Flame propagation in dust clouds. Numerical simulation and experimental investigation

Abstract

This dissertation describes the development and validation of a methodology for estimating the consequences of accidental dust explosions in complex geometries. The approach adopted entails the use of results from standardized tests in 20-litre explosion vessels as input to the combustion model in a computational fluid dynamics (CFD) code, and the subsequent validation of the model system by comparing with results from laboratory and large-scale experiments. The PhD project includes dedicated laboratory experiments designed to explore selected aspects of flame propagation in dust clouds, and to reveal similarities and differences between flame propagation in gaseous mixtures and mechanical suspensions of combustible powder in air. The research project represents a continuation of numerous efforts by various research groups, where the key underlying problem has been the scaling of results obtained in laboratory tests for predicting the consequences of dust explosion scenarios in industry. The traditional approach to the scaling problem entails the use of empirical correlations, typically represented as nomographs or formulas in relevant safety standards. It is generally accepted that empirical correlations may work reasonably well for simple geometries, such as isolated process vessels and silos. The need for more sophisticated methods arises for accident scenarios that involve complex geometrical boundary conditions, such as flame propagation in connected vessel systems and secondary dust explosions inside buildings. The European Commission (EC) supported the Dust Explosion Simulation Code (DESC) project under the Fifth Framework Programme. The goal was to develop and validate a CFD code for simulating industrial dust explosions in complex geometries. To this end, GexCon created the CFD code DESC (Dust Explosion Simulation Code) by modifying the existing CFD code FLACS (FLame ACceleration Simulator), originally developed for simulating gas explosions in congested offshore geometries. The specific contributions from the candidate with respect to the development of the CFD software is limited to the methodology for estimating combustion parameters for a given dust sample from experimental results, the validation of the resulting model system against experimental data, and general participation in the R&D team during the development process. The modelling of particle-laden flow and heterogeneous combustion in the CFD code DESC involves several simplifying assumptions. The flow model assumes thermal and kinetic equilibrium between the dispersed particles and the continuous phase, and the k-ε turbulence model in FLACS remains unchanged for multiphase flows. The empirical correlation for the turbulent burning velocity in dust clouds originates from experiments with premixed combustion in gaseous mixtures. The fraction of dust that takes part in the combustion reactions, as function of the nominal dust concentration, is estimated from the explosion pressures measured in a constant volume explosion vessel. The thermodynamic data available in FLACS limit the application area to materials containing the elements carbon, hydrogen, oxygen, nitrogen and sulphur. The simplifications limit the application area of DESC to certain classes of materials, and flame propagation in dust clouds with relatively high reactivity. DESC do not contain models for simulating phenomena such as agglomeration, gravitational settling, and selective separation of particles in flow through cyclones or along other curved paths. In spite of the simplicity of the model system, the results from the validation work show that the CFD code DESC can describe the course of dust explosions in relatively complex geometries with reasonable accuracy relative to the inherent spread in the experimental results. The results obtained for silo explosions reproduce trends observed for variation in vent area and ignition position from various experiments. Results obtained for flame propagation sustained by dust dispersion from a layer indicate that the empirical model for dust lifting in DESC is suitable for the purpose. Results obtained for dust explosions vented through ducts reproduce the experimental trends fairly well. Simulations of dust explosions in a system of two vented vessels connected by a pipe with a 90° bend indicate that the DESC can reproduce relatively complex chains of events, including dust lifting from a layer. The results for the connected vessel system also demonstrate how sensitive the results can be with respect to modest changes in the initial and boundary conditions. Finally, simulations of explosion experiments in elongated vessels with repeated obstacles reproduce the experimental trends fairly well. Although the results from the validation work indicate that CFD simulations can become a valuable tool for consequence modelling and design of industrial facilities, the modelling in DESC requires further improvements. An essential improvement entails fundamental changes to the numerical solver to reduce in the influence of the grid resolution on the results from the simulations. In the current versions of FLACS and DESC, simulation of explosion scenarios is subject to strict grid guidelines. The current versions of both codes use a structured Cartesian grid, with limited possibilities for local grid refinement. This poses a particular challenge for DESC, since the grid resolution required to resolve complex internal geometries on a structured Cartesian grid varies significantly from case to case. The long-term solution to these challenges will presumably entail the use of adaptive mesh refinement (AMR), and this is outside the scope of the present work. The model system may also benefit from various other improvements, such as turbulent burning velocity correlations specifically developed for dust explosions, an explicit model for turbulent flame thickness, radiation models, local grid refinement in the region where ignition occurs, reduced dependence on empirical input to the model system, and in general more realistic modelling of particleladen flow and heterogeneous combustion. There is, however, a fine balance between the level of detailed information that must be specified in the model, and the applicability and user-friendliness of the model system. For most industrial applications of a CFD tool for dust explosions, there are significant inherent uncertainties associated with initial and boundary conditions. Dust explosion experiments in transparent balloons show that the initial phase of flame propagation in turbulent dust clouds can progress in a distributed manner, with very limited energy output. This observation may explain some of the challenges associated with the analysis of pressure-time histories from 20-litre explosion vessels for dust explosions when using a weak ignition source. Experiments in a 3.6-m flame acceleration tube demonstrate the importance of explosion-generated turbulence for dust explosions, and illustrate the challenge associated with poor repeatability in dust explosion experiments. The results obtained for propane-air mixtures in the same apparatus indicate that FLACS under-predicts the rate of combustion for turbulent flame propagation in fuel-rich propane-air mixtures. The CFD code DESC represents a significant step forward for process safety related to dust explosions in the process industry. There is, however, significant room for further improvements to the model system, and dedicated experiments will play an important role for the future development of the code. Improved safety in the process industry requires reliable and well-documented consequence models, and future development of DESC should include an integrated framework for model validation, including verification and testing.