First choices of network composition and the topology.

First use of neural networks forautomotive application can be traced back to early 90s. In 1991, Marko testedvarious neural classifiers for online diagnosis of engine control defects(misfires) and proposed a direct control by inverse neural model of an activesuspension system 1. In 2, Puskorius andFeldkamp, summarizing one decade of research, proposed neural nets for varioussub-functions in engine control: AFR and idle speed control, misfire detection,catalyst monitoring, prediction of pollutant emissions.

Indeed, since thebeginning of the 90s, neural approaches have been proposed by numerous authors,for example:  • Vehicle control: Anti-lock braking system(ABS), active suspension, steering, speed control  • Engine modeling: Manifold pressure, air massflow, volumetric efficiency, Indicated pressure into cylinders, AFR,start-of-combustion for Homogeneous Charge Compression Ignition (HCCI), torqueor power • Engine control: Idle speed control, AFRcontrol, transient fuel compensation (TFC), cylinder air charge control withVVT, ignition timing control, throttle, turbocharger, EGR control, pollutantsreduction • Engine diagnosis: Misfire and knockdetection, spark voltage vector recognition systemsThe works are too numerous to bereferenced here. Nevertheless, the reader can consult the publications 3, 4, 5, 6, 7 and the referencestherein, for an overview. 1.

Don't waste your time
on finding examples

We can write the essay sample you need

    Architecturechoices of neural networks2.1 Introductionof architecturesThe choice of network architectureis dependent on the problem. Classification, linear or non-linear problems,with or without underlying system dynamics guides the choices of networkcomposition and the topology. In general it can be distinguished between threetypes of networks:• Single-Feedforward Networks (SLFN)• Multi-Layer Feedforward Networks(MLFN)• Recurrent Networks (RNN).Where the singlefeedforward network describes a simple mapping network it can be used inclassification or for mapping of simple input output functionality. It isdefined through a single layer of neurons. Hence, the knowledge storagecapacity is restricted and only simple logic relations can be mapped. Anextension of this is the multi-layer feedforward network, also found asmulti-layer perceptron.

This network architecture is defined through a minimumof one hidden layer of neurons. The number of hidden layers can be increaseddependent on the problem. However, literaturestates (reference) that a multi-layer perceptron with three hidden layers issufficient to map every continuous function by adding a certain number ofneurons to meet required complexity. However, big growing networks can beill-posed for overtraining and be difficult to implement in real-timeapplications.

Therefore, recurrent structures of networks are in place thatwill accommodate the underlying output dynamics, a feature that is ofparticular interest with engine applications. In turbocharged combustion enginesintake and exhaust shows related dynamics through the turbine and compressorconnection. Those dynamics can be taken into consideration with outputrecurrent network structures. The automotive sector has applied neural networksmodels in several different cases. Their main implementation is seen in controldesign in the area of engine operation. Hence, in engine development neuralnetworks are used for control problems such as fuel injection, outputperformance or speed 8.

In addition, advancedcontrol strategies as variable turbine geometry (VGT), exhaust gasrecirculation (EGR) or variable valve timing (VVT) have been in the focus ofANN modelling 9. Nevertheless, theapplication is also used for virtual sensing such as emissions 10, 11 or as described inProkhorov 12 for misfiredetection, torque monitoring or tyre pressure change detection.The combustion process itself has been investigated andparameters been modelled with neural networks by different authors 13. Potenza developed amodel estimating Air-to-Fuel Ratio (AFR) or in-cylinder pressure and temperatureon the basis of crankshaft kinematics and its vibrations. In the work of Hecombustion parameters and emissions are modelled under the consideration ofboost pressure and EGR.

x

Hi!
I'm Owen!

Would you like to get a custom essay? How about receiving a customized one?

Check it out