日野汽車股份有限公司

This article addresses the problem of optimal vehicle sampling for fleet-wide in-use emissions monitoring, a necessity driven by the absence of direct emissions sensors in modern production vehicles and the variable impact of in-use changes and operational factors (mileage, time-in-service, workload) on emissions performance across a fleet. Recognizing that comprehensive fleet testing is impractical due to significant downtime and cost, we propose a novel approach to identify a small, yet optimally informative subset of vehicles for sampling. The proposed approach leverages submodular function maximization, a technique rooted in optimal experimental design, specifically D-optimal design, to maximize the determinant of the information matrix (e.g., of XTX, where X is the regressor/design matrix in the case of a linear in parameters model). This approach ensures that the collected data yields maximum information for refining and building accurate models for emissions changes. We compare the submodular maximization strategy with conventional uniform and extreme sampling methods. Our simulation results demonstrate the potential for the submodular approach to outperform both alternatives by achieving lower variance (as measured by standard deviation and coefficient of variation) in estimating parameters for the assumed linear, quadratic, and simplified quadratic models for emission changes. The application of submodular function maximization is thus shown to be beneficial in vehicle fleet management for data collection in resource-constrained environments and leading to more accurate in-use emissions prediction. The envisioned process, in which a limited number of vehicles selected by our methodology are tested and the data are utilized to improve emissions models, can support the implementation of model-based strategies for engine emissions management.

This paper addresses the changes in engine emissions due to in-use component changes through the synergistic application of predictive control, machine learning, and onboard adaptation. In particular, we consider an adaptive economic Model Predictive Control (eMPC) strategy to mitigate the effects of performance drift on Nitrogen Oxides (NOx) and Soot emissions from compression ignition (diesel) engines. A performance drift block, which applies a multiplier and offset to nominal emissions, is integrated with a high-fidelity Neural Network (NN) plant model to simulate these characteristic changes. To counteract variability, two online adaptation methods are integrated within the eMPC framework: One is based on Recursive Least Squares (RLS) and another on a continuously updated online NN. The proposed control architecture is validated through simulations over standard transient cycles. Results demonstrate that while the rate-based eMPC possesses inherent robustness to performance drift, in particular, for the formulation of eMPC that involves NOx penalty in the cost function, online adaptation further facilitates satisfying emission constraints. In particular, both adaptive methods improve Soot limit enforcement compared to a non-adaptive controller, with the online NN providing superior performance by capturing the nonlinear dynamics of in-use changes.

This paper addresses the control of diesel engine nitrogen oxides (NOx) and Soot emissions through the application of Model Predictive Control (MPC). The developments described in the paper are based on a high-fidelity model of the engine airpath and torque response in GT-Power, which is extended with a feedforward neural network (FNN)-based model of engine out (feedgas) emissions identified from experimental engine data to enable the controller co-simulation and performance verification. A Recurrent Neural Network (RNN) is then identified for use as a prediction model in the implementation of a nonlinear economic MPC that adjusts intake manifold pressure and EGR rate set-points to the inner loop airpath controller as well as the engine fueling rate. Based on GT-Power engine model and FNN emissions model, the closed-loop simulations of the control system and the plant model, over different driving cycles, demonstrate the capability to shape engine out emissions response by adjusting weights and constraints in economic MPC formulation.

The dominant factors affecting porosity formation in laser powder bed fusion (PBF-LB/M) of an aluminum alloy were investigated through sparse modeling with the cross-sectional pore area ratio as the target variable and the process parameters of PBF-LB/M and the melting and solidification conditions of the alloy as the explanatory variables. A combination of a few explanatory variables that did not significantly increase the mean squared error for the relationship between the measured pore area ratios and the ratios estimated via the regression equations was found through lasso regression and backward elimination, which indicated that the energy density (one of the process parameters) and melt-pool depth (one of the melting conditions) were the dominant factors affecting the pore area ratio. The obtained regression coefficients for the energy density and melt-pool depth were negative and positive, respectively. In addition, the relationship between the energy density and melt-pool depth was curvilinear. These results suggest not only that the pore area ratio increases with the energy density and melt-pool depth but also that it decreases with an increase in the energy density or a decrease in the change rate of the pool depth under the range of the slow increase in the pool depth with an increase in the energy density.

This paper presents the results of developing a multi-layer Neural Network (NN) to represent diesel engine emissions and integrating this NN into control design. Firstly, a NN is trained and validated to simultaneously predict oxides of nitrogen (N Ox) and Soot using both transient and steady-state data. Based on the input-output correlation analysis, inputs to NN with the highest influence on the emissions are selected while keeping the NN structure simple. Secondly, a co-simulation framework is implemented to integrate the NN emissions model with a model of a diesel engine airpath system built in GT-Power and used to identify a low-order linear parameter-varying (LPV) model for emissions prediction. Finally, an economic supervisory model predictive controller (MPC) is developed using the LPV emissions model to adjust setpoints to an inner-loop airpath tracking MPC. Simulation results are reported illustrating the capability of the resulting controller to reduce N Ox, meet the target Soot limit, and track the adjusted intake manifold pressure and exhaust gas recirculation (EGR) rate targets.