partial least squares regression

Nowadays, due to the evaluation and high costs of maintenance and repair of sugarcane harvesting machines, it is necessary to monitor sugarcane harvester hydraulic oil using a faster and non-destructive method to determine contamination and TAN index. In this research, the ability of the visible spectroscopy method to non-destructively measure and predict the water content and TAN index in harvester Austoft 7000 hydraulic oil samples at different operating hours was investigated. For this purpose, spectra were taken from the samples in the spectral region of 400-780 nm. Multivariate Partial Least Squares (PLS) regression models were developed based on reference measurements and pre-processed spectra information by combining different pre-processing (Moving Average, Savitzky-Golay, Standard normal variate and First Derivative) methods to measure and to predict the water content and TAN index of hydraulic oil.  The results showed that the visible spectroscopy method could be used for quick and non-destructive measurement of water content and TAN index at different operating hours of harvester Austoft 7000 hydraulic oil. The best prediction results of water content in hydraulic oil were obtained with PLS model based on moving average (MA) preprocessing method (rcv=0.96, RMSECV=1.86, rp=0.89 and RMSEP=3.18), which had excellent accuracy (SDR=3.12). On the other hand, the PLS model based on the combination of moving average preprocessing and standard normal distribution (MA+SNV) was able to predict the TAN index with excellent accuracy (SDR=3.1) (rcv=0.94, RMSECV=0.007, rp=0.89 and RMSEP= 0.010). Therefore, the application of visible spectroscopy technology in agriculture and industries can be recommended for rapid monitoring of hydraulic oil quality and with the aim of controlling pollution.

The aims of this study were to determine the distribution of cadmium and lead in topsoil in Bahar city of Hamadan and to evaluate pollution of these metals in agricultural lands which cultivated potato. Therefore, systematic sampling was used, soil samples were taken from depth 0-45 cm. Visible and near-infrared (VIS-NIR) spectroscopy was provided as a quick and accurate method in 370 -2000 nm range. The reflectance spectrum of 95 soil samples were collected to estimate the concentration of cadmium and lead. To reduce noise on soil spectra, MSC, SNV and derivative preprocessing methods were used. The results showed contamination usually occurs due to combinations of sewage irrigation, it accumulates in the topsoil. Prediction of heavy metals were estimated by partial least squares regression and artificial neural network (R2PLSR = 0/90, R2BPNN = 0/95).