Multivariate optical computation for predictive spectroscopy

A novel optical approach to predicting chemical and physical properties based on principal component analysis (PCA) is proposed and evaluated using a data set from earlier work. In our approach, a regression vector produced by PCA is designed into the structure of a set of paired optical filters. Light passing through the paired filters produces an analog detector signal directly proportional to the chemical/physical property for which the regression vector was designed. This simple optical computational method for predictive spectroscopy is evaluated in several ways, using the example data for numeric simulation. First, we evaluate the sensitivity of the method to various types of spectroscopy errors commonly encountered, and find the method to have the same susceptibilities toward error as standard methods. Second, we use propagation of errors to determine the effects of detector noise on the predictive power of the method, finding the optical computation approach to have a large multiplex advantage over conventional methods. Third, we use two different design approaches to the construction of the paired filter set for the example measurement to evaluate manufacturability, finding that adequate methods exist to design appropriate optical devices. Fourth, we numerically simulate the predictive errors introduced by design errors in the paired filters, finding that predictive errors are not increased over conventional methods. Fifth, we consider how the performance of the method is affected by light intensities that are not linearly related to chemical composition, and find that the method is only marginally affected. In summary, we conclude that many types of predictive measurements based upon use of regression vectors and linear mathematics can be performed more rapidly, more effectively, and at considerably lower cost by the proposed optical computation method than by traditional dispersive or interferometric instrumentation. Although our simulations have used Raman experimental data, the method is equally applicable to NIR, UV-Vis, IR, fluorescence and other spectroscopies.© (1998) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.