Technical Note - Computerized approaches to coal blending

Four computerized approaches for coal blending are described and their strengths and weaknesses are explored. Spreadsheet analysis Other than word processing, spreadsheet software is the most commonly used software. Its wide availability and ease of use are two desirable features that have led many to try to adapt it to the purpose of coal blending. These are the two main advantages of this approach. To use a spreadsheet for coal blending, each row of data represents one source of coal. Each column represents a quality of the coal, with the last column specifying the tonnage of that coal in the blend. The calculated figures are the blended qualities, which can be programmed to appear in the bottom row. A trial-and-error approach is then used to develop the best blend. Inserting any blend in the last column provides the blend characteristics at the bottom of the page. If any specifications are not met, the blend can be changed, all calculations being immediately updated. This process is repeated until all specifications are satisfied. A third advantage of this method is that it is exactly the procedure that one would go through if doing this by hand. With a spreadsheet, many more blends can be evaluated in much less time, allowing the designer to try to improve over satisfactory blends to find the best of them. The resulting improved blends or cost savings are a fourth advantage of this approach. There is only one major disadvantage of using the spread-sheet. This is that most probably the best blend is not found. As an inexpensive introduction to computer-assisted blending, which can be easily understood and cost just a few hundred dollars, this is a recommended place to begin. Computerized search Computerized search approaches represent crude first attempts at letting the computer design the blend. Unfortunately, the blends found by these programs are very often less satisfactory than those developed manually. The way these techniques work is that they design, according to some predesigned pattern, a number of different blends and then quickly evaluate them. It is then straightforward to select the best from this sample set. The quality of the results depends on the number of blends designed, the way they differ from each other (the search procedure), the number of sources available, the number of quality specifications, and the severity of the specification limits. Good results are usually obtained for small, simple problems. At best, the results obtained are inconsistent from one blend to the next. Programs of this type are advertised in the mining literature. They usually mention how many different blends are evaluated, making it clear that this is more than can be tried manually. Expert systems Expert systems use a source of information known as a knowledge base. The knowledge base, usually a set of if-then rules, attempts to reconstruct the deductive reasoning process of an expert in some job capacity. For coal blending, the usual trial-and-error process has an easily recognized pattern of rules that are followed to try to find a better blend. By following these rules, it is possible to home in on the best blend without the exhaustive search required by the previous method and without the manual manipulation of the spreadsheet approach. The use of an expert system accomplishes this. One of the rules may say something like "if BTU is out of spec, then add more in the blend from a source having a high BTU and subtract some from a source having low BTU." In other words, it goes through the trial-and-error process. The process continues until all specifications are achieved and can continue further to try to optimize the cost. Of course, less simplistic rules are also needed. One set of rules would determine maximum changes in the tonnage added or deleted to the blend or from any one source. Other rules would be needed to determine which quality to improve if two or more are out of spec. At this level, the process of building the knowledge base provides the engineer with a deeper understanding of his or her own thought processes. The advantage of obtaining better blends is clear for this approach. Two more advantages should also be apparent. One is that it follows the same common sense logic that is used when doing the blends by hand or by spreadsheet, making it easy to understand. The other is that, like the search procedure, it designs the blend rather than just analyzing manually designed blends. Thus, it has all the good features of both approaches discussed thus far while providing improved results over each. A final advantage is the ease of use and ease of development. The expert system shells that are on the market are very easy to use and can be run on microcomputers. Linear programming Linear programming is an algebraic tool that solves a set of linear equations to yield an optimal solution to a problem. The optimal solution is based on the linear model so the results are only as good as the ability of the linear model to accurately portray the reality of the problem situation. It is fortunate for the coal blending problem that a linear model can be devel¬oped that reflects exactly the problem that is required. Unfortunately, the mining industry has been slow to use this tool. Coal blending applications available that use linear programming take one of three forms. The first, requiring exper-