Minerals Beneficiation - Five Variable Flotation Tests Using Factorial Design

Factorial design is a mathematical method of drawing valid conclusions from a series of tests made in a predetermined pattern. It is applied to flotation ore testing using, in this case, five variables, each varied twice, requiring a total of 32 tests. Representative conclusions are checked by repeat testing. THE factorial design method of testing is particu-J- larly applicable to mill-scale plant tests because 1—the conclusions take into consideration all variabilities of the plant, 2—there are no subjective guesses as to the conclusions, 3—the number of tests necessary to check large numbers of variables are at a minimum, and 4—interactions between variables can be discerned. It is equally applicable to laboratory testing because once the variables to be investigated have been decided upon, as well as the number of levels for each variable, a well-trained assistant can run the tests, since the test work follows a predetermined mathematical pattern, leaving the conclusions to be found by the person in charge. The conclusion takes into consideration all errors of the tests. To illustrate the use of factorial design, a sample of copper porphyry ore was tested, the copper content of the batches not being uniform. Five variables were investigated, each at two levels, requiring only 32 tests. After suitable arithmetical analysis, conclusions for copper recovery and concentrate grade are arrived at. Four cases of these conclusions are checked by repeat testing. Testing Methods Reproducibility of flotation tests varies widely, depending on the experimenter, the process, the equipment, and probably a host of unknown and perhaps interacting variables. If the reproducibility of tests is good, let us say to 0.2 pct recovery of a certain metal, and the minimum difference between tests upon the variation of the interested variables is, say 1 pct, then simple averaging and inspection is ample in interpreting results. However, often such a happy situation is not the case. This is particularly true in mill-scale operations where the ores are variable, and a host of minor operating fluctuations common to most milling, and familiar to all mill operators, is the normal state of affairs. Yet, mill scale tests are run, the results averaged, and conclusions drawn therefrom. The trustworthiness of such results should be measured by comparison to the experimental error—the measure of the degree of reproducibility of the results. This comparison is a test of significance. Thus, if daily results of recoveries average 80.0 pct, and a spread of 78.0 pct to 82.0 pct, is recorded, then, clearly, on changing mill operations, an 81 pct recovery cannot be construed as significant by using averages, since it is well within the previous test limits. By using rigorous statistical tests of significance, conclusions arrived at by inspection or simple averaging may be shown to be based on inadequate evidence or even wrong. In any event, a plant super-