Search Results

Cam-phasing is increasingly considered as a feasible Variable Valve Timing (VVT) technology for production engines. Additional independent control variables in a dual-independent VVT engine increase the complexity of the system, and achieving its full benefit depends critically on devising an optimum control strategy. A traditional approach relying on hardware experiments to generate set-point maps for all independent control variables leads to an exponential increase in the number of required tests and prohibitive cost. Instead, this work formulates the task of defining actuator set-points as an optimization problem. In our previous study, an optimization framework was developed and demonstrated with the objective of maximizing torque at full load. This study extends the technique and uses the optimization framework to minimize fuel consumption of a VVT engine at part load.

Variable Valve Actuation (VVA) technology provides high potential in achieving high performance, low fuel consumption and pollutant reduction. However, more degrees of freedom impose a big challenge for engine characterization and calibration. In this study, a simulation based approach and optimization framework is proposed to optimize the setpoints of multiple independent control variables. Since solving an optimization problem typically requires hundreds of function evaluations, a direct use of the high-fidelity simulation tool leads to the unbearably long computational time. Hence, the Artificial Neural Networks (ANN) are trained with high-fidelity simulation results and used as surrogate models, representing engine's response to different control variable combinations with greatly reduced computational time. To demonstrate the proposed methodology, the cam-phasing strategy at Wide Open Throttle (WOT) is optimized for a dual-independent Variable Valve Timing (VVT) engine.

An accurate air flow rate model is critical for high-quality air-fuel ratio control in Spark-Ignition engines using a Three-Way-Catalyst. Emerging Variable Valve Timing technology complicates cylinder air charge estimation by increasing the number of independent variables. In our previous study (SAE 2004-01-3054), an Artificial Neural Network (ANN) has been used successfully to represent the air flow rate as a function of four independent variables: intake camshaft position, exhaust camshaft position, engine speed and intake manifold pressure. However, in more general terms the air flow rate also depends on ambient temperature and pressure, the latter being largely a function of altitude. With arbitrary cam phasing combinations, the ambient pressure effects in particular can be very complex. In this study, we propose using a separate neural network to compensate the effects of altitude on the air flow rate.

Owing to the stochastic nature of engine knock, determination of the knock limited spark angle (KLSA) is difficult in engine cycle simulation. Therefore, the state-of-the-art knock modeling is mostly limited to either merely predicting knock onset (i.e. auto-ignition of end gas) or combining a simple unburned mass fraction (UMF) model representative of knock intensity (KI). In this study, a newly developed KLSA model, which takes both predictions of knock onset and intensity into account, is firstly introduced. Multiple variables including the excess air ratio, EGR ratio, cylinder pressure and the end gas temperature are included in the knock onset model. Based on the auto-ignition theory of hot spots in end gas, both the energy density and heat release rate in hot spots are taken into consideration in the KI model.

Visual Place Recognition (VPR) excels at providing a good location prior for autonomous vehicles to initialize the map-based visual SLAM system, especially when the environment changes after a long term. Condition change and viewpoint change, which influences features extracted from images, are two of the major challenges in recognizing a visited place. Existing VPR methods focus on developing the robustness of global feature to address them but ignore the benefits that local feature can auxiliarily offer. Therefore, we introduce a novel hierarchical place recognition method with both global and local features deriving from homologous VLAD to improve the VPR performance. Our model is weak supervised by GPS label and we design a fine-tuning strategy with a coupled triplet loss to make the model more suitable for extracting local features.

In this paper an estimation method on location of peak pressure (LPP) employing flame ionization measurement, with the spark plug as a sensor, was discussed to achieve combustion parameters estimation at quick start of HEV engines. Through the cycle-based ion signal analysis, the location of peak pressure can be extracted in individual cylinder for the optimization of engine quick start control of HEV engine. A series of quick start processes with different cranking speed and engine coolant temperature are tested for establishing the relationship between the ion signals and the combustion parameters. An Artificial Neural Network (ANN) algorithm is used in this study for estimating these two combustion parameters. The experiment results show that the location of peak pressure can be well established by this method.

In order to meet increasingly stringent emission regulations, it is significance to establish a control- oriented transient NOx and PM emission prediction model and improve the accuracy and real-time performance. In this study, the prediction model of transient PM emissions based on Transformer is established. In terms of model accuracy and real-time performance, Transformer emission prediction model is compared with Multilayer perceptron (MLP) neural network and Long-Short Term Memory (LSTM) emission prediction model. The results show that the performance of Transformer transient emission prediction model is superior to other model structures, it can be used for real-time prediction.