Converting existing compression ignition engines to spark ignition approach is a promising approach to increase the application of natural gas in the heavy-duty transportation sector. However, the diesel-like environment dramatically affects the engine performance and emissions. As a result, experimental tests are needed to investigate the characteristics of such converted engines. A machine learning model based on bagged decision trees algorithm was established in this study to reduce the experimental cost and identify the operating conditions of special interest for analysis. Preliminary engine tests that changed spark timing, mixture equivalence ratio, and engine speed (three key engine operation variables) but maintained intake and boundary conditions were applied as model input to train such a correlative model. The model output was the indicated mean effective pressure, which is an engine parameter generally used to assist in locating high engine efficiency regions at constant engine speed and fuel/air ratio. After training, the correlative model can provide acceptable prediction performance except few outliers. Subsequently, boosting ensemble learning approach was applied in this study to help improve the model performance. Furthermore, the results showed that the boosted decision trees algorithm better described the combustion process inside the cylinder, as least for the operating conditions investigated in this study.

Although many studies have been carried out using control-oriented engine models for advanced combustion technologies, methods of adapting model parameters for these models have still not been considerably studied. In this paper, we propose a method to adapt model parameters for an ignition model and a combustion model of a physics-based discrete diesel engine model using neural networks. Using experimental data on an engine bench as well, we examined how to perform adapting and training, such as optimization algorithms, how to select training data, appropriate feature values, and algorithms for switching the neural networks according to the operating regions. As a result, this adaptation method was able to improve the prediction accuracy of the heat release rate of pre-combustion better than when adapted using the previous adaptation method and when the model parameters were fixed.

The use of machine learning (ML) based surrogate models is a promising technique to significantly accelerate simulation-based design optimization of IC engines, due to the high computational cost of running computational fluid dynamics (CFD) simulations. However, surrogate-based optimization for IC engine applications suffers from two main issues. First, training ML models requires hyperparameter selection, often involving trial-and-error combined with domain expertise. The second issue is that the data required to train these models is often unknown a priori. In this work, we present an automated hyperparameter selection technique coupled with an active learning approach to address these challenges. The technique presented in this study involves the use of a Bayesian approach to optimize the hyperparameters of the base learners that make up a Super Learner model to obtain better test performance. In addition to performing hyperparameter optimization (HPO), an active learning approach is employed, where the process of data generation using simulations, ML training, and surrogate optimization, is performed repeatedly to refine the solution in the vicinity of the predicted optimum. The proposed approach is applied to the optimization of a compression ignition engine with control parameters relating to fuel injection, in-cylinder flow, and thermodynamic conditions. It is demonstrated that by automatically selecting the best values of the hyperparameters, a 1.6% improvement in merit value is obtained, compared to an improvement of 1.0% with default hyperparameters. Overall, the framework introduced in this study reduces the need for technical expertise in training ML models for optimization, while also reducing the number of simulations needed for performing surrogate-based design optimization.

A faster, more accurate replacement for existing collision algorithms has been developed. The method, called the NTC algorithm, is not grid dependent, and is much faster than older algorithms. Calculations with sixty thousand parcels required only a few CPU minutes. However, there is a significant need to develop mesh-independent momentum coupling between the gas and spray, so that the collision algorithm’s full accuracy can be fully realized.

With the increasing recognition that variable valve actuation (VVA) in its various forms is a powerful tool for optimizing the performance of internal combustion engines, more and more production systems are being designed and implemented throughout the industry. However, as these control systems become more capable of altering lift, timing, duration, and even the number of valve events, the complexity of designing algorithms and calibrating them becomes enormous. In addition, without prior knowledge of an engine’s response to these algorithms, designing a cost-effective mechanism which provides adequate but not over-reaching capability is difficult. Ricardo has developed methodology for timestep coupled simulations which enables the use of one-dimensional (1-D) gas dynamics simulation of engine performance (WAVE™) coupled to a simulation of the valve actuation mechanism constructed in MATLAB ® and AMESim ® . This arrangement allows valve motion input to the 1-D code to be controlled either manually or by a VVA controller simulation, allowing such engine parameters as torque, fuel consumption, NVH, and EGR rates to be monitored as a function of valve timing strategy. This method allows the examination of such engine development concerns as tolerances, valve velocities and accelerations, and interactions with other engine controls to be studied without the costs, leadtimes, or hardware reliability problems that are associated with prototyping a VVA system. In addition, the interfacing of the valve control/engine performance simulation combination with the Design of Experiments optimization software iSIGHT allows the control system space to be explored automatically, without the brute force numerical search required to examine all permutations of the control strategies. The output of this procedure is an array of requirements which can be quickly translated into a specification document which will guide hardware and controls design efforts.

