Design and Testing of a Multivariate Optical Element: The First Demonstration of Multivariate Optical Computing for Predictive Spectroscopy

Abstract

A demonstration of multivariate optical computing is presented using binary dye mixtures consisting of Bismarck Brown and Crystal Violet. Bismarck Brown was treated as the analyte, while Crystal Violet was treated as a random interfering species. First, a multilayer multivariate optical element (MOE) for the determination of Bismarck Brown was designed using a novel nonlinear optimization algorithm. Next, the MOE was fabricated by depositing alternating layers of two metal oxide films (Nb2O5 and SiO2) on a BK-7 glass substrate via reactive magnetron sputtering. Finally, the MOE was tested on 39 binary dye mixtures using a simple T-format prototype instrument constructed for this purpose. For each sample, measurements of the difference between transmittance through the MOE, and the reflectance from the MOE were made. By setting aside some of the samples for instrument calibration and then using the calibration model to predict the remaining samples, a standard error of prediction of 0.69 μM was obtained for Bismarck Brown using a linear regression model.

Precision in imaging multivariate optical computing

“…(2). A small gain value is desirable since this indicates an IMOE that is more sensitive to the analyte.…”

“…6 and is illustrated in Ref. 2, in which the IMOE is used in a 45°configuration, and both reflection and transmission are recorded simultaneously. This configuration has the disadvantages of constraining the design of the IMOE and enhancing model errors, but otherwise is most directly related to the calculations of Ref.…”

“…1,2 In chemistry this is typically done to determine the quantity of a given compound within a mixture of other interfering compounds, where the analyte and interferents have overlapping optical spectra in the spectral region used for the quantification.…”

“…The unscaled regression vector can then be described in terms of T Ϫ R, where T and R are the spectral transmission and reflection functions of the MOE. 2 If we include a gain or scaling factor, G, and an offset, O, to increase the design space, the MOC calculation corresponding to Eq. (1) is given by…”

Applications of integrated sensing and processing in spectroscopic imaging and sensing

“…These specific MOEs have been termed molecular optical filters. The second approach involves the deposition of alternating layers of two metal oxide films, Nb 2 O 5 and SiO 2 , on a 2.54 cm BK-7 glass substrate with reactive magnetron sputtering [28]. The interference effects of the layers are monitored as the layers are deposited to be sure the end product resembles the target transmission spectrum.…”