Scavenging Performance of a Two-Stroke Opposed-Piston Diesel Engine

Click the arrow below to view the video. The presentation is divided into two parts, so once you have finished Part 1, simply click the video player below to view Part 2. A written transcript is also provided.

Part 1

Part 2


Part 1

The contents of my presentation today will start with an introduction of the scavenging process in our engine, and then we’ll discuss, in detail, the simulation methodology we use to represent this scavenging process. We’ll break this down into how we represent the engine geometry in simulation, where the boundary conditions for the simulation come from, the validation processes we’d use to verify that the simulation results are accurate, and then we’ll talk a little bit more in detail about the scavenging results we can get out of simulations. Then we’ll switch gears a little and talk about scavenging measurements that we’ve done here in our test cell to gain an understanding of the scavenging process in our engine, but also to verify the trends that we see inside our simulations.

So this slide shows an animation of our engine in action. On the right-hand side, we have the exhaust ports. On the left-hand side, we have the intake ports. You’ll notice that the two pistons reach top dead center at the same time, and then due to the crankshaft phasing, the exhaust piston will uncover the exhaust ports first followed shortly thereafter by the intake piston uncovering the intake ports. As the exhaust ports uncover first, a blowdown of the burnt gases inside the cylinder commences into the exhaust manifold. Shortly thereafter, when the intake ports are uncovered, fresh air begins to flow into the cylinder, thereby pushing out a large quantity of the remaining burnt gases. And when the ports close and the pistons begin to head toward top dead center again, we have a relatively fresh charge of air inside the cylinder.

This slide outlines the methodology we use in order to set up the simulation for the scavenging process. We start with Pro/Engineer solid models of the engine components and take these to the Converge user interface in order to generate a surface file that can be used to represent the boundaries of the computational domain. This provides our geometry. We also need boundary conditions and these typically come from either 1-D system simulations or directly from our test cell. These boundary conditions include inlet and outlet pressure measurements and also the heat release profile that occurs inside the cylinder due to the combustion event.

Once we have the geometry and these boundary conditions, we can then run the Converge software. The Converge software package utilizes automatic mesh refinement so that complex regions of the flow are resolved with more cells. And so the number of cells during our simulation will change substantially from approximately 300,000 to 1.75 million, depending on what’s going on inside the computational domain. We do these computations, at Achates Power, on 16 cores of our cluster and this typically takes a little less than two days to run a full cycle of the engine. After the solution is complete, we then post-process the results. In particular, we’re interested in visualizing the in-cylinder residuals and flow patterns and also looking at port mass flow rates, pressures and temperatures, and generating the scavenging schedule, which will be used in 1-D system simulations. We can then feedback this information directly to our designers so that they can improve the engine components or we can take them back to 1-D simulations in order to generate new boundary conditions for further studies. The end result of this process is to provide recommendations for improved scavenging and flow through our engine.

This is a representation of the geometry that is used in the simulations. The red part on the right is the exhaust manifold. The intake manifold, which is partially transparent, is on the left-hand side and the gray cylinder is in the center of the picture.

Within this geometry, there are some regions that we’ll refer to later on in the presentation so we’ll point those out now. The inlet is shown there with the green part. At the end of the exhaust manifold is the outlet and inside the domain, there is a point in the intake manifold, and also a corresponding point in the exhaust manifold, where we measure pressures in the test cell, which we can then compare to pressures provided by the simulations at those same points.

This slide shows how that geometry is then represented in the Converge software. It’s a group of surface triangles, which outlines the fluid that we’ll be simulating. This particular surface representation is used for the compression heat addition, and combustion, followed by expansion part of the simulation. So what we have are three disconnected regions: the intake manifold, the combustion volume and the exhaust manifold. Now later on when we move into the scavenging process when the exhaust ports open and then the intake ports open, we’ll need to connect these three disconnected geometries so that flow can move from the intake manifold through the cylinder and into the exhaust manifold.

