Abaqus Smooth Particle Hydrodynamics
SPH. Smooth Particle Hydrodynamics. Long term readers of the blog will know that it’s something that TECHNIA has flirted with in the past. Not exactly a solution looking for a problem, but certainly one of those technologies which looks attractive and interesting enough for an initial hack around, before more pressing and above all more traditional work items replace it in the to do list.
TECHNIA has recently expanded and one of the newer team members has taken this technology by the proverbial scruff of the neck and applied it to a real project. And frankly the results are pretty impressive. 
“SPH technology, and its cousins CEL and DEM, has been embedded within Abaqus for some time now, offering the ability to model events where the gross deformation or motion of the mesh would lead to traditional Lagrangian elements giving up and going home long before the solution has reached the point that the question being asked could be answered. A consequence of this is that SPH and CEL offer the ability to include fluid-structure interactions into a FE problem relatively easily, where the effect of the fluid on the mechanical structures is required to accurately define the problem, but where coupling FE and CFD solutions is either overkill or the deformation or motion of the assembly make it impossible.
I’ve been investigating the possibilities – and limitations – of these ‘meshless methods’ for some time (depending on how argumentative you’re feeling at least one, and perhaps all, of them are not actually meshless. So, when the telephone rang with an enquiry which looked like it would lend itself perfectly to solution including SPH, a very rapid set of mental calculations were performed to check that there was, at least, no obvious technology gaps and there was a good chance of success before asking for more details.
The problem: BOS Solutions, as part of its range of services, supply decanter centrifuges to the oil & gas market for the separation of drilling mud into solid and liquid components. The machine works by accelerating an axial flow of the two phase fluid radially into a rapidly rotating vessel. Under centrifugal loading, the heavier solids are caked onto the inside of the vessel which are then scraped off and out of one end of the machine by a fluted screw, while the liquid is delivered to the other end of the machine – all as a continuous process (if you’re having trouble visualising that, search YouTube for ‘decanter centrifuge’ which will yield some helpful animations of similar machines in action). The velocities and forces involved lead to high wear of the mechanical components. So, could a numerical model be used to simulate the process and provide some prediction of the regions of wear to illuminate design process and so minimise the need for expensive prototyping (oh, and do it ‘blind’, so the experimental wear patterns validate the model and don’t become part of the model ‘optimisation’). Interesting? Yes. Challenging? Certainly. Possible……?
To meet the ultimate requirement, the problem was broken down into two phases. The first phase involved modeling the flow of the fluid through the centrifuge with sufficient fidelity that the pressures and flow velocities on the surfaces of the machine could be predicted. Due to the motion of the parts, this didn’t lend itself to a coupled FE/CFD approach and – of the available technologies within Abaqus – looked best suited to SPH. I’d like to say this is where I cut a long story short and go to the final model, but the reality was that the ‘story’ wasn’t that long anyway as the ability for Abaqus to automatically do things when including SPH in a model, such as handle the interaction between the particles and Lagrangian FE mesh, meant the model definition was relatively straight forward. The animation below shows a transverse section through the model as the machine rotates and the fluid flows through it.
With this done, the next challenge was to use the available output variables from the model to predict the regions where high wear was likely to occur (funnily enough, ‘wear’ isn’t a built-in field variable in Abaqus). To do this, a Python script was developed to work out the regions on the surface of the machine where contact had occurred between any particle and that surface and, if contact had occurred, extract the normal contact force and the slip velocity of the particle over the surface. The amount of wear for that contact was determined as the product of the contact force and the slip velocity. Sum this parameter for each output frame of the analysis and normalise it and – assuming the solution has reached a quasi steady-state condition – this normalised wear parameter can plotted on the model to provide a qualitative indicator of the relative amount of wear occurring on the surface of the machine. Simple… ish! 
With the predictions made, the acid test was comparison with the observed wear regions in the real machine. Did it work? In short, yes, it did. The correlation with the observed wear was remarkable given the variability that occurs in the machine compared to the limited range of effects that could realistically be included the model. More importantly, perhaps, a greater understanding of how the flow of the fluid through the machine influences wear in particular regions has been gained from the model than could be extracted from physical testing on its own. This, combined with the fact that the turn-around time for the model is days, not months for a physical testing programme, makes the tools developed and the impact the knowledge will have on future machine designs incredibly valuable to BOS Solutions.
As Laurence said at the start then, not exactly a solution looking for a problem – more a case of finding the right tool for the right job in an ever expanding virtual toolbox”.
