Top of page
Marcello M. Veigaa and John A. Meechb
a Madison do Brasil,
R. Barao do Flamengo 32, 8o andar,
Rio de Janeiro, 22220-060, Brasil
b The University of British Columbia,
Department of Mining and Mineral Process Engineering,
Vancouver, B.C., V6T 1Z4, Canada
ABSTRACT.
The application of a Fuzzy Logic-based expert system to assess mercury bioaccumulation risk in gold mining regions around the World is described. HgEx is a heuristic system which accommodates imprecise data for pH, Eh, water conductivity, biomass productivity, water transparency and contamination factor. Inaccurate data input can be handled for Hg background and sediment assays, or alternatively, measurements are transformed into linguistic expressions with respective Degrees of Belief to feed a heuristic model using neural equations. The paper will present a fuzzy adaptive method and show how it can be applied to model situations such as AIDS research, technological innovation and other risk-assessment problems.
1. INTRODUCTION
Risk assessment using artificial intelligence can reduce costs and confer elasticity to decision-making. When knowledge is intricate, fraught with uncertainty or when intervariable relationships are unknown, Fuzzy Expert Systems are useful to gather disperse information and accumulate evidence.[1,2]
HgEx [3,4,5] is a heuristic expert system which accommodates imprecise data input for variables such as pH, Eh, water conductivity, biomass productivity, water transparency and contamination factor.[6] The system can handle vague inputs for variables such as Hg background and sediment assays, or alternatively, measurements can be transformed into linguistic expressions with respective Degrees of Belief to feed a model based on neural equations.[7,8]
These models have been termed the Weighted Inference Method. Linguistic expressions that denote the degree of risk have the same practical effect as complex mathematical models which usually demand high cost, much data and skill to quantify the effect of numerous factors on bioaccumulation.
HgEx does not replace monitoring, but rather, complements such programs by reducing the need for accurate analytical results to obtain a preliminary, yet conclusive risk assessment. Expert Systems do not solve a pollution problem on their own, but they do bring expert and consistent information to diagnose dangerous situations and generate rapid decisions.
The system consists of three components: a HyperText-based [9] tutorial section; a toxicology section to evaluate the potential for mercurialism and a diagnostic section to establish bioaccumulation risk.
The tutorial section includes over 1500 pages of information on Hg chemistry, use and availability. Knowledge is available on conducting amalgamation to minimize Hg losses; directing field inspection and sampling programs; studying sources and methods of Hg-transformation; and instituting cleanup measures.
The toxicology section collects data on individuals exposed to Hg-emissions to establish the method and extent of contamination of an affected population. Poisoning can occur from either metallic vapor or through methyl-Hg acquired from food.
This paper focuses on Diagnostics. The modeling approach and its adaptability to contaminated sites in various settings around the world are described.
2. DIAGNOSTIC STRUCTURE
The structure of HgEx follows the uncomplicated concept that bioaccumulation is a function of levels of emission and transformations into dangerous species. Abatement by adsorption on ferruginous sediments is the only way that risk can decrease unless remedial efforts are instituted (see Figure 1).
Figure 1. Structure of the Diagnostic Part of HgEx.
The predicted risk is compared to analyzed samples if available, and conclusions about current and future bioaccumulation are provided. When high risk is indicated but biota do not show high Hg levels, the system suggests frequent monitoring programs and remedial procedures to avoid future problems. Other sources can also be accounted for to adapt erroneous diagnostics.[10]
When a variable is characterized in the form of linguistic expressions such as "high", "low", "acidic", "medium", "dark", etc., the meaning of such terms may be defined as a function of several input variables. Each expression receives a Degree of Belief (DoB) which derives from a discrete value and a defined belief function or is inferred by the system as in Figure 2. In HgEx, belief in these expressions with their respective DoBs are processed within the rules of the heuristic model using a series of Weighted Inference Equations for each fundamental concept.
Figure 2. Process by which HgEx deals with inputs.
The Weighted Inference Method derives a DoB in a conclusion by multiplying the importance of each piece of evidence (Wi) by the Degree of Belief (DoBi) that the evidence exists. This summation emerges from a single node as a Degree of Belief (DoBconc.) in the conclusion ranging in value from 0 to 100. The actual output displayed to the user is filtered from a function involving linguistic defuzzification of the DoBconc into terminology characterizing the spectrum over which the concept ranges (e.g. from non-existent to extremely high) as in Figure 3.
The main advantages of the technique are:
Figure 3. Linguistic Output of DoBHEF.
The heuristic model also converts inferred facts into discrete values as in Figure 2. For example, a user may not have Hg sediment analyses so a DoB can be inferred. Individual inference equations are used to calculate DoB in high, medium or low sediment or background levels based on mineral content and material color. Belief in these concepts are combined using weighted-average defuzzification to give a discrete value between 0.03 and 0.3 ppm Hg.
3. ADAPTATION OF THE MODEL
The model can be adapted to many heuristic or context-sensitive situations. Linkage between input DoBs and final output DoB occurs through an adaptive equation that is raised to an exponent value called alpha (acceptance factor) which ranges from 0 to 100. This factor represents the degree of acceptance by a particular Society for a process used to recover gold in which a dangerous material like Hg is used.
