Recurrent Neural Network to Estimate Intake Manifold O2 Concentration in a Diesel Engine

Emission regulations are becoming more and more stringent, especially on NOx pollutants, making diesel engines with their embedded control systems more and more complex. To ensure a correct and clean engine functioning, all the control strategies related to aftertreatment, fuel injection and air-path have to exploit or target the intake manifold O2 concentration. The O2 concentration is strictly related to engine-out NOx emissions and an accurate model, to be implemented in emission control systems, is essential. The paper addresses the modeling of the intake O2 concentration in a turbocharged diesel engine by means of a Recurrent Neural Network with simulation focus and fed with four inputs. The inputs are engine load, engine speed and the position of Exhaust Gas Recirculation and Variable Geometry Turbochargers valves. Training and validation data are generated using the engine simulation tool GT-Power implementing a detailed model of the engine while the training procedure is performed in MATLAB environment through NNSYSID toolbox. The performances of the obtained model are satisfactory in different tests and the model is able to account for the engine nonlinearities during transients.