NEURAL NETWORK BASED CONTROL SOLUTION OF AIR TO FUEL RATIO FOR 2NZFE FUEL INJECTION ENGINE

Abstract

In present research Air to Fuel Ratio for 2NZFE fuel injection engine is controlled by using Artificial Neural Network (ANN). ANN is an information processing paradigm which is inspired by a biological nervous system. ANN trains controller by passively accepts inputs and its corresponding desired outputs. Case study is done on 2NZFE fuel injection engine. All the required data for ANN training is collected by performing several active tests on 2NZFE fuel injection engine at TOYOTA EFI workshop located in I-9 sector of Islamabad. Total 100 data sets are recorded comprising of 11 parameters (10 inputs+01 output) in each set. Offline training of ANN is done by taking 10 parameters as input upon which AFR is dependent, whereas AFR is taken as an output parameter. ANN is modeled by using three layers including input, output and hidden layer with five neurons in hidden layer. “Gradient Descent Back-propagation with Momentum and with Adaptive learning rate” is used as a training algorithm with sigmoid tangent as an activation function between layers. 3000 iteration are carried out. Trained neural network predicted outputs which are then analyzed and compared with the actual targets. Based on this information electronic control unit of automobile engine can adjust engine parameters according to the stoichiometric AFR.