Many design parameters affect the performance of continuous variable transmissions. This paper presents the optimization of a continuous variable transmission by using the simulated annealing algorithm. The Bessel method of curve fitting and the tensor product method of surface fitting were used to facilitate the discrete fuel consumption, emissions of carbon monoxide (CO) and HC compound of experimental engine data. A compromise method was used to analyze the multi-objective functions. The values for design variables are recommended for further development.

A new algorithm to detect cylinder misfire in automotive gasoline engines is presented in this paper. It is based on processing the variations in crankshaft speed of the engine. In the algorithm, the crankshaft speed data is preprocessed and then modeled as a Mixture of Gaussian components. The appropriate number of components in the data is determined using a goodness-of-fit measure. When misfire is taking place, two or more Gaussian components will appear in the mixture model. Based on the estimated parameters (mean, variance) of the components and a specified probability of false alarms, a Generalized Likelihood decision rule is designed to classify every data point as a misfire or no-misfire behavior. The algorithm was tested on real engine data and the results were encouraging.

In this work, a novel design optimization technique based on active learning, which involves dynamic exploration and exploitation of the design space of interest using an ensemble of machine learning algorithms, is presented. In this approach, a hybrid methodology incorporating an explorative weak learner (regularized basis function model) which fits high-level information about the response surface, and an exploitative strong learner (based on committee machine) that fits finer details around promising regions identified by the weak learner, is employed. For each design iteration, an aristocratic approach is used to select a set of nominees, where points that meet a threshold merit value as predicted by the weak learner are selected to be evaluated using expensive function evaluation. In addition to these points, the global optimum as predicted by the strong learner is also evaluated to enable rapid convergence to the actual global optimum once the most promising region has been identified by the optimizer. This methodology is first tested by applying it to the optimization of a two-dimensional multi-modal surface. The performance of the new active learning approach is compared with traditional global optimization methods, namely micro-genetic algorithm (μGA) and particle swarm optimization (PSO). It is demonstrated that the new optimizer is able to reach the global optimum much faster, with a significantly fewer number of function evaluations. Subsequently, the new optimizer is also applied to a complex internal combustion (IC) engine combustion optimization case with nine control parameters related to fuel injection, initial thermodynamic conditions, and in-cylinder flow. It is again found that the new approach significantly lowers the number of function evaluations that are needed to reach the optimum design configuration (by up to 80%) when compared to particle swarm and genetic algorithm-based optimization techniques.

Modern high-efficiency engines utilize direct injection for charge preparation at extremely high pressures. At these conditions the scales of atomization become challenging to measure, as primary breakup occurs on the micrometer and nanosecond scales. As such, fuel sprays at these conditions have proven difficult to study via direct imaging. While high-speed cameras now exist that can shutter at tens to hundreds of nanoseconds, and long-range microscopes can be coupled to these cameras to provide high resolution images, the resolving power of these systems is typically limited by pixel size and field of view. The large pixel sizes make the realization of the diffraction-limited optical resolution quite challenging. On the other hand, limited data throughput under high repetition rate operation limits the field of view (FOV) due to reduced sensor area. Therefore, a novel measurement technique is critical to study fuel spray formation at engine-relevant conditions. In this work, we demonstrate a new high-resolution imaging technique, Spectral Microscopy, which aims to realize diffraction-limited imaging at effective framerates sufficient for capturing primary breakup in engine-relevant sprays. A spectral microscopy system utilizing a consumer-grade DSLR allows for significantly wider FOV with improved resolving power compared to high-speed cameras. Temporal shuttering is accomplished via separate and independently triggered back illumination sources, with wavelengths selected to overlap with the detection bands of the camera sensor’s RGB filter array. The RGB detection channels act as filters to capture independently timed red, green, and blue light pulses, enabling the capture of a three consecutive images at effective framerates exceeding 20 Million fps. To optimize system performance, a backlit illumination system is designed to maximize light throughput, a multi-lens setup is created, and an image processing algorithm is demonstrated that formulates a three-frame image from the camera sensor. The system capabilities are then demonstrated by imaging engine relevant diesel sprays. The spectral microscopy system detailed in this paper allows for micron-scale feature recognition at framerates exceeding 20 Million fps, thus expanding the capability for experimental research on primary breakup in fuel sprays for modern direct-injection engines.

