A stock of 300 pounds of 1 5/8-inch sillimanite balls was donated by the Champion Spark Plug Co. for grinding tests. They were true spheres, white, and all about the same size. Although native sillimanite has a density of 3.2, the artificial product had a density of only 2.56, this being the lowest density included throughout all the experiments. Another characteristic was the high resilience. As measured by the sclerescope, the hardness was 82. When dropped on a pavement from a height of 20 feet, they would bounce like rubber balls and not suffer damage. The flint pebbles had a higher sclerescope hardness than the sillimanite balls. Jasperoid pebbles, not used in these tests, had a hardness of 102. No effort was made to determine the resistance of the sillimanite balls to wear in the ball mill, but their effectiveness, when compared with steel balls, as a grinding medium was gratifying. As would be expected, the grinding capacity of the sillimanite load was low. This is evidenced by the low horsepower and the long grinding period.

Tests with sillimanite balls could not be carried to a conclusion, and no detailed results are given. The reason for this and for interpreting the tests with the flint pebbles not too rigorously is because of a principle not commonly recognized in grinding tests; that is, when media of different characteristics are to be compared tests must be made with several different sizes of each medium to determine which size is best for the particular particle size and hardness of the ore. This will be brought out in tables 30 and 31 covering the use of steel balls. For this test a variety of sizes of steel balls was available. It was not possible to make similar tests on the sillimanite because only one size of balls was available. At first thought it might seem that ore samples of a variety of particle sizes could be substituted for a variety of ball sizes, but such a transposition is incorrect because it is not known that different particle sizes have the same grindability; in fact, the evidence shows that they do not. Furthermore, it is difficult to be consistent in the amount of work applied to each sample.