Design and manufacture of Portland cement--application of sensitivity analysis in exploration and optimisation Part I: Exploration

摘要

Examples of application of different types of sensitivity analysis in explorations to be used in design and manufacture of Portland cement have been shown. The sensitivity analysis was based on results from partial least square regression. The sensitivity of an x-variable was defined as the predicted effect on y or y relative to confidence intervals of y. Exploration was performed as a two-step operation: simulation of variation in X and prediction of variation in y. For comparison, uncertainty testing on regression coefficients was performed. The goodness of the different methods of sensitivity was evaluated from the significance of the predicted values. The influence of a latent variable on y was as expected much more significant than that of a single x-variable. Application of simulated variation of X constrained by PLS-components will therefore be a better basis for the design of cement and will improve the quality control during production. Sensitivity analysis by jackknifing on the regression coefficients will not be of any use in the implementation in the design and manufacture in the cases where the latent structure in X has to be taken into consideration.