Proper Orthogonal Decomposition (POD) offers an approach to quantify cycle-to-cycle variation (CCV) of the flow field inside the internal combustion engine cylinder. POD decomposes instantaneous flow fields (also called snapshots) into a series of orthonormal flow patterns (called POD modes) and the corresponding mode coefficients. The POD modes are rank-ordered by decreasing kinetic energy content, and the low-order, high-energy modes are interpreted as constituting the large-scale coherent flow structure that varies from engine cycle to engine cycle. Various POD-based analysis techniques have thus been proposed to characterize engine flow field CCV using these low-order modes. The validity of such POD-based analyses rests, as a matter of course, on the reliability of the underlying POD results (modes and coefficients). Yet a POD mode can be disproportionately skewed by a single outlier snapshot within a large data set, and an algorithm exists to define and identify such outliers. In this paper, the effects of a candidate outlier snapshot on the results of POD-based conditional averaging and quadruple POD analyses are examined for two sets of crank angle-resolved flow fields on the mid-tumble plane of an optical engine cylinder recorded by high-speed particle image velocimetry. The results with and without the candidate outlier are compared and contrasted. In the case of POD-based conditional averaging, the presence of the outlier scrambles the composition of snapshot subsets that define large-scale flow pattern variations, and thus substantially alters the coherent flow structures that are identified; for quadruple POD, the shape of coherent structures as well as the number of modes to define them are not significantly affected by the outlier.

This work evaluates different optimization algorithms for Computational Fluid Dynamics (CFD) simulations of engine combustion. Due to the computational expense of CFD simulations, emulators built with machine learning algorithms were used as surrogates for the optimizers. Two types of emulators were used: a Gaussian Process (GP) and a weighted variety of machine learning methods called SuperLearner (SL). The emulators were trained using a dataset of 2048 CFD simulations that were run concurrently on a supercomputer. The Design of Experiments (DOE) for the CFD runs was obtained by perturbing nine input parameters using a Monte Carlo method. The CFD simulations were of a heavy duty engine running with a low octane gasoline-like fuel at a partially premixed compression ignition mode. Ten optimization algorithms were tested, including types typically used in research applications. Each optimizer was allowed 800 function evaluations and was randomly tested 100 times. The optimizers were evaluated for the median, minimum, and maximum merits obtained in the 100 attempts. Some optimizers required more sequential evaluations, thereby resulting in longer wall clock times to reach an optimum. The best performing optimization methods were particle swarm optimization (PSO), differential evolution (DE), GENOUD (an evolutionary algorithm), and Micro-Genetic Algorithm (GA). These methods found a high median optimum as well as a reasonable minimum optimum of the 100 trials. Moreover, all of these methods were able to operate with less than 100 successive iterations, which reduced the wall clock time required in practice. Two methods were found to be effective but required a much larger number of successive iterations: the DIRECT and MALSCHAINS algorithms. A random search method that completed in a single iteration performed poorly in finding 1 Currently at Southwest Research Institute, San Antonio, Texas optimum designs, but was included to illustrate the limitation of highly concurrent search methods. The last three methods, Nelder-Mead, BOBYQA, and COBYLA, did not perform as well.

