Development of a Robust Calibration Model for Nonlinear In-Line Process Data

Abstract

A comparative study involving a global linear method (partial least squares), a local linear method (locally weighted regression), and a nonlinear method (neural networks) has been performed in order to implement a calibration model on an industrial process. The models were designed to predict the water content in a reactor during a distillation process, using in-line measurements from a near-infrared analyzer. Curved effects due to changes in temperature and variations between the different batches make the problem particularly challenging. The influence of spectral range selection and data preprocessing has been studied. With each calibration method, specific procedures have been applied to promote model robustness. In particular, the use of a monitoring set with neural networks does not always prevent overfitting. Therefore, we developed a model selection criterion based on the determination of the median of monitoring error over replicate trials. The back-propagation neural network models selected were found to outperform the other methods on independent test data.

Transferência de calibração em métodos multivariados

“…Adicionalmente, para representar tais efeitos, os modelos globais tendem a ser de grandes dimensões, tornando-se, portanto, mais complexos que os modelos locais. Devese salientar que, se um determinado conjunto de amostras não foi substancialmente afetado pelas fontes de variações em questão, um modelo local tenderia a gerar melhores predições que o modelo global 5,23 . Contudo, o modelo global é mais robusto no sentido de que suas predições continuariam confiáveis em uma gama mais ampla de situações.…”

“…Pode-se também recorrer a técnicas de planejamento experimental como Convergência de Instrumentos ("Instrument Matching") 4 e Atualização de Modelos 5 . Para que essas soluções funcionem, porém, é necessário que o analista tenha condições de prever futuras fontes de variação e controlar os parâmetros experimentais.…”

Multivariate calibration transfer employing variable selection and subagging

“…(ii) Enhancement of model robustness. This approach encompasses the development of global models by inclusion of all the relevant sources of variation in the calibration data set, 5 as well as the use of preprocessing techniques to reduce the data variability that is not correlated to the y-property of interest. Examples of such techniques include baseline correction procedures, 6 Multiplicative Signal Correction (MSC), 7 Finite Impulse Response (FIR) filters, 8 Orthogonal Signal Correction (OSC), 9 wavelet decompositions, 10 Transfer by Orthogonal Projections (TOP) 11 and variable selection algorithms.…”

Bayesian linear regression and variable selection for spectroscopic calibration

“…These research have materialized in both enhanced pre-processing algorithms (such as extended multiplicative signal correction [6], extended inverted signal correction [7] and direct orthogonal signal correction [8]) and improved regression models (including locally weighted regression [9], neural networks [9,10], support vector machine [11] and Gaussian process [12]). This paper is mainly focused on the development of advanced regression models for calibration, and thus the pre-processing techniques are not discussed further.…”