Rugged spectroscopic calibration for process control

Multivariate spectroscopic calibration is now finding increased industrial use in the determination of mixture composition and product quality. Typically, these applications involve measurement of batch processes or process streams by UV, visible, near-infrared or infrared spectroscopy, followed by prediction of product composition or quality with multiple linear regression or partial least squares calibration models. Real time predictions of composition or quality measures may then be used to control the process to increase efficiency, purity, etc. One obstacle that limits widespread use of this strategy is the lack of calibration model ruggedness. Lack of ruggedness in calibration models may manifest itself in the form of large prediction errors following small perturbations in instrument response or slight changes in the sample system, fiber-optic probe, or process stream composition. In this paper, we describe a strategy for developing rugged calibration models using artificial neural networks and demonstrate the method on several NIR process data sets. Fourier transform or principal component preprocessing was used to reduce noise and the number of input measurements per sample. A large number of neural networks with different network architectures and random initializations were trained to predict composition using the pre-processed data. A sensitivity analysis was performed with monitoring data sets to screen the resulting networks for ones that were insensitive to simulated wavelength calibration errors, baseline offsets, path length changes or high levels of stray light. External validation data sets were used to demonstrate the ruggedness of selected neural network calibration models.