Standardisation of near infrared spectrometers using artificial neural networks

The use of chemometrics procedures with near infrared spectroscopic data to produce calibration equations for analytical chemistry has been very successful, A large increase in prediction error is observed when the calibration equation developed on one instrument is used directly on another. Since many spectral differences can exist between two spectrometers, a standardisation procedure is a requirement for the long-term use of quantitative or qualitative models, In this work, an original neural network approach is proposed in order to correct for spectral differences. Spectral response of a given instrument is modelled from another before the use of the calibration equations. In this way, the time-consuming step of recalibration for the second spectrometer is avoided and the initial error prediction level is retrieved.