Multivariate analysis on the influence of physicochemical parameters on the calorific value of oils used in the petrochemical industry

Multivariate statistical techniques were applied to identify the behavior of physicochemical parameters on calorific value in oil samples used in the industrial petrochemical sector of Mamonal in Cartagena de indias, Colombian Caribbean. Initially a physicochemical characterization of the samples was carried out through a principal components analysis, then a cluster analysis to identify similarities between the samples and to have a better description of the samples and finally a multiple regression model is fitted to determine the physicochemical factors that influence In the behavior of the calorific value of the oil samples. The results show that there are different associations between the physicochemical parameters; The cluster analysis yielded 4 groups of oil samples with different physicochemical behaviors and according to the criterion of loss of AIC information in the linear regression model, samples with low sulfur, sodium and ash content and high zinc content, Phosphorus and chlorine contribute to the increase of calorific value in the samples. KeywordOil reused, Principal components analysis, Heat power, Physicochemical variables.


II. EXPERIMENTAL Population and sample
The population on which it was worked are used oils from the industrial sector of Mamonal in Cartagena de indias. The statistical sample corresponds to 118 samples selected from 9 types of oil used, as described in Table  1 below. Statistic analysis Statistical analysis of the information is carried out in the following order: Initially, the descriptive measures (mean, median, mode, maximum, minimum and coefficient of variation) of the physicochemical parameters in the oil samples are calculated and then obtained The main component analysis (PCA), which seeks to measure associations between quantitative variables, and similarities between individuals with respect to these variables. From the analysis of main components, a cluster analysis is performed to identify the oil samples that present similar behaviors with respect to the physicochemical parameters and finally a multiple regression model is performed to explain the behavior of the calorific value of the oils used In order to identify the parameters that influence the increase in the calorific value of the oil samples, the Akaike information criterion (AIC) was used for them, which yields the best regression model in terms of Loss of information.
III. Table 2 contains the information of the descriptive statistics for the physicochemical parameters. The descriptive statistics show that the average calorific value of the oil samples is 11123.82, with little variability in the samples (CV = 9.51%), there is absence of some heavy metals in the oils; Such as copper, as well as sediments. It can be evidenced that there are samples with high content of iron, zinc and chlorine (maximum values of 1521.42, 1015.87 and 709.45 respectively). In general there is a lot of dispersion of the samples with respect to the different parameters, since there are coefficients of variation superior to 57%, even reaching values above 100%. The detected variations indicate that the samples present dispersed physicochemical behavior, so that there may be some with better conditions for effects of calorific power.

Principal component analysis
The correlation matrix between the physicochemical parameters for the oil samples is shown below. According to Table 3, important associations between the different parameters are observed. Wet oil samples have high sediment and ash content (correlations of 0.81 and 8.86 respectively). The sulfur content in the oil samples also shows an important and positive association with the zinc, copper, phosphorus, chlorine, calcium and ash content (correlations higher than 0.5). There is a positive association between the presence of copper and the sodium and zinc content (0.71 and 0.55), indicating that oil samples with a high copper content have high sodium and zinc contents. The phosphorus content also has strong positive linear associations with copper, silicon, chlorine and calcium contents (0.63, 0.55, 0.52 and 0.63 respectively). Oil samples with high ash presence are characterized by high zinc, copper and sediment content; While the density of the oils is directly related to the presence of zinc and sediments (correlations of 0.62, respectively). Table 4 shows that the first 4 factor axes accumulate 89% of the total variability of the information. Therefore, classification analysis is performed considering these axes in the analysis of principal components. Next, the resulting dendogram of Figure 1 is observed with the respective clusters. The dendogram shows 4 clusters, with respect to the behavior of the physicochemical parameters that were observed in them. The first cluster consists of samples 8, 3 and 4 (Oil samples applied to engines); The second for sample 1 (oil applied to double-shift dump trucks), the third for sample 9, 6 and 7 (oils used for the engine and forklift engine) and the last for samples 2 and 5 (oils used in engine Gearbox and Marine Diesel engine respectively).  According to the factorial plane and the direct data, the samples belonging to the first group present low levels of sulfur, iron, sodium, copper, silicon, phosphorus, calcium; As well as few ash and sediment residues. Group 2 is formed only by sample 1, this sample has low sodium level; While having high levels of silicon, sediment, ash and low humidity. Samples 9, 7 and 6 have high levels of sulfur, iron, sodium, zinc, copper, copper, phosphorus, chlorine, calcium and ash; As well as high humidity. The latter group consists of samples 2 and 5, which low levels of zinc and high levels of iron, sodium, silicon, phosphorus, chlorine and ash.
In Table 5 and according to the AIC criterion, the model that best explains the behavior of the calorific value according to the physicochemical parameters is the following one. p-value 0,0005*** 0,0241** 0,0814* 0,0468** 0,0161** 0,1629 0,0941* ***: Very significant, **: Highly significant, *: Significant The coefficients of the parameters that were significant to model the behavior of the calorific value of the samples, in terms of the AIC criterion. The results indicate that oil samples with low content of sulfur, sodium, ash and high zinc content, phosphorus, chlorine; Have a higher calorific value. The R2 adjusted for this model is 0.9227, indicating that 92.27% of the variability of calorific power is explained by the variables included in the model; While the p-value of the analysis of variance is 0.045, showing that the model is significant; That is, the explanatory variables considered in the yes influence the calorific value. Accordingly, samples 3, 4 and 8 are the closest to this ideal behavior, which correspond to oils used in engines.
Then, in Table 6, the calorific value of the samples with the forecast according to the model is shown. It is observed that group 1 (Samples 3, 4 and 8) are those that present the highest forecasts for calorific value and their real values are also within the highest. This indicates that the proposed model provides consistent and near-practical results.
III. CONCLUSION According to the results obtained, the following conclusions are observed in the physicochemical behavior of used oils and their influence on calorific value in the Cartagena industry: 1) Descriptive statistics show a high variability of the samples with respect to the parameters Physicochemical; 2) The most important associations are observed between the content of ash and moisture, sulfur, zinc and copper; As well as between copper, sodium and sulfur and between calcium and iron, among others; 3) Cluster analysis yielded four groups of oil samples; 4) According to Akaike's criterion, the parameters influencing the calorific value are sulfur, sodium, ash, zinc, phosphorus and chlorine; Which explain 92.27% of the calorific power behavior; 5) The samples of oil used in internal combustion engines, are those that present greater calorific power and those according to the adjusted model have the highest forecasts.
IV. REFERENCES