Search

The smart regenerative braking system in electric vehicles implements automatic control of the regeneration torque of motor to improve driver’s comfort and energy efficiency. To apply this system, the accurate prediction of the vehicle deceleration states is the preliminary step to reflect the driver’s behaviors. In this paper, we proposed a vehicle deceleration prediction model via deep neural network, which consists of a sequential recurrent neural network model with long-short term memory cell and a two-layer conventional neural network model. This model accommodates the physical constraint to designate the vehicle stop location in front of the traffic signals. The model is trained by vehicle experiment data with three drivers through the hyper-parameter optimization method. Using this model, the deceleration characteristics are characterized by two explicit parameters such that deceleration point, maximum point according to the initial slope and the shape of the braking profile. Using these two parameters as clustering variables through a K-means clustering method, the deceleration types are classified. These deceleration types to the input to the prediction model results in higher prediction accuracy of the vehicle states. The driving style of the three drivers at braking situations is analyzed according to the deceleration types as well.

A simulation model of a diesel engine is widely applied to evaluate the engine controller before a vehicle test because controller error can be observed by the engine model. For diesel engine modeling, an adiabatic model is commonly used to express intake manifold pressure because it is a powerful method to depict pressure dynamics. However, the adiabatic model is vulnerable to steady state error, as it neglects the heat exchange. In order to solve this issue, we propose a multi-model correction algorithm for the diesel engine model. The proposed diesel engine model consists of two submodels to calculate the exact intake manifold pressure. The first model is adiabatic, which is used for describing the pressure dynamics. The additional parameter that indicates the heat exchange of the intake manifold is added to compensate steady state error of the model. The second model is the pressure ratio model between intake and exhaust manifolds, used to adjust the additional parameter to reduce the steady state error. The proposed model is validated using 216 steady state conditions with a root mean square error of 3.17 %. The transient performance of the model is also demonstrated by a comparison with the engine experimental results.

for predictive powertrain control, accurate prediction of vehicle speed is required. As vehicle speed prediction is affected by the driver’s response to numerous driving conditions under uncertainty, the development of an accurate model is quite challenging. This paper proposes an ego-vehicle speed prediction model using a long short-term memory (LSTM) based recurrent neural network (RNN). The proposed model uses various inputs to increase the prediction accuracy: internal vehicle information, relative speed and distance to the vehicle ahead measured by a radar sensor, and the ego-vehicle location estimated by the GPS signal and B-spline roadway model. The LSTM based RNN model predicts the ego-vehicle speed for 15 seconds by using inputs from the past 30 seconds. The model was evaluated by real driving data for three scenarios: car-following, sharp curve road, and full path. In all scenarios, the radar sensor and the information of the location of the ego-vehicle contribute to improvement of the speed prediction accuracy. Thus, we conclude that for application of the predictive powertrain control, besides the internal vehicle information, the radar sensor, and the location of the ego-vehicle information are critical inputs to the speed prediction model.

The LP-EGR technology has drawn to further improve the fuel efficiency and reduce exhaust emissions in turbocharged GDI engines, so that the accurate model to estimate a LP-EGR rate is a prerequisite. For a precise model of the LP-EGR rate, the transport delay for each cycle and the highly nonlinear flow characteristics should be considered. This paper proposes an artificial neural network (ANN) architecture using multiple combustion parameters calculated from the in-cylinder pressure to precisely estimate the LP-EGR rate without time delay and capture the highly nonlinear flow characteristics. The ANN model was trained with Levenberg-Marquardt back-propagation using steady-state data of 9000 cycles and validated using the cross-validation technique. In this work, three optimization algorithms are introduced and compared to search for the optimal values of the ANN’s hyper-parameters: random search (RS), tree-structured Parzen estimator (TPE), and hyper-parameter optimization via radial basis function and dynamic coordinate search (HORD). As a result, all three algorithms effectively improved the efficiency to search multidimensional hyper-parameters, with the validation performance R 2 above 0.98 and a total RMSE of less than 0.76%. This means that all three methods were successful to find the hyper-parameters’ values for the precise LP-EGR rate estimation. Among the three algorithms the HORD showed the best performance with stable convergence, the R 2 values of 0.9896 or more, and a total RMSE of 0.63%.

