrix, color, elongation and extinction angle of anisotropic minerals, and optical angle of biaxial minerals. This feature is quite different from the conventional identification table.

View full abstract

A binary relation is a useful mathematical tool to describe geological relations. In the present paper, we give a brief explanation of binary relations using some geological examples, and try to formalize several geological concepts related to the Principle of Superposition with special attention to the hierarchical structure of the logical system in geology.
The Principle of Superposition is formalized as:
L∩L^-1=φ∩L⊂K
where φ is an empty set, and L and K are symbols representing relations on a set of strata; aLb : a is lower than b, and aKb : a is older than b. The formula indicates that when we code a program to find the order of strata in time from field observation on spatial distribution of strata, we must check whether given data satisfy a condition that L∩L^-1=φ or there are no pairs (a, b) which give both aLb and bLa.
Although the present study is a preliminary step to construct the Geology-Oriented Logical System (GEO-LOGICS), we have confidence that the mathematical formalization leads us to a deep understanding of geological knowledge as well as practical algorithm.

View full abstract

Although widely used today in Australia, fifteen years ago geostatistics was a new method for the evaluation of orebodies. At that time, a case study detailed the advantages it offered but was rejected; not on technical grounds, but rather because it confronted existing practice and seemed to pose a threat to company engineers. Such a reaction may reasonably be expected when innovative technology attempts to replace an older order. Careful interaction between researcher and practitioner is required if the adoption of a new technique is to be smooth.

View full abstract