From pharmaceutical powders to agricultural grains, a great proportion of the materials handled in industrial situations are granular or particulate in nature. The variety of stesses that the matierals may experience and the resulting bulk behaviours may be complex. In agricultural engineering, a better understanding into agricultural processes such as seeding, harvesting, transporting and storing will help to improve the handling of agricultural grains with optimised solutions. A detailed understanding of a granular system is crucial when attempting to model a system, whether it is on a micro (particle) or macro (bulk) scale. As numerical capabilities are ever increasing, the Discrete Element Method (DEM) is becoming an increasingly popular numerical technique for computing the behaviour of discrete particels for both industrial and scientific applications. A look into the literature shows a lack of validation of what DEM can predict, specifically with respect to bulk behaviour. In addition, when validation studies are conducted, discrepancies between bulk responses in physical tests and numerical predictions using measured particles properties may arise. The aire of this research is to develop a methodology to calibrate DEM models for agricultural grains using data meaured in bulk physical tests. The methodology will have a wider application to granular solids in general and will advance understanding in the area of DEM model calibration. A contrasting set of granular materials were used to develop the methodology including 3 inorganic solids (single and paired glass beads, and polyethylene terephthalate pellets) and two organic materials (black eyes beans and black kidney beans). The developed methodology consists of three steps: 1. The development of bulk physical tests to measure the bulk responses that will be used to calibrate the DEM models, 2. The creation of the numerical dataset that will describe how the DEM input parameters influence the bulk responses , and 3. The optimisation of the DEM parameters using a searching algorithm and the results from Step 1 and 2. Two laboratory devices were developed to provide calibration data for the proposed methodology: a rotating drum and an confined compression test. These devices were chosen as they can produce bulk responses that are repeatable and easy to quantify, as well as generate discriminating results in numerical simulations when DEM parameters are varied. The bulk response determined from the rotating drum device was the dynamic angle of repose Ør formed when the granular material in a 40% filled drum is rotating at a speed of 7 rpm. the confined compression apparatus was used to determine the bulk stiffness of a system by monitoring the change in void ratio from the stress applied during a loading and unloading cycle. The gradient of the loading and unloadng curves termed λ and κ respectively were chosen as the bulk responses to calibrate the DEM models. The experimental results revealed that the dynamic Ør was significantly influences by the particle aspect ration and boundary conditions. The stiffness parameters were found to be predominantly influences by the initial packing arrangement. The numerical dataset describing how the DEM input parameters influence the numerical bulk responses was created by simulating the bulk physical tests, varying selected DEM parameters and monitoring the effects on bulk parameters. To limit the number of simulations required, design of experiment (DOE) methods were used to determine a reduced factorial matrix of simulations. In additions, an extensive parametric investigation on the non-optimised parameters as well as a scaling sensitivity study was carried out. The final step in determining the optimised parameters is to use a searching algorithm to infer the DEM parameters based on the numerical dataset and used the experimental results as calibration data. To perform a comparative study, tow searching algorithms were explored: the first was a simple method based on Microsoft Excel's Solver algorithm coupled with a weighted inverse distance method. The second made used of the statistical analysis program Statistica. It was shown that the Excel Solver algorithm is simpler and quicker to use but for the present first implementation, could only perform an optimisation based on two bulk responses. Statistica required the creation of a staistical model based on the numerical dataset before using the profiling and desirability searching technique, but was able to optimise the parameter using all three bulk responses. A verification and validation of the optimisation methodology was conducted using the optimised parameters for the black eyed beans. A verification was cnducted by simulating the two calibration experiments using the optimsed parameters and comparing these with the experiments. In addition, a validation was peformed by predicting the response of ta shallow footing penetration on a bed of black eyed beans. It was found that DEM simulations using optimised parameters predicted vertical stress on the footing during penetration to an acceptable degree of accuracy for industrial applications (<10%) at penetration depths up to 30mm.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:562997 |
| Date | January 2010 |
| Creators | Johnstone, Mical William |
| Contributors | Ooi, Jin Y. |
| Publisher | University of Edinburgh |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://hdl.handle.net/1842/4655 |
Page generated in 0.002 seconds