Optimization of a response variable y constrained by principal directions of variations in the observation X-matrix

摘要: The purpose of the work is to enable design and manufacture cement with emphasis on quality properties cement. Data used in were collected from predictions characteristics predicted its production conditions kiln mill. some investigations conditions. was based sensitivity analysis focusing influence variables, x, properties, y. analyzed by y simulated variation x. In cases there correlation within observation matrix X, simulations constrained latent structure X. form optimization function for prediction optimal solutions variables then implemented models evaluated multivariate data using partial least square regression (PLS). PLS a member bilinear class methods. method compresses X most relevant factors these compressed as regressors thesis sometimes called PLS-components variables. Two types examination influences x-variables applied. first comparison size certainty coefficient scaled weighted second single x-variable or variable valued ranged relative confidence intervals development program an important part presented thesis. linear programming where optimized but upper lower limits at one x-variables. not focus only methods also apply real characterization performed pure observations Y. investigation three PLS-models properties; amount water required achieve standard consistency, setting time compressive strength 1 day investigation, PLS-model up 28 days made four submatrices representing different characteristics, mineralogy clinker cement, particle distribution third fourth superficial microstructure characterized X-ray diffraction (XRD) could be related kiln. microstructure, latter thermogravimetric analysis, From artificial predicting potential mineralogy. original new maintained, while other kept constant equal their mean values. represented XRDcurves, which are spectral data, two selected 2 _ ranges. Further, min max strength. addition, they interpreted qualitatively respect included PLS, differential mass loss curve recorded during analysis. addition examining blocks multi-block By application methods, that found directly Finally, mill properties. Amount consistency mill, values explain mechanistically chemically. main contributions papers this (the roman numerals refer list end chapter): 1. Developing model-based programming. A case demonstration optimizing [I-III] 2. Modelling [IV-V] 3. Presenting principles [VI-VII] 4. Optimizing [VIII-IX] use very useful enabled tailoring aiming target like strength, initial flow can well achieved giving achieving strengths. Optimal early ages microstructure.