Diesel engine control strategies use complex injection patterns which are designed to meet the increasing request for engine-out emissions and fuel consumption reduction. As a result of the large number of tuneable injection parameters in modern injection systems (such as start and duration of each injection), injection patterns can be designed with many degrees of freedom. Each variation of the injection parameters modifies the whole combustion process and, consequently, engine-out emissions. Aging of the injection system usually affects injection location within the cycle as well as the amount of injected fuel (compared to the target value), especially for small pre-injections. Since Diesel combustion is very sensitive to injection pattern variations, aging of injectors strongly affects engine behavior, both in terms of efficiency and pollutant emissions production. Moreover, such variations greatly affect other quantities related to the effectiveness of the combustion process, such as noise radiated by the engine. This work analyses the effects of pre-injection variations on combustion, pollutant emissions and noise radiated by the engine. In particular, several experimental tests were run on a 1.3L Common Rail Diesel engine varying the amount of fuel injected in pre-injections. Torque delivered by the engine and center of combustion (MFB50) were kept constant using a specifically designed closed-loop combustion controller. During the tests, noise radiated by the engine was measured by properly processing the signal coming from a microphone faced to the engine block. The investigation of the correlation between the combustion process and engine noise can be used to set up a closed-loop algorithm for detecting and recentering injectors’ drifts over time.

Tunnels represent one of the most severe operating conditions for diesel engines in diesel-electric locomotive applications, specifically for non-ventilated tunnels located at high elevation. High ambient air temperatures are observed in these tunnels due to heat rejected from the locomotive engines through the exhaust and engine cooling and lubrication systems. Engine protection algorithms cause the maximum allowable engine horsepower to be reduced due to these conditions leading to a reduction in train speed and occasionally train stall. A first law based model was developed to simulate the performance of a train pulled by GE diesel-electric locomotives equipped with medium speed diesel engines in a high altitude and non-ventilated tunnel. The model was compared against and calibrated to actual tunnel operation data of EPA Tier 2 compliant locomotives. The model was then used to study the impact of engine design changes required for EPA Tier 4 compliant locomotives, specifically the introduction of exhaust gas recirculation (EGR), on engine, locomotive, and train performance in the tunnel. Simulations were completed to evaluate engine control strategies targeting same or better train performance than the EPA Tier 2 compliant locomotive baseline case. Simulation results show that the introduction of EGR reduces train performance in the tunnel by increasing the required reduction in engine horsepower, but is slightly offset by improved performance from other engine design changes. The targeted engine and train performance could be obtained by disabling EGR during tunnel operation.

Stringent emission regulations require spark ignited (SI) engines to operate at stoichiometry to enable the use of a three way catalyst (TWC). Thus, accurate prediction of the intake charge mass flow rate is paramount. Current speed-density air mass-flow prediction techniques require extensive calibration for predicting volumetric efficiency, while mass air flow (MAF) meter based approaches suffer from a loss of accuracy during transients. This work aims to provide an alternative, i.e. a model based air charge estimation algorithm that can reduce calibration effort and provide a universal solution across engine platforms. An additional objective is to minimize the number of required sensors and associated cost. The foundation is established with a 0-D physics-based air charge model, where air flow through intake and exhaust valves is modeled on a crank-angle basis, without the need to measure in-cylinder pressure. The proposed algorithm solves differential equations for cylinder pressure and mass flow rate in/out of the cylinder to simultaneously obtain instantaneous pressure and mass-flow estimations, hence eliminating the need to install cylinder pressure transducers. An additional benefit is the robustness of the new model, due to its ability to self-compensate for an error in the intake runner pressure or initial estimation of the cylinder pressure. The model has been validated with GT-Power simulations and steady-state engine tests with multiple actuator sweeps. Transient tests and real-time implementations were performed as well.

