A technique is presented whereby the particle size of the hydrocyclone overflow product can be predicted by means of a mathematical model. The model uses hydrocyclone feed flowrate and density as well as hydrocyclone overflow density to calculate the required particle size. Various modelling techniques are investigated. Simple linear models are compared to neural network models. Special attention is given to the identification of significant model inputs. Simple linear and more complex neural network models, both utilising an extra model input, cyclone overflow density are identified. Error detection and analysis are explored, resulting in a robust soft-sensor, capable of predicting hydrocyclone product size accurately in the plant environment.
展开▼
