Search Results

The recent development of diesel engine fuel injection systems has been dominated by how to manage the degrees of freedom that common rail multi-pulse systems now offer. A number of production engines already use four injection events while in research, work based on up to eight injection events has been reported. It is the degrees of freedom that lead to a novel experimental requirements. There is a potentially complex experimental program needed to simply understand how injection parameters influence the combustion process in steady state. Combustion behavior is not a continuum and as both injection and EGR rates are adjusted, distinct combustion modes emerge. Conventional calibration processes are severely challenged in the face of large number of degrees of freedom and as a consequence new development approaches are needed.

One of the most critical challenges currently facing the diesel engine industry is how to improve fuel economy under emission regulations. Improvement in fuel economy can be achieved by precisely controlling Air/Fuel ratio and by monitoring fuel consumption in real time. Accurate and repeatable measurements of fuel rate play a critical role in successfully controlling air/fuel ratio and in monitoring fuel consumption. Volumetric and gravimetric measurements are well-known methods for measuring fuel consumption of internal combustion engines. However, these methods are not suitable for obtaining fuel flow rate data used in real-time control/measurement. In this paper, neural networks are used to solve the problem concerning discontinuous data of fuel flow rate measured by using an AVL 733 s fuel meter. The continuous parts of discontinuous fuel flow rate are used to train and validate a neural network, which can then be used to predict the discontinuous parts of the fuel flow rate.

In modern production diesel engine control systems, fuel path control is still largely conducted through a system of tables that set mode, timing and injection quantity and with common rail systems, rail pressure. In the hands of an experienced team, such systems have proved so far able to meet emissions standards, but they lack the analytical underpinning that lead to systematic solutions. In high degree of freedom systems typified by modern fuel injection, there is substantial scope to deploy optimising closed loop strategies during calibration and potentially in the delivered product. In an optimising controller, a digital algorithm will explicitly trade-off conflicting objectives and follow trajectories during transients that continue to meet a defined set of criteria. Such an optimising controller must be based on a model of the system behaviour which is used in real time to investigate the consequences of proposed control actions.

Most modern diesel engines are equipped with common fuel rail system. The common fuel rail pressure and start of injection are two important fuel path control variables which are needed to be carefully calibrated over all engine operation range. They both have big effects on engine emissions, fuel consumptions and combustion noise performance. Though there are mature techniques such as design of experiment, model based calibration together with optimization method for engine calibration task, the engine test points are still many and the calibration costs are still high. Besides, the outputs of the calibration are look up tables or maps which are used in engine open loop control strategy in engine control system. Open loop control system has no adaptive and disturbance rejection ability. So the initially optimally calibrated look up control tables will gradually become less and less optimal when the engine is aging.

In order to identify predictive models for a diesel engine combustion process, combustion cylinder pressure together with other fuel path variables such as rail pressure, injector current and sleeve pressure of 1000 continuous cycles were sampled and collected at high resolution. Using these engine steady state test data, three types of modeling approach have been studied. The first is the Auto-Regressive-Moving-Average (ARMA) model which had limited prediction ability for both peak combustion pressure (Pmax) and Indicated Mean Effective Pressure (IMEP). By applying correlation analysis, proper inputs were found for a linear predictive model of Pmax and IMEP respectively. The prediction performance of this linear model is excellent with a 30% fit number for both Pmax and IMEP. Further nonlinear modeling work shows that even a nonlinear Neural Network (NN) model does not have improved prediction performance compared to the linear predictive model.

A full calibration exercise of a diesel engine air path can take months to complete (depending on the number of variables). Model-based calibration approach can speed up the calibration process significantly. This paper discusses the overall calibration process of the air-path of the Cat® C7.1 engine using statistical machine learning tool. The standard Cat® C7.1 engine's twin-stage turbocharger was replaced by a VTG (Variable Turbine Geometry) as part of an evaluation of a novel air system. The changes made to the air-path system required a recalculation of the air path's boost set point and desired EGR set point maps. Statistical learning processes provided a firm basis to model and optimize the air path set point maps and allowed a healthy balance to be struck between the resources required for the exercise and the resulting data quality.