Optimization of the multivariate calibration of a Vis–NIR sensor for the on-line monitoring of marine diesel engine lubricating oil by variable selection methods

• Optimization process of a multivariate calibration of a new Vis–NIR sensor
• Variable selection methods were applied to obtain the finest calibration model.
• The sensor was developed for monitoring of marine diesel engine lubricating oil.
• The sensor measures the insoluble content in pentane of lubricating oil.

This paper deals with the description of the optimization by variable selection methods of the multivariate calibration process of a low-cost Visible–Near Infrared (400–1100 nm) sensor, developed for the on-line monitoring of the insoluble content in diesel marine engine lubricating oil. The performance of the calibration model developed for the Vis–NIR sensor was compared with the performance of the calibration model developed with spectra obtained with a UV/Vis–NIR laboratory spectrometer. The calibration results obtained with the two devices were compared to determine the limitations of the sensor system with respect to the laboratory equipment. First, the spectra were correlated with the insoluble content analyzed in Wearcheckiberica's oil laboratories obtaining a calibration model based on Partial Least Squares-regression (PLSR). Once the pre-processing strategy had been defined, the most significant predictor variables were chosen with the help of Martens uncertainty test, interval Partial Least Squares (iPLS) and Genetic Algorithms (GA) variable selection techniques. Finally, the two models were compared based on the number of latent variables of each model of the values of the Root Mean Square Error of the Cross Validation (RMSECV), the Standard Error of Performance (SECV) and the Ratio of Prediction to Deviation (RPD).