The use of neural networks to predict the particle-size distribution of coal ground in the PETC PRF hammermill

Sieve-analysis data obtained pursuant to grinding several types of coal using various settings of adjustable parameters on a hammer mill in PETC`s Process Research Facility were input to a neural-network simulation program in a variety of ways. The quality of the resulting particle-size predictions were tabulated and analyzed to determine which approach yielded the most accurate results. The simplicity of each approach was considered as a secondary factor. The most successful approaches yielded errors averaging 10 percent or less in the prediction of characteristic particle sizes such as the D-80 point. Generally speaking, the simpler approaches worked better, in that complicated manipulations of the input data produced either less accurate predictions or such small improvements in predictions that the extra effort seemed unjustified. Likely reasons for the success or failure of each approach are discussed. The methodology described could also prove useful for predicting coal particle sizes resulting from more complex grinding configurations.