Roelof L. J. Coetzer

A Further Study of Inconsistencies between Autoignition and Knock Intensity in the CFR Octane Rating Engine

Careful consideration of the development and operation of the ASTM knock detection system on the Cooperative Fuels Research (CFR) octane rating engine has shown that the pressure fluctuations, brought about by autoignition of the end-gas, do not contribute to measurement of knock intensity. The analyses of a variety of fuels at standard knock intensity revealed that knock intensity measured on the CFR engine is related to the rate of change of pressure prior to knocking and is consistent with the description and operation of, not only the original bouncing pin, but also the modern day electronic CFR knock measurement system. It was concluded that the use of octane number data to directly infer information about the autoignition behaviour of fuels should be done with caution.

Response surface strategies in constructing statistical bubble flow models for the development of a novel bubble column simulation approach

Abstract Bubble columns represent a substantial contribution to chemical industrial equipment and due to their complex hydrodynamics are difficult and computationally expensive to simulate. A model of the flow structure around a single bubble is required for the development of a computationally efficient alternative model. The flow structure data obtained from CFD is highly non-linear, which lends itself to a Design and Analysis of Computer Experiments (DACE) approach. The situation differs from conventional DACE applications since high resolution data is summarized, which allows different sampling criteria, correlation functions and optimization criteria to be evaluated in an assessable manner. A modified powered exponential correlation function is introduced, which in conjunction with a gradient based space-filling design resulted in an averaged 8 times smaller mean squared error compared to other correlation design combinations. Furthermore, sequentially fitting the data is shown to produce the best overall fits, when used in combination with space-filling designs.

Optimal Designs for Multiple-Mixture by Process Variable Experiments

In industry many processes include mixture components some or all of which are themselves made up of other sub-components and typically these variables are subject to lower and upper constraints. Process variables which may change the effect of the blending properties of the mixture components on the response of interest may also be involved. In this paper response surface models in more than two sets of mixture variables, including multiple-mixtures and sub-compositions, in combination with process variables are developed and attendant D-optimal designs are constructed. The work is motivated by an industrial problem involving the modelling of a coal gasification process.

Simulation of a coal stacking process using an online X-Ray Fluorescence analyser

The Sasol Coal Value Chain is a complex system consisting of blending, stacking and reclaiming of no fewer than six different coal sources with vastly different coal qualities. The amount and quality of the gas produced from coal depend crucially on the quality of the coal reclaimed from the coal stacking yards. In this paper the development of a real time coal quality simulation model using information from an online X-Ray Fluorescence analyser, integrated with various data sources from the Coal Supply Facility, is presented. The integration of different data sources is discussed to create a centralised and standardised data framework for input to the simulation model. The simulation of a heap profile of the coal quality for each heap stacked, together with the quality of the reclaimed coal, is discussed in detail. It is shown how the generated information from the model is utilised in the development of a reclaiming strategy.

Optimal Design for Compositional Data

In this paper optimal designs for experiments involving compositional data, specifically locally D-optimal designs for the additive logistic normal model and locally D S -optimal designs for Dirichlet regression, are investigated. The theory underpinning the construction of these designs is based on the appropriate information matrices and the development, while new, is relatively straightforward. The ideas are illustrated by means of a simple example, that of two consecutive reactions.