Comparison of Approaches to Predict Grinding in An Industrial Rod Mill Operating at Low Fillings

Abstract

ABSTRACT Several approaches have been proposed over the years that may be used to predict the performance of industrial rod mills. Unfortunately, studies that allow comparing the accuracy in their predictions are very scarce. The work compares four methods for predicting size reduction in an industrial rod mill. These were the Bond work index method (BW) and the Callcott–Lynch (CL) model, both originally proposed for rod mills, besides the Herbst–Fuerstenau (HF) and the Austin–Klimpel–Luckie (AKL) models, first proposed for ball mills, but here adapted and applied to rod milling. A total of 10 industrial surveys were conducted of the industrial mill in dry grinding of coke breeze, covering operation under a variety of conditions, all of which at lower fillings than commonly used in industry (<15%). Since most conditions surveyed were not within the typical operating range for rod mills, simulations using the Discrete Element Method were used to predict the mill power, used as input in both BW and HF models. At first, the fidelity of methods that allow predicting full-scale mill performance solely from bench-scale information, namely BW, HF, and AKL, is compared. Results showed that they all presented high biases in their estimates of product fineness, with AKL predicting finer products whereas BW and HF predicting coarser products. Using then data from a single industrial survey for model calibration (base case), in addition to the bench-scale test results, a comparison of the four approaches became possible. Taking the absolute deviations between measured and predicted P 80 values, it became evident that the HF provided the highest fidelity in the predictions, with mean deviations of 12.8%, AKL with 15.3%, with both BW and CL presenting deviations above 18%. The work thus shows that size-mass balance models that rely on fitting most model parameters from bench-scale tests (HF and AKL) can provide predictions with good confidence when data from an industrial base case are additionally used in fine-tuning the model, even when applied to describing rod milling under unusually low fillings.