Detection and identification of ore grindability in a semiautogenous grinding circuit model using wavelet transform variances of measured variables

A reduced dimension dynamic model subject to random disturbances for a semiautogenous grinding (SAG) circuit is developed that is able to handle changes in the characteristics of the new feed ore. This dynamic model, which is adjusted using data from a large mineral processing plant, has been developed with the objective of being useful for design and testing of fault identification and detection (FDI) systems, soft-sensors, automatic control systems, etc. The reduced dimension is a requirement in order to be able to obtain in a reasonable time statistical results to evaluate such systems. In this paper this SAG circuit dynamic model is used to test a method for detecting changes and identifying the grindability of the ore being processed by the SAG circuit. The method used — based on the variance of the continuous wavelet transform of measured circuit variables — incorporates improvements of a previous method. Results show that a step change of new feed ore is detected in about 30 to 80 min depending on the grindability change. This result may be considered to be adequate when taking into account that the response time for the mill hold-up to attain a new equilibrium value after the ensuing transient is of about 3 h. Identification of grindability in stationary operation gives near 100% of correct classification under the analysed conditions. The sensitivity of the FDI method to changes in circuit characteristics is also assessed and acceptable results are obtained.