MULTIVARIATE CALIBRATION AND SENSOR ARRAY DESIGN

The use of multivariate analysis for calibration and prediction in chemical systems is presented. The emphasis of the study is in the use of sensor arrays for multicomponent analysis.; The growing interest in the area of process analytical chemistry for control and optimization of chemical processes has been accompanied by an increased awareness of the utility of discrete chemical sensors. Much effort has been expended toward the development of highly selective sensors to use for process monitoring. An alternative approach, examined in this study, is to use nonspecific sensors and employ various chemometrics techniques to perform the analysis.; One specific application discussed is the use of an array of nonspecific ion selective electrodes to perform quantitative analysis on mixture samples. For this system, the response has a nonlinear relationship with the concentration of analytes in the samples. To estimate the parameters in the calibration model, two different approaches are developed. The first procedure assumes the sensors follow the relationship found in the set of extended Nicholski equations. The model parameters are estimated using simplex optimization coupled with simple linear regression and the derived calibration models are used to perform prediction on unknown samples. The second approach uses a relatively new technique called projection pursuit regression to develop the model without using prior knowledge as to the functional relationship that exists between the response of the sensor and the concentration of analytes in solution. The results of analyses using these two methods are presented.; Also discussed is a new method of sensor selection based on principal component analysis where the variance describing capability of the eigenvectors is used to aid in selection. Additionally, four new methods for selecting sensors based on the minimization of prediction error are presented.