Neural Network Based Models for Virtual NOx Sensing of Compression Ignition Engines Powertrain Division, Magneti Marelli SpA

Format:

Conference/Event

Author/Creator:

De Cesare, De Cesare, author.

Contributor:

Covassin, Federico

Conference Name:

10th International Conference on Engines & Vehicles (2011-09-11 : Naples, Italy)

Language:

English

Physical Description:

1 online resource

Place of Publication:

Warrendale, PA SAE International 2011

Summary:

The paper focuses on the experimental identification andvalidation of different neural networks for virtual sensing ofNOx emissions in combustion compression ignition engines(CI). A comparison of several neural network architectures (NN,TDNN and RNN) has been carried out in order to evaluate precisionand generalization in dynamic prediction of NOxformation. Furthermore the model complexity (number and types ofinputs, neuron and layer number, et cetera) has been considered to allowa future ECU implementation and on line training. Suited trainingprocedures and experimental tests are proposed to improve themodels.Several measurements of NOx emissions have beenperformed through different devices applied to the outlet of a EURO5 Common Rail diesel engine with EGR. The accuracy of the developedmodels is assessed by comparing simulated and experimentaltrajectories for a wide range of operating conditions.The study highlights that history and proper inputs aresignificant for the output estimation, and good results can beachieved either through Recursive Neural Networks (RNN) or throughNeural Networks (NN) with input history. A virtual NOxsensor will offer significant opportunities for implementingon-board feed-forward and feedback control strategies in order toimprove the performance and the diagnosis of the engine and of theafter-treatment devices

Notes:

Vendor supplied data

Publisher Number:

2011-24-0157

Access Restriction:

Restricted for use by site license