Calibration transfer between a benchtop FT-NIR and a handheld ultra-compact near infrared spectrometer for determination of gasoline quality parameters

Effective prediction of properties that guarantee the quality of gasoline can be obtained using near infrared (NIR) spectra and multivariate calibration models. The use of portable instruments for rapid field measurements (such as in fuel stations) is of increasing interest. In this work, the performance of a new handheld ultra-compact instrument (MicroNIR 1700, JDSU, CA, USA) was evaluated to predict important quality parameters of gasoline. Moreover, a reverse standardization (RS) technique was employed to transfer the calibration data set obtained with a benchtop instrument (master) to the MicroNIR (slave). RS transforms the spectra from the master instrument to resemble those from the slave instrument and then a model is built with the corrected spectra. Then this model can be directly used in the slave instrument. To perform this transformation, it is necessary to measure a number of representative samples (transfer samples) in both master and slave instruments. Models for five quality parameters (distillation temperatures at 10%, 50%, 90% and final boiling point (FBP) volume recovered – ASTM D86 and density – ASTM D1298) were built using partial least squares regression (PLS). One hundred and three gasoline samples were collected in Pernambuco State (Brazil) fuel stations. The transmittance spectra were acquired with a 20mm path length cuvette using the Frontier FT-NIR (PerkinElmer) and MicroNIRTM 1700 (JDSU) spectrometers. A labmade transmitance acessory was employed to acquire the spectra in the MicroNIR. From the pre-processing techniques tested (standard normal variate, multiplicative scatter correction and 1st derivative), the best results were obtained with base line correction. Calibration and external validation sets were composed by 70% and 30% of the samples, respectively. For the master instrument, root mean square error of cross validation (RMSECV) values were 0.863ºC (T10%), 0.544ºC (T50%), 2.797ºC (T90%), 3.060ºC (FBP) and 1.172 Kg.m-3 (density). For the external validation, the root mean square error of predictions (RMSEP) values were 0.399ºC (T10%), 0.367ºC (T50%), 1.765ºC (T90%), 2.445ºC (FBP), 0.810 Kg.m-3 (density). Twenty transfer samples, selected by the Kennard-Stone algorithm, were employed in RS procedure. The new models, built with the standardized spectra, were directly applied on the spectra of the external validation samples acquired in the slave instrument. RMSECV values obtained for these model were 0.994ºC (T10%), 0.650ºC (T50%), 3.628ºC (T90%), 3.853ºC (FBP) and 2.134 kg.m-3 (density). RMSEP values of 0.424ºC (T10%), 0.803ºC (T50%), 4.192ºC (T90%), 4.249ºC (FBP) and 2.586 Kg.m-3 (density). These RMSEP values, although obviously higher then the ones obtained with the master instrument, were all equivalent to reproducibility of the ASTM reference methods (3.2ºC - T10%, 1.88ºC - T50%, 4.1ºC - T90%, 6.78ºC - FBP and 1.9-3.2 Kg.m-3 - density). For comparison, models were also built using the data set acquired directly with the MicroNIR. The RMSEP vaules were 0.387ºC (T10%), 0.595ºC (T50%) , 6.021ºC (T90%), 4.869ºC (FBP) and 1.661 Kg.m-3 (density). RS procedure was successfully applied, showing that it is possible to transfer a spectral gasoline data set acquired with a high-resolution benchtop instrument to the portable MicroNIR.