Determining Parameters from Operating Data

When good data are available from a plant sampling campaign, the operating values of the parameters in the models for the unit operations can be estimated using conventional parameter estimation techniques. Typically plant streams are sampled and analysed and, if the feed and product streams from any particular unit are available, the parameters for that unit can be estimated using the MODSIM model directly. All MODSIM models accept the details of the unit feed as input and produce the simulated output of that unit. When experimental data are available for the feed and product streams, the least-squares best estimates of the unit parameters can be obtained by matching the simulated output to the measured output to minimize the sum of squared deviations. This is best done by embedding the code for the unit model as it is used in MODSIM into a least-squares minimization search program. This method is very effective and usually produces good estimates of the parameters for the unit in question. Sometimes data that are specific to the material inside the unit must also be used to estimate all the parameters for that unit. The size distribution in the load of a ball or SAG mill is an example when it is necessary to estimate the parameters that define the classification action of the grate discharge. Data of this sort are typically difficult to get and it is often necessary to use best guess values for these parameters.

When operating data are not available for all the process streams that are associated with a particular unit, data from a larger section of the plant must be used. This can be done by embedding the entire Modsim simulator inside a least squares minimization program. Typically, measured size distributions from several of the plant streams are matched against the simulator outputs for those streams. Modern optimization codes make this possible and the method is now used routinely.

Exercise 8-1 illustrates this method.