Increasingly stringent fuel economy and CO 2 emission regulations provide a strong impetus for development of high efficiency engine technologies. Diesel engines dominate the heavy duty market and significant segments of the global light duty market due to their intrinsically higher thermal efficiency compared to spark ignited (SI) engine counterparts. Predictive simulation tools can significantly reduce the time and cost associated with optimization of engine injection strategies, and enable investigation over a broad operating space unconstrained by availability of prototype hardware. In comparison with 0-D/1-D and 3-D simulations, Quasi-D models offer a balance between predictiveness and computational effort, thus making them very suitable for enhancing the fidelity of engine system simulation tools. A most widely used approach for diesel engine applications is a multi-zone spray and combustion model pioneered by Hiroyasu and his group. It divides diesel spray into packets and tracks fuel evaporation, air entrainment, gas properties and ignition delay (induction time) individually during the injection and combustion event. However, original sub-models are not well suited for modern diesel engines, and the main objective of this work is to develop a multi-zonal simulation capable of capturing the impact of high-injection pressures and Exhaust Gas Recirculation (EGR). In particular, a new spray tip penetration sub-model is developed based on measurements obtained in a high-pressure, high-temperature constant volume combustion vessel for pressures as high as 1450 bar. Next, ignition delay correlation is modified to capture the effect of reduced oxygen concentration in engines with EGR, and an algorithm considering the chemical reaction rate of hydrocarbon-oxygen mixture improves prediction of the heat release rates. Spray and combustion predictions were validated with experiments on a single-cylinder diesel engine with common rail fuel injection, charge boosting, and EGR.

Spark ignition engine in-cylinder air charge estimation is important for air-to-fuel ratio control, maintaining high after-treatment efficiency, and determination of current engine torque. Current cylinder air charge estimation methodologies generally depend upon either a mass air flow (MAF) sensor or a manifold absolute pressure (MAP) sensor individually. Methods based on either sensor have their own advantages and disadvantages. Some production vehicles are equipped with both MAF and MAP sensors to offer air charge estimation and other benefits. This research proposes several observer based cylinder air charge estimation methods that take advantage of both MAF and MAP sensors to potentially reduce calibration work while providing acceptable transient and steady-state accuracy with low computational load. This research also compares several common air estimation methods with the proposed observer based algorithms using steady-state and transient dynamometer tests and a rapid-prototype engine controller. With appropriate tuning the proposed observer based methods are able to estimate cylinder air charge mass under different engine operating conditions based on the manifold model and available sensors. Methods are validated and compared based on a continuous tip-in tip-out operating condition.

Durability is a prime concern in the design of hydraulic systems and fuel injectors [1–3] thus an accurate prediction of impact velocities between components and the flow through them is essential to assessing concepts. Simulation of these systems is difficult because the geometries are complex, some volumes go to zero as the components move, and the flow at a single operating condition generally spans Reynolds numbers less than 1 to more than 10 4 [4–8]. As a result of these challenges, experimental testing of prototypes is the dominant method for comparing concepts. This approach can be effective but is far more costly, time consuming, and less flexible than the ability to run simulations of concepts early in the design cycle. A validated model of a fuel injector built from publicly available data [1] is used to present a new approach to modelling hydraulic systems which overcomes many of these obstacles. This is accomplished by integrating several commercially available tools to solve the physics specific to each area within the fuel injector. First, the fuel injector is simulated using a 3D CFD simulation integrated with a 1D CFD system model. The flow in various regions of the injector is then analyzed to determine if the fluid models in these areas can be simplified based on the flow regime. Based on this analysis, a combination of models is assembled to improve the quality of the simulation while decreasing the time required to run the model. The fuel injector is simulated using a multibody dynamics model coupled to a reluctance network model of the solenoid and several fluid models. The first is a 3D CFD simulation which uses novel mesh refinement techniques during runtime to ensure high mesh quality throughout the motion of components, to resolve the velocity profile of laminar flows, and to satisfy the requirements of the RNG k-ε turbulence model and wall functions. This approach frees the analyst from defining the mesh before runtime and instead allows the mesh to adapt based on the flow conditions in the simulation. Due to the highly efficient meshing algorithm employed, it is possible to re-mesh at each timestep thus ensuring a high quality structured mesh throughout the simulation duration. Then a 3D FEM solution to the Reynolds Equation and a statistical contact model is employed to solve for the squeeze films between components and to allow separation and contact between bodies in the control valve. These detailed simulations are integrated with a 1D flow model of the fuel injection system. The results from the detailed coupled simulations are compared to the results from simpler 1D models and measured data to illustrate under which operating conditions a more advanced technique incorporating 3D CFD is worth the additional computational expense versus a traditional 1D model.

