Pushed by a competitive need for better fuel economy coupled with emissions regulations, engine developers are taking conventional internal-combustion engine (ICE) technologies to the limit. This means a greater risk of knock—that rattling, tinny sound of an engine in distress. What causes it? Uncontrolled ignition during combustion. Instead of a smooth flame front moving through the combustion chamber, hot spots will auto-ignite unpredictably in the remaining air-fuel mixture ahead of the flame front. Excessive temperature and pressures in localized spots result. Besides annoying drivers, excessive knock can wear engines faster and even catastrophically damage pistons and cylinders.
Boosting, higher compression ratios, geometry of the piston, injection strategies, spark timing, and octane levels of fuels are just some of the design factors that affect knocking. “All efficient engines have knock, it is a question of how much you are willing to tolerate,” said Eric Pomraning, Vice President of Convergent Science, in an interview with AEI. “But, engineers want to push as much as they can until they decide they cannot live with any more knock.”
Simulation aids design
Building experimental equipment to test a growing number of variables is expensive. It would be faster and cheaper to simulate on a computer, but the phenomenon can be expensive to model as well. Combustion phenomenon requires both CFD combined with chemical kinetics. To capture fine-grained combustion phenomenon requires fine-grained computational meshes, and each cell could require calculating thousands of reactions to mimic modern fuels. Even with the increasing speed and parallelism of 21st-century computers, it is time-consuming.
To reduce compute times to manageable proportions, providers of combined CFD/chemical reaction software attempt to make the computations simpler. They use adaptive meshing, reduce chemical mechanisms as much as possible, and group cells with similar or identical chemical reactions into single computations. The trick is maintaining accurate calculations while doing so. Pomraning also stressed that knock is highly variable, sometimes occurring sporadically from cycle to cycle, often in only a few cylinders of a multicylinder engine. “You need to produce statistical results, so it is important to run simulations over many cycles,” he said.
Speed vs. accuracy
One approach that Pomraning points to is a model they call the G-Equation that calculates the location of the flame using an empirical relationship. “This runs fast, but since it is based on empirical data, it often requires model tuning and is sensitive to the CFD mesh resolution,” he explained. Their more accurate approach for calculating the flame location is to use the SAGE chemistry solver while resolving the turbulent flame thickness using adaptive mesh refinement (AMR). SAGE is integrated into the company’s CFD solution, CONVERGE, as a standard feature and determines the flame front directly using the heat release and heat transfer from the reaction zone. In addition to added accuracy over the G-equation approach, calculating flame location using SAGE with AMR is grid convergent as the mesh is refined. Regardless of how the flame front location is determined (G-equation or SAGE directly), SAGE is always used ahead of the flame to determine the auto-ignition of end gases. “A few years ago, we started investing heavily in drastically reducing the run time associated with SAGE,” explained Pomraning. “The first step was implementing what we call multizone, which we developed in a collaboration with LLNL. SAGE-multizone (MZ) is currently available.”
The company also introduced faster mathematics with a new iterative solver for SAGE. In September 2013, a new chemical kinetics mechanism reduction will be introduced that will dynamically reduce such mechanisms at the beginning of each calculation step. Pomraning presented data that showed a speed up for SAGE of 75x with its multizone, iterative, and dynamic reduction mechanisms' improvements in place.
He reports that his customers use both approaches. Often the empirical G-Equation approach is useful in exploring a number of initial design decisions quickly. These customers reserve the more accurate and costly computations when their design choices are more mature or refined, according to Pomraning.
Model fuels for accurate simplification
To reduce complexity while maintaining accuracy, Reaction Design started and led the Model Fuels Consortium about seven years ago. The purpose of the MFC was to create a validated library of fuels that used a minimum number of representative constituents yet accurately modeled the chemical kinetic characteristics of commonly used fuels, from gasoline to biofuels. A fuel such as gasoline is composed of 10,000s of species, resulting in 1000s of individual reaction computations. The MFC modeled many common fuels into 100s of species with 100s of individual reaction computations. These were validated through experimentation, in collaboration with the companies in the MFC that range from automotive OEMs to oil companies. “They represent a quantum improvement in predictive accuracy compared to radically simplified models used with semi-empirical approaches,” explained Bernie Rosenthal, CEO of Reaction Design, in an interview with AEI.
After completing its work with the MFC, the company announced in February 2013 a predictive knock capability to its FORTÉ and CHEMKIN-PRO software solutions. (See Predictive Simulation of Knock http://www.sae.org/mags/aei/11818)
“Traditionally, when you look at how people have been trying to model combustion, they have focused on flow characteristics and turbulent mixing of fuel. While that is an important piece of the solution, just as important is understanding the kinetic chemistry,” said Rosenthal. “To simulate knock, it really comes down to: can you model the specific physics of interest?” As he describes it, knock often occurs in what is termed the Negative Temperature Coefficient (NTC) region of temperature during the combustion process. “As temperature increases, you would expect to see a linear relationship with ignition, but such is not the case,” he explained. Modeling that nonlinear, NTC is essential and a part of the FORTÉ and CHEMKIN-PRO solutions offered by Reaction Design. “It is a chemistry reaction problem, not a physics problem.”
Rosenthal presented data that showed simulations comparing favorably with experimental data for two fuels with an 87 Octane and 100 Octane rating.