In an opposed-piston two-stroke diesel engine, the areas of the ports change as the pistons move past them, and this is a geometric feature that Converge cannot currently capture—so we have a workaround for this. What we do is we shrink the piston diameters by .1% and then connect the pistons to the ports using what are called liner gaps. So this creates a very thin region around the piston that connects the exhaust ports to the cylinder. Now, due to the phasing of our crankshafts, we saw that the exhaust ports will open first, followed by the intake ports. And, so we would prefer that even though these liner gaps are in place, if the ports have not been opened at all, we’d like that there be no flow whatsoever. In order to accomplish this, we combine disconnect triangles, along with our liner gaps. The disconnect triangles prevent all flow through the ports up until a specified time where we know that the piston has actually uncovered the corresponding ports.

The pressure boundary conditions for our simulation are applied at the inlet and outlet, and these come directly from time-resolved measurements in our test cell. What I’m showing in the upper left here is a comparison of the inlet pressure that was provided by the test cell and applied as a boundary condition in our simulation to the actual inlet pressure that was calculated during the calculation. And you’ll notice that over most of the cycle, the two are in perfect agreement, as they should be. However, during the scavenging portion of the cycle, we start to see a slight discrepancy between the simulation value and the applied boundary condition. The reason for this is that the simulation requires a total pressure in order to be well posed and the experiment measured a static pressure. So during periods of strong flow, we start to develop a difference between the total and the static pressure and this shows up as a discrepancy in our plot here between the experimental pressure trace that was applied to the boundary condition, and the actual value of the pressure at the inlet in the simulation.

In the upper right, you’ll see the same comparison at the outlet of the computational domain. In this case, we see perfect agreement between the experimental pressure trace and the calculated pressure in the simulation. This is because there is no backflow into the outlet of the exhaust manifold. If there were strong backflow at any point, then the simulation would use this applied pressure as a total pressure just as it does at the inlet and we would see a discrepancy. So this indicates that, for this case, there is no backflow into the outlet of the exhaust manifold. The cylinder pressure trace measured in the experiment is used to correct our piston motions for flexibility and inertia and it is also used to determine the heat release rate profile. Converge is capable of simulating the combustion event inside the cylinder; however, for these purposes, we’re interested in looking at scavenging and so in order to reduce computational expense, we apply a heat release rate source into the computational domain instead of simulating the spray in combustion. So the chart there shows a representation of the heat release rate that is applied in the Converge simulations. The way this heat source is applied is as a sequence of spherical sources of fixed duration and strength, and we have 11 of them inside our cylinder. As the pistons move apart to create a larger volume inside the combustion region, we expand these spheres so that they fill more of the computational domain. Recently some other options have become available in Converge for applying the heat source. In particular, it is now possible to apply a heat source uniformly throughout the combustion chamber and it is also possible to apply a heat source that is required in order to exactly match a specified pressure trace.

Back to Top

Part 2

We apply the experimentally measured pressures at the inlet and outlet and, as a check of our simulation accuracy, we can then compare the pressures at the interior points that I mentioned previously to verify the dynamics inside the intake and exhaust manifold in our simulation are consistent with those that were obtained in the test cell.

On the upper-left hand side of the slide, there’s a comparison in the intake manifold between the experimentally measured pressure and the pressure from the simulation. What you see is an oscillating pressure in the manifold and this is due to expansion and compression of the contents of the intake manifold while the ports are closed and then when scavenging begins, we see that the intake manifold pressure drops substantially as flow starts to move from the inlet through the intake ports and through the cylinder out to the exhaust manifold. We see that over most of the cycle there is very good agreement between the simulation and the experiment. We do end up with a discrepancy during scavenging, once again, because of the inlet pressure discrepancy that was discussed previously.