Low pressure cooled exhaust gas recirculation (LP-EGR) system has been widely adopted to improve energy efficiency in turbocharged gasoline direct injection (GDI) engines. In order to utilize complete beneficial effects of the LP-EGR, a technique capable of accurately observing the LP-EGR flow into the cylinder in real-time is a prerequisite. To precisely estimate the LP-EGR rate in real-time, this paper proposes artificial neural network (ANN) models and its implementation on a real-time embedded system. As inputs for the ANN models, 12 combustion parameters physically correlated with the LP-EGR in the combustion process are selected and calculated from the in-cylinder pressure. The ANN models for the real-time LP-EGR estimation were trained with the steady-state data of 30,000 cycles and their hyper-parameters were searched by a hyper-parameter optimization method. Moreover, a model-based design procedure is introduced to implement the optimized ANN models on the real-time embedded system. Since the proposed implementation performs the validation procedure for each process, it provides a systematic and seamless process for creating ANN models for real-time embedded systems. In real-time experiments under eight steady-state engine operating points, the embedded ANN models show the estimation performance with R2 of above 0.9716. The operation time of each ANN was less than 1.285 ms meaning that the target system can operate in real-time sufficiently with a mass-produced 32 bit microprocessor up to 256 MHz.

The modern diesel engines equip the exhaust gas recirculation (EGR) system to suppress the NOx emissions. In addition, the variable geometric turbocharger (VGT) system is installed to improve the drivability and fuel efficiency. These EGR and VGT systems have cross-coupled behavior because they interact with the intake and the exhaust manifolds. Furthermore, the turbocharger time delay, gas flow dynamics through EGR pipe cause the nonlinearity characteristics. These nonlinear multi-input-multi-output characteristics cause the degradation of control accuracy, especially during the transient condition. In order to improve the control accuracy, this study proposes an iterative learning control (ILC) algorithm for feedforward controller of EGR and VGT systems. The feedforward controller obtains the values about EGR and VGT actuators using the previous control results in predefined transient states. The ILC algorithm consists of a PD-type learning function and a low-pass filter. The control gains of learning function are determined to guarantee the convergence of learning results. In addition, the low-pass filter is designed for robustness against plant disturbance. The proposed control algorithm was validated by engine experiment which repeated predefined transient states. The error was reduced by 15 % according to the gain. As a result, the proposed algorithm is affordable for improving the transient control performance.

The connected powertrain control, which uses intelligent transportation system information, has been widely researched to improve driver convenience and energy efficiency. The vehicle state prediction on decelerating driving conditions can be applied to automatic regenerative braking in electric vehicles. However, drivers can feel a sense of heterogeneity when regenerative control is performed based on prediction results from a general prediction model. As a result, a deceleration prediction model which represents individual driving characteristics is required to ensure a more comfortable experience with an automatic regenerative braking control. Thus, in this paper, we proposed a deceleration prediction model based on the parametric mathematical equation and explicit model parameters. The model is designed specifically for deceleration prediction by using the parametric equation that describes deceleration characteristics. Furthermore, the explicit model parameters are updated according to individual driver characteristics using the driver&rsquo, s braking data during real driving situations. The proposed algorithm was integrated and validated on a real-time embedded system, and then, it was applied to the model-based regenerative control algorithm as a case study.

A smart regenerative braking system, which is an advanced driver assistance system of electric vehicles, automatically controls the regeneration torque of the electric motor to brake the vehicle by recognizing the deceleration conditions. Thus, this autonomous braking system can provide driver convenience and energy efficiency by suppressing the frequent braking of the driver brake pedaling. In order to apply this assistance system, a deceleration planning algorithm should guarantee the safety deceleration under diverse driving situations. Furthermore, the planning algorithm suppresses a sense of heterogeneity by autonomous braking. To ensuring these requirements for deceleration planning, this study proposes a multi-level deceleration planning algorithm which consists of the two representative planning algorithms and one planning management. Two planning algorithms, which are the driver model-based planning and optimization-based planning, generate the deceleration profiles. Then, the planning management determines the optimal planning result among the deceleration profiles. To obtain an optimal result, planning management is updated based on the reinforcement learning algorithm. The proposed algorithm was learned and validated under a simulation environment using the real vehicle experimental data. As a result, the algorithm determines the optimal deceleration vehicle trajectory to autonomous regenerative braking.

