No abstract

Abstract. Water contains many chemical, physical, and biological impurities. Some impurities are benign while others are toxic. The quality of water is defined in terms of its physical, chemical, and biological parameters and ascertaining its quality is crucial before use for various intended purposes such as potable water, agricultural, industrial, etc. Various water analysis methods are employed to determine water quality parameters such as DO, COD, BOD, pH, TDS, salinity, chlorophyll-a, coli form, and organic contaminants such as pesticides. The list of potential water contaminants is exhaustive and impractical to test for in its entirety. Such water testing is sometimes costly and time consuming. This paper attempts to present application of data mining technique to build a model to predict a widely used gross water quality parameter called Biochemical oxygen demand (BOD). BOD is a measure of the amount of dissolved oxygen used by microbial oxidation of organic matter in wastewater. The standard method for measuring BOD is a 5-day process. Dilution of sample, constant pH and nutrient content besides the temperature of 20°C and dark area are required for correct results. High levels of nitrogen compounds yield false BOD results. Winkler titration which is also used to measure BOD is a chemical intensive process. Hence an automatic prediction model for BOD has been sought for accurate, cost-effective and time saving measurement. Based on data available for BOD measurements, this paper describes the development of a prediction model for BOD using a technique of data mining, namely, support vector machines (SVM). A correlation coefficient of 0.9471 and RMSE of 0.5019 was obtained for the BOD prediction model on river water quality data. The performance of the proposed model was also compared with two other models namely artificial neural network (ANN) and regression by discretization. Simulation results show that the proposed model performs better than the other two in terms of correlation coefficient and RMSE.

Multipass heat exchangers are often designed by using the rule of thumb FT⩾0.75, which is rather arbitrary. FT falls sharply with the increase in temperature cross. Hence, only a limited temperature cross can be allowed. The ability to accommodate temperature cross increases rapidly as the number of shell passes is increased. Though many investigators have emphasized the importance of temperature cross in exchanger design, it has as yet not been explicitly accounted for in the design. This paper introduces a new approach for estimating the number shells in a shell and tube exchanger which directly accounts for temperature cross, rather than routing this effect through FT or XP (Ahmad et al.’s parameter, which is again a correction factor not directly related to temperature cross). The approach is compatible with the established design procedures and bypasses the FT. It generates better designs by defining maximum permissible temperature cross, than the traditional designs based on specifying minimum permissible FT. Expressions have also been provided which correlate the present formulation with that of Ahmad et al. [S0022-1481(00)00803-3]

Abstract-Dead oil viscosity is a critical design factor for oilfields and refineries. From available literature, crude oil viscosity is found to be a strong function of pressure, temperature, bubble-point pressure, gas-oil ratio, gas gravity, and oil gravity. Oil viscosity is generally determined from laboratory experiments and empirically derived correlations.Reliable measurements of dead oil viscosity are difficult to obtain due to lack of lab equipment or liquid samples. Based on API oil gravity & temperature, various correlations have been used to predict dead oil viscosity. In addition to correlations, recently data mining techniques like Artificial Neural Networks (ANN) and Support Vector Machines (SVM) have been used to predict petroleum viscosity. The aim of this paper is to introduce the ensemble model of bagging as an important data mining technique to predict dead oil viscosity. The ensemble model predicted the viscosity accurately with a correlation coefficient of 0.99, an accuracy that is comparable to that of ANN as found in literature. It was also observed that bagging lowered the relative error of the base classifier (ANN) from 10% to about 8%, thereby stabilizing the ANN while retaining its accuracy.