The upper right-hand graph shows this same comparison, but now on the exhaust manifold side. Once again, while all ports are closed, we see the oscillations in the pressure indicating expansion and compression of the contents of the exhaust manifold. Then, around a crank angle of 120 degrees, we see a spike in the pressure in the manifold, which indicates the blowdown has occurred into the exhaust manifold from the cylinder. The bottom graph shows a comparison of the pressure in the cylinder measured from the experiment and the simulation and a comparison between the simulation value and the experimental value when the intake ports close, which indicates that we’re within 2.5% of the measured value and similar comparison at the peak cylinder pressure value, which puts us within 3.5% of the experimental value.

So, overall, we have excellent agreement between our simulation values and the experimentally measured values for the pressures in both manifolds and in the cylinder. The simulation can provide detailed information about the mass flow rates in various regions of the engine. So I’m showing here plots of the inlet mass flow rate in the upper left, the outlet mass flow rate in the upper right, the intake port mass flow rate in the lower left and the exhaust port mass flow rate in the lower right. So some things to see here, we see a spike in the outlet mass flow rate shortly after the blowdown as the contents of the cylinder have emptied into the exhaust manifold and then begin to exit through the outlet of the exhaust manifold. In the intake port mass flow rate diagram, we see that there are actually periods of backflow, where flow moves from the cylinder into the intake manifold instead of moving from the intake manifold into the cylinder. On the exhaust port side, we see the spike in mass flow rate due to the blowdown event and we see that there is no backflow through the exhaust ports over the entire simulation. From these time histories, we can calculate the cumulative mass flow rate through the engine and then make comparisons of that with the experimentally measured value. What we see is that the simulation values are within 10 to 15% of the experimentally measured mass flow rate values.

In order to visualize the scavenging process in the Achates Power engine, we use a passive scaler from the simulation. We initialize this passive scaler to 1 inside the cylinder and 0 inside the manifolds. So you see in the upper left-hand corner there, both manifolds are blue for 0 and the cylinder is red for the initialized value of 1. Shortly after that initial picture as we move to the right, the blowdown commences and we see some of the red plumes exiting into the exhaust manifold as burned gases begin to escape from the cylinder. As we move along and down to a crank angle of 145 degrees, we see the initial time when the intake ports open and the slight backflow that occurs as the intake ports first open into the intake manifold. However, shortly thereafter at a crank angle of 150 degrees, the flow reverses direction and fresh air begins to make its way into the cylinder. Moving further along, we see the fresh air progressing down the cylinder, mixing with the burned gases and then exiting through the exhaust ports. And then finally at the end, in the lower right, we see that we have a cylinder that’s primarily composed of fresh charge with a small amount of burned gases left.

From our simulations, we can calculate the scavenging schedule. The scavenging schedule is used in our 1-D system simulations, since those cannot capture the 3-D physics occurring inside the cylinder during the scavenging process. The scavenging schedule provides a relationship between the exhaust port concentration and the cylinder concentration. Initially, we start in the upper right-hand corner of the graph before the exhaust ports even open. Once the exhaust ports open, burned gases begin to flow into the exhaust manifold and then shortly thereafter, the intake ports open and fresh air comes in and begins to push out more burned gases. So this period of time where fresh air is pushing out pure burned gases is known as perfect displacement and this occurs, in this case, up until the point where the cylinder concentration has dropped to 35%. Thereafter, both burned gases and fresh air exit through the exhaust ports and so the scavenging schedule begins to drop as the exhaust port concentration falls with a higher concentration of fresh air leaving through the exhaust ports.

Now in this case, we see something interesting around a crank angle of 180 to 185 degrees, where the scavenging schedule levels off briefly. So, if we return to the previous slide, we can see the reason for this behavior. At around a crank angle of 180 to 185, we see that in the center of the cylinder, there is a trapped region of relatively high burnt gases still left in the cylinder, which then exits through the exhaust ports by the time we reach a crank angle of 190 degrees. So this indicates that that high concentration of burned gases made its way out of the exhaust ports in that timeframe, and this is the reason why the exhaust port concentration briefly remains around .5 before it then begins to drop again. Now a value of interest is the final value of the scavenging schedule, which is the .04% cylinder concentration or a 96% scavenging efficiency for this case.

