System for estimating quantities of pollutants emitted in exhaust gases of vehicle diesel engine using data recorded in neuron network

摘要

Data records stored in an engine control unit comprise one or more neuron networks (1); data entries (2) representative of the function of the engine and thus vehicle displacement, available in the electronic control unit without adding a detector, and means (4) of accumulating the estimated quantities (3). The network has 10-15 neurons and apprentice databases for the network which are vehicle running sequences for at least several minutes. The data entries used for the network(s) include at least one of the following parameters: the engine rotation speed at two consecutive instants t and t-delta t; the fuel flow at three consecutive instants t, t-delta t and t-2delta t; the temperature of the engine cooling water and the speed of the vehicle at instant t, or the richness of the mixture, where delta t represents the preset time interval between two consecutive measurement times. The system includes means of estimating the cumulative quantity of soot in the exhaust gases which are retained in the particle filter. With an injection engine with a common injection line, the parameters used are the fuel flow in the engine pre-injection; the fuel flow in the engine main injection; the displacement of a piston relative to top dead center when the last fuel injection is started; the displacement of a piston relative to top dead center when the last main fuel injection is started; the temperature of the cooling water; the air feed flow; the pressure inside the injection line and the engine rotation speed. One of the estimated quantities of pollutants is the cumulative quantity of soot retained in a particle filter and/or the cumulative quantity of NOx, carbon oxides and/or hydrocarbons in the exhaust gases. None of the system outlets returns to a system inlet to avoid all instability problems. The system includes means of reinitializing the estimated quantities of particles at zero independently of each other. The best configuration of the neuron network is evaluated by error calculations on the output quantities by accumulation on a sliding window. The sliding window is determined so its size is minimal with respect to an estimation error below a threshold value. In particular the sliding window corresponds to a vehicle displacement of between 0.5 and 1.5 km to respect an estimation error or more than 5g of solid particles emitted per 135 km travelled by the vehicle. Part of the data recorded at the apprenticeship of the network is spaced to allow validation against the data used. The system is used to control the particle filter regeneration when the estimated cumulative soot reaches a recorded threshold value. The system can also be used to aid calibration of engine control strategies from estimation of engine emissions over a rolling cycle of the vehicle.