Pattern recognition techniques for the classification of wastewater samples based on their UV-absorption spectra and their fractions after applying MW-fractionation techniques

Different category-wastewater samples were collected from the inlets of biological treatment plants (installed in hospitals, industries and municipality) and from the body of polluted natural surface water (PNSW) systems (lakes, rivers). UV absorption spectral data of each wastewater system or their fractions as obtained by fractionation with membranes of various pore openings or with gel permeation chromatography were treated by supervised (neural network) and unsupervised (cluster analysis) pattern recognition methods with the target to classify them in clusters that include exclusively samples of the same category. The results based on neural network method applied to log10(UV-absorption spectra) of 80 wastewater samples gave a prediction score of around 77% for all category-samples. The cluster analysis method applied to the 1st derivative of log10(UV-absorption spectra) of 79 wastewater samples gave a promising classification for one of the four category wastewater samples and the others were grouped in sub-clusters of the same cluster without clear separation. Fractionation through membrane dialysis of two extremely non-similar UV-spectra samples from each category showed that the cluster analysis was more successful when UV-absorption spectra of the high molecular weight fractions were used in cluster analysis. Fractionation with GPC-chromatography gave chromatographic peaks and peak-spectra that are different for each category of wastewater samples; this method revealed that the MW of absorbing species are different and the absorption intensities are significantly different between the inlet feeds of the three types of wastewater treatment plants.

From this investigation it is found that: (1) Pattern recognition techniques (cluster analysis and neural network) can give a rather good prediction score of the category of wastewater rapidly, just by obtaining and normalizing the UV-spectrum of the wastewater sample in question. (2) Cluster analysis of the normalized UV-spectra of wastewater category samples can also be used to remove outlier samples before the use of neural network technique for better prediction score. (3) Widely different samples of the same category treated by cluster analysis can be submitted to a fractionation technique combined with cluster analysis. (4) Cluster analysis of 1st derivative log10 or normalized (UV-absorption spectra) from fractions of higher molecular weight give better classification of the different category samples. (5) Fractionation with membrane dialysis was time consuming process and insufficient for an efficient fractionation. (6) GPC-fractionation is able to predict the category just from the GPC-chromatogram as well as from UV spectra of the fractions obtained.