The smart regenerative braking system (SRS) is an autonomous version of one-pedal driving in electric vehicles. To implement SRS, a deceleration planning algorithm is necessary to generate the deceleration used in automatic regenerative control. To reduce the discomfort from the automatic regeneration, the deceleration should be similar to human driving. In this paper, a deceleration planning algorithm based on multi-layer perceptron (MLP) is proposed. The MLP models can mimic the human driving behavior by learning the driving data. In addition, the proposed deceleration planning algorithm has a classified structure to improve the planning performance in each deceleration condition. Therefore, the individual MLP models were designed according to three different deceleration conditions: car-following, speed bump, and intersection. The proposed algorithm was validated through driving simulations. Then, time to collision and similarity to human driving were analyzed. The results show that the minimum time to collision was 1.443 s and the velocity root-mean-square error (RMSE) with human driving was 0.302 m/s. Through the driving simulation, it was validated that the vehicle moves safely with desirable velocity when SRS is in operation, based on the proposed algorithm. Furthermore, the classified structure has more advantages than the integrated structure in terms of planning performance.

A closed-loop control for torque balance using a cylinder pressure sensor can improve drivability and can suppress noise from light-duty diesel engines. However, the closed-loop control requires cylinder pressure sensor cost and computational time about pressure data. This paper proposes a torque balance control algorithm using a torque estimation model to solve these problems. The proposed algorithm consists of an indicated mean effective pressure (IMEP) estimation model and model reference controller. The estimation model estimates the IMEP of each cylinder. IMEP represents the indicated torque from the combustion of each cylinder that generates the crankshaft rotational dynamics. An IMEP estimation model is developed for all four cylinders based on the relationship of IMEP and crankshaft rotation. The proposed model also contains compensation parameters to solve dispersion and inequality problems of crankshaft acceleration. The proposed estimation algorithm only requires crankshaft acceleration and a single cylinder pressure sensor. This algorithm significantly reduces the cost required for four-cylinder torque control and is more suitable for real-time torque balance control because the IMEP computational time for other cylinders can be eliminated. The estimated IMEP by model is then controlled for real-time torque control which controls the IMEP of the individual cylinder for torque balance. The torque balance control algorithm is designed by a model reference adaptive control scheme and controller stability is derived by Lyapunov stability theorem. The IMEP estimation algorithm and torque balance controller were embedded and validated in a real-time engine management system. In conclusion, torque balance control can use estimated IMEP to reduce the IMEP variation of each cylinder.

This paper proposes a real-time empirical model of NOx emissions for diesel engines. The proposed model predicts the level of NOx emissions using an empirical model developed based on the thermal NO formation mechanism, the extended Zeldovich mechanism. Since it is difficult to consider the exact physical NO formation phenomena in real-time applications, the proposed algorithm adapts the key factors of the NO formation mechanism from the extended Zeldovich mechanism: temperature of the burned gas, concentration of the gas species, and combustion duration where NO is generated. These factors are considered in a prediction model as four parameters: exhaust gas recirculation rate (EGR rate), crank angle location of 50 % of mass fraction burned (MFB50), exhaust lambda value, and combustion acceleration. The proposed prediction model is validated with various steady engine experiments that showed a high linear correlation with the NOx emission measured by a NOx sensor. Furthermore, it is also validated for transient experiments.

Combustion phase variations in diesel engines due to production tolerance, engine aging, and different combustion conditions cause performance deterioration of fuel economy, torque, and emissions. Therefore, the combustion phase should be controlled by feedback information. A well-known method to control the combustion phase is mass fraction burnt 50% (MFB50) control using in-cylinder pressure. MFB50 control can retain performance through combustion phase compensation despite production tolerance, engine aging, and different combustion conditions. However, MFB50 can compensate only for variations caused by combustion phase changes; therefore, further study is required to supplement the limitation of MFB50 control. In this study, to analyze the effect of MFB50 control on combustion, MFB50 is controlled by adjusting the main injection timing using in-cylinder pressure at 1500 rpm and a BMEP of four bar according to engine parameter changes. The parameters are fuel rail pressure, boost pressure, mass air flow, swirl valve open, main injection quantity, pilot injection timing, and quantity. While the engine parameters were changing, the influence of MFB50 control on the combustion and emissions was analyzed to establish improvement points for combustion feedback control. Our experiments demonstrated that MFB50 control reduced NOx dispersions but at the cost of increasing PM dispersions. Therefore, to improve MFB50 control, a control algorithm that can handle PM emission dispersion needs to be considered.

