Abstract

Starting from a general approach for dynamic modelling, several classes of local dynamic models for soft-sensors are used to model the concentrate grade in a rougher flotation bank. Among the non-linear - but linear in the parameter - models are non-linear ARMAX, Takagi and Sugeno, fuzzy combinational, projection on latent states (PLS) and wavelet based models. The fully non-linear dynamic model studied is a multilayer perceptron. The models are identified using actual rougher plant data. This data - which is very noisy - is first analysed in order to detect apparent sporadic short term failures of the sensor system for measuring the concentrate grade and then to repair the failed measurements. The models are determined using an identification (training) data set. The root mean square error and the correlation coefficient are used to compare model performances using validation and cross validation data sets. Results show that the best dynamic models is PLS. followed by perceptron and wavelet based models. Non-linear ARMAX, fuzzy combination and Takagi and Sugeno dynamic models give somewhat larger errors.