Optimal combustion control has become a key factor in modern automotive applications to guarantee low engine out emissions and good driveability. In order to meet these goals, the engine management system has to guarantee an accurate control of torque delivered by the engine and optimal combustion phasing. Both quantities can be calculated through a proper processing of in-cylinder pressure signal. However, in-cylinder pressure on-board installation is still uncommon, mainly due to problems related to pressure sensors’ reliability and cost. Consequently, over the last years, the increasing request for combustion control optimization spawned a great amount of research in the development of remote combustion sensing methodologies, i.e. algorithms that allow extracting useful information about combustion effectiveness via low cost sensors, such as crankshaft speed, accelerometers or microphones. Based on the simultaneous acquisition of two crankshaft speed signals, this paper analyses the information that can be extracted about crankshaft’s torsional behavior through a proper processing of the acquired signals. In particular, the correlations existing between such information and indicated quantities (torque delivered by the engine and combustion phasing) have been analysed. In order to maximize the signal-to-noise ratio, each speed measurement has been performed at an end of the crankshaft, i.e. in correspondence of the flywheel and the distribution wheel. The presented approach has been applied to a light-duty L4 Diesel engine mounted in a test cell. Nevertheless, the methodology is general, and it can be applied to engines with a different number of cylinders, both CI and SI.

Combustion instabilities in dilute internal combustion engines are manifest in cyclic variability (CV) in engine performance measures such as integrated heat release or shaft work. Understanding the factors leading to CV is important in model-based control, especially with high dilution where experimental studies have demonstrated that deterministic effects can become more prominent. Observation of enough consecutive engine cycles for significant statistical analysis is standard in experimental studies but is largely wanting in numerical simulations because of the computational time required to compute hundreds or thousands of consecutive cycles. We have proposed and begun implementation of an alternative approach to allow rapid simulation of long series of engine dynamics based on a low-dimensional mapping of ensembles of single-cycle simulations which map input parameters to output engine performance. This paper details the use Titan at the Oak Ridge Leadership Computing Facility to investigate CV in a gasoline direct-injected spark-ignited engine with a moderately high rate of dilution achieved through external exhaust gas recirculation. The CONVERGE™ CFD software was used to perform single-cycle simulations with imposed variations of operating parameters and boundary conditions selected according to a sparse grid sampling of the parameter space. Using an uncertainty quantification technique, the sampling scheme is chosen similar to a design of experiments grid but uses algorithms designed to minimize the number of samples required to achieve a desired degree of accuracy. The simulations map input parameters to output metrics of engine performance for a single cycle, and by mapping over a large parameter space, results can be interpolated from within that space. This interpolation scheme forms the basis for a low-dimensional ‘metamodel’ (or model of a model) which can be used to mimic the dynamical behavior of corresponding high-dimensional simulations. Simulations of high-EGR spark-ignition combustion cycles within a parametric sampling grid were performed and analyzed statistically, and sensitivities of the physical factors leading to high CV are presented. With these results, the prospect of producing low-dimensional metamodels to describe engine dynamics at any point in the parameter space will be discussed. Additionally, modifications to the methodology to account for nondeterministic effects in the numerical solution environment are proposed.

A closed-cycle gasoline compression ignition engine simulation near top dead center (TDC) was used to profile the performance of a parallel commercial engine computational fluid dynamics code, as it was scaled on up to 4096 cores of an IBM Blue Gene/Q supercomputer. The test case has 9 million cells near TDC, with a fixed mesh size of 0.15 mm, and was run on configurations ranging from 128 to 4096 cores. Profiling was done for a small duration of 0.11 crank angle degrees near TDC during ignition. Optimization of input/output performance resulted in a significant speedup in reading restart files, and in an over 100-times speedup in writing restart files and files for post-processing. Improvements to communication resulted in a 1400-times speedup in the mesh load balancing operation during initialization, on 4096 cores. An improved, “stiffness-based” algorithm for load balancing chemical kinetics calculations was developed, which results in an over 3-times faster run-time near ignition on 4096 cores relative to the original load balancing scheme. With this improvement to load balancing, the code achieves over 78% scaling efficiency on 2048 cores, and over 65% scaling efficiency on 4096 cores, relative to 256 cores.