This paper proposes an in-cylinder pressure based real-time combustion control algorithm to reduce combustion dispersions from diesel engines. The proposed algorithm manipulates main injection quantity and timing as well as pilot injection quantity to control the Indicated Mean Effective Pressure (IMEP), crank angle location where 50% of Mass Fraction Burned (MFB50) and maximum value of Rate of Heat Release (ROHR max ). This control algorithm reduces the combustion dispersions by controlling the heat release curve for each operating condition. The proposed real-time combustion control structure mitigates combustion dispersions caused by operating condition changes, un-calibrated fuel injectors, and environmental condition variations, and it is validated with various experiments. In these experiments, various engine control variables were changed to validate the reduction in combustion dispersion when operating conditions change unintendedly. Changes in coolant temperature verified the compensation effect of the proposed algorithm under different external environments. It is validated with these experiments that there are 30% reduction in torque imbalances, a 10% emission dispersions reduction for NO x emissions, and a 65% reduction for PM emissions.

This study proposes a feedforward control method for exhaust gas recirculation (EGR) and variable geometry turbocharger (VGT) systems based on cylinder pressure information and air path model. The proposed method calculates EGR rate and exhaust manifold pressure using cylinder pressure information which provides accurate combustion state information. The accuracy of the air system model was improved according to the precise information of the EGR rate and exhaust pressure; subsequently, the performance of the proposed model based feedforward controller was enhanced. The proposed feedforward controller was implemented in an in-house developed engine management system and validated through the engine experiments under steady and transient conditions.

The start of combustion (SOC) in a compression ignition engine may not be consistent due to the engine operating conditions of each engine cycle. This is true even if fuel is injected at a consistent time. Variable factors include humidity, temperature, fuel quality, exhaust gas recirculation (EGR) rate, and boost pressure. These variations in SOC cause an increase in harmful emissions as well as deterioration of engine performance; therefore, SOC has to be detected and controlled. In this study, we present a simplified heat release (HR) equation to detect SOC for real-time closed-loop combustion control in a compression ignition engine. The proposed method extracts SOC using an initial heat release (IHR) equation that is a modified heat release equation. The detected SOC using IHR reveals a close correlation with one percent of heat release at various engine operating conditions such as main injection timing, injected fuel quantity, fuel-rail pressure, and EGR rate changes. In addition, the SOC detected from the proposed method shows great potential for real-time applications as it reduces of the amount of cylinder pressure data by 50.2% and computational load by 55.2%. Finally, SOC was well controlled in real-time with the estimated SOC despite engine operating condition changes.

Current diesel engines commonly employ a variable geometry turbocharger (VGT) to improve drivability and fuel efficiency, and this paper addresses model-based feedforward control of the VGT based on empirical models of compressor and turbine efficiencies to improve transient response. Though compressor and turbine efficiencies affect compressor air flow and turbine power, respectively, it is challenging to understand exactly the compressor and turbine efficiencies in real engines. In order to cope with this problem, we propose empirical models of compressor and turbine efficiencies. The input states are proposed based on maps of the compressor and turbine efficiencies, which are provided by turbocharger manufacturers. Parameters of the efficiency models are identified with 225 data of steady-state engine experiments to reflect engine operating conditions. The proposed compressor and turbine efficiency models were applied to the model-based VGT feedforward control algorithm to verify the effectiveness of the efficiency models. The proposed modelbased VGT feedforward control algorithm based on the compressor and turbine efficiency models was experimentally verified with 2.2 L common-rail-direct-injection diesel engines.

This paper investigates an air system model for a light duty diesel engine with a dual-loop exhaust gas recirculation (EGR) system and variable geometric turbocharger (VGT). The proposed model estimates unmeasurable engine states using only equipped sensors on mass product engines. The maximum error of thermodynamic states are only when the temperature, pressure and mass are less than 5% and the maximum error of the mass flow rate model is less than 0.003 kg/s. Temperature dynamics and transport delay are obtained to improve the estimation performance in transient engine operating conditions. The model is validated under engine experiments.

Particulate matters (PM) accumulation through a low-pressure exhaust gas recirculation (LP-EGR) path may hinder to obtain the desired LP-EGR rate and thus causes an increase of nitrogen oxides (NOx). The degree of lack of the LP-EGR rate should be detected, i.e., an LP-EGR fault, and a remedy to compensate for the lack of LP-EGR rate should be a mandate to suppress NOx emission, i.e., a fault management. In order to accomplish those objectives, this paper proposes an LP-EGR fault management system, which consists of a fault diagnosis algorithm, fault-tolerant control algorithm, and an LP-EGR rate model. The model applies a combustion parameter derived from in-cylinder pressure information to the conventional orifice valve model. Consequently, the LP-EGR rate estimation was improved to the maximum error of 2.38% and root-mean-square-error (RMSE) of 1.34% at various operating conditions even under the fault condition compared to that of the conventional model with the maximum error of 7.46% and RMSE of 5.39%. Using this LP-EGR rate model as a virtual sensor, the fault diagnosis algorithm determines an LP-EGR fault state. Based on the state, the fault-tolerant control determines whether or not to generate the offset of the exhaust throttle valve (ETV) position. This offset combines with the look-up table (LUT)-based feedforward controller to control an LP-EGR rate. As a result of real-time verification of the fault management system in the fault condition, the NOx emission decreased by up to about 15%.