Figure 1 shows that acceptance (alpha) depends on technical, economic and socio-political issues. These details vary from place to place and from time to time. For example, in the Amazon today, the value of a is about 1.0 signifying a transition from acceptance to rejection of amalgamation as Brazil completes its transformation from the Third World into the First. In North America today, alpha is about 0.01 indicating virtual rejection of Hg as a reagent to recover gold. About 150 years ago, alpha was 100 on a worldwide basis with total acceptance of amalgamation since Society was ignorant of the dangers of this metal.
This method gives elasticity to system output. For example, in Figure 3, when alpha = 1, a "high" emission derives for a DoBHEF between 50 and 80, but when alpha is 10, this range is interpreted as "very low". For alpha = 0.1, the term becomes "extremely-high".
The alpha factor derives from heuristic rules that account for: Au and Hg prices, current economic conditions, cultural values of the activity, relative political power of impacted populations, presence of environmental groups and a reliable media, etc. The rules are depicted in Figure 4. The boundaries indicate that regions in the map are fuzzy - 5 sets characterize each variable, requiring 10 rules in total.
Figure 4. Rule Structure to Determine alpha Factor.
The method allows one to study the evolution of a society toward acceptance or rejection of a particular process, idea or life-style. With Amalgamation, the initial position on the graph is half-way along the scale between Economic and Technical Group factors. Rejection influences from Society are virtually nil.
However, as dangers of Hg become well known or dependence on gold production declines, the value of a begins to decline. When alpha reaches 1.0, all forces are in balance, Hg use is economically attractive; it is technically sound but dangers to the environment are recognized. Alternatives and remedial measures are initiated and a continues to drop until the socio-political issues cause total rejection of the process when alpha reaches 0. Of course, in reality, 0 is never reached, (someone, somewhere will use Hg despite the pressure to desist). This is why a minimum alpha value of 0.01 is shown in all diagrams presented in the paper.
This straight and steady path would be the ideal evolutionary process, however economic and technical forces can also come to bear as time passes. New techniques to recover or control Hg may develop. Drastic downturns in an Economy may force the population back to the land in order to eke out an existence. Gold-mining and Hg use may rise again!
4. APPLICATION TO OTHER DOMAINS
It is considered that this method could be easily adapted to other domains. The starting position on a tertiary diagram would be one of 6 points: the vertices and the half-way positions on each scale. The initial situation may be favorable or may not (i.e. alpha may commence at 100 or 0). Figure 5 shows the 12 possible scenarios for a variety of issues or processes. An example for each is also presented below each diagram. The contour lines on these diagrams represent iso- a factor values. The synergistic effect of two groups produce the shape exhibited, i.e. a combined effect will influence a greater than will the weight of a single group. In some cases, all 3 groups combine to develop total acceptance or rejection of the concept, process or solution extremely rapidly - space exploration, AIDS research, etc.
5. CONCLUSION
This work has described development of a Heuristic System to diagnose Hg-polluted sites or regions in the Amazon. Using a Weighted-Inference Method, large amounts of data synthethize into belief in a conclusion about the extent of emissions even when exact measurements are lacking. Rapid diagnosis provides rapid solutions and can establish the best remedial action and the highest priority locations.
The system is adaptable to reflect different economic, technical and socio-political situations. As such, HgEx can apply to various locations and contexts to derive expert advice.
This approach can deal with other domains subject to complex synergistic inputs. Evolution of acceptance or rejection for a technique, concept or solution can be followed and perhaps, influenced by monitoring the change in the a factor through the use of the diagrams presented in this paper.
Acknowledgment
The Canadian Natural Sciences and Engineering Research Council supported this work through the Research Grants Program. Support from the Vancouver Branch of the Canadian Institute of Mining and Metallurgy in the form of a Graduate Research Scholarship is acknowledged by MMV.
References
1. L.A. Zadeh, "The Calculus of Fuzzy If/Then Rules", AI Expert, March 1992, 23-27.
2. J.V. Parkin, "Judgmental Model of Engineering Management", Journal of Management in Engineering, 10(1), 1994, 52-57.
3. J.A. Meech, M.M. Veiga, R. Hypolito, "Educational Measures to Address Hg Pollution in the Amazon", AMBIOS, 1995, (accepted Oct. 1994)
4. M.M. Veiga, J.A. Meech, "Heuristic Approach to Hg Pollution in the Amazon", Proc. Inter. Sym. on Treatment of Wastes, 123rd TMS Congress, San Francisco, 1994, 23-38.
5. M.M. Veiga, J.A. Meech, "Expert System for Risk Assessment of Mercury Discharge from Gold Mining Operations", Proc. Inter. Sym. on AI in Material Processing, CIM - Metal. Soc., Edmonton, 1992, 107-118.
6. J.A. Meech, M.M. Veiga, D. Tromans, "Mercury Emissions and Stability in the Amazon Region", Proc. Inter. Sym. on Waste Disposal, 34th Ann. Conf. of Metal. CIM, Vancouver, 1995, pp.16.
7. E. Khan, "Combining Fuzzy Logic and Neural Networks, in: Conf. Proc. Fuzzy '93, Computer Design, Burlingame, 1993, T31/1-48.
8. B. Kosko, Neural Networks and Fuzzy Systems, Prentice-Hall Inc., New Jersey, 1992, 449 p.
9. J.A. Meech & S. Kumar, Hypermanual on Expert Systems, CANMET, Ottawa, electronic text on Expert Systems, 1995, v.3.0, 4800 elec. pages.
10. M.M. Veiga, J.A. Meech, N. Oñate, "Mercury Pollution from Deforestation"; Nature 368, , 1994, 816-817.
Return to the publications index