M. Marjoniemi, CALIBRATION OF DYE SOLUTIONS WITH ARTIFICIAL NEURAL NETWORKS, The Journal of the American Leather Chemists Association, 89(2), 1994, pp. 39-48
In recent years artificial neural networks (ANN) have generated widesp
read interest and gained popularity. Numerous applications have been i
nvestigated, including pattern recognition, signal processing, process
control and modeling or calibrating. ANN as a nonlinear calibration m
ethod has been found to give superior results in calibration of nonlin
early behaving solutions. In the present study neural networks are app
lied to multivariate calibration of dye solutions using spectroscopic
data and for producing quantitative estimates of the concentrations of
a component in the mixture. Absorbance is linearly dependent on conce
ntration only in a narrow concentration range. There exists severe ove
rlapping of the absorption bands in multi-component dye solutions. The
solutions studied here were multi-component and the concentration ran
ge of the dye to be calibrated was wide (0-975 mg/l). Thus the absorba
nce of the solutions changes nonlinearly as a function of concentratio
n. The results are compared with the results obtained with other multi
variate calibration methods; principal component regression and partia
l least-squares regression. Metal complex dyes were used in making the
dye solutions. The absorbance spectra of the samples were measured in
the visible range of light. The sigmoid output function was used in t
he hidden layer of the network to perform nonlinear fitting.