An accurate estimation of the intake oxygen concentration (IOC) is a prerequisite to develop the optimal control strategy because it directly affects the combustion and emissions. Since the IOC is determined based on the mass conservation law in the intake manifold, estimating the mass flow rate of the exhaust gas recirculation (EGR) is most critical. However, to estimate the EGR mass flow rate, the conventional orifice valve model causes extrapolation error or inaccurate estimation results under transient operating conditions. In order to improve the estimation performance, this study proposes a correction algorithm for estimating IOC. A dynamic correction state is determined for the orifice valve model. In addition, the intake pressure dynamics is also derived based on the energy conservation law in the intake manifold. Using these dynamic models, a nonlinear parameter varying model is determined, and an extended Kalman filter (EKF) is applied to derive the value of correction state. Furthermore, unmeasurable physical states of the nonlinear parameter varying model are estimated from an air system model that only requires the engine-equipped sensors of mass production engines. The correction algorithm is validated through the engine experiments that clearly demonstrate high accuracy of the IOC estimation during transient conditions, which may apply for the vehicle application.

To preserve the fun of driving and enhance driving convenience, a smart regenerative braking system (SRS) is developed. The SRS provides automatic regeneration that is appropriate for the driving conditions, but the existing technology has a low level of acceptability and comfort. To solve this problem, this paper presents an automatic regenerative control system based on a deceleration model that reflects the driver&rsquo, s characteristics. The deceleration model is designed as a parametric model that mimics the driver&rsquo, s behavior. In addition, it consists of parameters that represent the driver&rsquo, s characteristics. These parameters are updated online by a learning algorithm. The validation results of the vehicle testing show that the vehicle maintained a safe distance from the leading car while simulating a driver&rsquo, s behavior. Of all the deceleration that occurred during the testing, 92% was conducted by the automatic regeneration system. In addition, the results of the online learning algorithm are different based on the driver&rsquo, s deceleration pattern. The presented automatic regenerative control system can be safely used in diverse car-following situations. Moreover, the system&rsquo, s acceptability is improved by updating the driver characteristics. In the future, the algorithm will be extended for use in more diverse deceleration situations by using intelligent transportation system information.

This paper proposes a real-time estimation algorithm of combustion parameters for the location of 50% of mass fraction burnt (MFB50), and indicated mean effective pressure (IMEP). The proposed estimation algorithm uses the difference pressure only instead of the in-cylinder pressure for calculation of the combustion parameters. Since the difference pressure is the pressure that is generated only by the combustion, it occurs between the start of combustion (SOC) and the end of combustion (EOC); this allows the proposed algorithm to estimate the combustion parameters with fewer cylinder pressure data samples and low computational load compared with the conventional method. The proposed algorithm estimates the IMEP with a result acquired during the MFB50 calculation and that can significantly reduce the computational load required to calculate the combustion parameters. Consequently, the proposed estimation algorithm requires only 51% of the execution time to calculate the combustion parameters compared to the conventional method. The proposed estimation algorithm is validated with an engine experiment under 131 operating conditions that showed high linear correlation with the original combustion parameters. In-cylinder pressure based combustion control using the estimated combustion parameters is introduced as a case study and the proposed estimation algorithm validated its significant potential for real-time applications.

This paper investigates the design of model-based feedforward compensators for exhaust gas recirculation (EGR) and variable geometry turbocharger (VGT) systems using air path models for a common-rail direct injection (CRDI) diesel engine to cope with the nonlinear control problem. The model-based feedforward compensators generate set-positions of the EGR valve and the VGT vane to track the desired mass air flow (MAF) and manifold absolute pressure (MAP) with consideration of the current engine operating conditions. In the best case, the rising time to reach 90% of the MAF set-point was reduced by 69.8% compared with the look-up table based feedforward compensators.