In addition to our simulations, we’ve also developed the capability to do measurements that get us similar information about scavenging. So this is a schematic of our experimental test set up. In our cylinder, we have two injectors and we replace one of these injectors with a sampling probe, which is basically just an injector blank with a small orifice on it. This injector blank is then attached to an NDIR analyzer, which can give us the measurement of CO2 inside the cylinder. So the valve will be opened 90 degrees before top dead center and held open for 45 degrees so that we can obtain a measurement of the charge in the cylinder after scavenging ends and before combustion occurs.

So from our experiment, we can measure the CO2 mass fraction in the cylinder after scavenging and before combustion. This measurement is vital in that combining it with other measurements that we already have, including the CO2 in the ambient air, the mass of fuel, the pressure in the cylinder and the temperatures of the intake and exhaust, we can then calculate some things that we can’t measure directly in the experiment. In particular, we can calculate scavenging efficiency. The assumptions that we use to do this include a two-zone non-isothermal thermal mixing model and we assume that the fuel is fully consumed by the combustion process.

Now there are some limits to the scavenging efficiency that you can obtain. This graph here shows the relationship between scavenging efficiency and the modified delivery ratio, which is just a normalized mass flow rate into the cylinder. A modified delivery ratio of 1 means that the amount of air that was delivered through the intake ports is equal to the amount of mass that is inside the cylinder. So at a delivery ratio of 2, you’re moving two times as much mass of air as you end up with in the cylinder at the end of the scavenging process.

There are some idealized cases to consider here. The first is perfect displacement. In perfect displacement, the air moves through the intake ports and pushes out the combustion product without any mixing occurring throughout the process. So this is the theoretical maximum that you can achieve in terms of scavenging efficiency. Perfect mixing is when the air enters through the intake ports and instantaneously mixes with all of the contents of the cylinder before the mixture then exits through the exhaust ports. And perfect short-circuiting is when the air enters the cylinder, does not mix with any of the combustion products and then exits through the exhaust ports.

Based on our measurements, we found that the scavenging efficiency for the Achates Power engine is a function of modified delivery ratio and engine speed. We did test a variety of other conditions, changing the fuel mass flow rate and the injection timing, but these changes did not produce any substantial change in the scavenging efficiency. So in the upper right, we show the data points from a low speed run as the symbols. On the plot are also shown the perfect displacement and the perfect mixing curve and then the dotted curve is our empirical correlation for scavenging that was obtained by considering all of our results together. So you see that all of the experimental results lie along our empirical correlation. In the lower left, we have the medium speed results, which show a similar agreement between the empirical correlation and the experimentally measured values. On the lower right, we have the high speed data. Now one thing that occurs here is that in the high speed data, some of the experimentally measured points lie above the perfect displacement model. This is likely an effect of experimental uncertainty and some of the simplifying assumptions that were used in order to calculate the scavenging efficiency from the experimentally measured values. However, these points occur at a relatively low delivery ratio and we would not be operating in this regime with the Achates Power engine.

As a further check of the computational results, we performed a correlation between the values measured in the test cell of scavenging efficiency and those calculated with our 3-D CFD technique. The plot here on the right shows these data points at the medium speed. The simulations were run over a variety of speeds, loads and pressure drops across the engine and what you see is very good agreement between our empirical correlation and the CFD results.

In conclusion, we’ve been able to successfully validate our Converge full-cycle simulation results with our experimental data. We’ve shown very good agreement of the pressure time histories and good agreement also of the bulk mass flow rate through the engine, and the simulations allow us to obtain detailed information that we couldn’t necessarily get with experiments. In particular, we can look at how the residuals are replaced in the cylinder by fresh air during the scavenging process.

The CO2 measurement technique that we’ve used has also provided insight into scavenging performance. We’ve been able to develop an empirical model that relates the scavenging efficiency to the speed and modified delivery ratio and we can use this simple formula in our 1-D system simulations.

In summary, it’s only with the combination of detailed simulations and measurements that we can understand and predict the scavenging performance of a better engine.

Back to Top